Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / fs / ocfs2 / super.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * super.c
5  *
6  * load/unload driver, mount/dismount volumes
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25
26 #include <linux/module.h>
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
42
43 #include <cluster/nodemanager.h>
44
45 #define MLOG_MASK_PREFIX ML_SUPER
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 /* this should be the only file to include a version 1 header */
51 #include "ocfs1_fs_compat.h"
52
53 #include "alloc.h"
54 #include "dlmglue.h"
55 #include "export.h"
56 #include "extent_map.h"
57 #include "heartbeat.h"
58 #include "inode.h"
59 #include "journal.h"
60 #include "localalloc.h"
61 #include "namei.h"
62 #include "slot_map.h"
63 #include "super.h"
64 #include "sysfile.h"
65 #include "uptodate.h"
66 #include "ver.h"
67 #include "vote.h"
68
69 #include "buffer_head_io.h"
70
71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
72
73 /* OCFS2 needs to schedule several differnt types of work which
74  * require cluster locking, disk I/O, recovery waits, etc. Since these
75  * types of work tend to be heavy we avoid using the kernel events
76  * workqueue and schedule on our own. */
77 struct workqueue_struct *ocfs2_wq = NULL;
78
79 static struct dentry *ocfs2_debugfs_root = NULL;
80
81 MODULE_AUTHOR("Oracle");
82 MODULE_LICENSE("GPL");
83
84 static int ocfs2_parse_options(struct super_block *sb, char *options,
85                                unsigned long *mount_opt, int is_remount);
86 static void ocfs2_put_super(struct super_block *sb);
87 static int ocfs2_mount_volume(struct super_block *sb);
88 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
89 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
90 static int ocfs2_initialize_mem_caches(void);
91 static void ocfs2_free_mem_caches(void);
92 static void ocfs2_delete_osb(struct ocfs2_super *osb);
93
94 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
95
96 static int ocfs2_sync_fs(struct super_block *sb, int wait);
97
98 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
99 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
100 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
101 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
102 static int ocfs2_check_volume(struct ocfs2_super *osb);
103 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
104                                struct buffer_head *bh,
105                                u32 sectsize);
106 static int ocfs2_initialize_super(struct super_block *sb,
107                                   struct buffer_head *bh,
108                                   int sector_size);
109 static int ocfs2_get_sector(struct super_block *sb,
110                             struct buffer_head **bh,
111                             int block,
112                             int sect_size);
113 static void ocfs2_write_super(struct super_block *sb);
114 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
115 static void ocfs2_destroy_inode(struct inode *inode);
116
117 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
118
119 static const struct super_operations ocfs2_sops = {
120         .statfs         = ocfs2_statfs,
121         .alloc_inode    = ocfs2_alloc_inode,
122         .destroy_inode  = ocfs2_destroy_inode,
123         .drop_inode     = ocfs2_drop_inode,
124         .clear_inode    = ocfs2_clear_inode,
125         .delete_inode   = ocfs2_delete_inode,
126         .sync_fs        = ocfs2_sync_fs,
127         .write_super    = ocfs2_write_super,
128         .put_super      = ocfs2_put_super,
129         .remount_fs     = ocfs2_remount,
130 };
131
132 enum {
133         Opt_barrier,
134         Opt_err_panic,
135         Opt_err_ro,
136         Opt_intr,
137         Opt_nointr,
138         Opt_hb_none,
139         Opt_hb_local,
140         Opt_data_ordered,
141         Opt_data_writeback,
142         Opt_atime_quantum,
143         Opt_err,
144 };
145
146 static match_table_t tokens = {
147         {Opt_barrier, "barrier=%u"},
148         {Opt_err_panic, "errors=panic"},
149         {Opt_err_ro, "errors=remount-ro"},
150         {Opt_intr, "intr"},
151         {Opt_nointr, "nointr"},
152         {Opt_hb_none, OCFS2_HB_NONE},
153         {Opt_hb_local, OCFS2_HB_LOCAL},
154         {Opt_data_ordered, "data=ordered"},
155         {Opt_data_writeback, "data=writeback"},
156         {Opt_atime_quantum, "atime_quantum=%u"},
157         {Opt_err, NULL}
158 };
159
160 /*
161  * write_super and sync_fs ripped right out of ext3.
162  */
163 static void ocfs2_write_super(struct super_block *sb)
164 {
165         if (mutex_trylock(&sb->s_lock) != 0)
166                 BUG();
167         sb->s_dirt = 0;
168 }
169
170 static int ocfs2_sync_fs(struct super_block *sb, int wait)
171 {
172         int status = 0;
173         tid_t target;
174         struct ocfs2_super *osb = OCFS2_SB(sb);
175
176         sb->s_dirt = 0;
177
178         if (ocfs2_is_hard_readonly(osb))
179                 return -EROFS;
180
181         if (wait) {
182                 status = ocfs2_flush_truncate_log(osb);
183                 if (status < 0)
184                         mlog_errno(status);
185         } else {
186                 ocfs2_schedule_truncate_log_flush(osb, 0);
187         }
188
189         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
190                 if (wait)
191                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
192                                         target);
193         }
194         return 0;
195 }
196
197 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
198 {
199         struct inode *new = NULL;
200         int status = 0;
201         int i;
202
203         mlog_entry_void();
204
205         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
206         if (IS_ERR(new)) {
207                 status = PTR_ERR(new);
208                 mlog_errno(status);
209                 goto bail;
210         }
211         osb->root_inode = new;
212
213         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
214         if (IS_ERR(new)) {
215                 status = PTR_ERR(new);
216                 mlog_errno(status);
217                 goto bail;
218         }
219         osb->sys_root_inode = new;
220
221         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
222              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
223                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
224                 if (!new) {
225                         ocfs2_release_system_inodes(osb);
226                         status = -EINVAL;
227                         mlog_errno(status);
228                         /* FIXME: Should ERROR_RO_FS */
229                         mlog(ML_ERROR, "Unable to load system inode %d, "
230                              "possibly corrupt fs?", i);
231                         goto bail;
232                 }
233                 // the array now has one ref, so drop this one
234                 iput(new);
235         }
236
237 bail:
238         mlog_exit(status);
239         return status;
240 }
241
242 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
243 {
244         struct inode *new = NULL;
245         int status = 0;
246         int i;
247
248         mlog_entry_void();
249
250         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
251              i < NUM_SYSTEM_INODES;
252              i++) {
253                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
254                 if (!new) {
255                         ocfs2_release_system_inodes(osb);
256                         status = -EINVAL;
257                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
258                              status, i, osb->slot_num);
259                         goto bail;
260                 }
261                 /* the array now has one ref, so drop this one */
262                 iput(new);
263         }
264
265 bail:
266         mlog_exit(status);
267         return status;
268 }
269
270 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
271 {
272         int status = 0, i;
273         struct inode *inode;
274
275         mlog_entry_void();
276
277         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
278                 inode = osb->system_inodes[i];
279                 if (inode) {
280                         iput(inode);
281                         osb->system_inodes[i] = NULL;
282                 }
283         }
284
285         inode = osb->sys_root_inode;
286         if (inode) {
287                 iput(inode);
288                 osb->sys_root_inode = NULL;
289         }
290
291         inode = osb->root_inode;
292         if (inode) {
293                 iput(inode);
294                 osb->root_inode = NULL;
295         }
296
297         mlog_exit(status);
298         return status;
299 }
300
301 /* We're allocating fs objects, use GFP_NOFS */
302 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
303 {
304         struct ocfs2_inode_info *oi;
305
306         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
307         if (!oi)
308                 return NULL;
309
310         return &oi->vfs_inode;
311 }
312
313 static void ocfs2_destroy_inode(struct inode *inode)
314 {
315         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
316 }
317
318 /* From xfs_super.c:xfs_max_file_offset
319  * Copyright (c) 2000-2004 Silicon Graphics, Inc.
