[GFS2] Replace rgrp "recent list" with mru list
[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/lm_interface.h>
21
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "bmap.h"
25 #include "daemon.h"
26 #include "glock.h"
27 #include "glops.h"
28 #include "inode.h"
29 #include "mount.h"
30 #include "ops_fstype.h"
31 #include "ops_dentry.h"
32 #include "ops_super.h"
33 #include "recovery.h"
34 #include "rgrp.h"
35 #include "super.h"
36 #include "sys.h"
37 #include "util.h"
38 #include "log.h"
39
40 #define DO 0
41 #define UNDO 1
42
43 static struct gfs2_sbd *init_sbd(struct super_block *sb)
44 {
45         struct gfs2_sbd *sdp;
46
47         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
48         if (!sdp)
49                 return NULL;
50
51         sb->s_fs_info = sdp;
52         sdp->sd_vfs = sb;
53
54         gfs2_tune_init(&sdp->sd_tune);
55
56         INIT_LIST_HEAD(&sdp->sd_reclaim_list);
57         spin_lock_init(&sdp->sd_reclaim_lock);
58         init_waitqueue_head(&sdp->sd_reclaim_wq);
59
60         mutex_init(&sdp->sd_inum_mutex);
61         spin_lock_init(&sdp->sd_statfs_spin);
62
63         spin_lock_init(&sdp->sd_rindex_spin);
64         mutex_init(&sdp->sd_rindex_mutex);
65         INIT_LIST_HEAD(&sdp->sd_rindex_list);
66         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
67
68         INIT_LIST_HEAD(&sdp->sd_jindex_list);
69         spin_lock_init(&sdp->sd_jindex_spin);
70         mutex_init(&sdp->sd_jindex_mutex);
71
72         INIT_LIST_HEAD(&sdp->sd_quota_list);
73         spin_lock_init(&sdp->sd_quota_spin);
74         mutex_init(&sdp->sd_quota_mutex);
75
76         spin_lock_init(&sdp->sd_log_lock);
77
78         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
79         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
80         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
81         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
82         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
83
84         mutex_init(&sdp->sd_log_reserve_mutex);
85         INIT_LIST_HEAD(&sdp->sd_ail1_list);
86         INIT_LIST_HEAD(&sdp->sd_ail2_list);
87
88         init_rwsem(&sdp->sd_log_flush_lock);
89         atomic_set(&sdp->sd_log_in_flight, 0);
90         init_waitqueue_head(&sdp->sd_log_flush_wait);
91
92         INIT_LIST_HEAD(&sdp->sd_revoke_list);
93
94         mutex_init(&sdp->sd_freeze_lock);
95
96         return sdp;
97 }
98
99 static void init_vfs(struct super_block *sb, unsigned noatime)
100 {
101         struct gfs2_sbd *sdp = sb->s_fs_info;
102
103         sb->s_magic = GFS2_MAGIC;
104         sb->s_op = &gfs2_super_ops;
105         sb->s_export_op = &gfs2_export_ops;
106         sb->s_time_gran = 1;
107         sb->s_maxbytes = MAX_LFS_FILESIZE;
108
109         if (sb->s_flags & (MS_NOATIME | MS_NODIRATIME))
110                 set_bit(noatime, &sdp->sd_flags);
111
112         /* Don't let the VFS update atimes.  GFS2 handles this itself. */
113         sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
114 }
115
116 static int init_names(struct gfs2_sbd *sdp, int silent)
117 {
118         char *proto, *table;
119         int error = 0;
120
121         proto = sdp->sd_args.ar_lockproto;
122         table = sdp->sd_args.ar_locktable;
123
124         /*  Try to autodetect  */
125
126         if (!proto[0] || !table[0]) {
127                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
128                 if (error)
129                         return error;
130
131                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
132                 if (error)
133                         goto out;
134
135                 if (!proto[0])
136                         proto = sdp->sd_sb.sb_lockproto;
137                 if (!table[0])
138                         table = sdp->sd_sb.sb_locktable;
139         }
140
141         if (!table[0])
142                 table = sdp->sd_vfs->s_id;
143
144         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
145         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
146
147         table = sdp->sd_table_name;
148         while ((table = strchr(table, '/')))
149                 *table = '_';
150
151 out:
152         return error;
153 }
154
155 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
156                         int undo)
157 {
158         struct task_struct *p;
159         int error = 0;
160
161         if (undo)
162                 goto fail_trans;
163
164         for (sdp->sd_glockd_num = 0;
165              sdp->sd_glockd_num < sdp->sd_args.ar_num_glockd;
166              sdp->sd_glockd_num++) {
167                 p = kthread_run(gfs2_glockd, sdp, "gfs2_glockd");
168                 error = IS_ERR(p);
169                 if (error) {
170                         fs_err(sdp, "can't start glockd thread: %d\n", error);
171                         goto fail;
172                 }
173                 sdp->sd_glockd_process[sdp->sd_glockd_num] = p;
174         }
175
176         error = gfs2_glock_nq_num(sdp,
177                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
