]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/ocfs2/super.c
ocfs2: Move o2hb functionality into the stack glue.
[karo-tx-linux.git] / 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 };
91
92 static int ocfs2_parse_options(struct super_block *sb, char *options,
93                                struct mount_options *mopt,
94                                int is_remount);
95 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
96 static void ocfs2_put_super(struct super_block *sb);
97 static int ocfs2_mount_volume(struct super_block *sb);
98 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
99 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
100 static int ocfs2_initialize_mem_caches(void);
101 static void ocfs2_free_mem_caches(void);
102 static void ocfs2_delete_osb(struct ocfs2_super *osb);
103
104 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
105
106 static int ocfs2_sync_fs(struct super_block *sb, int wait);
107
108 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
109 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
110 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
111 static int ocfs2_fill_local_node_info(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_err,
157 };
158
159 static match_table_t tokens = {
160         {Opt_barrier, "barrier=%u"},
161         {Opt_err_panic, "errors=panic"},
162         {Opt_err_ro, "errors=remount-ro"},
163         {Opt_intr, "intr"},
164         {Opt_nointr, "nointr"},
165         {Opt_hb_none, OCFS2_HB_NONE},
166         {Opt_hb_local, OCFS2_HB_LOCAL},
167         {Opt_data_ordered, "data=ordered"},
168         {Opt_data_writeback, "data=writeback"},
169         {Opt_atime_quantum, "atime_quantum=%u"},
170         {Opt_slot, "preferred_slot=%u"},
171         {Opt_commit, "commit=%u"},
172         {Opt_localalloc, "localalloc=%d"},
173         {Opt_localflocks, "localflocks"},
174         {Opt_err, NULL}
175 };
176
177 /*
178  * write_super and sync_fs ripped right out of ext3.
179  */
180 static void ocfs2_write_super(struct super_block *sb)
181 {
182         if (mutex_trylock(&sb->s_lock) != 0)
183                 BUG();
184         sb->s_dirt = 0;
185 }
186
187 static int ocfs2_sync_fs(struct super_block *sb, int wait)
188 {
189         int status;
190         tid_t target;
191         struct ocfs2_super *osb = OCFS2_SB(sb);
192
193         sb->s_dirt = 0;
194
195         if (ocfs2_is_hard_readonly(osb))
196                 return -EROFS;
197
198         if (wait) {
199                 status = ocfs2_flush_truncate_log(osb);
200                 if (status < 0)
201                         mlog_errno(status);
202         } else {
203                 ocfs2_schedule_truncate_log_flush(osb, 0);
204         }
205
206         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
207                 if (wait)
208                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
209                                         target);
210         }
211         return 0;
212 }
213
214 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
215 {
216         struct inode *new = NULL;
217         int status = 0;
218         int i;
219
220         mlog_entry_void();
221
222         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
223         if (IS_ERR(new)) {
224                 status = PTR_ERR(new);
225                 mlog_errno(status);
226                 goto bail;
227         }
228         osb->root_inode = new;
229
230         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
231         if (IS_ERR(new)) {
232                 status = PTR_ERR(new);
233                 mlog_errno(status);
234                 goto bail;
235         }
236         osb->sys_root_inode = new;
237
238         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
239              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
240                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
241                 if (!new) {
242                         ocfs2_release_system_inodes(osb);
243                         status = -EINVAL;
244                         mlog_errno(status);
245                         /* FIXME: Should ERROR_RO_FS */
246                         mlog(ML_ERROR, "Unable to load system inode %d, "
247                              "possibly corrupt fs?", i);
248                         goto bail;
249                 }
250                 // the array now has one ref, so drop this one
251                 iput(new);
252         }
253
254 bail:
255         mlog_exit(status);
256         return status;
257 }
258
259 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
260 {
261         struct inode *new = NULL;
262         int status = 0;
263         int i;
264
265         mlog_entry_void();
266
267         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
268              i < NUM_SYSTEM_INODES;
269              i++) {
270                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
271                 if (!new) {
272                         ocfs2_release_system_inodes(osb);
273                         status = -EINVAL;
274                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
275                              status, i, osb->slot_num);
276                         goto bail;
277                 }
278                 /* the array now has one ref, so drop this one */
279                 iput(new);
280         }
281
282 bail:
283         mlog_exit(status);
284         return status;
285 }
286
287 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
288 {
289         int i;
290         struct inode *inode;
291
292         mlog_entry_void();
293
294         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
295                 inode = osb->system_inodes[i];
296                 if (inode) {
297                         iput(inode);
298                         osb->system_inodes[i] = NULL;
299                 }
300         }
301
302         inode = osb->sys_root_inode;
303         if (inode) {
304                 iput(inode);
305                 osb->sys_root_inode = NULL;
306         }
307
308         inode = osb->root_inode;
309         if (inode) {
310                 iput(inode);
311                 osb->root_inode = NULL;
312         }
313
314         mlog_exit(0);
315 }
316
317 /* We're allocating fs objects, use GFP_NOFS */
318 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
319 {
320         struct ocfs2_inode_info *oi;
321
322         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
323         if (!oi)
324                 return NULL;
325
326         return &oi->vfs_inode;
327 }
328
329 static void ocfs2_destroy_inode(struct inode *inode)
330 {
331         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
332 }
333
334 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
335                                                 unsigned int cbits)
336 {
337         unsigned int bytes = 1 << cbits;
338         unsigned int trim = bytes;
339         unsigned int bitshift = 32;
340
341         /*
342          * i_size and all block offsets in ocfs2 are always 64 bits
343          * wide. i_clusters is 32 bits, in cluster-sized units. So on
344          * 64 bit platforms, cluster size will be the limiting factor.
345          */
346
347 #if BITS_PER_LONG == 32
348 # if defined(CONFIG_LBD)
349         BUILD_BUG_ON(sizeof(sector_t) != 8);
350         /*
351          * We might be limited by page cache size.
