ocfs2: Implement show_options()
[pandora-kernel.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
44 #include <cluster/nodemanager.h>
45
46 #define MLOG_MASK_PREFIX ML_SUPER
47 #include <cluster/masklog.h>
48
49 #include "ocfs2.h"
50
51 /* this should be the only file to include a version 1 header */
52 #include "ocfs1_fs_compat.h"
53
54 #include "alloc.h"
55 #include "dlmglue.h"
56 #include "export.h"
57 #include "extent_map.h"
58 #include "heartbeat.h"
59 #include "inode.h"
60 #include "journal.h"
61 #include "localalloc.h"
62 #include "namei.h"
63 #include "slot_map.h"
64 #include "super.h"
65 #include "sysfile.h"
66 #include "uptodate.h"
67 #include "ver.h"
68 #include "vote.h"
69
70 #include "buffer_head_io.h"
71
72 static struct kmem_cache *ocfs2_inode_cachep = NULL;
73
74 /* OCFS2 needs to schedule several differnt types of work which
75  * require cluster locking, disk I/O, recovery waits, etc. Since these
76  * types of work tend to be heavy we avoid using the kernel events
77  * workqueue and schedule on our own. */
78 struct workqueue_struct *ocfs2_wq = NULL;
79
80 static struct dentry *ocfs2_debugfs_root = NULL;
81
82 MODULE_AUTHOR("Oracle");
83 MODULE_LICENSE("GPL");
84
85 struct mount_options
86 {
87         unsigned long   mount_opt;
88         unsigned int    atime_quantum;
89         signed short    slot;
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 int 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_err,
154 };
155
156 static match_table_t tokens = {
157         {Opt_barrier, "barrier=%u"},
158         {Opt_err_panic, "errors=panic"},
159         {Opt_err_ro, "errors=remount-ro"},
160         {Opt_intr, "intr"},
161         {Opt_nointr, "nointr"},
162         {Opt_hb_none, OCFS2_HB_NONE},
163         {Opt_hb_local, OCFS2_HB_LOCAL},
164         {Opt_data_ordered, "data=ordered"},
165         {Opt_data_writeback, "data=writeback"},
166         {Opt_atime_quantum, "atime_quantum=%u"},
167         {Opt_slot, "preferred_slot=%u"},
168         {Opt_err, NULL}
169 };
170
171 /*
172  * write_super and sync_fs ripped right out of ext3.
173  */
174 static void ocfs2_write_super(struct super_block *sb)
175 {
176         if (mutex_trylock(&sb->s_lock) != 0)
177                 BUG();
178         sb->s_dirt = 0;
179 }
180
181 static int ocfs2_sync_fs(struct super_block *sb, int wait)
182 {
183         int status = 0;
184         tid_t target;
185         struct ocfs2_super *osb = OCFS2_SB(sb);
186
187         sb->s_dirt = 0;
188
189         if (ocfs2_is_hard_readonly(osb))
190                 return -EROFS;
191
192         if (wait) {
193                 status = ocfs2_flush_truncate_log(osb);
194                 if (status < 0)
195                         mlog_errno(status);
196         } else {
197                 ocfs2_schedule_truncate_log_flush(osb, 0);
198         }
199
200         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
201                 if (wait)
202                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
203                                         target);
204         }
205         return 0;
206 }
207
208 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
209 {
210         struct inode *new = NULL;
211         int status = 0;
212         int i;
213
214         mlog_entry_void();
215
216         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
217         if (IS_ERR(new)) {
218                 status = PTR_ERR(new);
219                 mlog_errno(status);
220                 goto bail;
221         }
222         osb->root_inode = new;
223
224         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
225         if (IS_ERR(new)) {
226                 status = PTR_ERR(new);
227                 mlog_errno(status);
228                 goto bail;
229         }
230         osb->sys_root_inode = new;
231
232         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
233              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
234                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
235                 if (!new) {
236                         ocfs2_release_system_inodes(osb);
237                         status = -EINVAL;
238                         mlog_errno(status);
239                         /* FIXME: Should ERROR_RO_FS */
240                         mlog(ML_ERROR, "Unable to load system inode %d, "
241                              "possibly corrupt fs?", i);
242                         goto bail;
243                 }
244                 // the array now has one ref, so drop this one
245                 iput(new);
246         }
247
248 bail:
249         mlog_exit(status);
250         return status;
251 }
252
253 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
254 {
255         struct inode *new = NULL;
256         int status = 0;
257         int i;
258
259         mlog_entry_void();
260
261         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
262              i < NUM_SYSTEM_INODES;
263              i++) {
264                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
265                 if (!new) {
266                         ocfs2_release_system_inodes(osb);
267                         status = -EINVAL;
268                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
269                              status, i, osb->slot_num);
270                         goto bail;
271                 }
272                 /* the array now has one ref, so drop this one */
273                 iput(new);
274         }
275
276 bail:
277         mlog_exit(status);
278         return status;
279 }
280
281 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
282 {
283         int status = 0, i;
284         struct inode *inode;
285
286         mlog_entry_void();
287
288         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
289                 inode = osb->system_inodes[i];
290                 if (inode) {
291                         iput(inode);
292                         osb->system_inodes[i] = NULL;
293                 }
294         }
295
296         inode = osb->sys_root_inode;
297         if (inode) {
298                 iput(inode);
299                 osb->sys_root_inode = NULL;
300         }
301
302         inode = osb->root_inode;
303         if (inode) {
304                 iput(inode);
305                 osb->root_inode = NULL;
306         }
307
308         mlog_exit(status);
309         return status;
310 }
311
312 /* We're allocating fs objects, use GFP_NOFS */
313 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
314 {
315         struct ocfs2_inode_info *oi;
316
317         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
318         if (!oi)
319                 return NULL;
320
321         return &oi->vfs_inode;
322 }
323
324 static void ocfs2_destroy_inode(struct inode *inode)
325 {
326         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
327 }
328
329 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
330                                                 unsigned int cbits)
331 {
332         unsigned int bytes = 1 << cbits;
333         unsigned int trim = bytes;
334         unsigned int bitshift = 32;
335
336         /*
337          * i_size and all block offsets in ocfs2 are always 64 bits
338          * wide. i_clusters is 32 bits, in cluster-sized units. So on
339          * 64 bit platforms, cluster size will be the limiting factor.