178                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
179                                   mount_gh);
180         if (error) {
181                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
182                 goto fail;
183         }
184
185         error = gfs2_glock_nq_num(sdp,
186                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
187                                   LM_ST_SHARED,
188                                   LM_FLAG_NOEXP | GL_EXACT,
189                                   &sdp->sd_live_gh);
190         if (error) {
191                 fs_err(sdp, "can't acquire live glock: %d\n", error);
192                 goto fail_mount;
193         }
194
195         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
196                                CREATE, &sdp->sd_rename_gl);
197         if (error) {
198                 fs_err(sdp, "can't create rename glock: %d\n", error);
199                 goto fail_live;
200         }
201
202         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
203                                CREATE, &sdp->sd_trans_gl);
204         if (error) {
205                 fs_err(sdp, "can't create transaction glock: %d\n", error);
206                 goto fail_rename;
207         }
208         set_bit(GLF_STICKY, &sdp->sd_trans_gl->gl_flags);
209
210         return 0;
211
212 fail_trans:
213         gfs2_glock_put(sdp->sd_trans_gl);
214 fail_rename:
215         gfs2_glock_put(sdp->sd_rename_gl);
216 fail_live:
217         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
218 fail_mount:
219         gfs2_glock_dq_uninit(mount_gh);
220 fail:
221         while (sdp->sd_glockd_num--)
222                 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
223
224         return error;
225 }
226
227 static inline struct inode *gfs2_lookup_root(struct super_block *sb,
228                                              u64 no_addr)
229 {
230         return gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
231 }
232
233 static int init_sb(struct gfs2_sbd *sdp, int silent, int undo)
234 {
235         struct super_block *sb = sdp->sd_vfs;
236         struct gfs2_holder sb_gh;
237         u64 no_addr;
238         struct inode *inode;
239         int error = 0;
240
241         if (undo) {
242                 if (sb->s_root) {
243                         dput(sb->s_root);
244                         sb->s_root = NULL;
245                 }
246                 return 0;
247         }
248
249         error = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
250                                  LM_ST_SHARED, 0, &sb_gh);
251         if (error) {
252                 fs_err(sdp, "can't acquire superblock glock: %d\n", error);
253                 return error;
254         }
255
256         error = gfs2_read_sb(sdp, sb_gh.gh_gl, silent);
257         if (error) {
258                 fs_err(sdp, "can't read superblock: %d\n", error);
259                 goto out;
260         }
261
262         /* Set up the buffer cache and SB for real */
263         if (sdp->sd_sb.sb_bsize < bdev_hardsect_size(sb->s_bdev)) {
264                 error = -EINVAL;
265                 fs_err(sdp, "FS block size (%u) is too small for device "
266                        "block size (%u)\n",
267                        sdp->sd_sb.sb_bsize, bdev_hardsect_size(sb->s_bdev));
268                 goto out;
269         }
270         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
271                 error = -EINVAL;
272                 fs_err(sdp, "FS block size (%u) is too big for machine "
273                        "page size (%u)\n",
274                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
275                 goto out;
276         }
277         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
278
279         /* Get the root inode */
280         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
281         if (sb->s_type == &gfs2meta_fs_type)
282                 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
283         inode = gfs2_lookup_root(sb, no_addr);
284         if (IS_ERR(inode)) {
285                 error = PTR_ERR(inode);
286                 fs_err(sdp, "can't read in root inode: %d\n", error);
287                 goto out;
288         }
289
290         sb->s_root = d_alloc_root(inode);
291         if (!sb->s_root) {
292                 fs_err(sdp, "can't get root dentry\n");
293                 error = -ENOMEM;
294                 iput(inode);
295         } else
296                 sb->s_root->d_op = &gfs2_dops;
297         
298 out:
299         gfs2_glock_dq_uninit(&sb_gh);
300         return error;
301 }
302
303 /**
304  * map_journal_extents - create a reusable "extent" mapping from all logical
305  * blocks to all physical blocks for the given journal.  This will save
306  * us time when writing journal blocks.  Most journals will have only one
307  * extent that maps all their logical blocks.  That's because gfs2.mkfs
308  * arranges the journal blocks sequentially to maximize performance.