320  */
321 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
322 {
323         unsigned int pagefactor = 1;
324         unsigned int bitshift = BITS_PER_LONG - 1;
325
326         /* Figure out maximum filesize, on Linux this can depend on
327          * the filesystem blocksize (on 32 bit platforms).
328          * __block_prepare_write does this in an [unsigned] long...
329          *      page->index << (PAGE_CACHE_SHIFT - bbits)
330          * So, for page sized blocks (4K on 32 bit platforms),
331          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
332          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
333          * but for smaller blocksizes it is less (bbits = log2 bsize).
334          * Note1: get_block_t takes a long (implicit cast from above)
335          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
336          * can optionally convert the [unsigned] long from above into
337          * an [unsigned] long long.
338          */
339
340 #if BITS_PER_LONG == 32
341 # if defined(CONFIG_LBD)
342         BUILD_BUG_ON(sizeof(sector_t) != 8);
343         pagefactor = PAGE_CACHE_SIZE;
344         bitshift = BITS_PER_LONG;
345 # else
346         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
347 # endif
348 #endif
349
350         return (((unsigned long long)pagefactor) << bitshift) - 1;
351 }
352
353 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
354 {
355         int incompat_features;
356         int ret = 0;
357         unsigned long parsed_options;
358         struct ocfs2_super *osb = OCFS2_SB(sb);
359
360         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
361                 ret = -EINVAL;
362                 goto out;
363         }
364
365         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
366             (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
367                 ret = -EINVAL;
368                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
369                 goto out;
370         }
371
372         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
373             (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
374                 ret = -EINVAL;
375                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
376                 goto out;
377         }
378
379         /* We're going to/from readonly mode. */
380         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
381                 /* Lock here so the check of HARD_RO and the potential
382                  * setting of SOFT_RO is atomic. */
383                 spin_lock(&osb->osb_lock);
384                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
385                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
386                         ret = -EROFS;
387                         goto unlock_osb;
388                 }
389
390                 if (*flags & MS_RDONLY) {
391                         mlog(0, "Going to ro mode.\n");
392                         sb->s_flags |= MS_RDONLY;
393                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
394                 } else {
395                         mlog(0, "Making ro filesystem writeable.\n");
396
397                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
398                                 mlog(ML_ERROR, "Cannot remount RDWR "
399                                      "filesystem due to previous errors.\n");
400                                 ret = -EROFS;
401                                 goto unlock_osb;
402                         }
403                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
404                         if (incompat_features) {
405                                 mlog(ML_ERROR, "Cannot remount RDWR because "
406                                      "of unsupported optional features "
407                                      "(%x).\n", incompat_features);
408                                 ret = -EINVAL;
409                                 goto unlock_osb;
410                         }
411                         sb->s_flags &= ~MS_RDONLY;
412                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
413                 }
414 unlock_osb:
415                 spin_unlock(&osb->osb_lock);
416         }
417
418         if (!ret) {
419                 if (!ocfs2_is_hard_readonly(osb))
420                         ocfs2_set_journal_params(osb);
421
422                 /* Only save off the new mount options in case of a successful
423                  * remount. */
424                 osb->s_mount_opt = parsed_options;
425         }
426 out:
427         return ret;
428 }
429
430 static int ocfs2_sb_probe(struct super_block *sb,
431                           struct buffer_head **bh,
432                           int *sector_size)
433 {
434         int status = 0, tmpstat;
435         struct ocfs1_vol_disk_hdr *hdr;
436         struct ocfs2_dinode *di;
437         int blksize;
438
439         *bh = NULL;
440
441         /* may be > 512 */
442         *sector_size = bdev_hardsect_size(sb->s_bdev);
443         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
444                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
445                      *sector_size, OCFS2_MAX_BLOCKSIZE);
446                 status = -EINVAL;
447                 goto bail;
448         }
449
450         /* Can this really happen? */
451         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
452                 *sector_size = OCFS2_MIN_BLOCKSIZE;
453
454         /* check block zero for old format */
455         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
456         if (status < 0) {
457                 mlog_errno(status);
458                 goto bail;
459         }
460         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
461         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
462                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
463                      hdr->major_version, hdr->minor_version);
464                 status = -EINVAL;
465         }
466         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
467                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
468                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
469                      hdr->signature);
470                 status = -EINVAL;
471         }
472         brelse(*bh);
473         *bh = NULL;
474         if (status < 0) {
475                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
476                      "upgraded before mounting with ocfs v2\n");
477                 goto bail;
478         }
479
480         /*
481          * Now check at magic offset for 512, 1024, 2048, 4096
482          * blocksizes.  4096 is the maximum blocksize because it is
483          * the minimum clustersize.
