Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile
[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         brelse(bh);
766
767         if (inode)
768                 iput(inode);
769
770         if (osb) {
771                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
772                 wake_up(&osb->osb_mount_event);
773                 ocfs2_dismount_volume(sb, 1);
774         }
775
776         mlog_exit(status);
777         return status;
778 }
779
780 static int ocfs2_get_sb(struct file_system_type *fs_type,
781                         int flags,
782                         const char *dev_name,
783                         void *data,
784                         struct vfsmount *mnt)
785 {
786         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
787                            mnt);
788 }
789
790 static struct file_system_type ocfs2_fs_type = {
791         .owner          = THIS_MODULE,
792         .name           = "ocfs2",
793         .get_sb         = ocfs2_get_sb, /* is this called when we mount
794                                         * the fs? */
795         .kill_sb        = kill_block_super, /* set to the generic one
796                                              * right now, but do we
797                                              * need to change that? */
798         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
799         .next           = NULL
800 };
801
802 static int ocfs2_parse_options(struct super_block *sb,
803                                char *options,
804                                struct mount_options *mopt,
805                                int is_remount)
806 {
807         int status;
808         char *p;
809
810         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
811                    options ? options : "(none)");
812
813         mopt->commit_interval = 0;
814         mopt->mount_opt = 0;
815         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
816         mopt->slot = OCFS2_INVALID_SLOT;
817         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
818         mopt->cluster_stack[0] = '\0';
819
820         if (!options) {
821                 status = 1;
822                 goto bail;
823         }
824
825         while ((p = strsep(&options, ",")) != NULL) {
826                 int token, option;
827                 substring_t args[MAX_OPT_ARGS];
828
829                 if (!*p)
830                         continue;
831
832                 token = match_token(p, tokens, args);
833                 switch (token) {
834                 case Opt_hb_local:
835                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
836                         break;
837                 case Opt_hb_none:
838                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
839                         break;
840                 case Opt_barrier:
841                         if (match_int(&args[0], &option)) {
842                                 status = 0;
843                                 goto bail;
844                         }
845                         if (option)
846                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
847                         else
848                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
849                         break;
850                 case Opt_intr:
851                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
852                         break;
853                 case Opt_nointr:
854                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
855                         break;
856                 case Opt_err_panic:
857                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
858                         break;
859                 case Opt_err_ro:
860                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
861                         break;
862                 case Opt_data_ordered:
863                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
864                         break;
865                 case Opt_data_writeback:
866                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
867                         break;
868                 case Opt_user_xattr:
869                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
870                         break;
871                 case Opt_nouser_xattr:
872                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
873                         break;
874                 case Opt_atime_quantum:
875                         if (match_int(&args[0], &option)) {
876                                 status = 0;
877                                 goto bail;
878                         }
879                         if (option >= 0)
880                                 mopt->atime_quantum = option;
881                         break;
882                 case Opt_slot:
883                         option = 0;
884                         if (match_int(&args[0], &option)) {
885                                 status = 0;
886                                 goto bail;
887                         }
888                         if (option)
889                                 mopt->slot = (s16)option;
890                         break;
891                 case Opt_commit:
892                         option = 0;
893                         if (match_int(&args[0], &option)) {
894                                 status = 0;
895                                 goto bail;
896                         }
897                         if (option < 0)
898                                 return 0;
899                         if (option == 0)
900                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
901                         mopt->commit_interval = HZ * option;
902                         break;
903                 case Opt_localalloc:
904                         option = 0;
905                         if (match_int(&args[0], &option)) {
906                                 status = 0;
907                                 goto bail;
908                         }
909                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
910                                 mopt->localalloc_opt = option;
911                         break;
912                 case Opt_localflocks:
913                         /*
914                          * Changing this during remount could race
915                          * flock() requests, or "unbalance" existing
916                          * ones (e.g., a lock is taken in one mode but
917                          * dropped in the other). If users care enough
918                          * to flip locking modes during remount, we
919                          * could add a "local" flag to individual
920                          * flock structures for proper tracking of
921                          * state.
922                          */
923                         if (!is_remount)
924                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
925                         break;
926                 case Opt_stack:
927                         /* Check both that the option we were passed
928                          * is of the right length and that it is a proper
929                          * string of the right length.