340          */
341
342 #if BITS_PER_LONG == 32
343 # if defined(CONFIG_LBD)
344         BUILD_BUG_ON(sizeof(sector_t) != 8);
345         /*
346          * We might be limited by page cache size.
347          */
348         if (bytes > PAGE_CACHE_SIZE) {
349                 bytes = PAGE_CACHE_SIZE;
350                 trim = 1;
351                 /*
352                  * Shift by 31 here so that we don't get larger than
353                  * MAX_LFS_FILESIZE
354                  */
355                 bitshift = 31;
356         }
357 # else
358         /*
359          * We are limited by the size of sector_t. Use block size, as
360          * that's what we expose to the VFS.
361          */
362         bytes = 1 << bbits;
363         trim = 1;
364         bitshift = 31;
365 # endif
366 #endif
367
368         /*
369          * Trim by a whole cluster when we can actually approach the
370          * on-disk limits. Otherwise we can overflow i_clusters when
371          * an extent start is at the max offset.
372          */
373         return (((unsigned long long)bytes) << bitshift) - trim;
374 }
375
376 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
377 {
378         int incompat_features;
379         int ret = 0;
380         struct mount_options parsed_options;
381         struct ocfs2_super *osb = OCFS2_SB(sb);
382
383         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
384                 ret = -EINVAL;
385                 goto out;
386         }
387
388         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
389             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
390                 ret = -EINVAL;
391                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
392                 goto out;
393         }
394
395         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
396             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
397                 ret = -EINVAL;
398                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
399                 goto out;
400         }
401
402         /* We're going to/from readonly mode. */
403         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
404                 /* Lock here so the check of HARD_RO and the potential
405                  * setting of SOFT_RO is atomic. */
406                 spin_lock(&osb->osb_lock);
407                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
408                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
409                         ret = -EROFS;
410                         goto unlock_osb;
411                 }
412
413                 if (*flags & MS_RDONLY) {
414                         mlog(0, "Going to ro mode.\n");
415                         sb->s_flags |= MS_RDONLY;
416                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
417                 } else {
418                         mlog(0, "Making ro filesystem writeable.\n");
419
420                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
421                                 mlog(ML_ERROR, "Cannot remount RDWR "
422                                      "filesystem due to previous errors.\n");
423                                 ret = -EROFS;
424                                 goto unlock_osb;
425                         }
426                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
427                         if (incompat_features) {
428                                 mlog(ML_ERROR, "Cannot remount RDWR because "
429                                      "of unsupported optional features "
430                                      "(%x).\n", incompat_features);
431                                 ret = -EINVAL;
432                                 goto unlock_osb;
433                         }
434                         sb->s_flags &= ~MS_RDONLY;
435                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
436                 }
437 unlock_osb:
438                 spin_unlock(&osb->osb_lock);
439         }
440
441         if (!ret) {
442                 if (!ocfs2_is_hard_readonly(osb))
443                         ocfs2_set_journal_params(osb);
444
445                 /* Only save off the new mount options in case of a successful
446                  * remount. */
447                 osb->s_mount_opt = parsed_options.mount_opt;
448                 osb->s_atime_quantum = parsed_options.atime_quantum;
449                 osb->preferred_slot = parsed_options.slot;
450         }
451 out:
452         return ret;
453 }
454
455 static int ocfs2_sb_probe(struct super_block *sb,
456                           struct buffer_head **bh,
457                           int *sector_size)
458 {
459         int status = 0, tmpstat;
460         struct ocfs1_vol_disk_hdr *hdr;
461         struct ocfs2_dinode *di;
462         int blksize;
463
464         *bh = NULL;
465
466         /* may be > 512 */
467         *sector_size = bdev_hardsect_size(sb->s_bdev);
468         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
469                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
470                      *sector_size, OCFS2_MAX_BLOCKSIZE);
471                 status = -EINVAL;
472                 goto bail;
473         }
474
475         /* Can this really happen? */
476         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
477                 *sector_size = OCFS2_MIN_BLOCKSIZE;
478
479         /* check block zero for old format */
480         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
481         if (status < 0) {
482                 mlog_errno(status);
483                 goto bail;
484         }
485         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
486         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
487                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
488                      hdr->major_version, hdr->minor_version);
489                 status = -EINVAL;
490         }
491         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
492                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
493                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
494                      hdr->signature);
495                 status = -EINVAL;
496         }
497         brelse(*bh);
498         *bh = NULL;
499         if (status < 0) {
500                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
501                      "upgraded before mounting with ocfs v2\n");
502                 goto bail;
503         }
504
505         /*
506          * Now check at magic offset for 512, 1024, 2048, 4096
507          * blocksizes.  4096 is the maximum blocksize because it is
508          * the minimum clustersize.
