ocfs2: Switch over to JBD2.
[safe/jmp/linux-2.6] / fs / ocfs2 / super.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * super.c
5  *
6  * load/unload driver, mount/dismount volumes
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25
26 #include <linux/module.h>
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
42 #include <linux/mount.h>
43 #include <linux/seq_file.h>
44
45 #define MLOG_MASK_PREFIX ML_SUPER
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 /* this should be the only file to include a version 1 header */
51 #include "ocfs1_fs_compat.h"
52
53 #include "alloc.h"
54 #include "dlmglue.h"
55 #include "export.h"
56 #include "extent_map.h"
57 #include "heartbeat.h"
58 #include "inode.h"
59 #include "journal.h"
60 #include "localalloc.h"
61 #include "namei.h"
62 #include "slot_map.h"
63 #include "super.h"
64 #include "sysfile.h"
65 #include "uptodate.h"
66 #include "ver.h"
67 #include "xattr.h"
68
69 #include "buffer_head_io.h"
70
71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
72
73 /* OCFS2 needs to schedule several differnt types of work which
74  * require cluster locking, disk I/O, recovery waits, etc. Since these
75  * types of work tend to be heavy we avoid using the kernel events
76  * workqueue and schedule on our own. */
77 struct workqueue_struct *ocfs2_wq = NULL;
78
79 static struct dentry *ocfs2_debugfs_root = NULL;
80
81 MODULE_AUTHOR("Oracle");
82 MODULE_LICENSE("GPL");
83
84 struct mount_options
85 {
86         unsigned long   commit_interval;
87         unsigned long   mount_opt;
88         unsigned int    atime_quantum;
89         signed short    slot;
90         unsigned int    localalloc_opt;
91         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
92 };
93
94 static int ocfs2_parse_options(struct super_block *sb, char *options,
95                                struct mount_options *mopt,
96                                int is_remount);
97 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
98 static void ocfs2_put_super(struct super_block *sb);
99 static int ocfs2_mount_volume(struct super_block *sb);
100 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
101 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
102 static int ocfs2_initialize_mem_caches(void);
103 static void ocfs2_free_mem_caches(void);
104 static void ocfs2_delete_osb(struct ocfs2_super *osb);
105
106 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
107
108 static int ocfs2_sync_fs(struct super_block *sb, int wait);
109
110 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
111 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
112 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
113 static int ocfs2_check_volume(struct ocfs2_super *osb);
114 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
115                                struct buffer_head *bh,
116                                u32 sectsize);
117 static int ocfs2_initialize_super(struct super_block *sb,
118                                   struct buffer_head *bh,
119                                   int sector_size);
120 static int ocfs2_get_sector(struct super_block *sb,
121                             struct buffer_head **bh,
122                             int block,
123                             int sect_size);
124 static void ocfs2_write_super(struct super_block *sb);
125 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
126 static void ocfs2_destroy_inode(struct inode *inode);
127
128 static const struct super_operations ocfs2_sops = {
129         .statfs         = ocfs2_statfs,
130         .alloc_inode    = ocfs2_alloc_inode,
131         .destroy_inode  = ocfs2_destroy_inode,
132         .drop_inode     = ocfs2_drop_inode,
133         .clear_inode    = ocfs2_clear_inode,
134         .delete_inode   = ocfs2_delete_inode,
135         .sync_fs        = ocfs2_sync_fs,
136         .write_super    = ocfs2_write_super,
137         .put_super      = ocfs2_put_super,
138         .remount_fs     = ocfs2_remount,
139         .show_options   = ocfs2_show_options,
140 };
141
142 enum {
143         Opt_barrier,
144         Opt_err_panic,
145         Opt_err_ro,
146         Opt_intr,
147         Opt_nointr,
148         Opt_hb_none,
149         Opt_hb_local,
150         Opt_data_ordered,
151         Opt_data_writeback,
152         Opt_atime_quantum,
153         Opt_slot,
154         Opt_commit,
155         Opt_localalloc,
156         Opt_localflocks,
157         Opt_stack,
158         Opt_user_xattr,
159         Opt_nouser_xattr,
160         Opt_inode64,
161         Opt_err,
162 };
163
164 static const match_table_t tokens = {
165         {Opt_barrier, "barrier=%u"},
166         {Opt_err_panic, "errors=panic"},
167         {Opt_err_ro, "errors=remount-ro"},
168         {Opt_intr, "intr"},
169         {Opt_nointr, "nointr"},
170         {Opt_hb_none, OCFS2_HB_NONE},
171         {Opt_hb_local, OCFS2_HB_LOCAL},
172         {Opt_data_ordered, "data=ordered"},
173         {Opt_data_writeback, "data=writeback"},
174         {Opt_atime_quantum, "atime_quantum=%u"},
175         {Opt_slot, "preferred_slot=%u"},
176         {Opt_commit, "commit=%u"},
177         {Opt_localalloc, "localalloc=%d"},
178         {Opt_localflocks, "localflocks"},
179         {Opt_stack, "cluster_stack=%s"},
180         {Opt_user_xattr, "user_xattr"},
181         {Opt_nouser_xattr, "nouser_xattr"},
182         {Opt_inode64, "inode64"},
183         {Opt_err, NULL}
184 };
185
186 /*
187  * write_super and sync_fs ripped right out of ext3.