309  * So the extent would map the first block for the entire file length.
310  * However, gfs2_jadd can happen while file activity is happening, so
311  * those journals may not be sequential.  Less likely is the case where
312  * the users created their own journals by mounting the metafs and
313  * laying it out.  But it's still possible.  These journals might have
314  * several extents.
315  *
316  * TODO: This should be done in bigger chunks rather than one block at a time,
317  *       but since it's only done at mount time, I'm not worried about the
318  *       time it takes.
319  */
320 static int map_journal_extents(struct gfs2_sbd *sdp)
321 {
322         struct gfs2_jdesc *jd = sdp->sd_jdesc;
323         unsigned int lb;
324         u64 db, prev_db; /* logical block, disk block, prev disk block */
325         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
326         struct gfs2_journal_extent *jext = NULL;
327         struct buffer_head bh;
328         int rc = 0;
329
330         prev_db = 0;
331
332         for (lb = 0; lb < ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift; lb++) {
333                 bh.b_state = 0;
334                 bh.b_blocknr = 0;
335                 bh.b_size = 1 << ip->i_inode.i_blkbits;
336                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
337                 db = bh.b_blocknr;
338                 if (rc || !db) {
339                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
340                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
341                         break;
342                 }
343                 if (!prev_db || db != prev_db + 1) {
344                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
345                                        GFP_KERNEL);
346                         if (!jext) {
347                                 printk(KERN_INFO "GFS2 error: out of memory "
348                                        "mapping journal extents.\n");
349                                 rc = -ENOMEM;
350                                 break;
351                         }
352                         jext->dblock = db;
353                         jext->lblock = lb;
354                         jext->blocks = 1;
355                         list_add_tail(&jext->extent_list, &jd->extent_list);
356                 } else {
357                         jext->blocks++;
358                 }
359                 prev_db = db;
360         }
361         return rc;
362 }
363
364 static void gfs2_lm_others_may_mount(struct gfs2_sbd *sdp)
365 {
366         if (!sdp->sd_lockstruct.ls_ops->lm_others_may_mount)
367                 return;
368         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
369                 sdp->sd_lockstruct.ls_ops->lm_others_may_mount(
370                                         sdp->sd_lockstruct.ls_lockspace);
371 }
372
373 static int init_journal(struct gfs2_sbd *sdp, int undo)
374 {
375         struct gfs2_holder ji_gh;
376         struct task_struct *p;
377         struct gfs2_inode *ip;
378         int jindex = 1;
379         int error = 0;
380
381         if (undo) {
382                 jindex = 0;
383                 goto fail_recoverd;
384         }
385
386         sdp->sd_jindex = gfs2_lookup_simple(sdp->sd_master_dir, "jindex");
387         if (IS_ERR(sdp->sd_jindex)) {
388                 fs_err(sdp, "can't lookup journal index: %d\n", error);
389                 return PTR_ERR(sdp->sd_jindex);
390         }
391         ip = GFS2_I(sdp->sd_jindex);
392         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
393
394         /* Load in the journal index special file */
395
396         error = gfs2_jindex_hold(sdp, &ji_gh);
397         if (error) {
398                 fs_err(sdp, "can't read journal index: %d\n", error);
399                 goto fail;
400         }
401
402         error = -EINVAL;
403         if (!gfs2_jindex_size(sdp)) {
404                 fs_err(sdp, "no journals!\n");
405                 goto fail_jindex;
406         }
407
408         if (sdp->sd_args.ar_spectator) {
409                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
410                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
411         } else {
412                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
413                         fs_err(sdp, "can't mount journal #%u\n",