509          */
510         status = -EINVAL;
511         for (blksize = *sector_size;
512              blksize <= OCFS2_MAX_BLOCKSIZE;
513              blksize <<= 1) {
514                 tmpstat = ocfs2_get_sector(sb, bh,
515                                            OCFS2_SUPER_BLOCK_BLKNO,
516                                            blksize);
517                 if (tmpstat < 0) {
518                         status = tmpstat;
519                         mlog_errno(status);
520                         goto bail;
521                 }
522                 di = (struct ocfs2_dinode *) (*bh)->b_data;
523                 status = ocfs2_verify_volume(di, *bh, blksize);
524                 if (status >= 0)
525                         goto bail;
526                 brelse(*bh);
527                 *bh = NULL;
528                 if (status != -EAGAIN)
529                         break;
530         }
531
532 bail:
533         return status;
534 }
535
536 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
537 {
538         if (ocfs2_mount_local(osb)) {
539                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
540                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
541                              "mounted device.\n");
542                         return -EINVAL;
543                 }
544         }
545
546         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
547                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
548                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
549                              "a read-write clustered device.\n");
550                         return -EINVAL;
551                 }
552         }
553
554         return 0;
555 }
556
557 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
558 {
559         struct dentry *root;
560         int status, sector_size;
561         struct mount_options parsed_options;
562         struct inode *inode = NULL;
563         struct ocfs2_super *osb = NULL;
564         struct buffer_head *bh = NULL;
565         char nodestr[8];
566
567         mlog_entry("%p, %p, %i", sb, data, silent);
568
569         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
570                 status = -EINVAL;
571                 goto read_super_error;
572         }
573
574         /* for now we only have one cluster/node, make sure we see it
575          * in the heartbeat universe */
576         if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
577                 if (!o2hb_check_local_node_heartbeating()) {
578                         status = -EINVAL;
579                         goto read_super_error;
580                 }
581         }
582
583         /* probe for superblock */
584         status = ocfs2_sb_probe(sb, &bh, &sector_size);
585         if (status < 0) {
586                 mlog(ML_ERROR, "superblock probe failed!\n");
587                 goto read_super_error;
588         }
589
590         status = ocfs2_initialize_super(sb, bh, sector_size);
591         osb = OCFS2_SB(sb);
592         if (status < 0) {
593                 mlog_errno(status);
594                 goto read_super_error;
595         }
596         brelse(bh);
597         bh = NULL;
598         osb->s_mount_opt = parsed_options.mount_opt;
599         osb->s_atime_quantum = parsed_options.atime_quantum;
600         osb->preferred_slot = parsed_options.slot;
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), "%d", 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->mount_opt = 0;
752         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
753         mopt->slot = OCFS2_INVALID_SLOT;
754
755         if (!options) {
756                 status = 1;
757                 goto bail;
758         }
759
760         while ((p = strsep(&options, ",")) != NULL) {
761                 int token, option;
762                 substring_t args[MAX_OPT_ARGS];
763
764                 if (!*p)
765                         continue;
766
767                 token = match_token(p, tokens, args);
768                 switch (token) {
769                 case Opt_hb_local:
770                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
771                         break;
772                 case Opt_hb_none:
773                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
774                         break;
775                 case Opt_barrier:
776                         if (match_int(&args[0], &option)) {
777                                 status = 0;
778                                 goto bail;
779                         }
780                         if (option)
781                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
782                         else
783                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
784                         break;
785                 case Opt_intr:
786                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
787                         break;
788                 case Opt_nointr:
789                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
790                         break;
791                 case Opt_err_panic:
792                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
793                         break;
794                 case Opt_err_ro:
795                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
796                         break;
797                 case Opt_data_ordered:
798                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
799                         break;
800                 case Opt_data_writeback:
801                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
802                         break;
803                 case Opt_atime_quantum:
804                         if (match_int(&args[0], &option)) {
805                                 status = 0;
806                                 goto bail;
807                         }
808                         if (option >= 0)
809                                 mopt->atime_quantum = option;
810                         break;
811                 case Opt_slot:
812                         option = 0;
813                         if (match_int(&args[0], &option)) {
814                                 status = 0;
815                                 goto bail;
816                         }
817                         if (option)
818                                 mopt->slot = (s16)option;
819                         break;
820                 default:
821                         mlog(ML_ERROR,
822                              "Unrecognized mount option \"%s\" "
823                              "or missing value\n", p);
824                         status = 0;
825                         goto bail;
826                 }
827         }
828
829         status = 1;
830
831 bail:
832         mlog_exit(status);
833         return status;
834 }
835