188  */
189 static void ocfs2_write_super(struct super_block *sb)
190 {
191         if (mutex_trylock(&sb->s_lock) != 0)
192                 BUG();
193         sb->s_dirt = 0;
194 }
195
196 static int ocfs2_sync_fs(struct super_block *sb, int wait)
197 {
198         int status;
199         tid_t target;
200         struct ocfs2_super *osb = OCFS2_SB(sb);
201
202         sb->s_dirt = 0;
203
204         if (ocfs2_is_hard_readonly(osb))
205                 return -EROFS;
206
207         if (wait) {
208                 status = ocfs2_flush_truncate_log(osb);
209                 if (status < 0)
210                         mlog_errno(status);
211         } else {
212                 ocfs2_schedule_truncate_log_flush(osb, 0);
213         }
214
215         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
216                                       &target)) {
217                 if (wait)
218                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
219                                              target);
220         }
221         return 0;
222 }
223
224 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
225 {
226         struct inode *new = NULL;
227         int status = 0;
228         int i;
229
230         mlog_entry_void();
231
232         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
233         if (IS_ERR(new)) {
234                 status = PTR_ERR(new);
235                 mlog_errno(status);
236                 goto bail;
237         }
238         osb->root_inode = new;
239
240         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
241         if (IS_ERR(new)) {
242                 status = PTR_ERR(new);
243                 mlog_errno(status);
244                 goto bail;
245         }
246         osb->sys_root_inode = new;
247
248         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
249              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
250                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
251                 if (!new) {
252                         ocfs2_release_system_inodes(osb);
253                         status = -EINVAL;
254                         mlog_errno(status);
255                         /* FIXME: Should ERROR_RO_FS */
256                         mlog(ML_ERROR, "Unable to load system inode %d, "
257                              "possibly corrupt fs?", i);
258                         goto bail;
259                 }
260                 // the array now has one ref, so drop this one
261                 iput(new);
262         }
263
264 bail:
265         mlog_exit(status);
266         return status;
267 }
268
269 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
270 {
271         struct inode *new = NULL;
272         int status = 0;
273         int i;
274
275         mlog_entry_void();
276
277         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
278              i < NUM_SYSTEM_INODES;
279              i++) {
280                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
281                 if (!new) {
282                         ocfs2_release_system_inodes(osb);
283                         status = -EINVAL;
284                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
285                              status, i, osb->slot_num);
286                         goto bail;
287                 }
288                 /* the array now has one ref, so drop this one */
289                 iput(new);
290         }
291
292 bail:
293         mlog_exit(status);
294         return status;
295 }
296
297 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
298 {
299         int i;
300         struct inode *inode;
301
302         mlog_entry_void();
303
304         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
305                 inode = osb->system_inodes[i];
306                 if (inode) {
307                         iput(inode);
308                         osb->system_inodes[i] = NULL;
309                 }
310         }
311
312         inode = osb->sys_root_inode;
313         if (inode) {
314                 iput(inode);
315                 osb->sys_root_inode = NULL;
316         }
317
318         inode = osb->root_inode;
319         if (inode) {
320                 iput(inode);
321                 osb->root_inode = NULL;
322         }
323
324         mlog_exit(0);
325 }
326
327 /* We're allocating fs objects, use GFP_NOFS */
328 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
329 {
330         struct ocfs2_inode_info *oi;
331
332         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
333         if (!oi)
334                 return NULL;
335
336         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
337         return &oi->vfs_inode;
338 }
339
340 static void ocfs2_destroy_inode(struct inode *inode)
341 {
342         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
343 }
344
345 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
346                                                 unsigned int cbits)
347 {
348         unsigned int bytes = 1 << cbits;
349         unsigned int trim = bytes;
350         unsigned int bitshift = 32;
351
352         /*
353          * i_size and all block offsets in ocfs2 are always 64 bits
354          * wide. i_clusters is 32 bits, in cluster-sized units. So on
355          * 64 bit platforms, cluster size will be the limiting factor.
356          */
357
358 #if BITS_PER_LONG == 32
359 # if defined(CONFIG_LBD)
360         BUILD_BUG_ON(sizeof(sector_t) != 8);
361         /*
362          * We might be limited by page cache size.
363          */
364         if (bytes > PAGE_CACHE_SIZE) {
365                 bytes = PAGE_CACHE_SIZE;
366                 trim = 1;
367                 /*
368                  * Shift by 31 here so that we don't get larger than
369                  * MAX_LFS_FILESIZE
370                  */
371                 bitshift = 31;
372         }
373 # else
374         /*
375          * We are limited by the size of sector_t. Use block size, as
376          * that's what we expose to the VFS.
377          */
378         bytes = 1 << bbits;
379         trim = 1;
380         bitshift = 31;
381 # endif
382 #endif
383
384         /*
385          * Trim by a whole cluster when we can actually approach the
386          * on-disk limits. Otherwise we can overflow i_clusters when
387          * an extent start is at the max offset.
388          */
389         return (((unsigned long long)bytes) << bitshift) - trim;
390 }
391
392 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
393 {
394         int incompat_features;
395         int ret = 0;
396         struct mount_options parsed_options;
397         struct ocfs2_super *osb = OCFS2_SB(sb);
398
399         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
400                 ret = -EINVAL;
401                 goto out;
402         }
403
404         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
405             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
406                 ret = -EINVAL;
407                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
408                 goto out;
409         }
410
411         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
412             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
413                 ret = -EINVAL;
414                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
415                 goto out;
416         }
417
418         /* Probably don't want this on remount; it might
419          * mess with other nodes */
420         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
421             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
422                 ret = -EINVAL;
423                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
424                 goto out;
425         }
426
427         /* We're going to/from readonly mode. */
428         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
429                 /* Lock here so the check of HARD_RO and the potential
430                  * setting of SOFT_RO is atomic. */
431                 spin_lock(&osb->osb_lock);
432                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
433                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
434                         ret = -EROFS;
435                         goto unlock_osb;
436                 }
437
438                 if (*flags & MS_RDONLY) {
439                         mlog(0, "Going to ro mode.\n");
440                         sb->s_flags |= MS_RDONLY;
441                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
442                 } else {
443                         mlog(0, "Making ro filesystem writeable.\n");
444
445                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
446                                 mlog(ML_ERROR, "Cannot remount RDWR "
447                                      "filesystem due to previous errors.\n");
448                                 ret = -EROFS;
449                                 goto unlock_osb;
450                         }
451                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
452                         if (incompat_features) {
453                                 mlog(ML_ERROR, "Cannot remount RDWR because "
454                                      "of unsupported optional features "
455                                      "(%x).\n", incompat_features);
456                                 ret = -EINVAL;
457                                 goto unlock_osb;
458                         }
459                         sb->s_flags &= ~MS_RDONLY;
460                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
461                 }
462 unlock_osb:
463                 spin_unlock(&osb->osb_lock);
464         }
465
466         if (!ret) {
467                 /* Only save off the new mount options in case of a successful
468                  * remount. */
469                 osb->s_mount_opt = parsed_options.mount_opt;
470                 osb->s_atime_quantum = parsed_options.atime_quantum;
471                 osb->preferred_slot = parsed_options.slot;
472                 if (parsed_options.commit_interval)
473                         osb->osb_commit_interval = parsed_options.commit_interval;
474
475                 if (!ocfs2_is_hard_readonly(osb))
476                         ocfs2_set_journal_params(osb);
477         }
478 out:
479         return ret;
480 }
481
482 static int ocfs2_sb_probe(struct super_block *sb,
483                           struct buffer_head **bh,
484                           int *sector_size)
485 {
486         int status, tmpstat;
487         struct ocfs1_vol_disk_hdr *hdr;
488         struct ocfs2_dinode *di;
489         int blksize;
490
491         *bh = NULL;
492
493         /* may be > 512 */
494         *sector_size = bdev_hardsect_size(sb->s_bdev);
495         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
496                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
497                      *sector_size, OCFS2_MAX_BLOCKSIZE);
498                 status = -EINVAL;
499                 goto bail;
500         }
501
502         /* Can this really happen? */
503         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
504                 *sector_size = OCFS2_MIN_BLOCKSIZE;
505
506         /* check block zero for old format */
507         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
508         if (status < 0) {
509                 mlog_errno(status);
510                 goto bail;
511         }
512         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
513         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
514                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
515                      hdr->major_version, hdr->minor_version);
516                 status = -EINVAL;
517         }
518         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
519                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
520                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
521                      hdr->signature);
522                 status = -EINVAL;
523         }
524         brelse(*bh);
525         *bh = NULL;
526         if (status < 0) {
527                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
528                      "upgraded before mounting with ocfs v2\n");
529                 goto bail;
530         }
531
532         /*
533          * Now check at magic offset for 512, 1024, 2048, 4096
534          * blocksizes.  4096 is the maximum blocksize because it is
535          * the minimum clustersize.