414                                sdp->sd_lockstruct.ls_jid);
415                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
416                                gfs2_jindex_size(sdp),
417                                gfs2_jindex_size(sdp) - 1);
418                         goto fail_jindex;
419                 }
420                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
421
422                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
423                                           &gfs2_journal_glops,
424                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
425                                           &sdp->sd_journal_gh);
426                 if (error) {
427                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
428                         goto fail_jindex;
429                 }
430
431                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
432                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
433                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
434                                            &sdp->sd_jinode_gh);
435                 if (error) {
436                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
437                                error);
438                         goto fail_journal_gh;
439                 }
440
441                 error = gfs2_jdesc_check(sdp->sd_jdesc);
442                 if (error) {
443                         fs_err(sdp, "my journal (%u) is bad: %d\n",
444                                sdp->sd_jdesc->jd_jid, error);
445                         goto fail_jinode_gh;
446                 }
447                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
448
449                 /* Map the extents for this journal's blocks */
450                 map_journal_extents(sdp);
451         }
452
453         if (sdp->sd_lockstruct.ls_first) {
454                 unsigned int x;
455                 for (x = 0; x < sdp->sd_journals; x++) {
456                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
457                         if (error) {
458                                 fs_err(sdp, "error recovering journal %u: %d\n",
459                                        x, error);
460                                 goto fail_jinode_gh;
461                         }
462                 }
463
464                 gfs2_lm_others_may_mount(sdp);
465         } else if (!sdp->sd_args.ar_spectator) {
466                 error = gfs2_recover_journal(sdp->sd_jdesc);
467                 if (error) {
468                         fs_err(sdp, "error recovering my journal: %d\n", error);
469                         goto fail_jinode_gh;
470                 }
471         }
472
473         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
474         gfs2_glock_dq_uninit(&ji_gh);
475         jindex = 0;
476
477         p = kthread_run(gfs2_recoverd, sdp, "gfs2_recoverd");
478         error = IS_ERR(p);
479         if (error) {
480                 fs_err(sdp, "can't start recoverd thread: %d\n", error);
481                 goto fail_jinode_gh;
482         }
483         sdp->sd_recoverd_process = p;
484
485         return 0;
486
487 fail_recoverd:
488         kthread_stop(sdp->sd_recoverd_process);
489 fail_jinode_gh:
490         if (!sdp->sd_args.ar_spectator)
491                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
492 fail_journal_gh:
493         if (!sdp->sd_args.ar_spectator)
494                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
495 fail_jindex:
496         gfs2_jindex_free(sdp);
497         if (jindex)
498                 gfs2_glock_dq_uninit(&ji_gh);
499 fail:
500         iput(sdp->sd_jindex);
501         return error;
502 }
503
504
505 static int init_inodes(struct gfs2_sbd *sdp, int undo)
506 {
507         int error = 0;
508         struct gfs2_inode *ip;
509         struct inode *inode;
510
511         if (undo)
512                 goto fail_qinode;
513
514         inode = gfs2_lookup_root(sdp->sd_vfs, sdp->sd_sb.sb_master_dir.no_addr);
515         if (IS_ERR(inode)) {
516                 error = PTR_ERR(inode);
517                 fs_err(sdp, "can't read in master directory: %d\n", error);
518                 goto fail;
519         }
520         sdp->sd_master_dir = inode;
521
522         error = init_journal(sdp, undo);
523         if (error)