930                          */
931                         if (((args[0].to - args[0].from) !=
932                              OCFS2_STACK_LABEL_LEN) ||
933                             (strnlen(args[0].from,
934                                      OCFS2_STACK_LABEL_LEN) !=
935                              OCFS2_STACK_LABEL_LEN)) {
936                                 mlog(ML_ERROR,
937                                      "Invalid cluster_stack option\n");
938                                 status = 0;
939                                 goto bail;
940                         }
941                         memcpy(mopt->cluster_stack, args[0].from,
942                                OCFS2_STACK_LABEL_LEN);
943                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
944                         break;
945                 case Opt_inode64:
946                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
947                         break;
948                 default:
949                         mlog(ML_ERROR,
950                              "Unrecognized mount option \"%s\" "
951                              "or missing value\n", p);
952                         status = 0;
953                         goto bail;
954                 }
955         }
956
957         status = 1;
958
959 bail:
960         mlog_exit(status);
961         return status;
962 }
963
964 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
965 {
966         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
967         unsigned long opts = osb->s_mount_opt;
968         unsigned int local_alloc_megs;
969
970         if (opts & OCFS2_MOUNT_HB_LOCAL)
971                 seq_printf(s, ",_netdev,heartbeat=local");
972         else
973                 seq_printf(s, ",heartbeat=none");
974
975         if (opts & OCFS2_MOUNT_NOINTR)
976                 seq_printf(s, ",nointr");
977
978         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
979                 seq_printf(s, ",data=writeback");
980         else
981                 seq_printf(s, ",data=ordered");
982
983         if (opts & OCFS2_MOUNT_BARRIER)
984                 seq_printf(s, ",barrier=1");
985
986         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
987                 seq_printf(s, ",errors=panic");
988         else
989                 seq_printf(s, ",errors=remount-ro");
990
991         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
992                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
993
994         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
995                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
996
997         if (osb->osb_commit_interval)
998                 seq_printf(s, ",commit=%u",
999                            (unsigned) (osb->osb_commit_interval / HZ));
1000
1001         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1002         if (local_alloc_megs != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
1003                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1004
1005         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1006                 seq_printf(s, ",localflocks,");
1007
1008         if (osb->osb_cluster_stack[0])
1009                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1010                            osb->osb_cluster_stack);
1011
1012         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1013                 seq_printf(s, ",nouser_xattr");
1014         else
1015                 seq_printf(s, ",user_xattr");
1016
1017         if (opts & OCFS2_MOUNT_INODE64)
1018                 seq_printf(s, ",inode64");
1019
1020         return 0;
1021 }
1022
1023 static int __init ocfs2_init(void)
1024 {
1025         int status;
1026
1027         mlog_entry_void();
1028
1029         ocfs2_print_version();
1030
1031         status = init_ocfs2_uptodate_cache();
1032         if (status < 0) {
1033                 mlog_errno(status);
1034                 goto leave;
1035         }
1036
1037         status = ocfs2_initialize_mem_caches();
1038         if (status < 0) {
1039                 mlog_errno(status);
1040                 goto leave;
1041         }
1042
1043         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1044         if (!ocfs2_wq) {
1045                 status = -ENOMEM;
1046                 goto leave;
1047         }
1048
1049         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1050         if (!ocfs2_debugfs_root) {
1051                 status = -EFAULT;
