Pull battery into release branch
[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, s16 *slot,
86                                int is_remount);
87 static void ocfs2_put_super(struct super_block *sb);
88 static int ocfs2_mount_volume(struct super_block *sb);
89 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
90 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
91 static int ocfs2_initialize_mem_caches(void);
92 static void ocfs2_free_mem_caches(void);
93 static void ocfs2_delete_osb(struct ocfs2_super *osb);
94
95 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
96
97 static int ocfs2_sync_fs(struct super_block *sb, int wait);
98
99 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
100 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
101 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
102 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
103 static int ocfs2_check_volume(struct ocfs2_super *osb);
104 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
105                                struct buffer_head *bh,
106                                u32 sectsize);
107 static int ocfs2_initialize_super(struct super_block *sb,
108                                   struct buffer_head *bh,
109                                   int sector_size);
110 static int ocfs2_get_sector(struct super_block *sb,
111                             struct buffer_head **bh,
112                             int block,
113                             int sect_size);
114 static void ocfs2_write_super(struct super_block *sb);
115 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
116 static void ocfs2_destroy_inode(struct inode *inode);
117
118 static const struct super_operations ocfs2_sops = {
119         .statfs         = ocfs2_statfs,
120         .alloc_inode    = ocfs2_alloc_inode,
121         .destroy_inode  = ocfs2_destroy_inode,
122         .drop_inode     = ocfs2_drop_inode,
123         .clear_inode    = ocfs2_clear_inode,
124         .delete_inode   = ocfs2_delete_inode,
125         .sync_fs        = ocfs2_sync_fs,
126         .write_super    = ocfs2_write_super,
127         .put_super      = ocfs2_put_super,
128         .remount_fs     = ocfs2_remount,
129 };
130
131 enum {
132         Opt_barrier,
133         Opt_err_panic,
134         Opt_err_ro,
135         Opt_intr,
136         Opt_nointr,
137         Opt_hb_none,
138         Opt_hb_local,
139         Opt_data_ordered,
140         Opt_data_writeback,
141         Opt_atime_quantum,
142         Opt_slot,
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_slot, "preferred_slot=%u"},
158         {Opt_err, NULL}
159 };
160
161 /*
162  * write_super and sync_fs ripped right out of ext3.
163  */
164 static void ocfs2_write_super(struct super_block *sb)
165 {
166         if (mutex_trylock(&sb->s_lock) != 0)
167                 BUG();
168         sb->s_dirt = 0;
169 }
170
171 static int ocfs2_sync_fs(struct super_block *sb, int wait)
172 {
173         int status = 0;
174         tid_t target;
175         struct ocfs2_super *osb = OCFS2_SB(sb);
176
177         sb->s_dirt = 0;
178
179         if (ocfs2_is_hard_readonly(osb))
180                 return -EROFS;
181
182         if (wait) {
183                 status = ocfs2_flush_truncate_log(osb);
184                 if (status < 0)
185                         mlog_errno(status);
186         } else {
187                 ocfs2_schedule_truncate_log_flush(osb, 0);
188         }
189
190         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
191                 if (wait)
192                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
193                                         target);
194         }
195         return 0;
196 }
197
198 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
199 {
200         struct inode *new = NULL;
201         int status = 0;
202         int i;
203
204         mlog_entry_void();
205
206         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
207         if (IS_ERR(new)) {
208                 status = PTR_ERR(new);
209                 mlog_errno(status);
210                 goto bail;
211         }
212         osb->root_inode = new;
213
214         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
215         if (IS_ERR(new)) {
216                 status = PTR_ERR(new);
217                 mlog_errno(status);
218                 goto bail;
219         }
220         osb->sys_root_inode = new;
221
222         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
223              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
224                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
225                 if (!new) {
226                         ocfs2_release_system_inodes(osb);
227                         status = -EINVAL;
228                         mlog_errno(status);
229                         /* FIXME: Should ERROR_RO_FS */
230                         mlog(ML_ERROR, "Unable to load system inode %d, "
231                              "possibly corrupt fs?", i);
232                         goto bail;
233                 }
234                 // the array now has one ref, so drop this one
235                 iput(new);
236         }
237
238 bail:
239         mlog_exit(status);
240         return status;
241 }
242
243 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
244 {
245         struct inode *new = NULL;
246         int status = 0;
247         int i;
248
249         mlog_entry_void();
250
251         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
252              i < NUM_SYSTEM_INODES;
253              i++) {
254                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
255                 if (!new) {
256                         ocfs2_release_system_inodes(osb);
257                         status = -EINVAL;
258                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
259                              status, i, osb->slot_num);
260                         goto bail;
261                 }
262                 /* the array now has one ref, so drop this one */
263                 iput(new);
264         }
265
266 bail:
267         mlog_exit(status);
268         return status;
269 }
270
271 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
272 {
273         int status = 0, i;
274         struct inode *inode;
275
276         mlog_entry_void();
277
278         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
279                 inode = osb->system_inodes[i];
280                 if (inode) {
281                         iput(inode);
282                         osb->system_inodes[i] = NULL;
283                 }
284         }
285
286         inode = osb->sys_root_inode;
287         if (inode) {
288                 iput(inode);
289                 osb->sys_root_inode = NULL;
290         }
291
292         inode = osb->root_inode;
293         if (inode) {
294                 iput(inode);
295                 osb->root_inode = NULL;
296         }
297
298         mlog_exit(status);
299         return status;
300 }
301
302 /* We're allocating fs objects, use GFP_NOFS */
303 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
304 {
305         struct ocfs2_inode_info *oi;
306
307         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
308         if (!oi)
309                 return NULL;
310
311         return &oi->vfs_inode;
312 }
313
314 static void ocfs2_destroy_inode(struct inode *inode)
315 {
316         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
317 }
318
319 /* From xfs_super.c:xfs_max_file_offset
320  * Copyright (c) 2000-2004 Silicon Graphics, Inc.