484          */
485         status = -EINVAL;
486         for (blksize = *sector_size;
487              blksize <= OCFS2_MAX_BLOCKSIZE;
488              blksize <<= 1) {
489                 tmpstat = ocfs2_get_sector(sb, bh,
490                                            OCFS2_SUPER_BLOCK_BLKNO,
491                                            blksize);
492                 if (tmpstat < 0) {
493                         status = tmpstat;
494                         mlog_errno(status);
495                         goto bail;
496                 }
497                 di = (struct ocfs2_dinode *) (*bh)->b_data;
498                 status = ocfs2_verify_volume(di, *bh, blksize);
499                 if (status >= 0)
500                         goto bail;
501                 brelse(*bh);
502                 *bh = NULL;
503                 if (status != -EAGAIN)
504                         break;
505         }
506
507 bail:
508         return status;
509 }
510
511 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
512 {
513         if (ocfs2_mount_local(osb)) {
514                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
515                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
516                              "mounted device.\n");
517                         return -EINVAL;
518                 }
519         }
520
521         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
522                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
523                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
524                              "a read-write clustered device.\n");
525                         return -EINVAL;
526                 }
527         }
528
529         return 0;
530 }
531
532 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
533 {
534         struct dentry *root;
535         int status, sector_size;
536         unsigned long parsed_opt;
537         struct inode *inode = NULL;
538         struct ocfs2_super *osb = NULL;
539         struct buffer_head *bh = NULL;
540         char nodestr[8];
541
542         mlog_entry("%p, %p, %i", sb, data, silent);
543
544         if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
545                 status = -EINVAL;
546                 goto read_super_error;
547         }
548
549         /* for now we only have one cluster/node, make sure we see it
550          * in the heartbeat universe */
551         if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
552                 if (!o2hb_check_local_node_heartbeating()) {
553                         status = -EINVAL;
554                         goto read_super_error;
555                 }
556         }
557
558         /* probe for superblock */
559         status = ocfs2_sb_probe(sb, &bh, &sector_size);
560         if (status < 0) {
561                 mlog(ML_ERROR, "superblock probe failed!\n");
562                 goto read_super_error;
563         }
564
565         status = ocfs2_initialize_super(sb, bh, sector_size);
566         osb = OCFS2_SB(sb);
567         if (status < 0) {
568                 mlog_errno(status);
569                 goto read_super_error;
570         }
571         brelse(bh);
572         bh = NULL;
573         osb->s_mount_opt = parsed_opt;
574
575         sb->s_magic = OCFS2_SUPER_MAGIC;
576
577         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
578          * heartbeat=none */
579         if (bdev_read_only(sb->s_bdev)) {
580                 if (!(sb->s_flags & MS_RDONLY)) {
581                         status = -EACCES;
582                         mlog(ML_ERROR, "Readonly device detected but readonly "
583                              "mount was not specified.\n");
584                         goto read_super_error;
585                 }
586
587                 /* You should not be able to start a local heartbeat
588                  * on a readonly device. */
589                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
590                         status = -EROFS;
591                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
592                              "device.\n");
593                         goto read_super_error;
594                 }
595
596                 status = ocfs2_check_journals_nolocks(osb);
597                 if (status < 0) {
598                         if (status == -EROFS)
599                                 mlog(ML_ERROR, "Recovery required on readonly "
600                                      "file system, but write access is "
601                                      "unavailable.\n");
602                         else
603                                 mlog_errno(status);                     
604                         goto read_super_error;
605                 }
606
607                 ocfs2_set_ro_flag(osb, 1);
608
609                 printk(KERN_NOTICE "Readonly device detected. No cluster "
610                        "services will be utilized for this mount. Recovery "
611                        "will be skipped.\n");
612         }
613
614         if (!ocfs2_is_hard_readonly(osb)) {
615                 if (sb->s_flags & MS_RDONLY)
616                         ocfs2_set_ro_flag(osb, 0);
617         }
618
619         status = ocfs2_verify_heartbeat(osb);
620         if (status < 0) {
621                 mlog_errno(status);
622                 goto read_super_error;
623         }
624
625         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
626                                                  ocfs2_debugfs_root);
627         if (!osb->osb_debug_root) {
628                 status = -EINVAL;
629                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
630                 goto read_super_error;
631         }
632
633         status = ocfs2_mount_volume(sb);
634         if (osb->root_inode)
635                 inode = igrab(osb->root_inode);
636
637         if (status < 0)
638                 goto read_super_error;
639
640         if (!inode) {
641                 status = -EIO;
642                 mlog_errno(status);
643                 goto read_super_error;
644         }
645
646         root = d_alloc_root(inode);
647         if (!root) {
648                 status = -ENOMEM;
649                 mlog_errno(status);
650                 goto read_super_error;
651         }
652
653         sb->s_root = root;
654
655         ocfs2_complete_mount_recovery(osb);
656
657         if (ocfs2_mount_local(osb))
658                 snprintf(nodestr, sizeof(nodestr), "local");
659         else
660                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
661
662         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
663                "with %s data mode.\n",
664                osb->dev_str, nodestr, osb->slot_num,
665                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
666                "ordered");
667
668         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
669         wake_up(&osb->osb_mount_event);
670
671         mlog_exit(status);
672         return status;
673
674 read_super_error:
675         if (bh != NULL)
676                 brelse(bh);
677
678         if (inode)
679                 iput(inode);
680
681         if (osb) {
682                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
683                 wake_up(&osb->osb_mount_event);
684                 ocfs2_dismount_volume(sb, 1);
685         }
686
687         mlog_exit(status);
688         return status;
689 }
690
691 static int ocfs2_get_sb(struct file_system_type *fs_type,
692                         int flags,
693                         const char *dev_name,
694                         void *data,
695                         struct vfsmount *mnt)
696 {
697         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
698                            mnt);
699 }
700