536          */
537         status = -EINVAL;
538         for (blksize = *sector_size;
539              blksize <= OCFS2_MAX_BLOCKSIZE;
540              blksize <<= 1) {
541                 tmpstat = ocfs2_get_sector(sb, bh,
542                                            OCFS2_SUPER_BLOCK_BLKNO,
543                                            blksize);
544                 if (tmpstat < 0) {
545                         status = tmpstat;
546                         mlog_errno(status);
547                         goto bail;
548                 }
549                 di = (struct ocfs2_dinode *) (*bh)->b_data;
550                 status = ocfs2_verify_volume(di, *bh, blksize);
551                 if (status >= 0)
552                         goto bail;
553                 brelse(*bh);
554                 *bh = NULL;
555                 if (status != -EAGAIN)
556                         break;
557         }
558
559 bail:
560         return status;
561 }
562
563 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
564 {
565         if (ocfs2_mount_local(osb)) {
566                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
567                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
568                              "mounted device.\n");
569                         return -EINVAL;
570                 }
571         }
572
573         if (ocfs2_userspace_stack(osb)) {
574                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
575                         mlog(ML_ERROR, "Userspace stack expected, but "
576                              "o2cb heartbeat arguments passed to mount\n");
577                         return -EINVAL;
578                 }
579         }
580
581         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
582                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
583                     !ocfs2_userspace_stack(osb)) {
584                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
585                              "a read-write clustered device.\n");
586                         return -EINVAL;
587                 }
588         }
589
590         return 0;
591 }
592
593 /*
594  * If we're using a userspace stack, mount should have passed
595  * a name that matches the disk.  If not, mount should not
596  * have passed a stack.
597  */
598 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
599                                         struct mount_options *mopt)
600 {
601         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
602                 mlog(ML_ERROR,
603                      "cluster stack passed to mount, but this filesystem "
604                      "does not support it\n");
605                 return -EINVAL;
606         }
607
608         if (ocfs2_userspace_stack(osb) &&
609             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
610                     OCFS2_STACK_LABEL_LEN)) {
611                 mlog(ML_ERROR,
612                      "cluster stack passed to mount (\"%s\") does not "
613                      "match the filesystem (\"%s\")\n",
614                      mopt->cluster_stack,
615                      osb->osb_cluster_stack);
616                 return -EINVAL;
617         }
618
619         return 0;
620 }
621
622 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
623 {
624         struct dentry *root;
625         int status, sector_size;
626         struct mount_options parsed_options;
627         struct inode *inode = NULL;
628         struct ocfs2_super *osb = NULL;
629         struct buffer_head *bh = NULL;
630         char nodestr[8];
631
632         mlog_entry("%p, %p, %i", sb, data, silent);
633
634         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
635                 status = -EINVAL;
636                 goto read_super_error;
637         }
638
639         /* probe for superblock */
640         status = ocfs2_sb_probe(sb, &bh, &sector_size);
641         if (status < 0) {
642                 mlog(ML_ERROR, "superblock probe failed!\n");
643                 goto read_super_error;
644         }
645
646         status = ocfs2_initialize_super(sb, bh, sector_size);
647         osb = OCFS2_SB(sb);
648         if (status < 0) {
649                 mlog_errno(status);
650                 goto read_super_error;
651         }
652         brelse(bh);
653         bh = NULL;
654         osb->s_mount_opt = parsed_options.mount_opt;
655         osb->s_atime_quantum = parsed_options.atime_quantum;
656         osb->preferred_slot = parsed_options.slot;
657         osb->osb_commit_interval = parsed_options.commit_interval;
658         osb->local_alloc_default_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
659         osb->local_alloc_bits = osb->local_alloc_default_bits;
660
661         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
662         if (status)
663                 goto read_super_error;
664
665         sb->s_magic = OCFS2_SUPER_MAGIC;
666
667         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
668          * heartbeat=none */
669         if (bdev_read_only(sb->s_bdev)) {
670                 if (!(sb->s_flags & MS_RDONLY)) {
671                         status = -EACCES;
672                         mlog(ML_ERROR, "Readonly device detected but readonly "
673                              "mount was not specified.\n");
674                         goto read_super_error;
675                 }
676
677                 /* You should not be able to start a local heartbeat
678                  * on a readonly device. */
679                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
680                         status = -EROFS;
681                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
682                              "device.\n");
683                         goto read_super_error;
684                 }
685
686                 status = ocfs2_check_journals_nolocks(osb);
687                 if (status < 0) {
688                         if (status == -EROFS)
689                                 mlog(ML_ERROR, "Recovery required on readonly "
690                                      "file system, but write access is "
691                                      "unavailable.\n");
692                         else
693                                 mlog_errno(status);                     
694                         goto read_super_error;
695                 }
696
697                 ocfs2_set_ro_flag(osb, 1);
698
699                 printk(KERN_NOTICE "Readonly device detected. No cluster "
700                        "services will be utilized for this mount. Recovery "
701                        "will be skipped.\n");
702         }
703
704         if (!ocfs2_is_hard_readonly(osb)) {
705                 if (sb->s_flags & MS_RDONLY)
706                         ocfs2_set_ro_flag(osb, 0);
707         }
708
709         status = ocfs2_verify_heartbeat(osb);
710         if (status < 0) {
711                 mlog_errno(status);
712                 goto read_super_error;
713         }
714
715         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
716                                                  ocfs2_debugfs_root);
717         if (!osb->osb_debug_root) {
718                 status = -EINVAL;
719                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
720                 goto read_super_error;
721         }
722
723         status = ocfs2_mount_volume(sb);
724         if (osb->root_inode)
725                 inode = igrab(osb->root_inode);
726
727         if (status < 0)
728                 goto read_super_error;
729
730         if (!inode) {
731                 status = -EIO;
732                 mlog_errno(status);
733                 goto read_super_error;
734         }
735
736         root = d_alloc_root(inode);
737         if (!root) {
738                 status = -ENOMEM;
739                 mlog_errno(status);
740                 goto read_super_error;
741         }
742
743         sb->s_root = root;
744
745         ocfs2_complete_mount_recovery(osb);
746
747         if (ocfs2_mount_local(osb))
748                 snprintf(nodestr, sizeof(nodestr), "local");
749         else
750                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
751
752         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
753                "with %s data mode.\n",
754                osb->dev_str, nodestr, osb->slot_num,
755                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
756                "ordered");
757
758         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
759         wake_up(&osb->osb_mount_event);
760
761         mlog_exit(status);
762         return status;
763
764 read_super_error:
765         if (bh != NULL)
766                 brelse(bh);
767
768         if (inode)