352          */
353         if (bytes > PAGE_CACHE_SIZE) {
354                 bytes = PAGE_CACHE_SIZE;
355                 trim = 1;
356                 /*
357                  * Shift by 31 here so that we don't get larger than
358                  * MAX_LFS_FILESIZE
359                  */
360                 bitshift = 31;
361         }
362 # else
363         /*
364          * We are limited by the size of sector_t. Use block size, as
365          * that's what we expose to the VFS.
366          */
367         bytes = 1 << bbits;
368         trim = 1;
369         bitshift = 31;
370 # endif
371 #endif
372
373         /*
374          * Trim by a whole cluster when we can actually approach the
375          * on-disk limits. Otherwise we can overflow i_clusters when
376          * an extent start is at the max offset.
377          */
378         return (((unsigned long long)bytes) << bitshift) - trim;
379 }
380
381 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
382 {
383         int incompat_features;
384         int ret = 0;
385         struct mount_options parsed_options;
386         struct ocfs2_super *osb = OCFS2_SB(sb);
387
388         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
389                 ret = -EINVAL;
390                 goto out;
391         }
392
393         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
394             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
395                 ret = -EINVAL;
396                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
397                 goto out;
398         }
399
400         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
401             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
402                 ret = -EINVAL;
403                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
404                 goto out;
405         }
406
407         /* We're going to/from readonly mode. */
408         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
409                 /* Lock here so the check of HARD_RO and the potential
410                  * setting of SOFT_RO is atomic. */
411                 spin_lock(&osb->osb_lock);
412                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
413                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
414                         ret = -EROFS;
415                         goto unlock_osb;
416                 }
417
418                 if (*flags & MS_RDONLY) {
419                         mlog(0, "Going to ro mode.\n");
420                         sb->s_flags |= MS_RDONLY;
421                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
422                 } else {
423                         mlog(0, "Making ro filesystem writeable.\n");
424
425                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
426                                 mlog(ML_ERROR, "Cannot remount RDWR "
427                                      "filesystem due to previous errors.\n");
428                                 ret = -EROFS;
429                                 goto unlock_osb;
430                         }
431                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
432                         if (incompat_features) {
433                                 mlog(ML_ERROR, "Cannot remount RDWR because "
434                                      "of unsupported optional features "
435                                      "(%x).\n", incompat_features);
436                                 ret = -EINVAL;
437                                 goto unlock_osb;
438                         }
439                         sb->s_flags &= ~MS_RDONLY;
440                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
441                 }
442 unlock_osb:
443                 spin_unlock(&osb->osb_lock);
444         }
445
446         if (!ret) {
447                 /* Only save off the new mount options in case of a successful
448                  * remount. */
449                 osb->s_mount_opt = parsed_options.mount_opt;
450                 osb->s_atime_quantum = parsed_options.atime_quantum;
451                 osb->preferred_slot = parsed_options.slot;
452                 if (parsed_options.commit_interval)
453                         osb->osb_commit_interval = parsed_options.commit_interval;
454
455                 if (!ocfs2_is_hard_readonly(osb))
456                         ocfs2_set_journal_params(osb);
457         }
458 out:
459         return ret;
460 }
461
462 static int ocfs2_sb_probe(struct super_block *sb,
463                           struct buffer_head **bh,
464                           int *sector_size)
465 {
466         int status, tmpstat;
467         struct ocfs1_vol_disk_hdr *hdr;
468         struct ocfs2_dinode *di;
469         int blksize;
470
471         *bh = NULL;
472
473         /* may be > 512 */
474         *sector_size = bdev_hardsect_size(sb->s_bdev);
475         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
476                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
477                      *sector_size, OCFS2_MAX_BLOCKSIZE);
478                 status = -EINVAL;
479                 goto bail;
480         }
481
482         /* Can this really happen? */
483         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
484                 *sector_size = OCFS2_MIN_BLOCKSIZE;
485
486         /* check block zero for old format */
487         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
488         if (status < 0) {
489                 mlog_errno(status);
490                 goto bail;
491         }
492         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
493         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
494                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
495                      hdr->major_version, hdr->minor_version);
496                 status = -EINVAL;
497         }
498         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
499                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
500                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
501                      hdr->signature);
502                 status = -EINVAL;
503         }
504         brelse(*bh);
505         *bh = NULL;
506         if (status < 0) {
507                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
508                      "upgraded before mounting with ocfs v2\n");
509                 goto bail;
510         }
511
512         /*
513          * Now check at magic offset for 512, 1024, 2048, 4096
514          * blocksizes.  4096 is the maximum blocksize because it is
515          * the minimum clustersize.