701 static struct file_system_type ocfs2_fs_type = {
702         .owner          = THIS_MODULE,
703         .name           = "ocfs2",
704         .get_sb         = ocfs2_get_sb, /* is this called when we mount
705                                         * the fs? */
706         .kill_sb        = kill_block_super, /* set to the generic one
707                                              * right now, but do we
708                                              * need to change that? */
709         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
710         .next           = NULL
711 };
712
713 static int ocfs2_parse_options(struct super_block *sb,
714                                char *options,
715                                unsigned long *mount_opt,
716                                int is_remount)
717 {
718         int status;
719         char *p;
720
721         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
722                    options ? options : "(none)");
723
724         *mount_opt = 0;
725
726         if (!options) {
727                 status = 1;
728                 goto bail;
729         }
730
731         while ((p = strsep(&options, ",")) != NULL) {
732                 int token, option;
733                 substring_t args[MAX_OPT_ARGS];
734                 struct ocfs2_super * osb = OCFS2_SB(sb);
735
736                 if (!*p)
737                         continue;
738
739                 token = match_token(p, tokens, args);
740                 switch (token) {
741                 case Opt_hb_local:
742                         *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
743                         break;
744                 case Opt_hb_none:
745                         *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
746                         break;
747                 case Opt_barrier:
748                         if (match_int(&args[0], &option)) {
749                                 status = 0;
750                                 goto bail;
751                         }
752                         if (option)
753                                 *mount_opt |= OCFS2_MOUNT_BARRIER;
754                         else
755                                 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
756                         break;
757                 case Opt_intr:
758                         *mount_opt &= ~OCFS2_MOUNT_NOINTR;
759                         break;
760                 case Opt_nointr:
761                         *mount_opt |= OCFS2_MOUNT_NOINTR;
762                         break;
763                 case Opt_err_panic:
764                         *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
765                         break;
766                 case Opt_err_ro:
767                         *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
768                         break;
769                 case Opt_data_ordered:
770                         *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
771                         break;
772                 case Opt_data_writeback:
773                         *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
774                         break;
775                 case Opt_atime_quantum:
776                         if (match_int(&args[0], &option)) {
777                                 status = 0;
778                                 goto bail;
779                         }
780                         if (option >= 0)
781                                 osb->s_atime_quantum = option;
782                         else
783                                 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
784                         break;
785                 default:
786                         mlog(ML_ERROR,
787                              "Unrecognized mount option \"%s\" "
788                              "or missing value\n", p);
789                         status = 0;
790                         goto bail;
791                 }
792         }
793
794         status = 1;
795
796 bail:
797         mlog_exit(status);
798         return status;
799 }
800
801 static int __init ocfs2_init(void)
802 {
803         int status;
804
805         mlog_entry_void();
806
807         ocfs2_print_version();
808
809         status = init_ocfs2_uptodate_cache();
810         if (status < 0) {
811                 mlog_errno(status);
812                 goto leave;
813         }
814
815         status = ocfs2_initialize_mem_caches();
816         if (status < 0) {
817                 mlog_errno(status);
818                 goto leave;
819         }
820
821         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
822         if (!ocfs2_wq) {
823                 status = -ENOMEM;
824                 goto leave;
825         }
826
827         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
828         if (!ocfs2_debugfs_root) {
829                 status = -EFAULT;
830                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
831         }
832
833 leave:
834         if (status < 0) {
835                 ocfs2_free_mem_caches();
836                 exit_ocfs2_uptodate_cache();
837         }
838
839         mlog_exit(status);
840
841         if (status >= 0) {
842                 return register_filesystem(&ocfs2_fs_type);
843         } else
844                 return -1;
845 }
846
847 static void __exit ocfs2_exit(void)
848 {
849         mlog_entry_void();
850
851         if (ocfs2_wq) {
852                 flush_workqueue(ocfs2_wq);
853                 destroy_workqueue(ocfs2_wq);
854         }
855
856         debugfs_remove(ocfs2_debugfs_root);
857
858         ocfs2_free_mem_caches();
859
860         unregister_filesystem(&ocfs2_fs_type);
861
862         exit_ocfs2_uptodate_cache();
863
864         mlog_exit_void();
865 }
866
867 static void ocfs2_put_super(struct super_block *sb)
868 {
869         mlog_entry("(0x%p)\n", sb);
870
871         ocfs2_sync_blockdev(sb);
872         ocfs2_dismount_volume(sb, 0);
873
874         mlog_exit_void();
875 }
876
877 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
878 {
879         struct ocfs2_super *osb;
880         u32 numbits, freebits;
881         int status;
882         struct ocfs2_dinode *bm_lock;
883         struct buffer_head *bh = NULL;
884         struct inode *inode = NULL;
885
886         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
887
888         osb = OCFS2_SB(dentry->d_sb);
889
890         inode = ocfs2_get_system_file_inode(osb,
891                                             GLOBAL_BITMAP_SYSTEM_INODE,
892                                             OCFS2_INVALID_SLOT);
893         if (!inode) {
894                 mlog(ML_ERROR, "failed to get bitmap inode\n");
895                 status = -EIO;
896                 goto bail;
897         }
898
899         status = ocfs2_meta_lock(inode, &bh, 0);
900         if (status < 0) {
901                 mlog_errno(status);
902                 goto bail;
903         }
904
905         bm_lock = (struct ocfs2_dinode *) bh->b_data;
906
907         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
908         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
909
910         buf->f_type = OCFS2_SUPER_MAGIC;
911         buf->f_bsize = dentry->d_sb->s_blocksize;
912         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
913         buf->f_blocks = ((sector_t) numbits) *
914                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
915         buf->f_bfree = ((sector_t) freebits) *
916                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
917         buf->f_bavail = buf->f_bfree;
918         buf->f_files = numbits;
919         buf->f_ffree = freebits;
920
921         brelse(bh);
922
923         ocfs2_meta_unlock(inode, 0);
924         status = 0;
925 bail:
926         if (inode)
927                 iput(inode);