524                 goto fail_master;
525
526         /* Read in the master inode number inode */
527         sdp->sd_inum_inode = gfs2_lookup_simple(sdp->sd_master_dir, "inum");
528         if (IS_ERR(sdp->sd_inum_inode)) {
529                 error = PTR_ERR(sdp->sd_inum_inode);
530                 fs_err(sdp, "can't read in inum inode: %d\n", error);
531                 goto fail_journal;
532         }
533
534
535         /* Read in the master statfs inode */
536         sdp->sd_statfs_inode = gfs2_lookup_simple(sdp->sd_master_dir, "statfs");
537         if (IS_ERR(sdp->sd_statfs_inode)) {
538                 error = PTR_ERR(sdp->sd_statfs_inode);
539                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
540                 goto fail_inum;
541         }
542
543         /* Read in the resource index inode */
544         sdp->sd_rindex = gfs2_lookup_simple(sdp->sd_master_dir, "rindex");
545         if (IS_ERR(sdp->sd_rindex)) {
546                 error = PTR_ERR(sdp->sd_rindex);
547                 fs_err(sdp, "can't get resource index inode: %d\n", error);
548                 goto fail_statfs;
549         }
550         ip = GFS2_I(sdp->sd_rindex);
551         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
552         sdp->sd_rindex_uptodate = 0;
553
554         /* Read in the quota inode */
555         sdp->sd_quota_inode = gfs2_lookup_simple(sdp->sd_master_dir, "quota");
556         if (IS_ERR(sdp->sd_quota_inode)) {
557                 error = PTR_ERR(sdp->sd_quota_inode);
558                 fs_err(sdp, "can't get quota file inode: %d\n", error);
559                 goto fail_rindex;
560         }
561         return 0;
562
563 fail_qinode:
564         iput(sdp->sd_quota_inode);
565 fail_rindex:
566         gfs2_clear_rgrpd(sdp);
567         iput(sdp->sd_rindex);
568 fail_statfs:
569         iput(sdp->sd_statfs_inode);
570 fail_inum:
571         iput(sdp->sd_inum_inode);
572 fail_journal:
573         init_journal(sdp, UNDO);
574 fail_master:
575         iput(sdp->sd_master_dir);
576 fail:
577         return error;
578 }
579
580 static int init_per_node(struct gfs2_sbd *sdp, int undo)
581 {
582         struct inode *pn = NULL;
583         char buf[30];
584         int error = 0;
585         struct gfs2_inode *ip;
586
587         if (sdp->sd_args.ar_spectator)
588                 return 0;
589
590         if (undo)
591                 goto fail_qc_gh;
592
593         pn = gfs2_lookup_simple(sdp->sd_master_dir, "per_node");
594         if (IS_ERR(pn)) {
595                 error = PTR_ERR(pn);
596                 fs_err(sdp, "can't find per_node directory: %d\n", error);
597                 return error;
598         }
599
600         sprintf(buf, "inum_range%u", sdp->sd_jdesc->jd_jid);
601         sdp->sd_ir_inode = gfs2_lookup_simple(pn, buf);
602         if (IS_ERR(sdp->sd_ir_inode)) {
603                 error = PTR_ERR(sdp->sd_ir_inode);
604                 fs_err(sdp, "can't find local \"ir\" file: %d\n", error);
605                 goto fail;
606         }
607
608         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
609         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
610         if (IS_ERR(sdp->sd_sc_inode)) {
611                 error = PTR_ERR(sdp->sd_sc_inode);
612                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
613                 goto fail_ir_i;
614         }
615
616         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
617         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
618         if (IS_ERR(sdp->sd_qc_inode)) {
619                 error = PTR_ERR(sdp->sd_qc_inode);
620                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
621                 goto fail_ut_i;
622         }
623
624         iput(pn);
625         pn = NULL;
626
627         ip = GFS2_I(sdp->sd_ir_inode);
628         error = gfs2_glock_nq_init(ip->i_gl,
629                                    LM_ST_EXCLUSIVE, 0,
630                                    &sdp->sd_ir_gh);
631         if (error) {
632                 fs_err(sdp, "can't lock local \"ir\" file: %d\n", error);
633                 goto fail_qc_i;
634         }
635
636         ip = GFS2_I(sdp->sd_sc_inode);
637         error = gfs2_glock_nq_init(ip->i_gl,
638                                    LM_ST_EXCLUSIVE, 0,
639                                    &sdp->sd_sc_gh);
640         if (error) {