516          */
517         status = -EINVAL;
518         for (blksize = *sector_size;
519              blksize <= OCFS2_MAX_BLOCKSIZE;
520              blksize <<= 1) {
521                 tmpstat = ocfs2_get_sector(sb, bh,
522                                            OCFS2_SUPER_BLOCK_BLKNO,
523                                            blksize);
524                 if (tmpstat < 0) {
525                         status = tmpstat;
526                         mlog_errno(status);
527                         goto bail;
528                 }
529                 di = (struct ocfs2_dinode *) (*bh)->b_data;
530                 status = ocfs2_verify_volume(di, *bh, blksize);
531                 if (status >= 0)
532                         goto bail;
533                 brelse(*bh);
534                 *bh = NULL;
535                 if (status != -EAGAIN)
536                         break;
537         }
538
539 bail:
540         return status;
541 }
542
543 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
544 {
545         if (ocfs2_mount_local(osb)) {
546                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
547                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
548                              "mounted device.\n");
549                         return -EINVAL;
550                 }
551         }
552
553         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
554                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
555                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
556                              "a read-write clustered device.\n");
557                         return -EINVAL;
558                 }
559         }
560
561         return 0;
562 }
563
564 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
565 {
566         struct dentry *root;
567         int status, sector_size;
568         struct mount_options parsed_options;
569         struct inode *inode = NULL;
570         struct ocfs2_super *osb = NULL;
571         struct buffer_head *bh = NULL;
572         char nodestr[8];
573
574         mlog_entry("%p, %p, %i", sb, data, silent);
575
576         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
577                 status = -EINVAL;
578                 goto read_super_error;
579         }
580
581         /* probe for superblock */
582         status = ocfs2_sb_probe(sb, &bh, &sector_size);
583         if (status < 0) {
584                 mlog(ML_ERROR, "superblock probe failed!\n");
585                 goto read_super_error;
586         }
587
588         status = ocfs2_initialize_super(sb, bh, sector_size);
589         osb = OCFS2_SB(sb);
590         if (status < 0) {
591                 mlog_errno(status);
592                 goto read_super_error;
593         }
594         brelse(bh);
595         bh = NULL;
596         osb->s_mount_opt = parsed_options.mount_opt;
597         osb->s_atime_quantum = parsed_options.atime_quantum;
598         osb->preferred_slot = parsed_options.slot;
599         osb->osb_commit_interval = parsed_options.commit_interval;
600         osb->local_alloc_size = parsed_options.localalloc_opt;
601
602         sb->s_magic = OCFS2_SUPER_MAGIC;
603
604         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
605          * heartbeat=none */
606         if (bdev_read_only(sb->s_bdev)) {
607                 if (!(sb->s_flags & MS_RDONLY)) {
608                         status = -EACCES;
609                         mlog(ML_ERROR, "Readonly device detected but readonly "
610                              "mount was not specified.\n");
611                         goto read_super_error;
612                 }
613
614                 /* You should not be able to start a local heartbeat
615                  * on a readonly device. */
616                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
617                         status = -EROFS;
618                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
619                              "device.\n");
620                         goto read_super_error;
621                 }
622
623                 status = ocfs2_check_journals_nolocks(osb);
624                 if (status < 0) {
625                         if (status == -EROFS)
626                                 mlog(ML_ERROR, "Recovery required on readonly "
627                                      "file system, but write access is "
628                                      "unavailable.\n");
629                         else
630                                 mlog_errno(status);                     
631                         goto read_super_error;
632                 }
633
634                 ocfs2_set_ro_flag(osb, 1);
635
636                 printk(KERN_NOTICE "Readonly device detected. No cluster "
637                        "services will be utilized for this mount. Recovery "
638                        "will be skipped.\n");
639         }
640
641         if (!ocfs2_is_hard_readonly(osb)) {
642                 if (sb->s_flags & MS_RDONLY)
643                         ocfs2_set_ro_flag(osb, 0);
644         }
645
646         status = ocfs2_verify_heartbeat(osb);
647         if (status < 0) {
648                 mlog_errno(status);
649                 goto read_super_error;
650         }
651
652         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
653                                                  ocfs2_debugfs_root);
654         if (!osb->osb_debug_root) {
655                 status = -EINVAL;
656                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
657                 goto read_super_error;
658         }
659
660         status = ocfs2_mount_volume(sb);
661         if (osb->root_inode)
662                 inode = igrab(osb->root_inode);
663
664         if (status < 0)
665                 goto read_super_error;
666
667         if (!inode) {
668                 status = -EIO;
669                 mlog_errno(status);
670                 goto read_super_error;
671         }
672
673         root = d_alloc_root(inode);
674         if (!root) {
675                 status = -ENOMEM;
676                 mlog_errno(status);
677                 goto read_super_error;
678         }
679
680         sb->s_root = root;
681
682         ocfs2_complete_mount_recovery(osb);
683
684         if (ocfs2_mount_local(osb))
685                 snprintf(nodestr, sizeof(nodestr), "local");
686         else
687                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
688
689         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
690                "with %s data mode.\n",
691                osb->dev_str, nodestr, osb->slot_num,
692                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
693                "ordered");
694
695         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
696         wake_up(&osb->osb_mount_event);
697
698         mlog_exit(status);
699         return status;
700
701 read_super_error:
702         if (bh != NULL)
703                 brelse(bh);
704
705         if (inode)
706                 iput(inode);
707
708         if (osb) {
709                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
710                 wake_up(&osb->osb_mount_event);
711                 ocfs2_dismount_volume(sb, 1);
712         }
713
714         mlog_exit(status);
715         return status;
716 }
717
718 static int ocfs2_get_sb(struct file_system_type *fs_type,
719                         int flags,
720                         const char *dev_name,
721                         void *data,
722                         struct vfsmount *mnt)
723 {
724         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
725                            mnt);
726 }
727
728 static struct file_system_type ocfs2_fs_type = {
729         .owner          = THIS_MODULE,
730         .name           = "ocfs2",
731         .get_sb         = ocfs2_get_sb, /* is this called when we mount
732                                         * the fs? */
733         .kill_sb        = kill_block_super, /* set to the generic one
734                                              * right now, but do we
735                                              * need to change that? */
736         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
737         .next           = NULL
738 };
739