769                 iput(inode);
770
771         if (osb) {
772                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
773                 wake_up(&osb->osb_mount_event);
774                 ocfs2_dismount_volume(sb, 1);
775         }
776
777         mlog_exit(status);
778         return status;
779 }
780
781 static int ocfs2_get_sb(struct file_system_type *fs_type,
782                         int flags,
783                         const char *dev_name,
784                         void *data,
785                         struct vfsmount *mnt)
786 {
787         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
788                            mnt);
789 }
790
791 static struct file_system_type ocfs2_fs_type = {
792         .owner          = THIS_MODULE,
793         .name           = "ocfs2",
794         .get_sb         = ocfs2_get_sb, /* is this called when we mount
795                                         * the fs? */
796         .kill_sb        = kill_block_super, /* set to the generic one
797                                              * right now, but do we
798                                              * need to change that? */
799         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
800         .next           = NULL
801 };
802
803 static int ocfs2_parse_options(struct super_block *sb,
804                                char *options,
805                                struct mount_options *mopt,
806                                int is_remount)
807 {
808         int status;
809         char *p;
810
811         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
812                    options ? options : "(none)");
813
814         mopt->commit_interval = 0;
815         mopt->mount_opt = 0;
816         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
817         mopt->slot = OCFS2_INVALID_SLOT;
818         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
819         mopt->cluster_stack[0] = '\0';
820
821         if (!options) {
822                 status = 1;
823                 goto bail;
824         }
825
826         while ((p = strsep(&options, ",")) != NULL) {
827                 int token, option;
828                 substring_t args[MAX_OPT_ARGS];
829
830                 if (!*p)
831                         continue;
832
833                 token = match_token(p, tokens, args);
834                 switch (token) {
835                 case Opt_hb_local:
836                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
837                         break;
838                 case Opt_hb_none:
839                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
840                         break;
841                 case Opt_barrier:
842                         if (match_int(&args[0], &option)) {
843                                 status = 0;
844                                 goto bail;
845                         }
846                         if (option)
847                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
848                         else
849                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
850                         break;
851                 case Opt_intr:
852                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
853                         break;
854                 case Opt_nointr:
855                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
856                         break;
857                 case Opt_err_panic:
858                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
859                         break;
860                 case Opt_err_ro:
861                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
862                         break;
863                 case Opt_data_ordered:
864                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
865                         break;
866                 case Opt_data_writeback:
867                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
868                         break;
869                 case Opt_user_xattr:
870                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
871                         break;
872                 case Opt_nouser_xattr:
873                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
874                         break;
875                 case Opt_atime_quantum:
876                         if (match_int(&args[0], &option)) {
877                                 status = 0;
878                                 goto bail;
879                         }
880                         if (option >= 0)
881                                 mopt->atime_quantum = option;
882                         break;
883                 case Opt_slot:
884                         option = 0;
885                         if (match_int(&args[0], &option)) {
886                                 status = 0;
887                                 goto bail;
888                         }
889                         if (option)
890                                 mopt->slot = (s16)option;
891                         break;
892                 case Opt_commit:
893                         option = 0;
894                         if (match_int(&args[0], &option)) {
895                                 status = 0;
896                                 goto bail;
897                         }
898                         if (option < 0)
899                                 return 0;
900                         if (option == 0)
901                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
902                         mopt->commit_interval = HZ * option;
903                         break;
904                 case Opt_localalloc:
905                         option = 0;
906                         if (match_int(&args[0], &option)) {
907                                 status = 0;
908                                 goto bail;
909                         }
910                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
911                                 mopt->localalloc_opt = option;
912                         break;
913                 case Opt_localflocks:
914                         /*
915                          * Changing this during remount could race
916                          * flock() requests, or "unbalance" existing
917                          * ones (e.g., a lock is taken in one mode but
918                          * dropped in the other). If users care enough
919                          * to flip locking modes during remount, we
920                          * could add a "local" flag to individual
921                          * flock structures for proper tracking of
922                          * state.
923                          */
924                         if (!is_remount)
925                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
926                         break;
927                 case Opt_stack:
928                         /* Check both that the option we were passed
929                          * is of the right length and that it is a proper
930                          * string of the right length.
931                          */
932                         if (((args[0].to - args[0].from) !=
933                              OCFS2_STACK_LABEL_LEN) ||
934                             (strnlen(args[0].from,
935                                      OCFS2_STACK_LABEL_LEN) !=
936                              OCFS2_STACK_LABEL_LEN)) {
937                                 mlog(ML_ERROR,
938                                      "Invalid cluster_stack option\n");
939                                 status = 0;
940                                 goto bail;
941                         }
942                         memcpy(mopt->cluster_stack, args[0].from,
943                                OCFS2_STACK_LABEL_LEN);
944                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
945                         break;
946                 case Opt_inode64:
947                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
948                         break;
949                 default:
950                         mlog(ML_ERROR,
951                              "Unrecognized mount option \"%s\" "
952                              "or missing value\n", p);
953                         status = 0;
954                         goto bail;
955                 }
956         }
957
958         status = 1;
959
960 bail:
961         mlog_exit(status);
962         return status;
963 }
964
965 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
966 {
967         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
968         unsigned long opts = osb->s_mount_opt;
969         unsigned int local_alloc_megs;