836 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
837 {
838         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
839         unsigned long opts = osb->s_mount_opt;
840
841         if (opts & OCFS2_MOUNT_HB_LOCAL)
842                 seq_printf(s, ",_netdev,heartbeat=local");
843         else
844                 seq_printf(s, ",heartbeat=none");
845
846         if (opts & OCFS2_MOUNT_NOINTR)
847                 seq_printf(s, ",nointr");
848
849         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
850                 seq_printf(s, ",data=writeback");
851         else
852                 seq_printf(s, ",data=ordered");
853
854         if (opts & OCFS2_MOUNT_BARRIER)
855                 seq_printf(s, ",barrier=1");
856
857         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
858                 seq_printf(s, ",errors=panic");
859         else
860                 seq_printf(s, ",errors=remount-ro");
861
862         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
863                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
864
865         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
866                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
867
868         return 0;
869 }
870
871 static int __init ocfs2_init(void)
872 {
873         int status;
874
875         mlog_entry_void();
876
877         ocfs2_print_version();
878
879         status = init_ocfs2_uptodate_cache();
880         if (status < 0) {
881                 mlog_errno(status);
882                 goto leave;
883         }
884
885         status = ocfs2_initialize_mem_caches();
886         if (status < 0) {
887                 mlog_errno(status);
888                 goto leave;
889         }
890
891         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
892         if (!ocfs2_wq) {
893                 status = -ENOMEM;
894                 goto leave;
895         }
896
897         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
898         if (!ocfs2_debugfs_root) {
899                 status = -EFAULT;
900                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
901         }
902
903 leave:
904         if (status < 0) {
905                 ocfs2_free_mem_caches();
906                 exit_ocfs2_uptodate_cache();
907         }
908
909         mlog_exit(status);
910
911         if (status >= 0) {
912                 return register_filesystem(&ocfs2_fs_type);
913         } else
914                 return -1;
915 }
916
917 static void __exit ocfs2_exit(void)
918 {
919         mlog_entry_void();
920
921         if (ocfs2_wq) {
922                 flush_workqueue(ocfs2_wq);
923                 destroy_workqueue(ocfs2_wq);
924         }
925
926         debugfs_remove(ocfs2_debugfs_root);
927
928         ocfs2_free_mem_caches();
929
930         unregister_filesystem(&ocfs2_fs_type);
931
932         exit_ocfs2_uptodate_cache();
933
934         mlog_exit_void();
935 }
936
937 static void ocfs2_put_super(struct super_block *sb)
938 {
939         mlog_entry("(0x%p)\n", sb);
940
941         ocfs2_sync_blockdev(sb);
942         ocfs2_dismount_volume(sb, 0);
943
944         mlog_exit_void();
945 }
946
947 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
948 {
949         struct ocfs2_super *osb;
950         u32 numbits, freebits;
951         int status;
952         struct ocfs2_dinode *bm_lock;
953         struct buffer_head *bh = NULL;
954         struct inode *inode = NULL;
955
956         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
957
958         osb = OCFS2_SB(dentry->d_sb);
959
960         inode = ocfs2_get_system_file_inode(osb,
961                                             GLOBAL_BITMAP_SYSTEM_INODE,
962                                             OCFS2_INVALID_SLOT);
963         if (!inode) {
964                 mlog(ML_ERROR, "failed to get bitmap inode\n");
965                 status = -EIO;
966                 goto bail;
967         }
968
969         status = ocfs2_meta_lock(inode, &bh, 0);
970         if (status < 0) {
971                 mlog_errno(status);
972                 goto bail;
973         }
974
975         bm_lock = (struct ocfs2_dinode *) bh->b_data;
976
977         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
978         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
979
980         buf->f_type = OCFS2_SUPER_MAGIC;
981         buf->f_bsize = dentry->d_sb->s_blocksize;
982         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
983         buf->f_blocks = ((sector_t) numbits) *
984                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
985         buf->f_bfree = ((sector_t) freebits) *
986                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
987         buf->f_bavail = buf->f_bfree;
988         buf->f_files = numbits;
989         buf->f_ffree = freebits;
990
991         brelse(bh);
992
993         ocfs2_meta_unlock(inode, 0);
994         status = 0;
995 bail:
996         if (inode)
997                 iput(inode);
998
999         mlog_exit(status);
1000
1001         return status;
1002 }
1003
1004 static void ocfs2_inode_init_once(void *data,
1005                                   struct kmem_cache *cachep,
1006                                   unsigned long flags)
1007 {
1008         struct ocfs2_inode_info *oi = data;
1009
1010         oi->ip_flags = 0;
1011         oi->ip_open_count = 0;
1012         spin_lock_init(&oi->ip_lock);
1013         ocfs2_extent_map_init(&oi->vfs_inode);
1014         INIT_LIST_HEAD(&oi->ip_io_markers);
1015         oi->ip_created_trans = 0;
1016         oi->ip_last_trans = 0;
1017         oi->ip_dir_start_lookup = 0;
1018
1019         init_rwsem(&oi->ip_alloc_sem);
1020         mutex_init(&oi->ip_io_mutex);
1021
1022         oi->ip_blkno = 0ULL;
1023         oi->ip_clusters = 0;
1024
1025         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1026         ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
1027         ocfs2_lock_res_init_once(&oi->ip_data_lockres);
1028         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1029
1030         ocfs2_metadata_cache_init(&oi->vfs_inode);
1031
1032         inode_init_once(&oi->vfs_inode);
1033 }
1034
1035 static int ocfs2_initialize_mem_caches(void)
1036 {
1037         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1038                                        sizeof(struct ocfs2_inode_info),
1039                                        0,
1040                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1041                                                 SLAB_MEM_SPREAD),
1042                                        ocfs2_inode_init_once);
1043         if (!ocfs2_inode_cachep)
1044                 return -ENOMEM;
1045
1046         return 0;
1047 }
1048
1049 static void ocfs2_free_mem_caches(void)
1050 {
1051         if (ocfs2_inode_cachep)
1052                 kmem_cache_destroy(ocfs2_inode_cachep);