740 static int ocfs2_parse_options(struct super_block *sb,
741                                char *options,
742                                struct mount_options *mopt,
743                                int is_remount)
744 {
745         int status;
746         char *p;
747
748         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
749                    options ? options : "(none)");
750
751         mopt->commit_interval = 0;
752         mopt->mount_opt = 0;
753         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
754         mopt->slot = OCFS2_INVALID_SLOT;
755         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
756
757         if (!options) {
758                 status = 1;
759                 goto bail;
760         }
761
762         while ((p = strsep(&options, ",")) != NULL) {
763                 int token, option;
764                 substring_t args[MAX_OPT_ARGS];
765
766                 if (!*p)
767                         continue;
768
769                 token = match_token(p, tokens, args);
770                 switch (token) {
771                 case Opt_hb_local:
772                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
773                         break;
774                 case Opt_hb_none:
775                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
776                         break;
777                 case Opt_barrier:
778                         if (match_int(&args[0], &option)) {
779                                 status = 0;
780                                 goto bail;
781                         }
782                         if (option)
783                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
784                         else
785                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
786                         break;
787                 case Opt_intr:
788                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
789                         break;
790                 case Opt_nointr:
791                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
792                         break;
793                 case Opt_err_panic:
794                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
795                         break;
796                 case Opt_err_ro:
797                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
798                         break;
799                 case Opt_data_ordered:
800                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
801                         break;
802                 case Opt_data_writeback:
803                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
804                         break;
805                 case Opt_atime_quantum:
806                         if (match_int(&args[0], &option)) {
807                                 status = 0;
808                                 goto bail;
809                         }
810                         if (option >= 0)
811                                 mopt->atime_quantum = option;
812                         break;
813                 case Opt_slot:
814                         option = 0;
815                         if (match_int(&args[0], &option)) {
816                                 status = 0;
817                                 goto bail;
818                         }
819                         if (option)
820                                 mopt->slot = (s16)option;
821                         break;
822                 case Opt_commit:
823                         option = 0;
824                         if (match_int(&args[0], &option)) {
825                                 status = 0;
826                                 goto bail;
827                         }
828                         if (option < 0)
829                                 return 0;
830                         if (option == 0)
831                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
832                         mopt->commit_interval = HZ * option;
833                         break;
834                 case Opt_localalloc:
835                         option = 0;
836                         if (match_int(&args[0], &option)) {
837                                 status = 0;
838                                 goto bail;
839                         }
840                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
841                                 mopt->localalloc_opt = option;
842                         break;
843                 case Opt_localflocks:
844                         /*
845                          * Changing this during remount could race
846                          * flock() requests, or "unbalance" existing
847                          * ones (e.g., a lock is taken in one mode but
848                          * dropped in the other). If users care enough
849                          * to flip locking modes during remount, we
850                          * could add a "local" flag to individual
851                          * flock structures for proper tracking of
852                          * state.
853                          */
854                         if (!is_remount)
855                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
856                         break;
857                 default:
858                         mlog(ML_ERROR,
859                              "Unrecognized mount option \"%s\" "
860                              "or missing value\n", p);
861                         status = 0;
862                         goto bail;
863                 }
864         }
865
866         status = 1;
867
868 bail:
869         mlog_exit(status);
870         return status;
871 }
872
873 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
874 {
875         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
876         unsigned long opts = osb->s_mount_opt;
877
878         if (opts & OCFS2_MOUNT_HB_LOCAL)
879                 seq_printf(s, ",_netdev,heartbeat=local");
880         else
881                 seq_printf(s, ",heartbeat=none");
882
883         if (opts & OCFS2_MOUNT_NOINTR)
884                 seq_printf(s, ",nointr");
885
886         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
887                 seq_printf(s, ",data=writeback");
888         else
889                 seq_printf(s, ",data=ordered");
890
891         if (opts & OCFS2_MOUNT_BARRIER)
892                 seq_printf(s, ",barrier=1");
893
894         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
895                 seq_printf(s, ",errors=panic");
896         else
897                 seq_printf(s, ",errors=remount-ro");
898
899         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
900                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
901
902         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
903                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
904
905         if (osb->osb_commit_interval)
906                 seq_printf(s, ",commit=%u",
907                            (unsigned) (osb->osb_commit_interval / HZ));
908
909         if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
910                 seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
911
912         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
913                 seq_printf(s, ",localflocks,");
914
915         return 0;
916 }
917
918 static int __init ocfs2_init(void)
919 {
920         int status;
921
922         mlog_entry_void();
923
924         ocfs2_print_version();
925
926         dlmglue_init_stack();
927
928         status = init_ocfs2_uptodate_cache();
929         if (status < 0) {
930                 mlog_errno(status);
931                 goto leave;
932         }
933
934         status = ocfs2_initialize_mem_caches();
935         if (status < 0) {
936                 mlog_errno(status);
937                 goto leave;
938         }
939
940         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
941         if (!ocfs2_wq) {
942                 status = -ENOMEM;
943                 goto leave;
944         }
945
946         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
947         if (!ocfs2_debugfs_root) {
948                 status = -EFAULT;
949                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
950         }
951