1052                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1053         }
1054
1055         ocfs2_set_locking_protocol();
1056
1057 leave:
1058         if (status < 0) {
1059                 ocfs2_free_mem_caches();
1060                 exit_ocfs2_uptodate_cache();
1061         }
1062
1063         mlog_exit(status);
1064
1065         if (status >= 0) {
1066                 return register_filesystem(&ocfs2_fs_type);
1067         } else
1068                 return -1;
1069 }
1070
1071 static void __exit ocfs2_exit(void)
1072 {
1073         mlog_entry_void();
1074
1075         if (ocfs2_wq) {
1076                 flush_workqueue(ocfs2_wq);
1077                 destroy_workqueue(ocfs2_wq);
1078         }
1079
1080         debugfs_remove(ocfs2_debugfs_root);
1081
1082         ocfs2_free_mem_caches();
1083
1084         unregister_filesystem(&ocfs2_fs_type);
1085
1086         exit_ocfs2_uptodate_cache();
1087
1088         mlog_exit_void();
1089 }
1090
1091 static void ocfs2_put_super(struct super_block *sb)
1092 {
1093         mlog_entry("(0x%p)\n", sb);
1094
1095         ocfs2_sync_blockdev(sb);
1096         ocfs2_dismount_volume(sb, 0);
1097
1098         mlog_exit_void();
1099 }
1100
1101 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1102 {
1103         struct ocfs2_super *osb;
1104         u32 numbits, freebits;
1105         int status;
1106         struct ocfs2_dinode *bm_lock;
1107         struct buffer_head *bh = NULL;
1108         struct inode *inode = NULL;
1109
1110         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1111
1112         osb = OCFS2_SB(dentry->d_sb);
1113
1114         inode = ocfs2_get_system_file_inode(osb,
1115                                             GLOBAL_BITMAP_SYSTEM_INODE,
1116                                             OCFS2_INVALID_SLOT);
1117         if (!inode) {
1118                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1119                 status = -EIO;
1120                 goto bail;
1121         }
1122
1123         status = ocfs2_inode_lock(inode, &bh, 0);
1124         if (status < 0) {
1125                 mlog_errno(status);
1126                 goto bail;
1127         }
1128
1129         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1130
1131         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1132         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1133
1134         buf->f_type = OCFS2_SUPER_MAGIC;
1135         buf->f_bsize = dentry->d_sb->s_blocksize;
1136         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1137         buf->f_blocks = ((sector_t) numbits) *
1138                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1139         buf->f_bfree = ((sector_t) freebits) *
1140                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1141         buf->f_bavail = buf->f_bfree;
1142         buf->f_files = numbits;
1143         buf->f_ffree = freebits;
1144
1145         brelse(bh);
1146
1147         ocfs2_inode_unlock(inode, 0);
1148         status = 0;
1149 bail:
1150         if (inode)
1151                 iput(inode);
1152
1153         mlog_exit(status);
1154
1155         return status;
1156 }
1157
1158 static void ocfs2_inode_init_once(void *data)
1159 {
1160         struct ocfs2_inode_info *oi = data;
1161
1162         oi->ip_flags = 0;
1163         oi->ip_open_count = 0;
1164         spin_lock_init(&oi->ip_lock);
1165         ocfs2_extent_map_init(&oi->vfs_inode);
1166         INIT_LIST_HEAD(&oi->ip_io_markers);
1167         oi->ip_created_trans = 0;
1168         oi->ip_last_trans = 0;
1169         oi->ip_dir_start_lookup = 0;
1170
1171         init_rwsem(&oi->ip_alloc_sem);
1172         init_rwsem(&oi->ip_xattr_sem);
1173         mutex_init(&oi->ip_io_mutex);
1174
1175         oi->ip_blkno = 0ULL;
1176         oi->ip_clusters = 0;
1177
1178         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1179         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1180         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1181
1182         ocfs2_metadata_cache_init(&oi->vfs_inode);
1183
1184         inode_init_once(&oi->vfs_inode);
1185 }
1186
1187 static int ocfs2_initialize_mem_caches(void)
1188 {
1189         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1190                                        sizeof(struct ocfs2_inode_info),
1191                                        0,
1192                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1193                                                 SLAB_MEM_SPREAD),