928
929         mlog_exit(status);
930
931         return status;
932 }
933
934 static void ocfs2_inode_init_once(void *data,
935                                   struct kmem_cache *cachep,
936                                   unsigned long flags)
937 {
938         struct ocfs2_inode_info *oi = data;
939
940         if (flags & SLAB_CTOR_CONSTRUCTOR) {
941                 oi->ip_flags = 0;
942                 oi->ip_open_count = 0;
943                 spin_lock_init(&oi->ip_lock);
944                 ocfs2_extent_map_init(&oi->vfs_inode);
945                 INIT_LIST_HEAD(&oi->ip_io_markers);
946                 oi->ip_created_trans = 0;
947                 oi->ip_last_trans = 0;
948                 oi->ip_dir_start_lookup = 0;
949
950                 init_rwsem(&oi->ip_alloc_sem);
951                 mutex_init(&oi->ip_io_mutex);
952
953                 oi->ip_blkno = 0ULL;
954                 oi->ip_clusters = 0;
955
956                 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
957                 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
958                 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
959                 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
960
961                 ocfs2_metadata_cache_init(&oi->vfs_inode);
962
963                 inode_init_once(&oi->vfs_inode);
964         }
965 }
966
967 static int ocfs2_initialize_mem_caches(void)
968 {
969         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
970                                        sizeof(struct ocfs2_inode_info),
971                                        0,
972                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
973                                                 SLAB_MEM_SPREAD),
974                                        ocfs2_inode_init_once, NULL);
975         if (!ocfs2_inode_cachep)
976                 return -ENOMEM;
977
978         return 0;
979 }
980
981 static void ocfs2_free_mem_caches(void)
982 {
983         if (ocfs2_inode_cachep)
984                 kmem_cache_destroy(ocfs2_inode_cachep);
985
986         ocfs2_inode_cachep = NULL;
987 }
988
989 static int ocfs2_get_sector(struct super_block *sb,
990                             struct buffer_head **bh,
991                             int block,
992                             int sect_size)
993 {
994         if (!sb_set_blocksize(sb, sect_size)) {
995                 mlog(ML_ERROR, "unable to set blocksize\n");
996                 return -EIO;
997         }
998
999         *bh = sb_getblk(sb, block);
1000         if (!*bh) {
1001                 mlog_errno(-EIO);
1002                 return -EIO;
1003         }
1004         lock_buffer(*bh);
1005         if (!buffer_dirty(*bh))
1006                 clear_buffer_uptodate(*bh);
1007         unlock_buffer(*bh);
1008         ll_rw_block(READ, 1, bh);
1009         wait_on_buffer(*bh);
1010         return 0;
1011 }
1012
1013 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1014 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1015 {
1016         int status;
1017
1018         /* XXX hold a ref on the node while mounte?  easy enough, if
1019          * desirable. */
1020         if (ocfs2_mount_local(osb))
1021                 osb->node_num = 0;
1022         else
1023                 osb->node_num = o2nm_this_node();
1024
1025         if (osb->node_num == O2NM_MAX_NODES) {
1026                 mlog(ML_ERROR, "could not find this host's node number\n");
1027                 status = -ENOENT;
1028                 goto bail;
1029         }
1030
1031         mlog(0, "I am node %d\n", osb->node_num);
1032
1033         status = 0;
1034 bail:
1035         return status;
1036 }
1037
1038 static int ocfs2_mount_volume(struct super_block *sb)
1039 {
1040         int status = 0;
1041         int unlock_super = 0;
1042         struct ocfs2_super *osb = OCFS2_SB(sb);
1043
1044         mlog_entry_void();
1045
1046         if (ocfs2_is_hard_readonly(osb))
1047                 goto leave;
1048
1049         status = ocfs2_fill_local_node_info(osb);
1050         if (status < 0) {
1051                 mlog_errno(status);
1052                 goto leave;
1053         }
1054
1055         status = ocfs2_register_hb_callbacks(osb);
1056         if (status < 0) {
1057                 mlog_errno(status);
1058                 goto leave;
1059         }
1060
1061         status = ocfs2_dlm_init(osb);
1062         if (status < 0) {
1063                 mlog_errno(status);
1064                 goto leave;
1065         }
1066
1067         /* requires vote_thread to be running. */
1068         status = ocfs2_register_net_handlers(osb);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto leave;
1072         }
1073
1074         status = ocfs2_super_lock(osb, 1);
1075         if (status < 0) {
1076                 mlog_errno(status);
1077                 goto leave;
1078         }
1079         unlock_super = 1;
1080
1081         /* This will load up the node map and add ourselves to it. */
1082         status = ocfs2_find_slot(osb);
1083         if (status < 0) {
1084                 mlog_errno(status);
1085                 goto leave;
1086         }
1087
1088         ocfs2_populate_mounted_map(osb);
1089
1090         /* load all node-local system inodes */
1091         status = ocfs2_init_local_system_inodes(osb);
1092         if (status < 0) {
1093                 mlog_errno(status);
1094                 goto leave;
1095         }
1096
1097         status = ocfs2_check_volume(osb);
1098         if (status < 0) {
1099                 mlog_errno(status);
1100                 goto leave;
1101         }
1102
1103         status = ocfs2_truncate_log_init(osb);
1104         if (status < 0) {
1105                 mlog_errno(status);
1106                 goto leave;
1107         }
1108
1109         if (ocfs2_mount_local(osb))
1110                 goto leave;
1111
1112         /* This should be sent *after* we recovered our journal as it
1113          * will cause other nodes to unmark us as needing
1114          * recovery. However, we need to send it *before* dropping the
1115          * super block lock as otherwise their recovery threads might
1116          * try to clean us up while we're live! */
1117         status = ocfs2_request_mount_vote(osb);
1118         if (status < 0)
1119                 mlog_errno(status);
1120
1121 leave:
1122         if (unlock_super)
1123                 ocfs2_super_unlock(osb, 1);
1124
1125         mlog_exit(status);
1126         return status;
1127 }
1128
1129 /* we can't grab the goofy sem lock from inside wait_event, so we use
1130  * memory barriers to make sure that we'll see the null task before
1131  * being woken up */
1132 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1133 {
1134         mb();
1135         return osb->recovery_thread_task != NULL;
1136 }
1137
1138 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1139 {
1140         int tmp;
1141         struct ocfs2_super *osb = NULL;
1142         char nodestr[8];
1143
1144         mlog_entry("(0x%p)\n", sb);
1145
1146         BUG_ON(!sb);
1147         osb = OCFS2_SB(sb);
1148         BUG_ON(!osb);
1149
1150         ocfs2_shutdown_local_alloc(osb);
1151
1152         ocfs2_truncate_log_shutdown(osb);
1153
1154         /* disable any new recovery threads and wait for any currently
1155          * running ones to exit. Do this before setting the vol_state. */
1156         mutex_lock(&osb->recovery_lock);
1157         osb->disable_recovery = 1;
1158         mutex_unlock(&osb->recovery_lock);
1159         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1160
1161         /* At this point, we know that no more recovery threads can be
1162          * launched, so wait for any recovery completion work to
1163          * complete. */
1164         flush_workqueue(ocfs2_wq);
1165
1166         ocfs2_journal_shutdown(osb);