321  */
322 unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
323 {
324         unsigned int pagefactor = 1;
325         unsigned int bitshift = BITS_PER_LONG - 1;
326
327         /* Figure out maximum filesize, on Linux this can depend on
328          * the filesystem blocksize (on 32 bit platforms).
329          * __block_prepare_write does this in an [unsigned] long...
330          *      page->index << (PAGE_CACHE_SHIFT - bbits)
331          * So, for page sized blocks (4K on 32 bit platforms),
332          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
333          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
334          * but for smaller blocksizes it is less (bbits = log2 bsize).
335          * Note1: get_block_t takes a long (implicit cast from above)
336          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
337          * can optionally convert the [unsigned] long from above into
338          * an [unsigned] long long.
339          */
340
341 #if BITS_PER_LONG == 32
342 # if defined(CONFIG_LBD)
343         BUILD_BUG_ON(sizeof(sector_t) != 8);
344         pagefactor = PAGE_CACHE_SIZE;
345         bitshift = BITS_PER_LONG;
346 # else
347         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
348 # endif
349 #endif
350
351         return (((unsigned long long)pagefactor) << bitshift) - 1;
352 }
353
354 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
355 {
356         int incompat_features;
357         int ret = 0;
358         unsigned long parsed_options;
359         s16 slot;
360         struct ocfs2_super *osb = OCFS2_SB(sb);
361
362         if (!ocfs2_parse_options(sb, data, &parsed_options, &slot, 1)) {
363                 ret = -EINVAL;
364                 goto out;
365         }
366
367         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
368             (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
369                 ret = -EINVAL;
370                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
371                 goto out;
372         }
373
374         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
375             (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
376                 ret = -EINVAL;
377                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
378                 goto out;
379         }
380
381         /* We're going to/from readonly mode. */
382         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
383                 /* Lock here so the check of HARD_RO and the potential
384                  * setting of SOFT_RO is atomic. */
385                 spin_lock(&osb->osb_lock);
386                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
387                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
388                         ret = -EROFS;
389                         goto unlock_osb;
390                 }
391
392                 if (*flags & MS_RDONLY) {
393                         mlog(0, "Going to ro mode.\n");
394                         sb->s_flags |= MS_RDONLY;
395                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
396                 } else {
397                         mlog(0, "Making ro filesystem writeable.\n");
398
399                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
400                                 mlog(ML_ERROR, "Cannot remount RDWR "
401                                      "filesystem due to previous errors.\n");
402                                 ret = -EROFS;
403                                 goto unlock_osb;
404                         }
405                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
406                         if (incompat_features) {
407                                 mlog(ML_ERROR, "Cannot remount RDWR because "
408                                      "of unsupported optional features "
409                                      "(%x).\n", incompat_features);
410                                 ret = -EINVAL;
411                                 goto unlock_osb;
412                         }
413                         sb->s_flags &= ~MS_RDONLY;
414                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
415                 }
416 unlock_osb:
417                 spin_unlock(&osb->osb_lock);
418         }
419
420         if (!ret) {
421                 if (!ocfs2_is_hard_readonly(osb))
422                         ocfs2_set_journal_params(osb);
423
424                 /* Only save off the new mount options in case of a successful
425                  * remount. */
426                 osb->s_mount_opt = parsed_options;
427         }
428 out:
429         return ret;
430 }
431
432 static int ocfs2_sb_probe(struct super_block *sb,
433                           struct buffer_head **bh,
434                           int *sector_size)
435 {
436         int status = 0, tmpstat;
437         struct ocfs1_vol_disk_hdr *hdr;
438         struct ocfs2_dinode *di;
439         int blksize;
440
441         *bh = NULL;
442
443         /* may be > 512 */
444         *sector_size = bdev_hardsect_size(sb->s_bdev);
445         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
446                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
447                      *sector_size, OCFS2_MAX_BLOCKSIZE);
448                 status = -EINVAL;
449                 goto bail;
450         }
451
452         /* Can this really happen? */
453         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
454                 *sector_size = OCFS2_MIN_BLOCKSIZE;
455
456         /* check block zero for old format */
457         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
458         if (status < 0) {
459                 mlog_errno(status);
460                 goto bail;
461         }
462         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
463         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
464                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
465                      hdr->major_version, hdr->minor_version);
466                 status = -EINVAL;
467         }
468         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
469                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
470                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
471                      hdr->signature);
472                 status = -EINVAL;
473         }
474         brelse(*bh);
475         *bh = NULL;
476         if (status < 0) {
477                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
478                      "upgraded before mounting with ocfs v2\n");
479                 goto bail;
480         }
481
482         /*
483          * Now check at magic offset for 512, 1024, 2048, 4096
484          * blocksizes.  4096 is the maximum blocksize because it is
485          * the minimum clustersize.