1194                                        ocfs2_inode_init_once);
1195         if (!ocfs2_inode_cachep)
1196                 return -ENOMEM;
1197
1198         return 0;
1199 }
1200
1201 static void ocfs2_free_mem_caches(void)
1202 {
1203         if (ocfs2_inode_cachep)
1204                 kmem_cache_destroy(ocfs2_inode_cachep);
1205
1206         ocfs2_inode_cachep = NULL;
1207 }
1208
1209 static int ocfs2_get_sector(struct super_block *sb,
1210                             struct buffer_head **bh,
1211                             int block,
1212                             int sect_size)
1213 {
1214         if (!sb_set_blocksize(sb, sect_size)) {
1215                 mlog(ML_ERROR, "unable to set blocksize\n");
1216                 return -EIO;
1217         }
1218
1219         *bh = sb_getblk(sb, block);
1220         if (!*bh) {
1221                 mlog_errno(-EIO);
1222                 return -EIO;
1223         }
1224         lock_buffer(*bh);
1225         if (!buffer_dirty(*bh))
1226                 clear_buffer_uptodate(*bh);
1227         unlock_buffer(*bh);
1228         ll_rw_block(READ, 1, bh);
1229         wait_on_buffer(*bh);
1230         return 0;
1231 }
1232
1233 static int ocfs2_mount_volume(struct super_block *sb)
1234 {
1235         int status = 0;
1236         int unlock_super = 0;
1237         struct ocfs2_super *osb = OCFS2_SB(sb);
1238
1239         mlog_entry_void();
1240
1241         if (ocfs2_is_hard_readonly(osb))
1242                 goto leave;
1243
1244         status = ocfs2_dlm_init(osb);
1245         if (status < 0) {
1246                 mlog_errno(status);
1247                 goto leave;
1248         }
1249
1250         status = ocfs2_super_lock(osb, 1);
1251         if (status < 0) {
1252                 mlog_errno(status);
1253                 goto leave;
1254         }
1255         unlock_super = 1;
1256
1257         /* This will load up the node map and add ourselves to it. */
1258         status = ocfs2_find_slot(osb);
1259         if (status < 0) {
1260                 mlog_errno(status);
1261                 goto leave;
1262         }
1263
1264         /* load all node-local system inodes */
1265         status = ocfs2_init_local_system_inodes(osb);
1266         if (status < 0) {
1267                 mlog_errno(status);
1268                 goto leave;
1269         }
1270
1271         status = ocfs2_check_volume(osb);
1272         if (status < 0) {
1273                 mlog_errno(status);
1274                 goto leave;
1275         }
1276
1277         status = ocfs2_truncate_log_init(osb);
1278         if (status < 0) {
1279                 mlog_errno(status);
1280                 goto leave;
1281         }
1282
1283         if (ocfs2_mount_local(osb))
1284                 goto leave;
1285
1286 leave:
1287         if (unlock_super)
1288                 ocfs2_super_unlock(osb, 1);
1289
1290         mlog_exit(status);
1291         return status;
1292 }
1293
1294 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1295 {
1296         int tmp, hangup_needed = 0;
1297         struct ocfs2_super *osb = NULL;
1298         char nodestr[8];
1299
1300         mlog_entry("(0x%p)\n", sb);
1301
1302         BUG_ON(!sb);
1303         osb = OCFS2_SB(sb);
1304         BUG_ON(!osb);
1305
1306         ocfs2_shutdown_local_alloc(osb);
1307
1308         ocfs2_truncate_log_shutdown(osb);
1309
1310         /* This will disable recovery and flush any recovery work. */
1311         ocfs2_recovery_exit(osb);
1312
1313         ocfs2_journal_shutdown(osb);
1314
1315         ocfs2_sync_blockdev(sb);
1316
1317         /* No cluster connection means we've failed during mount, so skip
1318          * all the steps which depended on that to complete. */
1319         if (osb->cconn) {
1320                 tmp = ocfs2_super_lock(osb, 1);
1321                 if (tmp < 0) {
1322                         mlog_errno(tmp);
1323                         return;
1324                 }
1325         }
1326
1327         if (osb->slot_num != OCFS2_INVALID_SLOT)
1328                 ocfs2_put_slot(osb);
1329
1330         if (osb->cconn)
1331                 ocfs2_super_unlock(osb, 1);
1332
1333         ocfs2_release_system_inodes(osb);
1334
1335         /*
1336          * If we're dismounting due to mount error, mount.ocfs2 will clean
1337          * up heartbeat.  If we're a local mount, there is no heartbeat.
1338          * If we failed before we got a uuid_str yet, we can't stop
1339          * heartbeat.  Otherwise, do it.
