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