952 leave:
953         if (status < 0) {
954                 ocfs2_free_mem_caches();
955                 exit_ocfs2_uptodate_cache();
956         }
957
958         mlog_exit(status);
959
960         if (status >= 0) {
961                 return register_filesystem(&ocfs2_fs_type);
962         } else
963                 return -1;
964 }
965
966 static void __exit ocfs2_exit(void)
967 {
968         mlog_entry_void();
969
970         if (ocfs2_wq) {
971                 flush_workqueue(ocfs2_wq);
972                 destroy_workqueue(ocfs2_wq);
973         }
974
975         debugfs_remove(ocfs2_debugfs_root);
976
977         ocfs2_free_mem_caches();
978
979         unregister_filesystem(&ocfs2_fs_type);
980
981         exit_ocfs2_uptodate_cache();
982
983         dlmglue_exit_stack();
984
985         mlog_exit_void();
986 }
987
988 static void ocfs2_put_super(struct super_block *sb)
989 {
990         mlog_entry("(0x%p)\n", sb);
991
992         ocfs2_sync_blockdev(sb);
993         ocfs2_dismount_volume(sb, 0);
994
995         mlog_exit_void();
996 }
997
998 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
999 {
1000         struct ocfs2_super *osb;
1001         u32 numbits, freebits;
1002         int status;
1003         struct ocfs2_dinode *bm_lock;
1004         struct buffer_head *bh = NULL;
1005         struct inode *inode = NULL;
1006
1007         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1008
1009         osb = OCFS2_SB(dentry->d_sb);
1010
1011         inode = ocfs2_get_system_file_inode(osb,
1012                                             GLOBAL_BITMAP_SYSTEM_INODE,
1013                                             OCFS2_INVALID_SLOT);
1014         if (!inode) {
1015                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1016                 status = -EIO;
1017                 goto bail;
1018         }
1019
1020         status = ocfs2_inode_lock(inode, &bh, 0);
1021         if (status < 0) {
1022                 mlog_errno(status);
1023                 goto bail;
1024         }
1025
1026         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1027
1028         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1029         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1030
1031         buf->f_type = OCFS2_SUPER_MAGIC;
1032         buf->f_bsize = dentry->d_sb->s_blocksize;
1033         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1034         buf->f_blocks = ((sector_t) numbits) *
1035                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1036         buf->f_bfree = ((sector_t) freebits) *
1037                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1038         buf->f_bavail = buf->f_bfree;
1039         buf->f_files = numbits;
1040         buf->f_ffree = freebits;
1041
1042         brelse(bh);
1043
1044         ocfs2_inode_unlock(inode, 0);
1045         status = 0;
1046 bail:
1047         if (inode)
1048                 iput(inode);
1049
1050         mlog_exit(status);
1051
1052         return status;
1053 }
1054
1055 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1056 {
1057         struct ocfs2_inode_info *oi = data;
1058
1059         oi->ip_flags = 0;
1060         oi->ip_open_count = 0;
1061         spin_lock_init(&oi->ip_lock);
1062         ocfs2_extent_map_init(&oi->vfs_inode);
1063         INIT_LIST_HEAD(&oi->ip_io_markers);
1064         oi->ip_created_trans = 0;
1065         oi->ip_last_trans = 0;
1066         oi->ip_dir_start_lookup = 0;
1067
1068         init_rwsem(&oi->ip_alloc_sem);
1069         mutex_init(&oi->ip_io_mutex);
1070
1071         oi->ip_blkno = 0ULL;
1072         oi->ip_clusters = 0;
1073
1074         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1075         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1076         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1077
1078         ocfs2_metadata_cache_init(&oi->vfs_inode);
1079
1080         inode_init_once(&oi->vfs_inode);
1081 }
1082
1083 static int ocfs2_initialize_mem_caches(void)
1084 {
1085         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1086                                        sizeof(struct ocfs2_inode_info),
1087                                        0,
1088                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1089                                                 SLAB_MEM_SPREAD),
1090                                        ocfs2_inode_init_once);
1091         if (!ocfs2_inode_cachep)
1092                 return -ENOMEM;
1093
1094         return 0;
1095 }
1096
1097 static void ocfs2_free_mem_caches(void)
1098 {
1099         if (ocfs2_inode_cachep)
1100                 kmem_cache_destroy(ocfs2_inode_cachep);
1101
1102         ocfs2_inode_cachep = NULL;
1103 }
1104
1105 static int ocfs2_get_sector(struct super_block *sb,
1106                             struct buffer_head **bh,
1107                             int block,
1108                             int sect_size)
1109 {
1110         if (!sb_set_blocksize(sb, sect_size)) {
1111                 mlog(ML_ERROR, "unable to set blocksize\n");
1112                 return -EIO;
1113         }
1114
1115         *bh = sb_getblk(sb, block);
1116         if (!*bh) {
1117                 mlog_errno(-EIO);
1118                 return -EIO;
1119         }
1120         lock_buffer(*bh);
1121         if (!buffer_dirty(*bh))
1122                 clear_buffer_uptodate(*bh);
1123         unlock_buffer(*bh);
1124         ll_rw_block(READ, 1, bh);
1125         wait_on_buffer(*bh);
1126         return 0;
1127 }
1128
1129 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1130 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1131 {
1132         int status;
1133
1134         /* XXX hold a ref on the node while mounte?  easy enough, if
1135          * desirable. */
1136         if (ocfs2_mount_local(osb))
1137                 osb->node_num = 0;
1138         else {
1139                 status = ocfs2_cluster_this_node(&osb->node_num);
1140                 if (status < 0) {
1141                         mlog_errno(status);
1142                         mlog(ML_ERROR,
1143                              "could not find this host's node number\n");
1144                         goto bail;
1145                 }
1146         }
1147
1148         mlog(0, "I am node %u\n", osb->node_num);
1149
1150         status = 0;
1151 bail:
1152         return status;
1153 }
1154
1155 static int ocfs2_mount_volume(struct super_block *sb)
1156 {
1157         int status = 0;
1158         int unlock_super = 0;
1159         struct ocfs2_super *osb = OCFS2_SB(sb);
1160
1161         mlog_entry_void();
1162
1163         if (ocfs2_is_hard_readonly(osb))
1164                 goto leave;
1165
1166         status = ocfs2_fill_local_node_info(osb);
1167         if (status < 0) {
1168                 mlog_errno(status);
1169                 goto leave;
1170         }
1171
1172         status = ocfs2_dlm_init(osb);
1173         if (status < 0) {
1174                 mlog_errno(status);
1175                 goto leave;
1176         }
1177
1178         status = ocfs2_super_lock(osb, 1);
1179         if (status < 0) {
1180                 mlog_errno(status);
1181                 goto leave;
1182         }
1183         unlock_super = 1;
1184
1185         /* This will load up the node map and add ourselves to it. */
1186         status = ocfs2_find_slot(osb);
1187         if (status < 0) {
1188                 mlog_errno(status);
1189                 goto leave;
1190         }
1191
1192         /* load all node-local system inodes */
1193         status = ocfs2_init_local_system_inodes(osb);
1194         if (status < 0) {
1195                 mlog_errno(status);
1196                 goto leave;
1197         }
1198
1199         status = ocfs2_check_volume(osb);
1200         if (status < 0) {
1201                 mlog_errno(status);
1202                 goto leave;
1203         }
1204
1205         status = ocfs2_truncate_log_init(osb);
1206         if (status < 0) {
1207                 mlog_errno(status);
1208                 goto leave;
1209         }
1210
1211         if (ocfs2_mount_local(osb))
1212                 goto leave;
1213
1214 leave:
1215         if (unlock_super)
1216                 ocfs2_super_unlock(osb, 1);
1217
1218         mlog_exit(status);
1219         return status;
1220 }
1221