1340          */
1341         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1342                 hangup_needed = 1;
1343
1344         if (osb->cconn)
1345                 ocfs2_dlm_shutdown(osb, hangup_needed);
1346
1347         debugfs_remove(osb->osb_debug_root);
1348
1349         if (hangup_needed)
1350                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1351
1352         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1353
1354         if (ocfs2_mount_local(osb))
1355                 snprintf(nodestr, sizeof(nodestr), "local");
1356         else
1357                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1358
1359         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1360                osb->dev_str, nodestr);
1361
1362         ocfs2_delete_osb(osb);
1363         kfree(osb);
1364         sb->s_dev = 0;
1365         sb->s_fs_info = NULL;
1366 }
1367
1368 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1369                                 unsigned uuid_bytes)
1370 {
1371         int i, ret;
1372         char *ptr;
1373
1374         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1375
1376         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1377         if (osb->uuid_str == NULL)
1378                 return -ENOMEM;
1379
1380         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1381                 /* print with null */
1382                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1383                 if (ret != 2) /* drop super cleans up */
1384                         return -EINVAL;
1385                 /* then only advance past the last char */
1386                 ptr += 2;
1387         }
1388
1389         return 0;
1390 }
1391
1392 static int ocfs2_initialize_super(struct super_block *sb,
1393                                   struct buffer_head *bh,
1394                                   int sector_size)
1395 {
1396         int status;
1397         int i, cbits, bbits;
1398         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1399         struct inode *inode = NULL;
1400         struct ocfs2_journal *journal;
1401         __le32 uuid_net_key;
1402         struct ocfs2_super *osb;
1403
1404         mlog_entry_void();
1405
1406         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1407         if (!osb) {
1408                 status = -ENOMEM;
1409                 mlog_errno(status);
1410                 goto bail;
1411         }
1412
1413         sb->s_fs_info = osb;
1414         sb->s_op = &ocfs2_sops;
1415         sb->s_export_op = &ocfs2_export_ops;
1416         sb->s_xattr = ocfs2_xattr_handlers;
1417         sb->s_time_gran = 1;
1418         sb->s_flags |= MS_NOATIME;
1419         /* this is needed to support O_LARGEFILE */
1420         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1421         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1422         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1423
1424         osb->sb = sb;
1425         /* Save off for ocfs2_rw_direct */
1426         osb->s_sectsize_bits = blksize_bits(sector_size);
1427         BUG_ON(!osb->s_sectsize_bits);
1428
1429         spin_lock_init(&osb->dc_task_lock);
1430         init_waitqueue_head(&osb->dc_event);
1431         osb->dc_work_sequence = 0;
1432         osb->dc_wake_sequence = 0;
1433         INIT_LIST_HEAD(&osb->blocked_lock_list);
1434         osb->blocked_lock_count = 0;
1435         spin_lock_init(&osb->osb_lock);
1436         ocfs2_init_inode_steal_slot(osb);
1437
1438         atomic_set(&osb->alloc_stats.moves, 0);
1439         atomic_set(&osb->alloc_stats.local_data, 0);
1440         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1441         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1442         atomic_set(&osb->alloc_stats.bg_extends, 0);
1443
1444         ocfs2_init_node_maps(osb);
1445
1446         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1447                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1448
1449         status = ocfs2_recovery_init(osb);
1450         if (status) {
1451                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1452                 mlog_errno(status);
1453                 goto bail;
1454         }
1455
1456         init_waitqueue_head(&osb->checkpoint_event);
1457         atomic_set(&osb->needs_checkpoint, 0);
1458
1459         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1460
1461         osb->slot_num = OCFS2_INVALID_SLOT;
1462
1463         osb->s_xattr_inline_size = le16_to_cpu(
1464                                         di->id2.i_super.s_xattr_inline_size);
1465
1466         osb->local_alloc_state = OCFS2_LA_UNUSED;
1467         osb->local_alloc_bh = NULL;
1468         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
1469
1470         init_waitqueue_head(&osb->osb_mount_event);
1471
1472         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1473         if (!osb->vol_label) {
1474                 mlog(ML_ERROR, "unable to alloc vol label\n");
1475                 status = -ENOMEM;
1476                 goto bail;
1477         }
1478
1479         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1480         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1481                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1482                      osb->max_slots);
1483                 status = -EINVAL;
1484                 goto bail;
1485         }
1486         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1487
1488         osb->slot_recovery_generations =
1489                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
1490                         GFP_KERNEL);
1491         if (!osb->slot_recovery_generations) {
1492                 status = -ENOMEM;
1493                 mlog_errno(status);
1494                 goto bail;
1495         }
1496
1497         init_waitqueue_head(&osb->osb_wipe_event);
1498         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1499                                         sizeof(*osb->osb_orphan_wipes),
1500                                         GFP_KERNEL);
1501         if (!osb->osb_orphan_wipes) {
1502                 status = -ENOMEM;
1503                 mlog_errno(status);
1504                 goto bail;
1505         }
1506
1507         osb->s_feature_compat =
1508                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1509         osb->s_feature_ro_compat =