641                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
642                 goto fail_ir_gh;
643         }
644
645         ip = GFS2_I(sdp->sd_qc_inode);
646         error = gfs2_glock_nq_init(ip->i_gl,
647                                    LM_ST_EXCLUSIVE, 0,
648                                    &sdp->sd_qc_gh);
649         if (error) {
650                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
651                 goto fail_ut_gh;
652         }
653
654         return 0;
655
656 fail_qc_gh:
657         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
658 fail_ut_gh:
659         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
660 fail_ir_gh:
661         gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
662 fail_qc_i:
663         iput(sdp->sd_qc_inode);
664 fail_ut_i:
665         iput(sdp->sd_sc_inode);
666 fail_ir_i:
667         iput(sdp->sd_ir_inode);
668 fail:
669         if (pn)
670                 iput(pn);
671         return error;
672 }
673
674 static int init_threads(struct gfs2_sbd *sdp, int undo)
675 {
676         struct task_struct *p;
677         int error = 0;
678
679         if (undo)
680                 goto fail_quotad;
681
682         sdp->sd_log_flush_time = jiffies;
683         sdp->sd_jindex_refresh_time = jiffies;
684
685         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
686         error = IS_ERR(p);
687         if (error) {
688                 fs_err(sdp, "can't start logd thread: %d\n", error);
689                 return error;
690         }
691         sdp->sd_logd_process = p;
692
693         sdp->sd_statfs_sync_time = jiffies;
694         sdp->sd_quota_sync_time = jiffies;
695
696         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
697         error = IS_ERR(p);
698         if (error) {
699                 fs_err(sdp, "can't start quotad thread: %d\n", error);
700                 goto fail;
701         }
702         sdp->sd_quotad_process = p;
703
704         return 0;
705
706
707 fail_quotad:
708         kthread_stop(sdp->sd_quotad_process);
709 fail:
710         kthread_stop(sdp->sd_logd_process);
711         return error;
712 }
713
714 /**
715  * gfs2_lm_mount - mount a locking protocol
716  * @sdp: the filesystem
717  * @args: mount arguements
718  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
719  *
720  * Returns: errno
721  */
722
723 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
724 {
725         char *proto = sdp->sd_proto_name;
726         char *table = sdp->sd_table_name;
727         int flags = LM_MFLAG_CONV_NODROP;
728         int error;
729
730         if (sdp->sd_args.ar_spectator)
731                 flags |= LM_MFLAG_SPECTATOR;
732
733         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
734
735         error = gfs2_mount_lockproto(proto, table, sdp->sd_args.ar_hostdata,
736                                      gfs2_glock_cb, sdp,
737                                      GFS2_MIN_LVB_SIZE, flags,
738                                      &sdp->sd_lockstruct, &sdp->sd_kobj);
739         if (error) {
740                 fs_info(sdp, "can't mount proto=%s, table=%s, hostdata=%s\n",
741                         proto, table, sdp->sd_args.ar_hostdata);
742                 goto out;
743         }
744
745         if (gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_ops) ||
746             gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lvb_size >=
747                                   GFS2_MIN_LVB_SIZE)) {
748                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
749                 goto out;
750         }
751
752         if (sdp->sd_args.ar_spectator)
753                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
754         else
755                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
756                          sdp->sd_lockstruct.ls_jid);
757
758         fs_info(sdp, "Joined cluster. Now mounting FS...\n");
759
760         if ((sdp->sd_lockstruct.ls_flags & LM_LSFLAG_LOCAL) &&
761             !sdp->sd_args.ar_ignore_local_fs) {
762                 sdp->sd_args.ar_localflocks = 1;
763                 sdp->sd_args.ar_localcaching = 1;
764         }
765
766 out:
767         return error;
768 }
769
770 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
771 {
772         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