1222 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1223 {
1224         int tmp;
1225         struct ocfs2_super *osb = NULL;
1226         char nodestr[8];
1227
1228         mlog_entry("(0x%p)\n", sb);
1229
1230         BUG_ON(!sb);
1231         osb = OCFS2_SB(sb);
1232         BUG_ON(!osb);
1233
1234         ocfs2_shutdown_local_alloc(osb);
1235
1236         ocfs2_truncate_log_shutdown(osb);
1237
1238         /* This will disable recovery and flush any recovery work. */
1239         ocfs2_recovery_exit(osb);
1240
1241         ocfs2_journal_shutdown(osb);
1242
1243         ocfs2_sync_blockdev(sb);
1244
1245         /* No cluster connection means we've failed during mount, so skip
1246          * all the steps which depended on that to complete. */
1247         if (osb->cconn) {
1248                 tmp = ocfs2_super_lock(osb, 1);
1249                 if (tmp < 0) {
1250                         mlog_errno(tmp);
1251                         return;
1252                 }
1253         }
1254
1255         if (osb->slot_num != OCFS2_INVALID_SLOT)
1256                 ocfs2_put_slot(osb);
1257
1258         if (osb->cconn)
1259                 ocfs2_super_unlock(osb, 1);
1260
1261         ocfs2_release_system_inodes(osb);
1262
1263         if (osb->cconn)
1264                 ocfs2_dlm_shutdown(osb);
1265
1266         debugfs_remove(osb->osb_debug_root);
1267
1268         /*
1269          * This is a small hack to move ocfs2_hb_ctl into stackglue.
1270          * If we're dismounting due to mount error, mount.ocfs2 will clean
1271          * up heartbeat.  If we're a local mount, there is no heartbeat.
1272          * If we failed before we got a uuid_str yet, we can't stop
1273          * heartbeat.  Otherwise, do it.
1274          */
1275         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1276                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1277
1278         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1279
1280         if (ocfs2_mount_local(osb))
1281                 snprintf(nodestr, sizeof(nodestr), "local");
1282         else
1283                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1284
1285         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1286                osb->dev_str, nodestr);
1287
1288         ocfs2_delete_osb(osb);
1289         kfree(osb);
1290         sb->s_dev = 0;
1291         sb->s_fs_info = NULL;
1292 }
1293
1294 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1295                                 unsigned uuid_bytes)
1296 {
1297         int i, ret;
1298         char *ptr;
1299
1300         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1301
1302         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1303         if (osb->uuid_str == NULL)
1304                 return -ENOMEM;
1305
1306         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1307                 /* print with null */
1308                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1309                 if (ret != 2) /* drop super cleans up */
1310                         return -EINVAL;
1311                 /* then only advance past the last char */
1312                 ptr += 2;
1313         }
1314
1315         return 0;
1316 }
1317
1318 static int ocfs2_initialize_super(struct super_block *sb,
1319                                   struct buffer_head *bh,
1320                                   int sector_size)
1321 {
1322         int status;
1323         int i, cbits, bbits;
1324         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1325         struct inode *inode = NULL;
1326         struct ocfs2_journal *journal;
1327         __le32 uuid_net_key;
1328         struct ocfs2_super *osb;
1329
1330         mlog_entry_void();
1331
1332         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1333         if (!osb) {
1334                 status = -ENOMEM;
1335                 mlog_errno(status);
1336                 goto bail;
1337         }
1338
1339         sb->s_fs_info = osb;
1340         sb->s_op = &ocfs2_sops;
1341         sb->s_export_op = &ocfs2_export_ops;
1342         sb->s_time_gran = 1;
1343         sb->s_flags |= MS_NOATIME;
1344         /* this is needed to support O_LARGEFILE */
1345         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1346         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1347         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1348
1349         osb->sb = sb;
1350         /* Save off for ocfs2_rw_direct */
1351         osb->s_sectsize_bits = blksize_bits(sector_size);
1352         BUG_ON(!osb->s_sectsize_bits);
1353
1354         spin_lock_init(&osb->dc_task_lock);
1355         init_waitqueue_head(&osb->dc_event);
1356         osb->dc_work_sequence = 0;
1357         osb->dc_wake_sequence = 0;
1358         INIT_LIST_HEAD(&osb->blocked_lock_list);
1359         osb->blocked_lock_count = 0;
1360         spin_lock_init(&osb->osb_lock);
1361
1362         atomic_set(&osb->alloc_stats.moves, 0);
1363         atomic_set(&osb->alloc_stats.local_data, 0);
1364         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1365         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1366         atomic_set(&osb->alloc_stats.bg_extends, 0);
1367
1368         ocfs2_init_node_maps(osb);
1369
1370         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1371                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1372
1373         status = ocfs2_recovery_init(osb);
1374         if (status) {
1375                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1376                 mlog_errno(status);
1377                 goto bail;
1378         }
1379
1380         init_waitqueue_head(&osb->checkpoint_event);
1381         atomic_set(&osb->needs_checkpoint, 0);
1382
1383         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1384
1385         osb->slot_num = OCFS2_INVALID_SLOT;
1386
1387         osb->local_alloc_state = OCFS2_LA_UNUSED;
1388         osb->local_alloc_bh = NULL;
1389
1390         init_waitqueue_head(&osb->osb_mount_event);
1391
1392         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1393         if (!osb->vol_label) {
1394                 mlog(ML_ERROR, "unable to alloc vol label\n");
1395                 status = -ENOMEM;
1396                 goto bail;
1397         }
1398
1399         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1400         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1401                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1402                      osb->max_slots);
1403                 status = -EINVAL;
1404                 goto bail;
1405         }
1406         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1407
1408         init_waitqueue_head(&osb->osb_wipe_event);
1409         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1410                                         sizeof(*osb->osb_orphan_wipes),
1411                                         GFP_KERNEL);
1412         if (!osb->osb_orphan_wipes) {
1413                 status = -ENOMEM;
1414                 mlog_errno(status);
1415                 goto bail;
1416         }
1417
1418         osb->s_feature_compat =
1419                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1420         osb->s_feature_ro_compat =
1421                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1422         osb->s_feature_incompat =
1423                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1424
1425         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1426                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1427                      "optional features (%x).\n", i);
1428                 status = -EINVAL;
1429                 goto bail;
1430         }
1431         if (!(osb->sb->s_flags & MS_RDONLY) &&
1432             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1433                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1434                      "unsupported optional features (%x).\n", i);
1435                 status = -EINVAL;
1436                 goto bail;
1437         }
1438
1439         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1440
1441         /* FIXME
1442          * This should be done in ocfs2_journal_init(), but unknown
1443          * ordering issues will cause the filesystem to crash.