1510                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1511         osb->s_feature_incompat =
1512                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1513
1514         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1515                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1516                      "optional features (%x).\n", i);
1517                 status = -EINVAL;
1518                 goto bail;
1519         }
1520         if (!(osb->sb->s_flags & MS_RDONLY) &&
1521             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1522                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1523                      "unsupported optional features (%x).\n", i);
1524                 status = -EINVAL;
1525                 goto bail;
1526         }
1527
1528         if (ocfs2_userspace_stack(osb)) {
1529                 memcpy(osb->osb_cluster_stack,
1530                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
1531                        OCFS2_STACK_LABEL_LEN);
1532                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1533                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
1534                         mlog(ML_ERROR,
1535                              "couldn't mount because of an invalid "
1536                              "cluster stack label (%s) \n",
1537                              osb->osb_cluster_stack);
1538                         status = -EINVAL;
1539                         goto bail;
1540                 }
1541         } else {
1542                 /* The empty string is identical with classic tools that
1543                  * don't know about s_cluster_info. */
1544                 osb->osb_cluster_stack[0] = '\0';
1545         }
1546
1547         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1548
1549         /* FIXME
1550          * This should be done in ocfs2_journal_init(), but unknown
1551          * ordering issues will cause the filesystem to crash.
1552          * If anyone wants to figure out what part of the code
1553          * refers to osb->journal before ocfs2_journal_init() is run,
1554          * be my guest.
1555          */
1556         /* initialize our journal structure */
1557
1558         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1559         if (!journal) {
1560                 mlog(ML_ERROR, "unable to alloc journal\n");
1561                 status = -ENOMEM;
1562                 goto bail;
1563         }
1564         osb->journal = journal;
1565         journal->j_osb = osb;
1566
1567         atomic_set(&journal->j_num_trans, 0);
1568         init_rwsem(&journal->j_trans_barrier);
1569         init_waitqueue_head(&journal->j_checkpointed);
1570         spin_lock_init(&journal->j_lock);
1571         journal->j_trans_id = (unsigned long) 1;
1572         INIT_LIST_HEAD(&journal->j_la_cleanups);
1573         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1574         journal->j_state = OCFS2_JOURNAL_FREE;
1575
1576         /* get some pseudo constants for clustersize bits */
1577         osb->s_clustersize_bits =
1578                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1579         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1580         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1581
1582         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1583             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1584                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1585                      osb->s_clustersize);
1586                 status = -EINVAL;
1587                 goto bail;
1588         }
1589
1590         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1591             > (u32)~0UL) {
1592                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1593                      "what jbd can address in 32 bits.\n");
1594                 status = -EINVAL;
1595                 goto bail;
1596         }
1597
1598         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1599                                  sizeof(di->id2.i_super.s_uuid))) {
1600                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1601                 status = -ENOMEM;
1602                 goto bail;
1603         }
1604
1605         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1606
1607         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1608         osb->vol_label[63] = '\0';
1609         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1610         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1611         osb->first_cluster_group_blkno =
1612                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1613         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1614         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
1615         mlog(0, "vol_label: %s\n", osb->vol_label);
1616         mlog(0, "uuid: %s\n", osb->uuid_str);
1617         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1618              (unsigned long long)osb->root_blkno,
1619              (unsigned long long)osb->system_dir_blkno);
1620
1621         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1622         if (!osb->osb_dlm_debug) {
1623                 status = -ENOMEM;
1624                 mlog_errno(status);
1625                 goto bail;
1626         }
1627
1628         atomic_set(&osb->vol_state, VOLUME_INIT);
1629
1630         /* load root, system_dir, and all global system inodes */
1631         status = ocfs2_init_global_system_inodes(osb);
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto bail;
1635         }
1636
1637         /*
1638          * global bitmap
1639          */
1640         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1641                                             OCFS2_INVALID_SLOT);
1642         if (!inode) {
1643                 status = -EINVAL;
1644                 mlog_errno(status);
1645                 goto bail;
1646         }
1647
1648         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1649         iput(inode);
1650
1651         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1652
1653         status = ocfs2_init_slot_info(osb);
1654         if (status < 0) {
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659 bail:
1660         mlog_exit(status);
1661         return status;
1662 }
1663
1664 /*
1665  * will return: -EAGAIN if it is ok to keep searching for superblocks