970
971         if (opts & OCFS2_MOUNT_HB_LOCAL)
972                 seq_printf(s, ",_netdev,heartbeat=local");
973         else
974                 seq_printf(s, ",heartbeat=none");
975
976         if (opts & OCFS2_MOUNT_NOINTR)
977                 seq_printf(s, ",nointr");
978
979         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
980                 seq_printf(s, ",data=writeback");
981         else
982                 seq_printf(s, ",data=ordered");
983
984         if (opts & OCFS2_MOUNT_BARRIER)
985                 seq_printf(s, ",barrier=1");
986
987         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
988                 seq_printf(s, ",errors=panic");
989         else
990                 seq_printf(s, ",errors=remount-ro");
991
992         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
993                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
994
995         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
996                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
997
998         if (osb->osb_commit_interval)
999                 seq_printf(s, ",commit=%u",
1000                            (unsigned) (osb->osb_commit_interval / HZ));
1001
1002         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1003         if (local_alloc_megs != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
1004                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1005
1006         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1007                 seq_printf(s, ",localflocks,");
1008
1009         if (osb->osb_cluster_stack[0])
1010                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1011                            osb->osb_cluster_stack);
1012
1013         if (opts & OCFS2_MOUNT_INODE64)
1014                 seq_printf(s, ",inode64");
1015
1016         return 0;
1017 }
1018
1019 static int __init ocfs2_init(void)
1020 {
1021         int status;
1022
1023         mlog_entry_void();
1024
1025         ocfs2_print_version();
1026
1027         status = init_ocfs2_uptodate_cache();
1028         if (status < 0) {
1029                 mlog_errno(status);
1030                 goto leave;
1031         }
1032
1033         status = ocfs2_initialize_mem_caches();
1034         if (status < 0) {
1035                 mlog_errno(status);
1036                 goto leave;
1037         }
1038
1039         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1040         if (!ocfs2_wq) {
1041                 status = -ENOMEM;
1042                 goto leave;
1043         }
1044
1045         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1046         if (!ocfs2_debugfs_root) {
1047                 status = -EFAULT;
1048                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1049         }
1050
1051         ocfs2_set_locking_protocol();
1052
1053 leave:
1054         if (status < 0) {
1055                 ocfs2_free_mem_caches();
1056                 exit_ocfs2_uptodate_cache();
1057         }
1058
1059         mlog_exit(status);
1060
1061         if (status >= 0) {
1062                 return register_filesystem(&ocfs2_fs_type);
1063         } else
1064                 return -1;
1065 }
1066
1067 static void __exit ocfs2_exit(void)
1068 {
1069         mlog_entry_void();
1070
1071         if (ocfs2_wq) {
1072                 flush_workqueue(ocfs2_wq);
1073                 destroy_workqueue(ocfs2_wq);
1074         }
1075
1076         debugfs_remove(ocfs2_debugfs_root);
1077
1078         ocfs2_free_mem_caches();
1079
1080         unregister_filesystem(&ocfs2_fs_type);
1081
1082         exit_ocfs2_uptodate_cache();
1083
1084         mlog_exit_void();
1085 }
1086
1087 static void ocfs2_put_super(struct super_block *sb)
1088 {
1089         mlog_entry("(0x%p)\n", sb);
1090
1091         ocfs2_sync_blockdev(sb);
1092         ocfs2_dismount_volume(sb, 0);
1093
1094         mlog_exit_void();
1095 }
1096
1097 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1098 {
1099         struct ocfs2_super *osb;
1100         u32 numbits, freebits;
1101         int status;
1102         struct ocfs2_dinode *bm_lock;
1103         struct buffer_head *bh = NULL;
1104         struct inode *inode = NULL;
1105
1106         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1107
1108         osb = OCFS2_SB(dentry->d_sb);
1109
1110         inode = ocfs2_get_system_file_inode(osb,
1111                                             GLOBAL_BITMAP_SYSTEM_INODE,
1112                                             OCFS2_INVALID_SLOT);
1113         if (!inode) {
1114                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1115                 status = -EIO;
1116                 goto bail;
1117         }
1118
1119         status = ocfs2_inode_lock(inode, &bh, 0);
1120         if (status < 0) {
1121                 mlog_errno(status);
1122                 goto bail;
1123         }
1124
1125         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1126
1127         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1128         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1129
1130         buf->f_type = OCFS2_SUPER_MAGIC;
1131         buf->f_bsize = dentry->d_sb->s_blocksize;
1132         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1133         buf->f_blocks = ((sector_t) numbits) *
1134                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1135         buf->f_bfree = ((sector_t) freebits) *
1136                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1137         buf->f_bavail = buf->f_bfree;
1138         buf->f_files = numbits;
1139         buf->f_ffree = freebits;
1140
1141         brelse(bh);
1142
1143         ocfs2_inode_unlock(inode, 0);
1144         status = 0;
1145 bail:
1146         if (inode)
1147                 iput(inode);
1148
1149         mlog_exit(status);
1150
1151         return status;
1152 }
1153
1154 static void ocfs2_inode_init_once(void *data)
1155 {
1156         struct ocfs2_inode_info *oi = data;
1157
1158         oi->ip_flags = 0;
1159         oi->ip_open_count = 0;
1160         spin_lock_init(&oi->ip_lock);
1161         ocfs2_extent_map_init(&oi->vfs_inode);
1162         INIT_LIST_HEAD(&oi->ip_io_markers);
1163         oi->ip_created_trans = 0;
1164         oi->ip_last_trans = 0;
1165         oi->ip_dir_start_lookup = 0;
1166
1167         init_rwsem(&oi->ip_alloc_sem);
1168         init_rwsem(&oi->ip_xattr_sem);
1169         mutex_init(&oi->ip_io_mutex);
1170
1171         oi->ip_blkno = 0ULL;
1172         oi->ip_clusters = 0;
1173
1174         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1175         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1176         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1177
1178         ocfs2_metadata_cache_init(&oi->vfs_inode);
1179
1180         inode_init_once(&oi->vfs_inode);
1181 }
1182
1183 static int ocfs2_initialize_mem_caches(void)
1184 {
1185         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1186                                        sizeof(struct ocfs2_inode_info),
1187                                        0,
1188                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1189                                                 SLAB_MEM_SPREAD),
1190                                        ocfs2_inode_init_once);
1191         if (!ocfs2_inode_cachep)
1192                 return -ENOMEM;
1193
1194         return 0;
1195 }
1196
1197 static void ocfs2_free_mem_caches(void)
1198 {
1199         if (ocfs2_inode_cachep)
1200                 kmem_cache_destroy(ocfs2_inode_cachep);
1201
1202         ocfs2_inode_cachep = NULL;
1203 }
1204
1205 static int ocfs2_get_sector(struct super_block *sb,
1206                             struct buffer_head **bh,
1207                             int block,
1208                             int sect_size)
1209 {
1210         if (!sb_set_blocksize(sb, sect_size)) {
1211                 mlog(ML_ERROR, "unable to set blocksize\n");
1212                 return -EIO;
1213         }
1214
1215         *bh = sb_getblk(sb, block);
1216         if (!*bh) {
1217                 mlog_errno(-EIO);
1218                 return -EIO;
1219         }
1220         lock_buffer(*bh);
1221         if (!buffer_dirty(*bh))
1222                 clear_buffer_uptodate(*bh);
1223         unlock_buffer(*bh);
1224         ll_rw_block(READ, 1, bh);
1225         wait_on_buffer(*bh);
1226         return 0;
1227 }
1228
1229 static int ocfs2_mount_volume(struct super_block *sb)
1230 {
1231         int status = 0;
1232         int unlock_super = 0;
1233         struct ocfs2_super *osb = OCFS2_SB(sb);
1234
1235         mlog_entry_void();
1236
1237         if (ocfs2_is_hard_readonly(osb))
1238                 goto leave;
1239
1240         status = ocfs2_dlm_init(osb);