486          */
487         status = -EINVAL;
488         for (blksize = *sector_size;
489              blksize <= OCFS2_MAX_BLOCKSIZE;
490              blksize <<= 1) {
491                 tmpstat = ocfs2_get_sector(sb, bh,
492                                            OCFS2_SUPER_BLOCK_BLKNO,
493                                            blksize);
494                 if (tmpstat < 0) {
495                         status = tmpstat;
496                         mlog_errno(status);
497                         goto bail;
498                 }
499                 di = (struct ocfs2_dinode *) (*bh)->b_data;
500                 status = ocfs2_verify_volume(di, *bh, blksize);
501                 if (status >= 0)
502                         goto bail;
503                 brelse(*bh);
504                 *bh = NULL;
505                 if (status != -EAGAIN)
506                         break;
507         }
508
509 bail:
510         return status;
511 }
512
513 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
514 {
515         if (ocfs2_mount_local(osb)) {
516                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
517                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
518                              "mounted device.\n");
519                         return -EINVAL;
520                 }
521         }
522
523         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
524                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
525                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
526                              "a read-write clustered device.\n");
527                         return -EINVAL;
528                 }
529         }
530
531         return 0;
532 }
533
534 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
535 {
536         struct dentry *root;
537         int status, sector_size;
538         unsigned long parsed_opt;
539         s16 slot;
540         struct inode *inode = NULL;
541         struct ocfs2_super *osb = NULL;
542         struct buffer_head *bh = NULL;
543         char nodestr[8];
544
545         mlog_entry("%p, %p, %i", sb, data, silent);
546
547         if (!ocfs2_parse_options(sb, data, &parsed_opt, &slot, 0)) {
548                 status = -EINVAL;
549                 goto read_super_error;
550         }
551
552         /* for now we only have one cluster/node, make sure we see it
553          * in the heartbeat universe */
554         if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
555                 if (!o2hb_check_local_node_heartbeating()) {
556                         status = -EINVAL;
557                         goto read_super_error;
558                 }
559         }
560
561         /* probe for superblock */
562         status = ocfs2_sb_probe(sb, &bh, &sector_size);
563         if (status < 0) {
564                 mlog(ML_ERROR, "superblock probe failed!\n");
565                 goto read_super_error;
566         }
567
568         status = ocfs2_initialize_super(sb, bh, sector_size);
569         osb = OCFS2_SB(sb);
570         if (status < 0) {
571                 mlog_errno(status);
572                 goto read_super_error;
573         }
574         brelse(bh);
575         bh = NULL;
576         osb->s_mount_opt = parsed_opt;
577         osb->preferred_slot = slot;
578
579         sb->s_magic = OCFS2_SUPER_MAGIC;
580
581         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
582          * heartbeat=none */
583         if (bdev_read_only(sb->s_bdev)) {
584                 if (!(sb->s_flags & MS_RDONLY)) {
585                         status = -EACCES;
586                         mlog(ML_ERROR, "Readonly device detected but readonly "
587                              "mount was not specified.\n");
588                         goto read_super_error;
589                 }
590
591                 /* You should not be able to start a local heartbeat
592                  * on a readonly device. */
593                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
594                         status = -EROFS;
595                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
596                              "device.\n");
597                         goto read_super_error;
598                 }
599
600                 status = ocfs2_check_journals_nolocks(osb);
601                 if (status < 0) {
602                         if (status == -EROFS)
603                                 mlog(ML_ERROR, "Recovery required on readonly "
604                                      "file system, but write access is "
605                                      "unavailable.\n");
606                         else
607                                 mlog_errno(status);                     
608                         goto read_super_error;
609                 }
610
611                 ocfs2_set_ro_flag(osb, 1);
612
613                 printk(KERN_NOTICE "Readonly device detected. No cluster "
614                        "services will be utilized for this mount. Recovery "
615                        "will be skipped.\n");
616         }
617
618         if (!ocfs2_is_hard_readonly(osb)) {
619                 if (sb->s_flags & MS_RDONLY)
620                         ocfs2_set_ro_flag(osb, 0);
621         }
622
623         status = ocfs2_verify_heartbeat(osb);
624         if (status < 0) {
625                 mlog_errno(status);
626                 goto read_super_error;
627         }
628
629         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
630                                                  ocfs2_debugfs_root);
631         if (!osb->osb_debug_root) {
632                 status = -EINVAL;
633                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
634                 goto read_super_error;
635         }
636
637         status = ocfs2_mount_volume(sb);
638         if (osb->root_inode)
639                 inode = igrab(osb->root_inode);
640
641         if (status < 0)
642                 goto read_super_error;
643
644         if (!inode) {
645                 status = -EIO;
646                 mlog_errno(status);
647                 goto read_super_error;
648         }
649
650         root = d_alloc_root(inode);
651         if (!root) {
652                 status = -ENOMEM;
653                 mlog_errno(status);
654                 goto read_super_error;
655         }
656
657         sb->s_root = root;
658
659         ocfs2_complete_mount_recovery(osb);
660
661         if (ocfs2_mount_local(osb))
662                 snprintf(nodestr, sizeof(nodestr), "local");
663         else
664                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
665
666         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
667                "with %s data mode.\n",
668                osb->dev_str, nodestr, osb->slot_num,
669                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
670                "ordered");
671
672         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
673         wake_up(&osb->osb_mount_event);
674
675         mlog_exit(status);
676         return status;
677
678 read_super_error:
679         if (bh != NULL)
680                 brelse(bh);
681
682         if (inode)
683                 iput(inode);
684
685         if (osb) {
686                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
687                 wake_up(&osb->osb_mount_event);
688                 ocfs2_dismount_volume(sb, 1);
689         }
690
691         mlog_exit(status);
692         return status;
693 }
694
695 static int ocfs2_get_sb(struct file_system_type *fs_type,
696                         int flags,
697                         const char *dev_name,
698                         void *data,
699                         struct vfsmount *mnt)
700 {
701         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
702                            mnt);
703 }
704
705 static struct file_system_type ocfs2_fs_type = {