1666  *              -EINVAL if there is a bad superblock
1667  *              0 on success
1668  */
1669 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1670                                struct buffer_head *bh,
1671                                u32 blksz)
1672 {
1673         int status = -EAGAIN;
1674
1675         mlog_entry_void();
1676
1677         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1678                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1679                 status = -EINVAL;
1680                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1681                         mlog(ML_ERROR, "found superblock with incorrect block "
1682                              "size: found %u, should be %u\n",
1683                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1684                                blksz);
1685                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1686                            OCFS2_MAJOR_REV_LEVEL ||
1687                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1688                            OCFS2_MINOR_REV_LEVEL) {
1689                         mlog(ML_ERROR, "found superblock with bad version: "
1690                              "found %u.%u, should be %u.%u\n",
1691                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1692                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1693                              OCFS2_MAJOR_REV_LEVEL,
1694                              OCFS2_MINOR_REV_LEVEL);
1695                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1696                         mlog(ML_ERROR, "bad block number on superblock: "
1697                              "found %llu, should be %llu\n",
1698                              (unsigned long long)le64_to_cpu(di->i_blkno),
1699                              (unsigned long long)bh->b_blocknr);
1700                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1701                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1702                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1703                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1704                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1705                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1706                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1707                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1708                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1709                         mlog(ML_ERROR,
1710                              "Superblock slots found greater than file system "
1711                              "maximum: found %u, max %u\n",
1712                              le16_to_cpu(di->id2.i_super.s_max_slots),
1713                              OCFS2_MAX_SLOTS);
1714                 } else {
1715                         /* found it! */
1716                         status = 0;
1717                 }
1718         }
1719
1720         mlog_exit(status);
1721         return status;
1722 }
1723
1724 static int ocfs2_check_volume(struct ocfs2_super *osb)
1725 {
1726         int status;
1727         int dirty;
1728         int local;
1729         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1730                                                   * recover
1731                                                   * ourselves. */
1732
1733         mlog_entry_void();
1734
1735         /* Init our journal object. */
1736         status = ocfs2_journal_init(osb->journal, &dirty);
1737         if (status < 0) {
1738                 mlog(ML_ERROR, "Could not initialize journal!\n");
1739                 goto finally;
1740         }
1741
1742         /* If the journal was unmounted cleanly then we don't want to
1743          * recover anything. Otherwise, journal_load will do that
1744          * dirty work for us :) */
1745         if (!dirty) {
1746                 status = ocfs2_journal_wipe(osb->journal, 0);
1747                 if (status < 0) {
1748                         mlog_errno(status);
1749                         goto finally;
1750                 }
1751         } else {
1752                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1753                      "recovering volume.\n");
1754         }
1755
1756         local = ocfs2_mount_local(osb);
1757
1758         /* will play back anything left in the journal. */
1759         status = ocfs2_journal_load(osb->journal, local, dirty);
1760         if (status < 0) {
1761                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
1762                 goto finally;
1763         }
1764
1765         if (dirty) {
1766                 /* recover my local alloc if we didn't unmount cleanly. */
1767                 status = ocfs2_begin_local_alloc_recovery(osb,
1768                                                           osb->slot_num,
1769                                                           &local_alloc);
1770                 if (status < 0) {
1771                         mlog_errno(status);
1772                         goto finally;
1773                 }
1774                 /* we complete the recovery process after we've marked
1775                  * ourselves as mounted. */
1776         }
1777
1778         mlog(0, "Journal loaded.\n");
1779
1780         status = ocfs2_load_local_alloc(osb);
1781         if (status < 0) {
1782                 mlog_errno(status);
1783                 goto finally;
1784         }
1785
1786         if (dirty) {
1787                 /* Recovery will be completed after we've mounted the
1788                  * rest of the volume. */
1789                 osb->dirty = 1;
1790                 osb->local_alloc_copy = local_alloc;
1791                 local_alloc = NULL;
1792         }
1793
1794         /* go through each journal, trylock it and if you get the
1795          * lock, and it's marked as dirty, set the bit in the recover
1796          * map and launch a recovery thread for it. */
1797         status = ocfs2_mark_dead_nodes(osb);
1798         if (status < 0)
1799                 mlog_errno(status);
1800
1801 finally:
1802         if (local_alloc)
1803                 kfree(local_alloc);
1804
1805         mlog_exit(status);
1806         return status;
1807 }
1808
1809 /*
1810  * The routine gets called from dismount or close whenever a dismount on
1811  * volume is requested and the osb open count becomes 1.