1241         if (status < 0) {
1242                 mlog_errno(status);
1243                 goto leave;
1244         }
1245
1246         status = ocfs2_super_lock(osb, 1);
1247         if (status < 0) {
1248                 mlog_errno(status);
1249                 goto leave;
1250         }
1251         unlock_super = 1;
1252
1253         /* This will load up the node map and add ourselves to it. */
1254         status = ocfs2_find_slot(osb);
1255         if (status < 0) {
1256                 mlog_errno(status);
1257                 goto leave;
1258         }
1259
1260         /* load all node-local system inodes */
1261         status = ocfs2_init_local_system_inodes(osb);
1262         if (status < 0) {
1263                 mlog_errno(status);
1264                 goto leave;
1265         }
1266
1267         status = ocfs2_check_volume(osb);
1268         if (status < 0) {
1269                 mlog_errno(status);
1270                 goto leave;
1271         }
1272
1273         status = ocfs2_truncate_log_init(osb);
1274         if (status < 0) {
1275                 mlog_errno(status);
1276                 goto leave;
1277         }
1278
1279         if (ocfs2_mount_local(osb))
1280                 goto leave;
1281
1282 leave:
1283         if (unlock_super)
1284                 ocfs2_super_unlock(osb, 1);
1285
1286         mlog_exit(status);
1287         return status;
1288 }
1289
1290 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1291 {
1292         int tmp, hangup_needed = 0;
1293         struct ocfs2_super *osb = NULL;
1294         char nodestr[8];
1295
1296         mlog_entry("(0x%p)\n", sb);
1297
1298         BUG_ON(!sb);
1299         osb = OCFS2_SB(sb);
1300         BUG_ON(!osb);
1301
1302         ocfs2_shutdown_local_alloc(osb);
1303
1304         ocfs2_truncate_log_shutdown(osb);
1305
1306         /* This will disable recovery and flush any recovery work. */
1307         ocfs2_recovery_exit(osb);
1308
1309         ocfs2_journal_shutdown(osb);
1310
1311         ocfs2_sync_blockdev(sb);
1312
1313         /* No cluster connection means we've failed during mount, so skip
1314          * all the steps which depended on that to complete. */
1315         if (osb->cconn) {
1316                 tmp = ocfs2_super_lock(osb, 1);
1317                 if (tmp < 0) {
1318                         mlog_errno(tmp);
1319                         return;
1320                 }
1321         }
1322
1323         if (osb->slot_num != OCFS2_INVALID_SLOT)
1324                 ocfs2_put_slot(osb);
1325
1326         if (osb->cconn)
1327                 ocfs2_super_unlock(osb, 1);
1328
1329         ocfs2_release_system_inodes(osb);
1330
1331         /*
1332          * If we're dismounting due to mount error, mount.ocfs2 will clean
1333          * up heartbeat.  If we're a local mount, there is no heartbeat.
1334          * If we failed before we got a uuid_str yet, we can't stop
1335          * heartbeat.  Otherwise, do it.
1336          */
1337         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1338                 hangup_needed = 1;
1339
1340         if (osb->cconn)
1341                 ocfs2_dlm_shutdown(osb, hangup_needed);
1342
1343         debugfs_remove(osb->osb_debug_root);
1344
1345         if (hangup_needed)
1346                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1347
1348         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1349
1350         if (ocfs2_mount_local(osb))
1351                 snprintf(nodestr, sizeof(nodestr), "local");
1352         else
1353                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1354
1355         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1356                osb->dev_str, nodestr);
1357
1358         ocfs2_delete_osb(osb);
1359         kfree(osb);
1360         sb->s_dev = 0;
1361         sb->s_fs_info = NULL;
1362 }
1363
1364 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1365                                 unsigned uuid_bytes)
1366 {
1367         int i, ret;
1368         char *ptr;
1369
1370         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1371
1372         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1373         if (osb->uuid_str == NULL)
1374                 return -ENOMEM;
1375
1376         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1377                 /* print with null */
1378                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1379                 if (ret != 2) /* drop super cleans up */
1380                         return -EINVAL;
1381                 /* then only advance past the last char */
1382                 ptr += 2;
1383         }
1384
1385         return 0;
1386 }
1387
1388 static int ocfs2_initialize_super(struct super_block *sb,
1389                                   struct buffer_head *bh,
1390                                   int sector_size)
1391 {
1392         int status;
1393         int i, cbits, bbits;
1394         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1395         struct inode *inode = NULL;
1396         struct ocfs2_journal *journal;
1397         __le32 uuid_net_key;
1398         struct ocfs2_super *osb;
1399
1400         mlog_entry_void();
1401
1402         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1403         if (!osb) {
1404                 status = -ENOMEM;
1405                 mlog_errno(status);
1406                 goto bail;
1407         }
1408
1409         sb->s_fs_info = osb;
1410         sb->s_op = &ocfs2_sops;
1411         sb->s_export_op = &ocfs2_export_ops;
1412         sb->s_xattr = ocfs2_xattr_handlers;
1413         sb->s_time_gran = 1;
1414         sb->s_flags |= MS_NOATIME;
1415         /* this is needed to support O_LARGEFILE */
1416         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1417         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1418         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1419
1420         osb->sb = sb;
1421         /* Save off for ocfs2_rw_direct */
1422         osb->s_sectsize_bits = blksize_bits(sector_size);
1423         BUG_ON(!osb->s_sectsize_bits);
1424
1425         spin_lock_init(&osb->dc_task_lock);
1426         init_waitqueue_head(&osb->dc_event);
1427         osb->dc_work_sequence = 0;
1428         osb->dc_wake_sequence = 0;
1429         INIT_LIST_HEAD(&osb->blocked_lock_list);
1430         osb->blocked_lock_count = 0;
1431         spin_lock_init(&osb->osb_lock);
1432         ocfs2_init_inode_steal_slot(osb);
1433
1434         atomic_set(&osb->alloc_stats.moves, 0);
1435         atomic_set(&osb->alloc_stats.local_data, 0);
1436         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1437         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1438         atomic_set(&osb->alloc_stats.bg_extends, 0);
1439
1440         ocfs2_init_node_maps(osb);
1441
1442         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1443                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1444
1445         status = ocfs2_recovery_init(osb);
1446         if (status) {
1447                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1448                 mlog_errno(status);
1449                 goto bail;
1450         }
1451
1452         init_waitqueue_head(&osb->checkpoint_event);
1453         atomic_set(&osb->needs_checkpoint, 0);
1454
1455         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1456
1457         osb->slot_num = OCFS2_INVALID_SLOT;
1458
1459         osb->s_xattr_inline_size = le16_to_cpu(
1460                                         di->id2.i_super.s_xattr_inline_size);
1461
1462         osb->local_alloc_state = OCFS2_LA_UNUSED;
1463         osb->local_alloc_bh = NULL;
1464         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
1465
1466         init_waitqueue_head(&osb->osb_mount_event);
1467
1468         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1469         if (!osb->vol_label) {
1470                 mlog(ML_ERROR, "unable to alloc vol label\n");
1471                 status = -ENOMEM;
1472                 goto bail;
1473         }
1474
1475         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1476         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1477                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1478                      osb->max_slots);
1479                 status = -EINVAL;
1480                 goto bail;
1481         }
1482         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1483
1484         osb->slot_recovery_generations =
1485                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
1486                         GFP_KERNEL);
1487         if (!osb->slot_recovery_generations) {
1488                 status = -ENOMEM;
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492