1053
1054         ocfs2_inode_cachep = NULL;
1055 }
1056
1057 static int ocfs2_get_sector(struct super_block *sb,
1058                             struct buffer_head **bh,
1059                             int block,
1060                             int sect_size)
1061 {
1062         if (!sb_set_blocksize(sb, sect_size)) {
1063                 mlog(ML_ERROR, "unable to set blocksize\n");
1064                 return -EIO;
1065         }
1066
1067         *bh = sb_getblk(sb, block);
1068         if (!*bh) {
1069                 mlog_errno(-EIO);
1070                 return -EIO;
1071         }
1072         lock_buffer(*bh);
1073         if (!buffer_dirty(*bh))
1074                 clear_buffer_uptodate(*bh);
1075         unlock_buffer(*bh);
1076         ll_rw_block(READ, 1, bh);
1077         wait_on_buffer(*bh);
1078         return 0;
1079 }
1080
1081 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1082 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1083 {
1084         int status;
1085
1086         /* XXX hold a ref on the node while mounte?  easy enough, if
1087          * desirable. */
1088         if (ocfs2_mount_local(osb))
1089                 osb->node_num = 0;
1090         else
1091                 osb->node_num = o2nm_this_node();
1092
1093         if (osb->node_num == O2NM_MAX_NODES) {
1094                 mlog(ML_ERROR, "could not find this host's node number\n");
1095                 status = -ENOENT;
1096                 goto bail;
1097         }
1098
1099         mlog(0, "I am node %d\n", osb->node_num);
1100
1101         status = 0;
1102 bail:
1103         return status;
1104 }
1105
1106 static int ocfs2_mount_volume(struct super_block *sb)
1107 {
1108         int status = 0;
1109         int unlock_super = 0;
1110         struct ocfs2_super *osb = OCFS2_SB(sb);
1111
1112         mlog_entry_void();
1113
1114         if (ocfs2_is_hard_readonly(osb))
1115                 goto leave;
1116
1117         status = ocfs2_fill_local_node_info(osb);
1118         if (status < 0) {
1119                 mlog_errno(status);
1120                 goto leave;
1121         }
1122
1123         status = ocfs2_register_hb_callbacks(osb);
1124         if (status < 0) {
1125                 mlog_errno(status);
1126                 goto leave;
1127         }
1128
1129         status = ocfs2_dlm_init(osb);
1130         if (status < 0) {
1131                 mlog_errno(status);
1132                 goto leave;
1133         }
1134
1135         /* requires vote_thread to be running. */
1136         status = ocfs2_register_net_handlers(osb);
1137         if (status < 0) {
1138                 mlog_errno(status);
1139                 goto leave;
1140         }
1141
1142         status = ocfs2_super_lock(osb, 1);
1143         if (status < 0) {
1144                 mlog_errno(status);
1145                 goto leave;
1146         }
1147         unlock_super = 1;
1148
1149         /* This will load up the node map and add ourselves to it. */
1150         status = ocfs2_find_slot(osb);
1151         if (status < 0) {
1152                 mlog_errno(status);
1153                 goto leave;
1154         }
1155
1156         ocfs2_populate_mounted_map(osb);
1157
1158         /* load all node-local system inodes */
1159         status = ocfs2_init_local_system_inodes(osb);
1160         if (status < 0) {
1161                 mlog_errno(status);
1162                 goto leave;
1163         }
1164
1165         status = ocfs2_check_volume(osb);
1166         if (status < 0) {
1167                 mlog_errno(status);
1168                 goto leave;
1169         }
1170
1171         status = ocfs2_truncate_log_init(osb);
1172         if (status < 0) {
1173                 mlog_errno(status);
1174                 goto leave;
1175         }
1176
1177         if (ocfs2_mount_local(osb))
1178                 goto leave;
1179
1180         /* This should be sent *after* we recovered our journal as it
1181          * will cause other nodes to unmark us as needing
1182          * recovery. However, we need to send it *before* dropping the
1183          * super block lock as otherwise their recovery threads might
1184          * try to clean us up while we're live! */
1185         status = ocfs2_request_mount_vote(osb);
1186         if (status < 0)
1187                 mlog_errno(status);
1188
1189 leave:
1190         if (unlock_super)
1191                 ocfs2_super_unlock(osb, 1);
1192
1193         mlog_exit(status);
1194         return status;
1195 }
1196
1197 /* we can't grab the goofy sem lock from inside wait_event, so we use
1198  * memory barriers to make sure that we'll see the null task before
1199  * being woken up */
1200 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1201 {
1202         mb();
1203         return osb->recovery_thread_task != NULL;
1204 }
1205
1206 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1207 {
1208         int tmp;
1209         struct ocfs2_super *osb = NULL;
1210         char nodestr[8];
1211
1212         mlog_entry("(0x%p)\n", sb);
1213
1214         BUG_ON(!sb);
1215         osb = OCFS2_SB(sb);
1216         BUG_ON(!osb);
1217
1218         ocfs2_shutdown_local_alloc(osb);
1219
1220         ocfs2_truncate_log_shutdown(osb);
1221
1222         /* disable any new recovery threads and wait for any currently
1223          * running ones to exit. Do this before setting the vol_state. */
1224         mutex_lock(&osb->recovery_lock);
1225         osb->disable_recovery = 1;
1226         mutex_unlock(&osb->recovery_lock);
1227         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1228
1229         /* At this point, we know that no more recovery threads can be
1230          * launched, so wait for any recovery completion work to
1231          * complete. */
1232         flush_workqueue(ocfs2_wq);
1233
1234         ocfs2_journal_shutdown(osb);
1235
1236         ocfs2_sync_blockdev(sb);
1237
1238         /* No dlm means we've failed during mount, so skip all the
1239          * steps which depended on that to complete. */
1240         if (osb->dlm) {
1241                 tmp = ocfs2_super_lock(osb, 1);
1242                 if (tmp < 0) {
1243                         mlog_errno(tmp);
1244                         return;
1245                 }
1246
1247                 tmp = ocfs2_request_umount_vote(osb);
1248                 if (tmp < 0)
1249                         mlog_errno(tmp);
1250         }
1251
1252         if (osb->slot_num != OCFS2_INVALID_SLOT)
1253                 ocfs2_put_slot(osb);
1254
1255         if (osb->dlm)
1256                 ocfs2_super_unlock(osb, 1);
1257
1258         ocfs2_release_system_inodes(osb);
1259
1260         if (osb->dlm) {
1261                 ocfs2_unregister_net_handlers(osb);
1262
1263                 ocfs2_dlm_shutdown(osb);
1264         }
1265
1266         ocfs2_clear_hb_callbacks(osb);
1267
1268         debugfs_remove(osb->osb_debug_root);
1269
1270         if (!mnt_err)