706         .owner          = THIS_MODULE,
707         .name           = "ocfs2",
708         .get_sb         = ocfs2_get_sb, /* is this called when we mount
709                                         * the fs? */
710         .kill_sb        = kill_block_super, /* set to the generic one
711                                              * right now, but do we
712                                              * need to change that? */
713         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
714         .next           = NULL
715 };
716
717 static int ocfs2_parse_options(struct super_block *sb,
718                                char *options,
719                                unsigned long *mount_opt,
720                                s16 *slot,
721                                int is_remount)
722 {
723         int status;
724         char *p;
725
726         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
727                    options ? options : "(none)");
728
729         *mount_opt = 0;
730         *slot = OCFS2_INVALID_SLOT;
731
732         if (!options) {
733                 status = 1;
734                 goto bail;
735         }
736
737         while ((p = strsep(&options, ",")) != NULL) {
738                 int token, option;
739                 substring_t args[MAX_OPT_ARGS];
740                 struct ocfs2_super * osb = OCFS2_SB(sb);
741
742                 if (!*p)
743                         continue;
744
745                 token = match_token(p, tokens, args);
746                 switch (token) {
747                 case Opt_hb_local:
748                         *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
749                         break;
750                 case Opt_hb_none:
751                         *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
752                         break;
753                 case Opt_barrier:
754                         if (match_int(&args[0], &option)) {
755                                 status = 0;
756                                 goto bail;
757                         }
758                         if (option)
759                                 *mount_opt |= OCFS2_MOUNT_BARRIER;
760                         else
761                                 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
762                         break;
763                 case Opt_intr:
764                         *mount_opt &= ~OCFS2_MOUNT_NOINTR;
765                         break;
766                 case Opt_nointr:
767                         *mount_opt |= OCFS2_MOUNT_NOINTR;
768                         break;
769                 case Opt_err_panic:
770                         *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
771                         break;
772                 case Opt_err_ro:
773                         *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
774                         break;
775                 case Opt_data_ordered:
776                         *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
777                         break;
778                 case Opt_data_writeback:
779                         *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
780                         break;
781                 case Opt_atime_quantum:
782                         if (match_int(&args[0], &option)) {
783                                 status = 0;
784                                 goto bail;
785                         }
786                         if (option >= 0)
787                                 osb->s_atime_quantum = option;
788                         else
789                                 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
790                         break;
791                 case Opt_slot:
792                         option = 0;
793                         if (match_int(&args[0], &option)) {
794                                 status = 0;
795                                 goto bail;
796                         }
797                         if (option)
798                                 *slot = (s16)option;
799                         break;
800                 default:
801                         mlog(ML_ERROR,
802                              "Unrecognized mount option \"%s\" "
803                              "or missing value\n", p);
804                         status = 0;
805                         goto bail;
806                 }
807         }
808
809         status = 1;
810
811 bail:
812         mlog_exit(status);
813         return status;
814 }
815
816 static int __init ocfs2_init(void)
817 {
818         int status;
819
820         mlog_entry_void();
821
822         ocfs2_print_version();
823
824         status = init_ocfs2_uptodate_cache();
825         if (status < 0) {
826                 mlog_errno(status);
827                 goto leave;
828         }
829
830         status = ocfs2_initialize_mem_caches();
831         if (status < 0) {
832                 mlog_errno(status);
833                 goto leave;
834         }
835
836         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
837         if (!ocfs2_wq) {
838                 status = -ENOMEM;
839                 goto leave;
840         }
841
842         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
843         if (!ocfs2_debugfs_root) {
844                 status = -EFAULT;
845                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
846         }
847
848 leave:
849         if (status < 0) {
850                 ocfs2_free_mem_caches();
851                 exit_ocfs2_uptodate_cache();
852         }
853
854         mlog_exit(status);
855
856         if (status >= 0) {
857                 return register_filesystem(&ocfs2_fs_type);
858         } else
859                 return -1;
860 }
861
862 static void __exit ocfs2_exit(void)
863 {
864         mlog_entry_void();
865
866         if (ocfs2_wq) {
867                 flush_workqueue(ocfs2_wq);
868                 destroy_workqueue(ocfs2_wq);
869         }
870
871         debugfs_remove(ocfs2_debugfs_root);
872
873         ocfs2_free_mem_caches();
874
875         unregister_filesystem(&ocfs2_fs_type);
876
877         exit_ocfs2_uptodate_cache();
878
879         mlog_exit_void();
880 }
881
882 static void ocfs2_put_super(struct super_block *sb)
883 {
884         mlog_entry("(0x%p)\n", sb);
885
886         ocfs2_sync_blockdev(sb);
887         ocfs2_dismount_volume(sb, 0);
888
889         mlog_exit_void();
890 }
891
892 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
893 {
894         struct ocfs2_super *osb;
895         u32 numbits, freebits;
896         int status;
897         struct ocfs2_dinode *bm_lock;
898         struct buffer_head *bh = NULL;
899         struct inode *inode = NULL;
900
901         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
902
903         osb = OCFS2_SB(dentry->d_sb);
904
905         inode = ocfs2_get_system_file_inode(osb,
906                                             GLOBAL_BITMAP_SYSTEM_INODE,
907                                             OCFS2_INVALID_SLOT);
908         if (!inode) {
909                 mlog(ML_ERROR, "failed to get bitmap inode\n");
910                 status = -EIO;
911                 goto bail;
912         }
913
914         status = ocfs2_meta_lock(inode, &bh, 0);
915         if (status < 0) {
916                 mlog_errno(status);
917                 goto bail;
918         }
919
920         bm_lock = (struct ocfs2_dinode *) bh->b_data;
921
922         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
923         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
924
925         buf->f_type = OCFS2_SUPER_MAGIC;
926         buf->f_bsize = dentry->d_sb->s_blocksize;
927         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
928         buf->f_blocks = ((sector_t) numbits) *
929                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