773                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
774 }
775
776 /**
777  * fill_super - Read in superblock
778  * @sb: The VFS superblock
779  * @data: Mount options
780  * @silent: Don't complain if it's not a GFS2 filesystem
781  *
782  * Returns: errno
783  */
784
785 static int fill_super(struct super_block *sb, void *data, int silent)
786 {
787         struct gfs2_sbd *sdp;
788         struct gfs2_holder mount_gh;
789         int error;
790
791         sdp = init_sbd(sb);
792         if (!sdp) {
793                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
794                 return -ENOMEM;
795         }
796
797         error = gfs2_mount_args(sdp, (char *)data, 0);
798         if (error) {
799                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
800                 goto fail;
801         }
802
803         init_vfs(sb, SDF_NOATIME);
804
805         /* Set up the buffer cache and fill in some fake block size values
806            to allow us to read-in the on-disk superblock. */
807         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
808         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
809         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
810                                GFS2_BASIC_BLOCK_SHIFT;
811         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
812
813         error = init_names(sdp, silent);
814         if (error)
815                 goto fail;
816
817         gfs2_create_debugfs_file(sdp);
818
819         error = gfs2_sys_fs_add(sdp);
820         if (error)
821                 goto fail;
822
823         error = gfs2_lm_mount(sdp, silent);
824         if (error)
825                 goto fail_sys;
826
827         error = init_locking(sdp, &mount_gh, DO);
828         if (error)
829                 goto fail_lm;
830
831         error = init_sb(sdp, silent, DO);
832         if (error)
833                 goto fail_locking;
834
835         error = init_inodes(sdp, DO);
836         if (error)
837                 goto fail_sb;
838
839         error = init_per_node(sdp, DO);
840         if (error)
841                 goto fail_inodes;
842
843         error = gfs2_statfs_init(sdp);
844         if (error) {
845                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
846                 goto fail_per_node;
847         }
848
849         error = init_threads(sdp, DO);
850         if (error)
851                 goto fail_per_node;
852
853         if (!(sb->s_flags & MS_RDONLY)) {
854                 error = gfs2_make_fs_rw(sdp);
855                 if (error) {
856                         fs_err(sdp, "can't make FS RW: %d\n", error);
857                         goto fail_threads;
858                 }
859         }
860
861         gfs2_glock_dq_uninit(&mount_gh);
862
863         return 0;
864
865 fail_threads:
866         init_threads(sdp, UNDO);
867 fail_per_node:
868         init_per_node(sdp, UNDO);
869 fail_inodes:
870         init_inodes(sdp, UNDO);
871 fail_sb:
872         init_sb(sdp, 0, UNDO);
873 fail_locking:
874         init_locking(sdp, &mount_gh, UNDO);
875 fail_lm:
876         gfs2_gl_hash_clear(sdp);
877         gfs2_lm_unmount(sdp);
878         while (invalidate_inodes(sb))
879                 yield();
880 fail_sys:
881         gfs2_sys_fs_del(sdp);
882 fail:
883         gfs2_delete_debugfs_file(sdp);
884         kfree(sdp);
885         sb->s_fs_info = NULL;
886         return error;
887 }
888
889 static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
890                 const char *dev_name, void *data, struct vfsmount *mnt)
891 {
892         struct super_block *sb;
893         struct gfs2_sbd *sdp;
894         int error = get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
895         if (error)
896                 goto out;
897         sb = mnt->mnt_sb;
898         sdp = sb->s_fs_info;
899         sdp->sd_gfs2mnt = mnt;
900 out:
901         return error;
902 }
903
904 static int fill_super_meta(struct super_block *sb, struct super_block *new,
905                            void *data, int silent)
906 {
907         struct gfs2_sbd *sdp = sb->s_fs_info;
908         struct inode *inode;
909         int error = 0;
910
911         new->s_fs_info = sdp;
912         sdp->sd_vfs_meta = sb;
913
914         init_vfs(new, SDF_NOATIME);
915
916         /* Get the master inode */
917         inode = igrab(sdp->sd_master_dir);
918