1493         init_waitqueue_head(&osb->osb_wipe_event);
1494         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1495                                         sizeof(*osb->osb_orphan_wipes),
1496                                         GFP_KERNEL);
1497         if (!osb->osb_orphan_wipes) {
1498                 status = -ENOMEM;
1499                 mlog_errno(status);
1500                 goto bail;
1501         }
1502
1503         osb->s_feature_compat =
1504                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1505         osb->s_feature_ro_compat =
1506                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1507         osb->s_feature_incompat =
1508                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1509
1510         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1511                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1512                      "optional features (%x).\n", i);
1513                 status = -EINVAL;
1514                 goto bail;
1515         }
1516         if (!(osb->sb->s_flags & MS_RDONLY) &&
1517             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1518                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1519                      "unsupported optional features (%x).\n", i);
1520                 status = -EINVAL;
1521                 goto bail;
1522         }
1523
1524         if (ocfs2_userspace_stack(osb)) {
1525                 memcpy(osb->osb_cluster_stack,
1526                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
1527                        OCFS2_STACK_LABEL_LEN);
1528                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1529                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
1530                         mlog(ML_ERROR,
1531                              "couldn't mount because of an invalid "
1532                              "cluster stack label (%s) \n",
1533                              osb->osb_cluster_stack);
1534                         status = -EINVAL;
1535                         goto bail;
1536                 }
1537         } else {
1538                 /* The empty string is identical with classic tools that
1539                  * don't know about s_cluster_info. */
1540                 osb->osb_cluster_stack[0] = '\0';
1541         }
1542
1543         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1544
1545         /* FIXME
1546          * This should be done in ocfs2_journal_init(), but unknown
1547          * ordering issues will cause the filesystem to crash.
1548          * If anyone wants to figure out what part of the code
1549          * refers to osb->journal before ocfs2_journal_init() is run,
1550          * be my guest.
1551          */
1552         /* initialize our journal structure */
1553
1554         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1555         if (!journal) {
1556                 mlog(ML_ERROR, "unable to alloc journal\n");
1557                 status = -ENOMEM;
1558                 goto bail;
1559         }
1560         osb->journal = journal;
1561         journal->j_osb = osb;
1562
1563         atomic_set(&journal->j_num_trans, 0);
1564         init_rwsem(&journal->j_trans_barrier);
1565         init_waitqueue_head(&journal->j_checkpointed);
1566         spin_lock_init(&journal->j_lock);
1567         journal->j_trans_id = (unsigned long) 1;
1568         INIT_LIST_HEAD(&journal->j_la_cleanups);
1569         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1570         journal->j_state = OCFS2_JOURNAL_FREE;
1571
1572         /* get some pseudo constants for clustersize bits */
1573         osb->s_clustersize_bits =
1574                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1575         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1576         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1577
1578         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1579             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1580                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1581                      osb->s_clustersize);
1582                 status = -EINVAL;
1583                 goto bail;
1584         }
1585
1586         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1587             > (u32)~0UL) {
1588                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1589                      "what jbd can address in 32 bits.\n");
1590                 status = -EINVAL;
1591                 goto bail;
1592         }
1593
1594         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1595                                  sizeof(di->id2.i_super.s_uuid))) {
1596                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1597                 status = -ENOMEM;
1598                 goto bail;
1599         }
1600
1601         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1602
1603         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1604         osb->vol_label[63] = '\0';
1605         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1606         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1607         osb->first_cluster_group_blkno =
1608                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1609         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1610         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
1611         mlog(0, "vol_label: %s\n", osb->vol_label);
1612         mlog(0, "uuid: %s\n", osb->uuid_str);
1613         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1614              (unsigned long long)osb->root_blkno,
1615              (unsigned long long)osb->system_dir_blkno);
1616
1617         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1618         if (!osb->osb_dlm_debug) {
1619                 status = -ENOMEM;
1620                 mlog_errno(status);
1621                 goto bail;
1622         }
1623
1624         atomic_set(&osb->vol_state, VOLUME_INIT);
1625
1626         /* load root, system_dir, and all global system inodes */
1627         status = ocfs2_init_global_system_inodes(osb);
1628         if (status < 0) {
1629                 mlog_errno(status);
1630                 goto bail;
1631         }
1632
1633         /*
1634          * global bitmap
1635          */
1636         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1637                                             OCFS2_INVALID_SLOT);
1638         if (!inode) {
1639                 status = -EINVAL;
1640                 mlog_errno(status);
1641                 goto bail;
1642         }
1643
1644         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1645         iput(inode);
1646
1647         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1648
1649         status = ocfs2_init_slot_info(osb);
1650         if (status < 0) {
1651                 mlog_errno(status);
1652                 goto bail;
1653         }
1654
1655 bail:
1656         mlog_exit(status);
1657         return status;
1658 }
1659
1660 /*
1661  * will return: -EAGAIN if it is ok to keep searching for superblocks
1662  *              -EINVAL if there is a bad superblock
1663  *              0 on success
1664  */
1665 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1666                                struct buffer_head *bh,
1667                                u32 blksz)
1668 {
1669         int status = -EAGAIN;
1670
1671         mlog_entry_void();
1672
1673         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1674                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1675                 status = -EINVAL;
1676                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1677                         mlog(ML_ERROR, "found superblock with incorrect block "
1678                              "size: found %u, should be %u\n",
1679                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1680                                blksz);
1681                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1682                            OCFS2_MAJOR_REV_LEVEL ||
1683                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1684                            OCFS2_MINOR_REV_LEVEL) {
1685                         mlog(ML_ERROR, "found superblock with bad version: "
1686                              "found %u.%u, should be %u.%u\n",
1687                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1688                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1689                              OCFS2_MAJOR_REV_LEVEL,
1690                              OCFS2_MINOR_REV_LEVEL);
1691                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1692                         mlog(ML_ERROR, "bad block number on superblock: "