930         buf->f_bfree = ((sector_t) freebits) *
931                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
932         buf->f_bavail = buf->f_bfree;
933         buf->f_files = numbits;
934         buf->f_ffree = freebits;
935
936         brelse(bh);
937
938         ocfs2_meta_unlock(inode, 0);
939         status = 0;
940 bail:
941         if (inode)
942                 iput(inode);
943
944         mlog_exit(status);
945
946         return status;
947 }
948
949 static void ocfs2_inode_init_once(void *data,
950                                   struct kmem_cache *cachep,
951                                   unsigned long flags)
952 {
953         struct ocfs2_inode_info *oi = data;
954
955         oi->ip_flags = 0;
956         oi->ip_open_count = 0;
957         spin_lock_init(&oi->ip_lock);
958         ocfs2_extent_map_init(&oi->vfs_inode);
959         INIT_LIST_HEAD(&oi->ip_io_markers);
960         oi->ip_created_trans = 0;
961         oi->ip_last_trans = 0;
962         oi->ip_dir_start_lookup = 0;
963
964         init_rwsem(&oi->ip_alloc_sem);
965         mutex_init(&oi->ip_io_mutex);
966
967         oi->ip_blkno = 0ULL;
968         oi->ip_clusters = 0;
969
970         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
971         ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
972         ocfs2_lock_res_init_once(&oi->ip_data_lockres);
973         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
974
975         ocfs2_metadata_cache_init(&oi->vfs_inode);
976
977         inode_init_once(&oi->vfs_inode);
978 }
979
980 static int ocfs2_initialize_mem_caches(void)
981 {
982         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
983                                        sizeof(struct ocfs2_inode_info),
984                                        0,
985                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
986                                                 SLAB_MEM_SPREAD),
987                                        ocfs2_inode_init_once);
988         if (!ocfs2_inode_cachep)
989                 return -ENOMEM;
990
991         return 0;
992 }
993
994 static void ocfs2_free_mem_caches(void)
995 {
996         if (ocfs2_inode_cachep)
997                 kmem_cache_destroy(ocfs2_inode_cachep);
998
999         ocfs2_inode_cachep = NULL;
1000 }
1001
1002 static int ocfs2_get_sector(struct super_block *sb,
1003                             struct buffer_head **bh,
1004                             int block,
1005                             int sect_size)
1006 {
1007         if (!sb_set_blocksize(sb, sect_size)) {
1008                 mlog(ML_ERROR, "unable to set blocksize\n");
1009                 return -EIO;
1010         }
1011
1012         *bh = sb_getblk(sb, block);
1013         if (!*bh) {
1014                 mlog_errno(-EIO);
1015                 return -EIO;
1016         }
1017         lock_buffer(*bh);
1018         if (!buffer_dirty(*bh))
1019                 clear_buffer_uptodate(*bh);
1020         unlock_buffer(*bh);
1021         ll_rw_block(READ, 1, bh);
1022         wait_on_buffer(*bh);
1023         return 0;
1024 }
1025
1026 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1027 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1028 {
1029         int status;
1030
1031         /* XXX hold a ref on the node while mounte?  easy enough, if
1032          * desirable. */
1033         if (ocfs2_mount_local(osb))
1034                 osb->node_num = 0;
1035         else
1036                 osb->node_num = o2nm_this_node();
1037
1038         if (osb->node_num == O2NM_MAX_NODES) {
1039                 mlog(ML_ERROR, "could not find this host's node number\n");
1040                 status = -ENOENT;
1041                 goto bail;
1042         }
1043
1044         mlog(0, "I am node %d\n", osb->node_num);
1045
1046         status = 0;
1047 bail:
1048         return status;
1049 }
1050
1051 static int ocfs2_mount_volume(struct super_block *sb)
1052 {
1053         int status = 0;
1054         int unlock_super = 0;
1055         struct ocfs2_super *osb = OCFS2_SB(sb);
1056
1057         mlog_entry_void();
1058
1059         if (ocfs2_is_hard_readonly(osb))
1060                 goto leave;
1061
1062         status = ocfs2_fill_local_node_info(osb);
1063         if (status < 0) {
1064                 mlog_errno(status);
1065                 goto leave;
1066         }
1067
1068         status = ocfs2_register_hb_callbacks(osb);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto leave;
1072         }
1073
1074         status = ocfs2_dlm_init(osb);
1075         if (status < 0) {
1076                 mlog_errno(status);
1077                 goto leave;
1078         }
1079
1080         /* requires vote_thread to be running. */
1081         status = ocfs2_register_net_handlers(osb);
1082         if (status < 0) {
1083                 mlog_errno(status);
1084                 goto leave;
1085         }
1086
1087         status = ocfs2_super_lock(osb, 1);
1088         if (status < 0) {
1089                 mlog_errno(status);
1090                 goto leave;
1091         }
1092         unlock_super = 1;
1093
1094         /* This will load up the node map and add ourselves to it. */
1095         status = ocfs2_find_slot(osb);
1096         if (status < 0) {
1097                 mlog_errno(status);
1098                 goto leave;
1099         }
1100
1101         ocfs2_populate_mounted_map(osb);
1102
1103         /* load all node-local system inodes */
1104         status = ocfs2_init_local_system_inodes(osb);
1105         if (status < 0) {
1106                 mlog_errno(status);
1107                 goto leave;
1108         }
1109
1110         status = ocfs2_check_volume(osb);
1111         if (status < 0) {
1112                 mlog_errno(status);
1113                 goto leave;
1114         }
1115
1116         status = ocfs2_truncate_log_init(osb);
1117         if (status < 0) {
1118                 mlog_errno(status);
1119                 goto leave;
1120         }
1121
1122         if (ocfs2_mount_local(osb))
1123                 goto leave;
1124
1125         /* This should be sent *after* we recovered our journal as it
1126          * will cause other nodes to unmark us as needing
1127          * recovery. However, we need to send it *before* dropping the
1128          * super block lock as otherwise their recovery threads might
1129          * try to clean us up while we're live! */
1130         status = ocfs2_request_mount_vote(osb);
1131         if (status < 0)
1132                 mlog_errno(status);
1133
1134 leave:
1135         if (unlock_super)
1136                 ocfs2_super_unlock(osb, 1);
1137
1138         mlog_exit(status);
1139         return status;
1140 }
1141
1142 /* we can't grab the goofy sem lock from inside wait_event, so we use
1143  * memory barriers to make sure that we'll see the null task before
1144  * being woken up */
1145 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1146 {
1147         mb();
1148         return osb->recovery_thread_task != NULL;
1149 }
1150
1151 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1152 {
1153         int tmp;
1154         struct ocfs2_super *osb = NULL;
1155         char nodestr[8];
1156
1157         mlog_entry("(0x%p)\n", sb);
1158
1159         BUG_ON(!sb);
1160         osb = OCFS2_SB(sb);
1161         BUG_ON(!osb);
1162
1163         ocfs2_shutdown_local_alloc(osb);
1164
1165         ocfs2_truncate_log_shutdown(osb);
1166
1167         /* disable any new recovery threads and wait for any currently
1168          * running ones to exit. Do this before setting the vol_state. */
1169         mutex_lock(&osb->recovery_lock);
1170         osb->disable_recovery = 1;
1171         mutex_unlock(&osb->recovery_lock);
1172         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1173
1174         /* At this point, we know that no more recovery threads can be
1175          * launched, so wait for any recovery completion work to