1167
1168         ocfs2_sync_blockdev(sb);
1169
1170         /* No dlm means we've failed during mount, so skip all the
1171          * steps which depended on that to complete. */
1172         if (osb->dlm) {
1173                 tmp = ocfs2_super_lock(osb, 1);
1174                 if (tmp < 0) {
1175                         mlog_errno(tmp);
1176                         return;
1177                 }
1178
1179                 tmp = ocfs2_request_umount_vote(osb);
1180                 if (tmp < 0)
1181                         mlog_errno(tmp);
1182
1183                 if (osb->slot_num != OCFS2_INVALID_SLOT)
1184                         ocfs2_put_slot(osb);
1185
1186                 ocfs2_super_unlock(osb, 1);
1187         }
1188
1189         ocfs2_release_system_inodes(osb);
1190
1191         if (osb->dlm) {
1192                 ocfs2_unregister_net_handlers(osb);
1193
1194                 ocfs2_dlm_shutdown(osb);
1195         }
1196
1197         ocfs2_clear_hb_callbacks(osb);
1198
1199         debugfs_remove(osb->osb_debug_root);
1200
1201         if (!mnt_err)
1202                 ocfs2_stop_heartbeat(osb);
1203
1204         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1205
1206         if (ocfs2_mount_local(osb))
1207                 snprintf(nodestr, sizeof(nodestr), "local");
1208         else
1209                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1210
1211         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1212                osb->dev_str, nodestr);
1213
1214         ocfs2_delete_osb(osb);
1215         kfree(osb);
1216         sb->s_dev = 0;
1217         sb->s_fs_info = NULL;
1218 }
1219
1220 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1221                                 unsigned uuid_bytes)
1222 {
1223         int i, ret;
1224         char *ptr;
1225
1226         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1227
1228         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1229         if (osb->uuid_str == NULL)
1230                 return -ENOMEM;
1231
1232         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1233                 /* print with null */
1234                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1235                 if (ret != 2) /* drop super cleans up */
1236                         return -EINVAL;
1237                 /* then only advance past the last char */
1238                 ptr += 2;
1239         }
1240
1241         return 0;
1242 }
1243
1244 static int ocfs2_initialize_super(struct super_block *sb,
1245                                   struct buffer_head *bh,
1246                                   int sector_size)
1247 {
1248         int status = 0;
1249         int i;
1250         struct ocfs2_dinode *di = NULL;
1251         struct inode *inode = NULL;
1252         struct buffer_head *bitmap_bh = NULL;
1253         struct ocfs2_journal *journal;
1254         __le32 uuid_net_key;
1255         struct ocfs2_super *osb;
1256
1257         mlog_entry_void();
1258
1259         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1260         if (!osb) {
1261                 status = -ENOMEM;
1262                 mlog_errno(status);
1263                 goto bail;
1264         }
1265
1266         sb->s_fs_info = osb;
1267         sb->s_op = &ocfs2_sops;
1268         sb->s_export_op = &ocfs2_export_ops;
1269         sb->s_flags |= MS_NOATIME;
1270         /* this is needed to support O_LARGEFILE */
1271         sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1272
1273         osb->sb = sb;
1274         /* Save off for ocfs2_rw_direct */
1275         osb->s_sectsize_bits = blksize_bits(sector_size);
1276         BUG_ON(!osb->s_sectsize_bits);
1277
1278         osb->net_response_ids = 0;
1279         spin_lock_init(&osb->net_response_lock);
1280         INIT_LIST_HEAD(&osb->net_response_list);
1281
1282         INIT_LIST_HEAD(&osb->osb_net_handlers);
1283         init_waitqueue_head(&osb->recovery_event);
1284         spin_lock_init(&osb->vote_task_lock);
1285         init_waitqueue_head(&osb->vote_event);
1286         osb->vote_work_sequence = 0;
1287         osb->vote_wake_sequence = 0;
1288         INIT_LIST_HEAD(&osb->blocked_lock_list);
1289         osb->blocked_lock_count = 0;
1290         INIT_LIST_HEAD(&osb->vote_list);
1291         spin_lock_init(&osb->osb_lock);
1292
1293         atomic_set(&osb->alloc_stats.moves, 0);
1294         atomic_set(&osb->alloc_stats.local_data, 0);
1295         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1296         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1297         atomic_set(&osb->alloc_stats.bg_extends, 0);
1298
1299         ocfs2_init_node_maps(osb);
1300
1301         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1302                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1303
1304         mutex_init(&osb->recovery_lock);
1305
1306         osb->disable_recovery = 0;
1307         osb->recovery_thread_task = NULL;
1308
1309         init_waitqueue_head(&osb->checkpoint_event);
1310         atomic_set(&osb->needs_checkpoint, 0);
1311
1312         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1313
1314         osb->node_num = O2NM_INVALID_NODE_NUM;
1315         osb->slot_num = OCFS2_INVALID_SLOT;
1316
1317         osb->local_alloc_state = OCFS2_LA_UNUSED;
1318         osb->local_alloc_bh = NULL;
1319
1320         ocfs2_setup_hb_callbacks(osb);
1321
1322         init_waitqueue_head(&osb->osb_mount_event);
1323
1324         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1325         if (!osb->vol_label) {
1326                 mlog(ML_ERROR, "unable to alloc vol label\n");
1327                 status = -ENOMEM;
1328                 goto bail;
1329         }
1330
1331         di = (struct ocfs2_dinode *)bh->b_data;
1332
1333         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1334         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1335                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1336                      osb->max_slots);
1337                 status = -EINVAL;
1338                 goto bail;
1339         }
1340         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1341
1342         init_waitqueue_head(&osb->osb_wipe_event);
1343         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1344                                         sizeof(*osb->osb_orphan_wipes),
1345                                         GFP_KERNEL);
1346         if (!osb->osb_orphan_wipes) {
1347                 status = -ENOMEM;
1348                 mlog_errno(status);
1349                 goto bail;
1350         }
1351
1352         osb->s_feature_compat =
1353                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1354         osb->s_feature_ro_compat =
1355                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1356         osb->s_feature_incompat =
1357                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1358
1359         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1360                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1361                      "optional features (%x).\n", i);
1362                 status = -EINVAL;
1363                 goto bail;
1364         }
1365         if (!(osb->sb->s_flags & MS_RDONLY) &&
1366             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1367                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1368                      "unsupported optional features (%x).\n", i);
1369                 status = -EINVAL;
1370                 goto bail;
1371         }
1372
1373         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1374
1375         /* FIXME
1376          * This should be done in ocfs2_journal_init(), but unknown
1377          * ordering issues will cause the filesystem to crash.