1693                              "found %llu, should be %llu\n",
1694                              (unsigned long long)le64_to_cpu(di->i_blkno),
1695                              (unsigned long long)bh->b_blocknr);
1696                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1697                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1698                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1699                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1700                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1701                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1702                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1703                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1704                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1705                         mlog(ML_ERROR,
1706                              "Superblock slots found greater than file system "
1707                              "maximum: found %u, max %u\n",
1708                              le16_to_cpu(di->id2.i_super.s_max_slots),
1709                              OCFS2_MAX_SLOTS);
1710                 } else {
1711                         /* found it! */
1712                         status = 0;
1713                 }
1714         }
1715
1716         mlog_exit(status);
1717         return status;
1718 }
1719
1720 static int ocfs2_check_volume(struct ocfs2_super *osb)
1721 {
1722         int status;
1723         int dirty;
1724         int local;
1725         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1726                                                   * recover
1727                                                   * ourselves. */
1728
1729         mlog_entry_void();
1730
1731         /* Init our journal object. */
1732         status = ocfs2_journal_init(osb->journal, &dirty);
1733         if (status < 0) {
1734                 mlog(ML_ERROR, "Could not initialize journal!\n");
1735                 goto finally;
1736         }
1737
1738         /* If the journal was unmounted cleanly then we don't want to
1739          * recover anything. Otherwise, journal_load will do that
1740          * dirty work for us :) */
1741         if (!dirty) {
1742                 status = ocfs2_journal_wipe(osb->journal, 0);
1743                 if (status < 0) {
1744                         mlog_errno(status);
1745                         goto finally;
1746                 }
1747         } else {
1748                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1749                      "recovering volume.\n");
1750         }
1751
1752         local = ocfs2_mount_local(osb);
1753
1754         /* will play back anything left in the journal. */
1755         status = ocfs2_journal_load(osb->journal, local, dirty);
1756         if (status < 0) {
1757                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
1758                 goto finally;
1759         }
1760
1761         if (dirty) {
1762                 /* recover my local alloc if we didn't unmount cleanly. */
1763                 status = ocfs2_begin_local_alloc_recovery(osb,
1764                                                           osb->slot_num,
1765                                                           &local_alloc);
1766                 if (status < 0) {
1767                         mlog_errno(status);
1768                         goto finally;
1769                 }
1770                 /* we complete the recovery process after we've marked
1771                  * ourselves as mounted. */
1772         }
1773
1774         mlog(0, "Journal loaded.\n");
1775
1776         status = ocfs2_load_local_alloc(osb);
1777         if (status < 0) {
1778                 mlog_errno(status);
1779                 goto finally;
1780         }
1781
1782         if (dirty) {
1783                 /* Recovery will be completed after we've mounted the
1784                  * rest of the volume. */
1785                 osb->dirty = 1;
1786                 osb->local_alloc_copy = local_alloc;
1787                 local_alloc = NULL;
1788         }
1789
1790         /* go through each journal, trylock it and if you get the
1791          * lock, and it's marked as dirty, set the bit in the recover
1792          * map and launch a recovery thread for it. */
1793         status = ocfs2_mark_dead_nodes(osb);
1794         if (status < 0)
1795                 mlog_errno(status);
1796
1797 finally:
1798         if (local_alloc)
1799                 kfree(local_alloc);
1800
1801         mlog_exit(status);
1802         return status;
1803 }
1804
1805 /*
1806  * The routine gets called from dismount or close whenever a dismount on
1807  * volume is requested and the osb open count becomes 1.
1808  * It will remove the osb from the global list and also free up all the
1809  * initialized resources and fileobject.
1810  */
1811 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1812 {
1813         mlog_entry_void();
1814
1815         /* This function assumes that the caller has the main osb resource */
1816
1817         ocfs2_free_slot_info(osb);
1818
1819         kfree(osb->osb_orphan_wipes);
1820         kfree(osb->slot_recovery_generations);
1821         /* FIXME
1822          * This belongs in journal shutdown, but because we have to
1823          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1824          * we free it here.
1825          */
1826         kfree(osb->journal);
1827         if (osb->local_alloc_copy)
1828                 kfree(osb->local_alloc_copy);
1829         kfree(osb->uuid_str);
1830         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1831         memset(osb, 0, sizeof(struct ocfs2_super));
1832
1833         mlog_exit_void();
1834 }
1835
1836 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1837  * panic(). We do not support continue-on-error operation. */
1838 static void ocfs2_handle_error(struct super_block *sb)
1839 {
1840         struct ocfs2_super *osb = OCFS2_SB(sb);
1841
1842         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1843                 panic("OCFS2: (device %s): panic forced after error\n",
1844                       sb->s_id);
1845
1846         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1847
1848         if (sb->s_flags & MS_RDONLY &&
1849             (ocfs2_is_soft_readonly(osb) ||
1850              ocfs2_is_hard_readonly(osb)))
1851                 return;
1852
1853         printk(KERN_CRIT "File system is now read-only due to the potential "
1854                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1855                "system is unmounted.\n");
1856         sb->s_flags |= MS_RDONLY;
1857         ocfs2_set_ro_flag(osb, 0);
1858 }
1859
1860 static char error_buf[1024];
1861
1862 void __ocfs2_error(struct super_block *sb,
1863                    const char *function,
1864                    const char *fmt, ...)
1865 {
1866         va_list args;
1867
1868         va_start(args, fmt);
1869         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1870         va_end(args);
1871
1872         /* Not using mlog here because we want to show the actual
1873          * function the error came from. */
1874         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1875                sb->s_id, function, error_buf);
1876
1877         ocfs2_handle_error(sb);
1878 }
1879
1880 /* Handle critical errors. This is intentionally more drastic than
1881  * ocfs2_handle_error, so we only use for things like journal errors,
1882  * etc. */
1883 void __ocfs2_abort(struct super_block* sb,
1884                    const char *function,
1885                    const char *fmt, ...)
1886 {
1887         va_list args;
1888
1889         va_start(args, fmt);
1890         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1891         va_end(args);
1892
1893         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1894                sb->s_id, function, error_buf);
1895
1896         /* We don't have the cluster support yet to go straight to
1897          * hard readonly in here. Until then, we want to keep
1898          * ocfs2_abort() so that we can at least mark critical
1899          * errors.
1900          *
1901          * TODO: This should abort the journal and alert other nodes
1902          * that our slot needs recovery. */
1903
1904         /* Force a panic(). This stinks, but it's better than letting
1905          * things continue without having a proper hard readonly
1906          * here. */
1907         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1908         ocfs2_handle_error(sb);
1909 }
1910
1911 module_init(ocfs2_init);
1912 module_exit(ocfs2_exit);