1444          * If anyone wants to figure out what part of the code
1445          * refers to osb->journal before ocfs2_journal_init() is run,
1446          * be my guest.
1447          */
1448         /* initialize our journal structure */
1449
1450         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1451         if (!journal) {
1452                 mlog(ML_ERROR, "unable to alloc journal\n");
1453                 status = -ENOMEM;
1454                 goto bail;
1455         }
1456         osb->journal = journal;
1457         journal->j_osb = osb;
1458
1459         atomic_set(&journal->j_num_trans, 0);
1460         init_rwsem(&journal->j_trans_barrier);
1461         init_waitqueue_head(&journal->j_checkpointed);
1462         spin_lock_init(&journal->j_lock);
1463         journal->j_trans_id = (unsigned long) 1;
1464         INIT_LIST_HEAD(&journal->j_la_cleanups);
1465         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1466         journal->j_state = OCFS2_JOURNAL_FREE;
1467
1468         /* get some pseudo constants for clustersize bits */
1469         osb->s_clustersize_bits =
1470                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1471         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1472         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1473
1474         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1475             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1476                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1477                      osb->s_clustersize);
1478                 status = -EINVAL;
1479                 goto bail;
1480         }
1481
1482         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1483             > (u32)~0UL) {
1484                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1485                      "what jbd can address in 32 bits.\n");
1486                 status = -EINVAL;
1487                 goto bail;
1488         }
1489
1490         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1491                                  sizeof(di->id2.i_super.s_uuid))) {
1492                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1493                 status = -ENOMEM;
1494                 goto bail;
1495         }
1496
1497         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1498
1499         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1500         osb->vol_label[63] = '\0';
1501         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1502         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1503         osb->first_cluster_group_blkno =
1504                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1505         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1506         mlog(0, "vol_label: %s\n", osb->vol_label);
1507         mlog(0, "uuid: %s\n", osb->uuid_str);
1508         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1509              (unsigned long long)osb->root_blkno,
1510              (unsigned long long)osb->system_dir_blkno);
1511
1512         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1513         if (!osb->osb_dlm_debug) {
1514                 status = -ENOMEM;
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         atomic_set(&osb->vol_state, VOLUME_INIT);
1520
1521         /* load root, system_dir, and all global system inodes */
1522         status = ocfs2_init_global_system_inodes(osb);
1523         if (status < 0) {
1524                 mlog_errno(status);
1525                 goto bail;
1526         }
1527
1528         /*
1529          * global bitmap
1530          */
1531         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1532                                             OCFS2_INVALID_SLOT);
1533         if (!inode) {
1534                 status = -EINVAL;
1535                 mlog_errno(status);
1536                 goto bail;
1537         }
1538
1539         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1540         iput(inode);
1541
1542         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1543
1544         status = ocfs2_init_slot_info(osb);
1545         if (status < 0) {
1546                 mlog_errno(status);
1547                 goto bail;
1548         }
1549
1550 bail:
1551         mlog_exit(status);
1552         return status;
1553 }
1554
1555 /*
1556  * will return: -EAGAIN if it is ok to keep searching for superblocks
1557  *              -EINVAL if there is a bad superblock
1558  *              0 on success
1559  */
1560 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1561                                struct buffer_head *bh,
1562                                u32 blksz)
1563 {
1564         int status = -EAGAIN;
1565
1566         mlog_entry_void();
1567
1568         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1569                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1570                 status = -EINVAL;
1571                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1572                         mlog(ML_ERROR, "found superblock with incorrect block "
1573                              "size: found %u, should be %u\n",
1574                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1575                                blksz);
1576                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1577                            OCFS2_MAJOR_REV_LEVEL ||
1578                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1579                            OCFS2_MINOR_REV_LEVEL) {
1580                         mlog(ML_ERROR, "found superblock with bad version: "
1581                              "found %u.%u, should be %u.%u\n",
1582                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1583                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1584                              OCFS2_MAJOR_REV_LEVEL,
1585                              OCFS2_MINOR_REV_LEVEL);
1586                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1587                         mlog(ML_ERROR, "bad block number on superblock: "
1588                              "found %llu, should be %llu\n",
1589                              (unsigned long long)le64_to_cpu(di->i_blkno),
1590                              (unsigned long long)bh->b_blocknr);
1591                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1592                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1593                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1594                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1595                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1596                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1597                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1598                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1599                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1600                         mlog(ML_ERROR,
1601                              "Superblock slots found greater than file system "
1602                              "maximum: found %u, max %u\n",
1603                              le16_to_cpu(di->id2.i_super.s_max_slots),
1604                              OCFS2_MAX_SLOTS);