1271                 ocfs2_stop_heartbeat(osb);
1272
1273         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1274
1275         if (ocfs2_mount_local(osb))
1276                 snprintf(nodestr, sizeof(nodestr), "local");
1277         else
1278                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1279
1280         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1281                osb->dev_str, nodestr);
1282
1283         ocfs2_delete_osb(osb);
1284         kfree(osb);
1285         sb->s_dev = 0;
1286         sb->s_fs_info = NULL;
1287 }
1288
1289 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1290                                 unsigned uuid_bytes)
1291 {
1292         int i, ret;
1293         char *ptr;
1294
1295         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1296
1297         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1298         if (osb->uuid_str == NULL)
1299                 return -ENOMEM;
1300
1301         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1302                 /* print with null */
1303                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1304                 if (ret != 2) /* drop super cleans up */
1305                         return -EINVAL;
1306                 /* then only advance past the last char */
1307                 ptr += 2;
1308         }
1309
1310         return 0;
1311 }
1312
1313 static int ocfs2_initialize_super(struct super_block *sb,
1314                                   struct buffer_head *bh,
1315                                   int sector_size)
1316 {
1317         int status = 0;
1318         int i, cbits, bbits;
1319         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1320         struct inode *inode = NULL;
1321         struct buffer_head *bitmap_bh = NULL;
1322         struct ocfs2_journal *journal;
1323         __le32 uuid_net_key;
1324         struct ocfs2_super *osb;
1325
1326         mlog_entry_void();
1327
1328         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1329         if (!osb) {
1330                 status = -ENOMEM;
1331                 mlog_errno(status);
1332                 goto bail;
1333         }
1334
1335         sb->s_fs_info = osb;
1336         sb->s_op = &ocfs2_sops;
1337         sb->s_export_op = &ocfs2_export_ops;
1338         sb->s_time_gran = 1;
1339         sb->s_flags |= MS_NOATIME;
1340         /* this is needed to support O_LARGEFILE */
1341         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1342         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1343         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1344
1345         osb->sb = sb;
1346         /* Save off for ocfs2_rw_direct */
1347         osb->s_sectsize_bits = blksize_bits(sector_size);
1348         BUG_ON(!osb->s_sectsize_bits);
1349
1350         osb->net_response_ids = 0;
1351         spin_lock_init(&osb->net_response_lock);
1352         INIT_LIST_HEAD(&osb->net_response_list);
1353
1354         INIT_LIST_HEAD(&osb->osb_net_handlers);
1355         init_waitqueue_head(&osb->recovery_event);
1356         spin_lock_init(&osb->vote_task_lock);
1357         init_waitqueue_head(&osb->vote_event);
1358         osb->vote_work_sequence = 0;
1359         osb->vote_wake_sequence = 0;
1360         INIT_LIST_HEAD(&osb->blocked_lock_list);
1361         osb->blocked_lock_count = 0;
1362         INIT_LIST_HEAD(&osb->vote_list);
1363         spin_lock_init(&osb->osb_lock);
1364
1365         atomic_set(&osb->alloc_stats.moves, 0);
1366         atomic_set(&osb->alloc_stats.local_data, 0);
1367         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1368         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1369         atomic_set(&osb->alloc_stats.bg_extends, 0);
1370
1371         ocfs2_init_node_maps(osb);
1372
1373         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1374                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1375
1376         mutex_init(&osb->recovery_lock);
1377
1378         osb->disable_recovery = 0;
1379         osb->recovery_thread_task = NULL;
1380
1381         init_waitqueue_head(&osb->checkpoint_event);
1382         atomic_set(&osb->needs_checkpoint, 0);
1383
1384         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1385
1386         osb->node_num = O2NM_INVALID_NODE_NUM;
1387         osb->slot_num = OCFS2_INVALID_SLOT;
1388
1389         osb->local_alloc_state = OCFS2_LA_UNUSED;
1390         osb->local_alloc_bh = NULL;
1391
1392         ocfs2_setup_hb_callbacks(osb);
1393
1394         init_waitqueue_head(&osb->osb_mount_event);
1395
1396         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1397         if (!osb->vol_label) {
1398                 mlog(ML_ERROR, "unable to alloc vol label\n");
1399                 status = -ENOMEM;
1400                 goto bail;
1401         }
1402
1403         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1404         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1405                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1406                      osb->max_slots);
1407                 status = -EINVAL;
1408                 goto bail;
1409         }
1410         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1411
1412         init_waitqueue_head(&osb->osb_wipe_event);
1413         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1414                                         sizeof(*osb->osb_orphan_wipes),
1415                                         GFP_KERNEL);
1416         if (!osb->osb_orphan_wipes) {
1417                 status = -ENOMEM;
1418                 mlog_errno(status);
1419                 goto bail;
1420         }
1421
1422         osb->s_feature_compat =
1423                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1424         osb->s_feature_ro_compat =
1425                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1426         osb->s_feature_incompat =
1427                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1428
1429         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1430                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1431                      "optional features (%x).\n", i);
1432                 status = -EINVAL;
1433                 goto bail;
1434         }
1435         if (!(osb->sb->s_flags & MS_RDONLY) &&
1436             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1437                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1438                      "unsupported optional features (%x).\n", i);
1439                 status = -EINVAL;
1440                 goto bail;
1441         }
1442
1443         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1444
1445         /* FIXME
1446          * This should be done in ocfs2_journal_init(), but unknown
1447          * ordering issues will cause the filesystem to crash.