1176          * complete. */
1177         flush_workqueue(ocfs2_wq);
1178
1179         ocfs2_journal_shutdown(osb);
1180
1181         ocfs2_sync_blockdev(sb);
1182
1183         /* No dlm means we've failed during mount, so skip all the
1184          * steps which depended on that to complete. */
1185         if (osb->dlm) {
1186                 tmp = ocfs2_super_lock(osb, 1);
1187                 if (tmp < 0) {
1188                         mlog_errno(tmp);
1189                         return;
1190                 }
1191
1192                 tmp = ocfs2_request_umount_vote(osb);
1193                 if (tmp < 0)
1194                         mlog_errno(tmp);
1195
1196                 if (osb->slot_num != OCFS2_INVALID_SLOT)
1197                         ocfs2_put_slot(osb);
1198
1199                 ocfs2_super_unlock(osb, 1);
1200         }
1201
1202         ocfs2_release_system_inodes(osb);
1203
1204         if (osb->dlm) {
1205                 ocfs2_unregister_net_handlers(osb);
1206
1207                 ocfs2_dlm_shutdown(osb);
1208         }
1209
1210         ocfs2_clear_hb_callbacks(osb);
1211
1212         debugfs_remove(osb->osb_debug_root);
1213
1214         if (!mnt_err)
1215                 ocfs2_stop_heartbeat(osb);
1216
1217         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1218
1219         if (ocfs2_mount_local(osb))
1220                 snprintf(nodestr, sizeof(nodestr), "local");
1221         else
1222                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1223
1224         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1225                osb->dev_str, nodestr);
1226
1227         ocfs2_delete_osb(osb);
1228         kfree(osb);
1229         sb->s_dev = 0;
1230         sb->s_fs_info = NULL;
1231 }
1232
1233 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1234                                 unsigned uuid_bytes)
1235 {
1236         int i, ret;
1237         char *ptr;
1238
1239         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1240
1241         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1242         if (osb->uuid_str == NULL)
1243                 return -ENOMEM;
1244
1245         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1246                 /* print with null */
1247                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1248                 if (ret != 2) /* drop super cleans up */
1249                         return -EINVAL;
1250                 /* then only advance past the last char */
1251                 ptr += 2;
1252         }
1253
1254         return 0;
1255 }
1256
1257 static int ocfs2_initialize_super(struct super_block *sb,
1258                                   struct buffer_head *bh,
1259                                   int sector_size)
1260 {
1261         int status = 0;
1262         int i;
1263         struct ocfs2_dinode *di = NULL;
1264         struct inode *inode = NULL;
1265         struct buffer_head *bitmap_bh = NULL;
1266         struct ocfs2_journal *journal;
1267         __le32 uuid_net_key;
1268         struct ocfs2_super *osb;
1269
1270         mlog_entry_void();
1271
1272         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1273         if (!osb) {
1274                 status = -ENOMEM;
1275                 mlog_errno(status);
1276                 goto bail;
1277         }
1278
1279         sb->s_fs_info = osb;
1280         sb->s_op = &ocfs2_sops;
1281         sb->s_export_op = &ocfs2_export_ops;
1282         sb->s_flags |= MS_NOATIME;
1283         /* this is needed to support O_LARGEFILE */
1284         sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1285
1286         osb->sb = sb;
1287         /* Save off for ocfs2_rw_direct */
1288         osb->s_sectsize_bits = blksize_bits(sector_size);
1289         BUG_ON(!osb->s_sectsize_bits);
1290
1291         osb->net_response_ids = 0;
1292         spin_lock_init(&osb->net_response_lock);
1293         INIT_LIST_HEAD(&osb->net_response_list);
1294
1295         INIT_LIST_HEAD(&osb->osb_net_handlers);
1296         init_waitqueue_head(&osb->recovery_event);
1297         spin_lock_init(&osb->vote_task_lock);
1298         init_waitqueue_head(&osb->vote_event);
1299         osb->vote_work_sequence = 0;
1300         osb->vote_wake_sequence = 0;
1301         INIT_LIST_HEAD(&osb->blocked_lock_list);
1302         osb->blocked_lock_count = 0;
1303         INIT_LIST_HEAD(&osb->vote_list);
1304         spin_lock_init(&osb->osb_lock);
1305
1306         atomic_set(&osb->alloc_stats.moves, 0);
1307         atomic_set(&osb->alloc_stats.local_data, 0);
1308         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1309         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1310         atomic_set(&osb->alloc_stats.bg_extends, 0);
1311
1312         ocfs2_init_node_maps(osb);
1313
1314         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1315                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1316
1317         mutex_init(&osb->recovery_lock);
1318
1319         osb->disable_recovery = 0;
1320         osb->recovery_thread_task = NULL;
1321
1322         init_waitqueue_head(&osb->checkpoint_event);
1323         atomic_set(&osb->needs_checkpoint, 0);
1324
1325         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1326
1327         osb->node_num = O2NM_INVALID_NODE_NUM;
1328         osb->slot_num = OCFS2_INVALID_SLOT;
1329
1330         osb->local_alloc_state = OCFS2_LA_UNUSED;
1331         osb->local_alloc_bh = NULL;
1332
1333         ocfs2_setup_hb_callbacks(osb);
1334
1335         init_waitqueue_head(&osb->osb_mount_event);
1336
1337         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1338         if (!osb->vol_label) {
1339                 mlog(ML_ERROR, "unable to alloc vol label\n");
1340                 status = -ENOMEM;
1341                 goto bail;
1342         }
1343
1344         di = (struct ocfs2_dinode *)bh->b_data;
1345
1346         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1347         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1348                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1349                      osb->max_slots);
1350                 status = -EINVAL;
1351                 goto bail;
1352         }
1353         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1354
1355         init_waitqueue_head(&osb->osb_wipe_event);
1356         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1357                                         sizeof(*osb->osb_orphan_wipes),
1358                                         GFP_KERNEL);
1359         if (!osb->osb_orphan_wipes) {
1360                 status = -ENOMEM;
1361                 mlog_errno(status);
1362                 goto bail;
1363         }
1364
1365         osb->s_feature_compat =
1366                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1367         osb->s_feature_ro_compat =
1368                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1369         osb->s_feature_incompat =
1370                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1371
1372         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1373                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1374                      "optional features (%x).\n", i);
1375                 status = -EINVAL;
1376                 goto bail;
1377         }
1378         if (!(osb->sb->s_flags & MS_RDONLY) &&
1379             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1380                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1381                      "unsupported optional features (%x).\n", i);
1382                 status = -EINVAL;
1383                 goto bail;
1384         }
1385
1386         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1387
1388         /* FIXME
1389          * This should be done in ocfs2_journal_init(), but unknown
1390          * ordering issues will cause the filesystem to crash.