1605                 } else {
1606                         /* found it! */
1607                         status = 0;
1608                 }
1609         }
1610
1611         mlog_exit(status);
1612         return status;
1613 }
1614
1615 static int ocfs2_check_volume(struct ocfs2_super *osb)
1616 {
1617         int status;
1618         int dirty;
1619         int local;
1620         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1621                                                   * recover
1622                                                   * ourselves. */
1623
1624         mlog_entry_void();
1625
1626         /* Init our journal object. */
1627         status = ocfs2_journal_init(osb->journal, &dirty);
1628         if (status < 0) {
1629                 mlog(ML_ERROR, "Could not initialize journal!\n");
1630                 goto finally;
1631         }
1632
1633         /* If the journal was unmounted cleanly then we don't want to
1634          * recover anything. Otherwise, journal_load will do that
1635          * dirty work for us :) */
1636         if (!dirty) {
1637                 status = ocfs2_journal_wipe(osb->journal, 0);
1638                 if (status < 0) {
1639                         mlog_errno(status);
1640                         goto finally;
1641                 }
1642         } else {
1643                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1644                      "recovering volume.\n");
1645         }
1646
1647         local = ocfs2_mount_local(osb);
1648
1649         /* will play back anything left in the journal. */
1650         ocfs2_journal_load(osb->journal, local);
1651
1652         if (dirty) {
1653                 /* recover my local alloc if we didn't unmount cleanly. */
1654                 status = ocfs2_begin_local_alloc_recovery(osb,
1655                                                           osb->slot_num,
1656                                                           &local_alloc);
1657                 if (status < 0) {
1658                         mlog_errno(status);
1659                         goto finally;
1660                 }
1661                 /* we complete the recovery process after we've marked
1662                  * ourselves as mounted. */
1663         }
1664
1665         mlog(0, "Journal loaded.\n");
1666
1667         status = ocfs2_load_local_alloc(osb);
1668         if (status < 0) {
1669                 mlog_errno(status);
1670                 goto finally;
1671         }
1672
1673         if (dirty) {
1674                 /* Recovery will be completed after we've mounted the
1675                  * rest of the volume. */
1676                 osb->dirty = 1;
1677                 osb->local_alloc_copy = local_alloc;
1678                 local_alloc = NULL;
1679         }
1680
1681         /* go through each journal, trylock it and if you get the
1682          * lock, and it's marked as dirty, set the bit in the recover
1683          * map and launch a recovery thread for it. */
1684         status = ocfs2_mark_dead_nodes(osb);
1685         if (status < 0)
1686                 mlog_errno(status);
1687
1688 finally:
1689         if (local_alloc)
1690                 kfree(local_alloc);
1691
1692         mlog_exit(status);
1693         return status;
1694 }
1695
1696 /*
1697  * The routine gets called from dismount or close whenever a dismount on
1698  * volume is requested and the osb open count becomes 1.
1699  * It will remove the osb from the global list and also free up all the
1700  * initialized resources and fileobject.
1701  */
1702 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1703 {
1704         mlog_entry_void();
1705
1706         /* This function assumes that the caller has the main osb resource */
1707
1708         ocfs2_free_slot_info(osb);
1709
1710         kfree(osb->osb_orphan_wipes);
1711         /* FIXME
1712          * This belongs in journal shutdown, but because we have to
1713          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1714          * we free it here.
1715          */
1716         kfree(osb->journal);
1717         if (osb->local_alloc_copy)
1718                 kfree(osb->local_alloc_copy);
1719         kfree(osb->uuid_str);
1720         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1721         memset(osb, 0, sizeof(struct ocfs2_super));
1722
1723         mlog_exit_void();
1724 }
1725
1726 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1727  * panic(). We do not support continue-on-error operation. */
1728 static void ocfs2_handle_error(struct super_block *sb)
1729 {
1730         struct ocfs2_super *osb = OCFS2_SB(sb);
1731
1732         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1733                 panic("OCFS2: (device %s): panic forced after error\n",
1734                       sb->s_id);
1735
1736         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1737
1738         if (sb->s_flags & MS_RDONLY &&
1739             (ocfs2_is_soft_readonly(osb) ||
1740              ocfs2_is_hard_readonly(osb)))
1741                 return;
1742
1743         printk(KERN_CRIT "File system is now read-only due to the potential "
1744                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1745                "system is unmounted.\n");
1746         sb->s_flags |= MS_RDONLY;
1747         ocfs2_set_ro_flag(osb, 0);
1748 }
1749
1750 static char error_buf[1024];
1751
1752 void __ocfs2_error(struct super_block *sb,
1753                    const char *function,
1754                    const char *fmt, ...)
1755 {
1756         va_list args;
1757
1758         va_start(args, fmt);
1759         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1760         va_end(args);
1761
1762         /* Not using mlog here because we want to show the actual
1763          * function the error came from. */
1764         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1765                sb->s_id, function, error_buf);
1766
1767         ocfs2_handle_error(sb);
1768 }
1769
1770 /* Handle critical errors. This is intentionally more drastic than
1771  * ocfs2_handle_error, so we only use for things like journal errors,
1772  * etc. */
1773 void __ocfs2_abort(struct super_block* sb,
1774                    const char *function,
1775                    const char *fmt, ...)
1776 {
1777         va_list args;
1778
1779         va_start(args, fmt);
1780         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1781         va_end(args);
1782
1783         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1784                sb->s_id, function, error_buf);
1785
1786         /* We don't have the cluster support yet to go straight to
1787          * hard readonly in here. Until then, we want to keep
1788          * ocfs2_abort() so that we can at least mark critical
1789          * errors.
1790          *
1791          * TODO: This should abort the journal and alert other nodes
1792          * that our slot needs recovery. */
1793
1794         /* Force a panic(). This stinks, but it's better than letting
1795          * things continue without having a proper hard readonly
1796          * here. */
1797         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1798         ocfs2_handle_error(sb);
1799 }
1800
1801 module_init(ocfs2_init);
1802 module_exit(ocfs2_exit);