1448          * If anyone wants to figure out what part of the code
1449          * refers to osb->journal before ocfs2_journal_init() is run,
1450          * be my guest.
1451          */
1452         /* initialize our journal structure */
1453
1454         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1455         if (!journal) {
1456                 mlog(ML_ERROR, "unable to alloc journal\n");
1457                 status = -ENOMEM;
1458                 goto bail;
1459         }
1460         osb->journal = journal;
1461         journal->j_osb = osb;
1462
1463         atomic_set(&journal->j_num_trans, 0);
1464         init_rwsem(&journal->j_trans_barrier);
1465         init_waitqueue_head(&journal->j_checkpointed);
1466         spin_lock_init(&journal->j_lock);
1467         journal->j_trans_id = (unsigned long) 1;
1468         INIT_LIST_HEAD(&journal->j_la_cleanups);
1469         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1470         journal->j_state = OCFS2_JOURNAL_FREE;
1471
1472         /* get some pseudo constants for clustersize bits */
1473         osb->s_clustersize_bits =
1474                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1475         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1476         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1477
1478         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1479             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1480                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1481                      osb->s_clustersize);
1482                 status = -EINVAL;
1483                 goto bail;
1484         }
1485
1486         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1487             > (u32)~0UL) {
1488                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1489                      "what jbd can address in 32 bits.\n");
1490                 status = -EINVAL;
1491                 goto bail;
1492         }
1493
1494         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1495                                  sizeof(di->id2.i_super.s_uuid))) {
1496                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1497                 status = -ENOMEM;
1498                 goto bail;
1499         }
1500
1501         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1502         osb->net_key = le32_to_cpu(uuid_net_key);
1503
1504         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1505         osb->vol_label[63] = '\0';
1506         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1507         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1508         osb->first_cluster_group_blkno =
1509                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1510         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1511         mlog(0, "vol_label: %s\n", osb->vol_label);
1512         mlog(0, "uuid: %s\n", osb->uuid_str);
1513         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1514              (unsigned long long)osb->root_blkno,
1515              (unsigned long long)osb->system_dir_blkno);
1516
1517         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1518         if (!osb->osb_dlm_debug) {
1519                 status = -ENOMEM;
1520                 mlog_errno(status);
1521                 goto bail;
1522         }
1523
1524         atomic_set(&osb->vol_state, VOLUME_INIT);
1525
1526         /* load root, system_dir, and all global system inodes */
1527         status = ocfs2_init_global_system_inodes(osb);
1528         if (status < 0) {
1529                 mlog_errno(status);
1530                 goto bail;
1531         }
1532
1533         /*
1534          * global bitmap
1535          */
1536         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1537                                             OCFS2_INVALID_SLOT);
1538         if (!inode) {
1539                 status = -EINVAL;
1540                 mlog_errno(status);
1541                 goto bail;
1542         }
1543
1544         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1545
1546         /* We don't have a cluster lock on the bitmap here because
1547          * we're only interested in static information and the extra
1548          * complexity at mount time isn't worht it. Don't pass the
1549          * inode in to the read function though as we don't want it to
1550          * be put in the cache. */
1551         status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1552                                   NULL);
1553         iput(inode);
1554         if (status < 0) {
1555                 mlog_errno(status);
1556                 goto bail;
1557         }
1558
1559         di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1560         osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1561         brelse(bitmap_bh);
1562         mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1563              (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1564
1565         status = ocfs2_init_slot_info(osb);
1566         if (status < 0) {
1567                 mlog_errno(status);
1568                 goto bail;
1569         }
1570
1571 bail:
1572         mlog_exit(status);
1573         return status;
1574 }
1575
1576 /*
1577  * will return: -EAGAIN if it is ok to keep searching for superblocks
1578  *              -EINVAL if there is a bad superblock
1579  *              0 on success
1580  */
1581 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1582                                struct buffer_head *bh,
1583                                u32 blksz)
1584 {
1585         int status = -EAGAIN;
1586
1587         mlog_entry_void();
1588
1589         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1590                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1591                 status = -EINVAL;
1592                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1593                         mlog(ML_ERROR, "found superblock with incorrect block "
1594                              "size: found %u, should be %u\n",
1595                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1596                                blksz);
1597                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1598                            OCFS2_MAJOR_REV_LEVEL ||
1599                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1600                            OCFS2_MINOR_REV_LEVEL) {
1601                         mlog(ML_ERROR, "found superblock with bad version: "
1602                              "found %u.%u, should be %u.%u\n",
1603                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1604                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1605                              OCFS2_MAJOR_REV_LEVEL,
1606                              OCFS2_MINOR_REV_LEVEL);
1607                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1608                         mlog(ML_ERROR, "bad block number on superblock: "
1609                              "found %llu, should be %llu\n",
1610                              (unsigned long long)le64_to_cpu(di->i_blkno),
1611                              (unsigned long long)bh->b_blocknr);
1612                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1613                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1614                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1615                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1616                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1617                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1618                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1619                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1620                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1621                         mlog(ML_ERROR,