1391          * If anyone wants to figure out what part of the code
1392          * refers to osb->journal before ocfs2_journal_init() is run,
1393          * be my guest.
1394          */
1395         /* initialize our journal structure */
1396
1397         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1398         if (!journal) {
1399                 mlog(ML_ERROR, "unable to alloc journal\n");
1400                 status = -ENOMEM;
1401                 goto bail;
1402         }
1403         osb->journal = journal;
1404         journal->j_osb = osb;
1405
1406         atomic_set(&journal->j_num_trans, 0);
1407         init_rwsem(&journal->j_trans_barrier);
1408         init_waitqueue_head(&journal->j_checkpointed);
1409         spin_lock_init(&journal->j_lock);
1410         journal->j_trans_id = (unsigned long) 1;
1411         INIT_LIST_HEAD(&journal->j_la_cleanups);
1412         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1413         journal->j_state = OCFS2_JOURNAL_FREE;
1414
1415         /* get some pseudo constants for clustersize bits */
1416         osb->s_clustersize_bits =
1417                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1418         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1419         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1420
1421         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1422             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1423                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1424                      osb->s_clustersize);
1425                 status = -EINVAL;
1426                 goto bail;
1427         }
1428
1429         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1430             > (u32)~0UL) {
1431                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1432                      "what jbd can address in 32 bits.\n");
1433                 status = -EINVAL;
1434                 goto bail;
1435         }
1436
1437         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1438                                  sizeof(di->id2.i_super.s_uuid))) {
1439                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1440                 status = -ENOMEM;
1441                 goto bail;
1442         }
1443
1444         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1445         osb->net_key = le32_to_cpu(uuid_net_key);
1446
1447         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1448         osb->vol_label[63] = '\0';
1449         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1450         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1451         osb->first_cluster_group_blkno =
1452                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1453         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1454         mlog(0, "vol_label: %s\n", osb->vol_label);
1455         mlog(0, "uuid: %s\n", osb->uuid_str);
1456         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1457              (unsigned long long)osb->root_blkno,
1458              (unsigned long long)osb->system_dir_blkno);
1459
1460         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1461         if (!osb->osb_dlm_debug) {
1462                 status = -ENOMEM;
1463                 mlog_errno(status);
1464                 goto bail;
1465         }
1466
1467         atomic_set(&osb->vol_state, VOLUME_INIT);
1468
1469         /* load root, system_dir, and all global system inodes */
1470         status = ocfs2_init_global_system_inodes(osb);
1471         if (status < 0) {
1472                 mlog_errno(status);
1473                 goto bail;
1474         }
1475
1476         /*
1477          * global bitmap
1478          */
1479         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1480                                             OCFS2_INVALID_SLOT);
1481         if (!inode) {
1482                 status = -EINVAL;
1483                 mlog_errno(status);
1484                 goto bail;
1485         }
1486
1487         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1488
1489         /* We don't have a cluster lock on the bitmap here because
1490          * we're only interested in static information and the extra
1491          * complexity at mount time isn't worht it. Don't pass the
1492          * inode in to the read function though as we don't want it to
1493          * be put in the cache. */
1494         status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1495                                   NULL);
1496         iput(inode);
1497         if (status < 0) {
1498                 mlog_errno(status);
1499                 goto bail;
1500         }
1501
1502         di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1503         osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1504         brelse(bitmap_bh);
1505         mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1506              (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1507
1508         status = ocfs2_init_slot_info(osb);
1509         if (status < 0) {
1510                 mlog_errno(status);
1511                 goto bail;
1512         }
1513
1514 bail:
1515         mlog_exit(status);
1516         return status;
1517 }
1518
1519 /*
1520  * will return: -EAGAIN if it is ok to keep searching for superblocks
1521  *              -EINVAL if there is a bad superblock
1522  *              0 on success
1523  */
1524 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1525                                struct buffer_head *bh,
1526                                u32 blksz)
1527 {
1528         int status = -EAGAIN;
1529
1530         mlog_entry_void();
1531
1532         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1533                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1534                 status = -EINVAL;
1535                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1536                         mlog(ML_ERROR, "found superblock with incorrect block "
1537                              "size: found %u, should be %u\n",
1538                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1539                                blksz);
1540                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1541                            OCFS2_MAJOR_REV_LEVEL ||
1542                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1543                            OCFS2_MINOR_REV_LEVEL) {
1544                         mlog(ML_ERROR, "found superblock with bad version: "
1545                              "found %u.%u, should be %u.%u\n",
1546                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1547                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1548                              OCFS2_MAJOR_REV_LEVEL,
1549                              OCFS2_MINOR_REV_LEVEL);
1550                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1551                         mlog(ML_ERROR, "bad block number on superblock: "
1552                              "found %llu, should be %llu\n",
1553                              (unsigned long long)le64_to_cpu(di->i_blkno),
1554                              (unsigned long long)bh->b_blocknr);
1555                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1556                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1557                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1558                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1559                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1560                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1561                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1562                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1563                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1564                         mlog(ML_ERROR,