919         new->s_root = d_alloc_root(inode);
920         if (!new->s_root) {
921                 fs_err(sdp, "can't get root dentry\n");
922                 error = -ENOMEM;
923                 iput(inode);
924         } else
925                 new->s_root->d_op = &gfs2_dops;
926
927         return error;
928 }
929
930 static int set_bdev_super(struct super_block *s, void *data)
931 {
932         s->s_bdev = data;
933         s->s_dev = s->s_bdev->bd_dev;
934         return 0;
935 }
936
937 static int test_bdev_super(struct super_block *s, void *data)
938 {
939         return s->s_bdev == data;
940 }
941
942 static struct super_block* get_gfs2_sb(const char *dev_name)
943 {
944         struct kstat stat;
945         struct nameidata nd;
946         struct super_block *sb = NULL, *s;
947         int error;
948
949         error = path_lookup(dev_name, LOOKUP_FOLLOW, &nd);
950         if (error) {
951                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error\n",
952                        dev_name);
953                 goto out;
954         }
955         error = vfs_getattr(nd.path.mnt, nd.path.dentry, &stat);
956
957         list_for_each_entry(s, &gfs2_fs_type.fs_supers, s_instances) {
958                 if ((S_ISBLK(stat.mode) && s->s_dev == stat.rdev) ||
959                     (S_ISDIR(stat.mode) &&
960                      s == nd.path.dentry->d_inode->i_sb)) {
961                         sb = s;
962                         goto free_nd;
963                 }
964         }
965
966         printk(KERN_WARNING "GFS2: Unrecognized block device or "
967                "mount point %s\n", dev_name);
968
969 free_nd:
970         path_put(&nd.path);
971 out:
972         return sb;
973 }
974
975 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
976                             const char *dev_name, void *data, struct vfsmount *mnt)
977 {
978         int error = 0;
979         struct super_block *sb = NULL, *new;
980         struct gfs2_sbd *sdp;
981
982         sb = get_gfs2_sb(dev_name);
983         if (!sb) {
984                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
985                 error = -ENOENT;
986                 goto error;
987         }
988         sdp = sb->s_fs_info;
989         if (sdp->sd_vfs_meta) {
990                 printk(KERN_WARNING "GFS2: gfs2meta mount already exists\n");
991                 error = -EBUSY;
992                 goto error;
993         }
994         down(&sb->s_bdev->bd_mount_sem);
995         new = sget(fs_type, test_bdev_super, set_bdev_super, sb->s_bdev);
996         up(&sb->s_bdev->bd_mount_sem);
997         if (IS_ERR(new)) {
998                 error = PTR_ERR(new);
999                 goto error;
1000         }
1001         new->s_flags = flags;
1002         strlcpy(new->s_id, sb->s_id, sizeof(new->s_id));
1003         sb_set_blocksize(new, sb->s_blocksize);
1004         error = fill_super_meta(sb, new, data, flags & MS_SILENT ? 1 : 0);
1005         if (error) {
1006                 up_write(&new->s_umount);
1007                 deactivate_super(new);
1008                 goto error;
1009         }
1010
1011         new->s_flags |= MS_ACTIVE;
1012
1013         /* Grab a reference to the gfs2 mount point */
1014         atomic_inc(&sdp->sd_gfs2mnt->mnt_count);
1015         return simple_set_mnt(mnt, new);
1016 error:
1017         return error;
1018 }
1019
1020 static void gfs2_kill_sb(struct super_block *sb)
1021 {
1022         if (sb->s_fs_info) {
1023                 gfs2_delete_debugfs_file(sb->s_fs_info);
1024                 gfs2_meta_syncfs(sb->s_fs_info);
1025         }
1026         kill_block_super(sb);
1027 }
1028
1029 static void gfs2_kill_sb_meta(struct super_block *sb)
1030 {
1031         struct gfs2_sbd *sdp = sb->s_fs_info;
1032         generic_shutdown_super(sb);
1033         sdp->sd_vfs_meta = NULL;
1034         atomic_dec(&sdp->sd_gfs2mnt->mnt_count);
1035 }
1036
1037 struct file_system_type gfs2_fs_type = {
1038         .name = "gfs2",
1039         .fs_flags = FS_REQUIRES_DEV,
1040         .get_sb = gfs2_get_sb,
1041         .kill_sb = gfs2_kill_sb,
1042         .owner = THIS_MODULE,
1043 };
1044
1045 struct file_system_type gfs2meta_fs_type = {
1046         .name = "gfs2meta",
1047         .fs_flags = FS_REQUIRES_DEV,
1048         .get_sb = gfs2_get_sb_meta,
1049         .kill_sb = gfs2_kill_sb_meta,
1050         .owner = THIS_MODULE,
1051 };
1052