1378          * If anyone wants to figure out what part of the code
1379          * refers to osb->journal before ocfs2_journal_init() is run,
1380          * be my guest.
1381          */
1382         /* initialize our journal structure */
1383
1384         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1385         if (!journal) {
1386                 mlog(ML_ERROR, "unable to alloc journal\n");
1387                 status = -ENOMEM;
1388                 goto bail;
1389         }
1390         osb->journal = journal;
1391         journal->j_osb = osb;
1392
1393         atomic_set(&journal->j_num_trans, 0);
1394         init_rwsem(&journal->j_trans_barrier);
1395         init_waitqueue_head(&journal->j_checkpointed);
1396         spin_lock_init(&journal->j_lock);
1397         journal->j_trans_id = (unsigned long) 1;
1398         INIT_LIST_HEAD(&journal->j_la_cleanups);
1399         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1400         journal->j_state = OCFS2_JOURNAL_FREE;
1401
1402         /* get some pseudo constants for clustersize bits */
1403         osb->s_clustersize_bits =
1404                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1405         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1406         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1407
1408         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1409             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1410                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1411                      osb->s_clustersize);
1412                 status = -EINVAL;
1413                 goto bail;
1414         }
1415
1416         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1417             > (u32)~0UL) {
1418                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1419                      "what jbd can address in 32 bits.\n");
1420                 status = -EINVAL;
1421                 goto bail;
1422         }
1423
1424         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1425                                  sizeof(di->id2.i_super.s_uuid))) {
1426                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1427                 status = -ENOMEM;
1428                 goto bail;
1429         }
1430
1431         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1432         osb->net_key = le32_to_cpu(uuid_net_key);
1433
1434         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1435         osb->vol_label[63] = '\0';
1436         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1437         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1438         osb->first_cluster_group_blkno =
1439                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1440         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1441         mlog(0, "vol_label: %s\n", osb->vol_label);
1442         mlog(0, "uuid: %s\n", osb->uuid_str);
1443         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1444              (unsigned long long)osb->root_blkno,
1445              (unsigned long long)osb->system_dir_blkno);
1446
1447         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1448         if (!osb->osb_dlm_debug) {
1449                 status = -ENOMEM;
1450                 mlog_errno(status);
1451                 goto bail;
1452         }
1453
1454         atomic_set(&osb->vol_state, VOLUME_INIT);
1455
1456         /* load root, system_dir, and all global system inodes */
1457         status = ocfs2_init_global_system_inodes(osb);
1458         if (status < 0) {
1459                 mlog_errno(status);
1460                 goto bail;
1461         }
1462
1463         /*
1464          * global bitmap
1465          */
1466         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1467                                             OCFS2_INVALID_SLOT);
1468         if (!inode) {
1469                 status = -EINVAL;
1470                 mlog_errno(status);
1471                 goto bail;
1472         }
1473
1474         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1475
1476         /* We don't have a cluster lock on the bitmap here because
1477          * we're only interested in static information and the extra
1478          * complexity at mount time isn't worht it. Don't pass the
1479          * inode in to the read function though as we don't want it to
1480          * be put in the cache. */
1481         status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1482                                   NULL);
1483         iput(inode);
1484         if (status < 0) {
1485                 mlog_errno(status);
1486                 goto bail;
1487         }
1488
1489         di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1490         osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1491         brelse(bitmap_bh);
1492         mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1493              (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1494
1495         status = ocfs2_init_slot_info(osb);
1496         if (status < 0) {
1497                 mlog_errno(status);
1498                 goto bail;
1499         }
1500
1501 bail:
1502         mlog_exit(status);
1503         return status;
1504 }
1505
1506 /*
1507  * will return: -EAGAIN if it is ok to keep searching for superblocks
1508  *              -EINVAL if there is a bad superblock
1509  *              0 on success
1510  */
1511 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1512                                struct buffer_head *bh,
1513                                u32 blksz)
1514 {
1515         int status = -EAGAIN;
1516
1517         mlog_entry_void();
1518
1519         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1520                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1521                 status = -EINVAL;
1522                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1523                         mlog(ML_ERROR, "found superblock with incorrect block "
1524                              "size: found %u, should be %u\n",
1525                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1526                                blksz);
1527                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1528                            OCFS2_MAJOR_REV_LEVEL ||
1529                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1530                            OCFS2_MINOR_REV_LEVEL) {
1531                         mlog(ML_ERROR, "found superblock with bad version: "
1532                              "found %u.%u, should be %u.%u\n",
1533                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1534                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1535                              OCFS2_MAJOR_REV_LEVEL,
1536                              OCFS2_MINOR_REV_LEVEL);
1537                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1538                         mlog(ML_ERROR, "bad block number on superblock: "
1539                              "found %llu, should be %llu\n",
1540                              (unsigned long long)le64_to_cpu(di->i_blkno),
1541                              (unsigned long long)bh->b_blocknr);
1542                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1543                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1544                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1545                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1546                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1547                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1548                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1549                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1550                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1551                         mlog(ML_ERROR,