1565                              "Superblock slots found greater than file system "
1566                              "maximum: found %u, max %u\n",
1567                              le16_to_cpu(di->id2.i_super.s_max_slots),
1568                              OCFS2_MAX_SLOTS);
1569                 } else {
1570                         /* found it! */
1571                         status = 0;
1572                 }
1573         }
1574
1575         mlog_exit(status);
1576         return status;
1577 }
1578
1579 static int ocfs2_check_volume(struct ocfs2_super *osb)
1580 {
1581         int status = 0;
1582         int dirty;
1583         int local;
1584         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1585                                                   * recover
1586                                                   * ourselves. */
1587
1588         mlog_entry_void();
1589
1590         /* Init our journal object. */
1591         status = ocfs2_journal_init(osb->journal, &dirty);
1592         if (status < 0) {
1593                 mlog(ML_ERROR, "Could not initialize journal!\n");
1594                 goto finally;
1595         }
1596
1597         /* If the journal was unmounted cleanly then we don't want to
1598          * recover anything. Otherwise, journal_load will do that
1599          * dirty work for us :) */
1600         if (!dirty) {
1601                 status = ocfs2_journal_wipe(osb->journal, 0);
1602                 if (status < 0) {
1603                         mlog_errno(status);
1604                         goto finally;
1605                 }
1606         } else {
1607                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1608                      "recovering volume.\n");
1609         }
1610
1611         local = ocfs2_mount_local(osb);
1612
1613         /* will play back anything left in the journal. */
1614         ocfs2_journal_load(osb->journal, local);
1615
1616         if (dirty) {
1617                 /* recover my local alloc if we didn't unmount cleanly. */
1618                 status = ocfs2_begin_local_alloc_recovery(osb,
1619                                                           osb->slot_num,
1620                                                           &local_alloc);
1621                 if (status < 0) {
1622                         mlog_errno(status);
1623                         goto finally;
1624                 }
1625                 /* we complete the recovery process after we've marked
1626                  * ourselves as mounted. */
1627         }
1628
1629         mlog(0, "Journal loaded.\n");
1630
1631         status = ocfs2_load_local_alloc(osb);
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto finally;
1635         }
1636
1637         if (dirty) {
1638                 /* Recovery will be completed after we've mounted the
1639                  * rest of the volume. */
1640                 osb->dirty = 1;
1641                 osb->local_alloc_copy = local_alloc;
1642                 local_alloc = NULL;
1643         }
1644
1645         /* go through each journal, trylock it and if you get the
1646          * lock, and it's marked as dirty, set the bit in the recover
1647          * map and launch a recovery thread for it. */
1648         status = ocfs2_mark_dead_nodes(osb);
1649         if (status < 0)
1650                 mlog_errno(status);
1651
1652 finally:
1653         if (local_alloc)
1654                 kfree(local_alloc);
1655
1656         mlog_exit(status);
1657         return status;
1658 }
1659
1660 /*
1661  * The routine gets called from dismount or close whenever a dismount on
1662  * volume is requested and the osb open count becomes 1.
1663  * It will remove the osb from the global list and also free up all the
1664  * initialized resources and fileobject.
1665  */
1666 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1667 {
1668         mlog_entry_void();
1669
1670         /* This function assumes that the caller has the main osb resource */
1671
1672         if (osb->slot_info)
1673                 ocfs2_free_slot_info(osb->slot_info);
1674
1675         kfree(osb->osb_orphan_wipes);
1676         /* FIXME
1677          * This belongs in journal shutdown, but because we have to
1678          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1679          * we free it here.
1680          */
1681         kfree(osb->journal);
1682         if (osb->local_alloc_copy)
1683                 kfree(osb->local_alloc_copy);
1684         kfree(osb->uuid_str);
1685         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1686         memset(osb, 0, sizeof(struct ocfs2_super));
1687
1688         mlog_exit_void();
1689 }
1690
1691 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1692  * panic(). We do not support continue-on-error operation. */
1693 static void ocfs2_handle_error(struct super_block *sb)
1694 {
1695         struct ocfs2_super *osb = OCFS2_SB(sb);
1696
1697         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1698                 panic("OCFS2: (device %s): panic forced after error\n",
1699                       sb->s_id);
1700
1701         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1702
1703         if (sb->s_flags & MS_RDONLY &&
1704             (ocfs2_is_soft_readonly(osb) ||
1705              ocfs2_is_hard_readonly(osb)))
1706                 return;
1707
1708         printk(KERN_CRIT "File system is now read-only due to the potential "
1709                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1710                "system is unmounted.\n");
1711         sb->s_flags |= MS_RDONLY;
1712         ocfs2_set_ro_flag(osb, 0);
1713 }
1714
1715 static char error_buf[1024];
1716
1717 void __ocfs2_error(struct super_block *sb,
1718                    const char *function,
1719                    const char *fmt, ...)
1720 {
1721         va_list args;
1722
1723         va_start(args, fmt);
1724         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1725         va_end(args);
1726
1727         /* Not using mlog here because we want to show the actual
1728          * function the error came from. */
1729         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1730                sb->s_id, function, error_buf);
1731
1732         ocfs2_handle_error(sb);
1733 }
1734
1735 /* Handle critical errors. This is intentionally more drastic than
1736  * ocfs2_handle_error, so we only use for things like journal errors,
1737  * etc. */
1738 void __ocfs2_abort(struct super_block* sb,
1739                    const char *function,
1740                    const char *fmt, ...)
1741 {
1742         va_list args;
1743
1744         va_start(args, fmt);
1745         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1746         va_end(args);
1747
1748         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1749                sb->s_id, function, error_buf);
1750
1751         /* We don't have the cluster support yet to go straight to
1752          * hard readonly in here. Until then, we want to keep
1753          * ocfs2_abort() so that we can at least mark critical
1754          * errors.
1755          *
1756          * TODO: This should abort the journal and alert other nodes
1757          * that our slot needs recovery. */
1758
1759         /* Force a panic(). This stinks, but it's better than letting
1760          * things continue without having a proper hard readonly
1761          * here. */
1762         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1763         ocfs2_handle_error(sb);
1764 }
1765
1766 module_init(ocfs2_init);
1767 module_exit(ocfs2_exit);