1812  * It will remove the osb from the global list and also free up all the
1813  * initialized resources and fileobject.
1814  */
1815 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1816 {
1817         mlog_entry_void();
1818
1819         /* This function assumes that the caller has the main osb resource */
1820
1821         ocfs2_free_slot_info(osb);
1822
1823         kfree(osb->osb_orphan_wipes);
1824         kfree(osb->slot_recovery_generations);
1825         /* FIXME
1826          * This belongs in journal shutdown, but because we have to
1827          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1828          * we free it here.
1829          */
1830         kfree(osb->journal);
1831         if (osb->local_alloc_copy)
1832                 kfree(osb->local_alloc_copy);
1833         kfree(osb->uuid_str);
1834         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1835         memset(osb, 0, sizeof(struct ocfs2_super));
1836
1837         mlog_exit_void();
1838 }
1839
1840 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1841  * panic(). We do not support continue-on-error operation. */
1842 static void ocfs2_handle_error(struct super_block *sb)
1843 {
1844         struct ocfs2_super *osb = OCFS2_SB(sb);
1845
1846         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1847                 panic("OCFS2: (device %s): panic forced after error\n",
1848                       sb->s_id);
1849
1850         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1851
1852         if (sb->s_flags & MS_RDONLY &&
1853             (ocfs2_is_soft_readonly(osb) ||
1854              ocfs2_is_hard_readonly(osb)))
1855                 return;
1856
1857         printk(KERN_CRIT "File system is now read-only due to the potential "
1858                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1859                "system is unmounted.\n");
1860         sb->s_flags |= MS_RDONLY;
1861         ocfs2_set_ro_flag(osb, 0);
1862 }
1863
1864 static char error_buf[1024];
1865
1866 void __ocfs2_error(struct super_block *sb,
1867                    const char *function,
1868                    const char *fmt, ...)
1869 {
1870         va_list args;
1871
1872         va_start(args, fmt);
1873         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1874         va_end(args);
1875
1876         /* Not using mlog here because we want to show the actual
1877          * function the error came from. */
1878         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1879                sb->s_id, function, error_buf);
1880
1881         ocfs2_handle_error(sb);
1882 }
1883
1884 /* Handle critical errors. This is intentionally more drastic than
1885  * ocfs2_handle_error, so we only use for things like journal errors,
1886  * etc. */
1887 void __ocfs2_abort(struct super_block* sb,
1888                    const char *function,
1889                    const char *fmt, ...)
1890 {
1891         va_list args;
1892
1893         va_start(args, fmt);
1894         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1895         va_end(args);
1896
1897         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1898                sb->s_id, function, error_buf);
1899
1900         /* We don't have the cluster support yet to go straight to
1901          * hard readonly in here. Until then, we want to keep
1902          * ocfs2_abort() so that we can at least mark critical
1903          * errors.
1904          *
1905          * TODO: This should abort the journal and alert other nodes
1906          * that our slot needs recovery. */
1907
1908         /* Force a panic(). This stinks, but it's better than letting
1909          * things continue without having a proper hard readonly
1910          * here. */
1911         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1912         ocfs2_handle_error(sb);
1913 }
1914
1915 module_init(ocfs2_init);
1916 module_exit(ocfs2_exit);