1622                              "Superblock slots found greater than file system "
1623                              "maximum: found %u, max %u\n",
1624                              le16_to_cpu(di->id2.i_super.s_max_slots),
1625                              OCFS2_MAX_SLOTS);
1626                 } else {
1627                         /* found it! */
1628                         status = 0;
1629                 }
1630         }
1631
1632         mlog_exit(status);
1633         return status;
1634 }
1635
1636 static int ocfs2_check_volume(struct ocfs2_super *osb)
1637 {
1638         int status = 0;
1639         int dirty;
1640         int local;
1641         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1642                                                   * recover
1643                                                   * ourselves. */
1644
1645         mlog_entry_void();
1646
1647         /* Init our journal object. */
1648         status = ocfs2_journal_init(osb->journal, &dirty);
1649         if (status < 0) {
1650                 mlog(ML_ERROR, "Could not initialize journal!\n");
1651                 goto finally;
1652         }
1653
1654         /* If the journal was unmounted cleanly then we don't want to
1655          * recover anything. Otherwise, journal_load will do that
1656          * dirty work for us :) */
1657         if (!dirty) {
1658                 status = ocfs2_journal_wipe(osb->journal, 0);
1659                 if (status < 0) {
1660                         mlog_errno(status);
1661                         goto finally;
1662                 }
1663         } else {
1664                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1665                      "recovering volume.\n");
1666         }
1667
1668         local = ocfs2_mount_local(osb);
1669
1670         /* will play back anything left in the journal. */
1671         ocfs2_journal_load(osb->journal, local);
1672
1673         if (dirty) {
1674                 /* recover my local alloc if we didn't unmount cleanly. */
1675                 status = ocfs2_begin_local_alloc_recovery(osb,
1676                                                           osb->slot_num,
1677                                                           &local_alloc);
1678                 if (status < 0) {
1679                         mlog_errno(status);
1680                         goto finally;
1681                 }
1682                 /* we complete the recovery process after we've marked
1683                  * ourselves as mounted. */
1684         }
1685
1686         mlog(0, "Journal loaded.\n");
1687
1688         status = ocfs2_load_local_alloc(osb);
1689         if (status < 0) {
1690                 mlog_errno(status);
1691                 goto finally;
1692         }
1693
1694         if (dirty) {
1695                 /* Recovery will be completed after we've mounted the
1696                  * rest of the volume. */
1697                 osb->dirty = 1;
1698                 osb->local_alloc_copy = local_alloc;
1699                 local_alloc = NULL;
1700         }
1701
1702         /* go through each journal, trylock it and if you get the
1703          * lock, and it's marked as dirty, set the bit in the recover
1704          * map and launch a recovery thread for it. */
1705         status = ocfs2_mark_dead_nodes(osb);
1706         if (status < 0)
1707                 mlog_errno(status);
1708
1709 finally:
1710         if (local_alloc)
1711                 kfree(local_alloc);
1712
1713         mlog_exit(status);
1714         return status;
1715 }
1716
1717 /*
1718  * The routine gets called from dismount or close whenever a dismount on
1719  * volume is requested and the osb open count becomes 1.
1720  * It will remove the osb from the global list and also free up all the
1721  * initialized resources and fileobject.
1722  */
1723 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1724 {
1725         mlog_entry_void();
1726
1727         /* This function assumes that the caller has the main osb resource */
1728
1729         if (osb->slot_info)
1730                 ocfs2_free_slot_info(osb->slot_info);
1731
1732         kfree(osb->osb_orphan_wipes);
1733         /* FIXME
1734          * This belongs in journal shutdown, but because we have to
1735          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1736          * we free it here.
1737          */
1738         kfree(osb->journal);
1739         if (osb->local_alloc_copy)
1740                 kfree(osb->local_alloc_copy);
1741         kfree(osb->uuid_str);
1742         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1743         memset(osb, 0, sizeof(struct ocfs2_super));
1744
1745         mlog_exit_void();
1746 }
1747
1748 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1749  * panic(). We do not support continue-on-error operation. */
1750 static void ocfs2_handle_error(struct super_block *sb)
1751 {
1752         struct ocfs2_super *osb = OCFS2_SB(sb);
1753
1754         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1755                 panic("OCFS2: (device %s): panic forced after error\n",
1756                       sb->s_id);
1757
1758         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1759
1760         if (sb->s_flags & MS_RDONLY &&
1761             (ocfs2_is_soft_readonly(osb) ||
1762              ocfs2_is_hard_readonly(osb)))
1763                 return;
1764
1765         printk(KERN_CRIT "File system is now read-only due to the potential "
1766                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1767                "system is unmounted.\n");
1768         sb->s_flags |= MS_RDONLY;
1769         ocfs2_set_ro_flag(osb, 0);
1770 }
1771
1772 static char error_buf[1024];
1773
1774 void __ocfs2_error(struct super_block *sb,
1775                    const char *function,
1776                    const char *fmt, ...)
1777 {
1778         va_list args;
1779
1780         va_start(args, fmt);
1781         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1782         va_end(args);
1783
1784         /* Not using mlog here because we want to show the actual
1785          * function the error came from. */
1786         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1787                sb->s_id, function, error_buf);
1788
1789         ocfs2_handle_error(sb);
1790 }
1791
1792 /* Handle critical errors. This is intentionally more drastic than
1793  * ocfs2_handle_error, so we only use for things like journal errors,
1794  * etc. */
1795 void __ocfs2_abort(struct super_block* sb,
1796                    const char *function,
1797                    const char *fmt, ...)
1798 {
1799         va_list args;
1800
1801         va_start(args, fmt);
1802         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1803         va_end(args);
1804
1805         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1806                sb->s_id, function, error_buf);
1807
1808         /* We don't have the cluster support yet to go straight to
1809          * hard readonly in here. Until then, we want to keep
1810          * ocfs2_abort() so that we can at least mark critical
1811          * errors.
1812          *
1813          * TODO: This should abort the journal and alert other nodes
1814          * that our slot needs recovery. */
1815
1816         /* Force a panic(). This stinks, but it's better than letting
1817          * things continue without having a proper hard readonly
1818          * here. */
1819         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1820         ocfs2_handle_error(sb);
1821 }
1822
1823 module_init(ocfs2_init);
1824 module_exit(ocfs2_exit);