1552                              "Superblock slots found greater than file system "
1553                              "maximum: found %u, max %u\n",
1554                              le16_to_cpu(di->id2.i_super.s_max_slots),
1555                              OCFS2_MAX_SLOTS);
1556                 } else {
1557                         /* found it! */
1558                         status = 0;
1559                 }
1560         }
1561
1562         mlog_exit(status);
1563         return status;
1564 }
1565
1566 static int ocfs2_check_volume(struct ocfs2_super *osb)
1567 {
1568         int status = 0;
1569         int dirty;
1570         int local;
1571         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1572                                                   * recover
1573                                                   * ourselves. */
1574
1575         mlog_entry_void();
1576
1577         /* Init our journal object. */
1578         status = ocfs2_journal_init(osb->journal, &dirty);
1579         if (status < 0) {
1580                 mlog(ML_ERROR, "Could not initialize journal!\n");
1581                 goto finally;
1582         }
1583
1584         /* If the journal was unmounted cleanly then we don't want to
1585          * recover anything. Otherwise, journal_load will do that
1586          * dirty work for us :) */
1587         if (!dirty) {
1588                 status = ocfs2_journal_wipe(osb->journal, 0);
1589                 if (status < 0) {
1590                         mlog_errno(status);
1591                         goto finally;
1592                 }
1593         } else {
1594                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1595                      "recovering volume.\n");
1596         }
1597
1598         local = ocfs2_mount_local(osb);
1599
1600         /* will play back anything left in the journal. */
1601         ocfs2_journal_load(osb->journal, local);
1602
1603         if (dirty) {
1604                 /* recover my local alloc if we didn't unmount cleanly. */
1605                 status = ocfs2_begin_local_alloc_recovery(osb,
1606                                                           osb->slot_num,
1607                                                           &local_alloc);
1608                 if (status < 0) {
1609                         mlog_errno(status);
1610                         goto finally;
1611                 }
1612                 /* we complete the recovery process after we've marked
1613                  * ourselves as mounted. */
1614         }
1615
1616         mlog(0, "Journal loaded.\n");
1617
1618         status = ocfs2_load_local_alloc(osb);
1619         if (status < 0) {
1620                 mlog_errno(status);
1621                 goto finally;
1622         }
1623
1624         if (dirty) {
1625                 /* Recovery will be completed after we've mounted the
1626                  * rest of the volume. */
1627                 osb->dirty = 1;
1628                 osb->local_alloc_copy = local_alloc;
1629                 local_alloc = NULL;
1630         }
1631
1632         /* go through each journal, trylock it and if you get the
1633          * lock, and it's marked as dirty, set the bit in the recover
1634          * map and launch a recovery thread for it. */
1635         status = ocfs2_mark_dead_nodes(osb);
1636         if (status < 0)
1637                 mlog_errno(status);
1638
1639 finally:
1640         if (local_alloc)
1641                 kfree(local_alloc);
1642
1643         mlog_exit(status);
1644         return status;
1645 }
1646
1647 /*
1648  * The routine gets called from dismount or close whenever a dismount on
1649  * volume is requested and the osb open count becomes 1.
1650  * It will remove the osb from the global list and also free up all the
1651  * initialized resources and fileobject.
1652  */
1653 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1654 {
1655         mlog_entry_void();
1656
1657         /* This function assumes that the caller has the main osb resource */
1658
1659         if (osb->slot_info)
1660                 ocfs2_free_slot_info(osb->slot_info);
1661
1662         kfree(osb->osb_orphan_wipes);
1663         /* FIXME
1664          * This belongs in journal shutdown, but because we have to
1665          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1666          * we free it here.
1667          */
1668         kfree(osb->journal);
1669         if (osb->local_alloc_copy)
1670                 kfree(osb->local_alloc_copy);
1671         kfree(osb->uuid_str);
1672         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1673         memset(osb, 0, sizeof(struct ocfs2_super));
1674
1675         mlog_exit_void();
1676 }
1677
1678 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1679  * panic(). We do not support continue-on-error operation. */
1680 static void ocfs2_handle_error(struct super_block *sb)
1681 {
1682         struct ocfs2_super *osb = OCFS2_SB(sb);
1683
1684         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1685                 panic("OCFS2: (device %s): panic forced after error\n",
1686                       sb->s_id);
1687
1688         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1689
1690         if (sb->s_flags & MS_RDONLY &&
1691             (ocfs2_is_soft_readonly(osb) ||
1692              ocfs2_is_hard_readonly(osb)))
1693                 return;
1694
1695         printk(KERN_CRIT "File system is now read-only due to the potential "
1696                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1697                "system is unmounted.\n");
1698         sb->s_flags |= MS_RDONLY;
1699         ocfs2_set_ro_flag(osb, 0);
1700 }
1701
1702 static char error_buf[1024];
1703
1704 void __ocfs2_error(struct super_block *sb,
1705                    const char *function,
1706                    const char *fmt, ...)
1707 {
1708         va_list args;
1709
1710         va_start(args, fmt);
1711         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1712         va_end(args);
1713
1714         /* Not using mlog here because we want to show the actual
1715          * function the error came from. */
1716         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1717                sb->s_id, function, error_buf);
1718
1719         ocfs2_handle_error(sb);
1720 }
1721
1722 /* Handle critical errors. This is intentionally more drastic than
1723  * ocfs2_handle_error, so we only use for things like journal errors,
1724  * etc. */
1725 void __ocfs2_abort(struct super_block* sb,
1726                    const char *function,
1727                    const char *fmt, ...)
1728 {
1729         va_list args;
1730
1731         va_start(args, fmt);
1732         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1733         va_end(args);
1734
1735         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1736                sb->s_id, function, error_buf);
1737
1738         /* We don't have the cluster support yet to go straight to
1739          * hard readonly in here. Until then, we want to keep
1740          * ocfs2_abort() so that we can at least mark critical
1741          * errors.
1742          *
1743          * TODO: This should abort the journal and alert other nodes
1744          * that our slot needs recovery. */
1745
1746         /* Force a panic(). This stinks, but it's better than letting
1747          * things continue without having a proper hard readonly
1748          * here. */
1749         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1750         ocfs2_handle_error(sb);
1751 }
1752
1753 module_init(ocfs2_init);
1754 module_exit(ocfs2_exit);