xref: /linux/fs/ocfs2/super.c (revision 2b8232ce512105e28453f301d1510de8363bccd1)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * super.c
5  *
6  * load/unload driver, mount/dismount volumes
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/module.h>
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
42 #include <linux/mount.h>
43 
44 #include <cluster/nodemanager.h>
45 
46 #define MLOG_MASK_PREFIX ML_SUPER
47 #include <cluster/masklog.h>
48 
49 #include "ocfs2.h"
50 
51 /* this should be the only file to include a version 1 header */
52 #include "ocfs1_fs_compat.h"
53 
54 #include "alloc.h"
55 #include "dlmglue.h"
56 #include "export.h"
57 #include "extent_map.h"
58 #include "heartbeat.h"
59 #include "inode.h"
60 #include "journal.h"
61 #include "localalloc.h"
62 #include "namei.h"
63 #include "slot_map.h"
64 #include "super.h"
65 #include "sysfile.h"
66 #include "uptodate.h"
67 #include "ver.h"
68 #include "vote.h"
69 
70 #include "buffer_head_io.h"
71 
72 static struct kmem_cache *ocfs2_inode_cachep = NULL;
73 
74 /* OCFS2 needs to schedule several differnt types of work which
75  * require cluster locking, disk I/O, recovery waits, etc. Since these
76  * types of work tend to be heavy we avoid using the kernel events
77  * workqueue and schedule on our own. */
78 struct workqueue_struct *ocfs2_wq = NULL;
79 
80 static struct dentry *ocfs2_debugfs_root = NULL;
81 
82 MODULE_AUTHOR("Oracle");
83 MODULE_LICENSE("GPL");
84 
85 struct mount_options
86 {
87 	unsigned long	mount_opt;
88 	unsigned int	atime_quantum;
89 	signed short	slot;
90 };
91 
92 static int ocfs2_parse_options(struct super_block *sb, char *options,
93 			       struct mount_options *mopt,
94 			       int is_remount);
95 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
96 static void ocfs2_put_super(struct super_block *sb);
97 static int ocfs2_mount_volume(struct super_block *sb);
98 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
99 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
100 static int ocfs2_initialize_mem_caches(void);
101 static void ocfs2_free_mem_caches(void);
102 static void ocfs2_delete_osb(struct ocfs2_super *osb);
103 
104 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
105 
106 static int ocfs2_sync_fs(struct super_block *sb, int wait);
107 
108 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
109 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
110 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
111 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
112 static int ocfs2_check_volume(struct ocfs2_super *osb);
113 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
114 			       struct buffer_head *bh,
115 			       u32 sectsize);
116 static int ocfs2_initialize_super(struct super_block *sb,
117 				  struct buffer_head *bh,
118 				  int sector_size);
119 static int ocfs2_get_sector(struct super_block *sb,
120 			    struct buffer_head **bh,
121 			    int block,
122 			    int sect_size);
123 static void ocfs2_write_super(struct super_block *sb);
124 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
125 static void ocfs2_destroy_inode(struct inode *inode);
126 
127 static const struct super_operations ocfs2_sops = {
128 	.statfs		= ocfs2_statfs,
129 	.alloc_inode	= ocfs2_alloc_inode,
130 	.destroy_inode	= ocfs2_destroy_inode,
131 	.drop_inode	= ocfs2_drop_inode,
132 	.clear_inode	= ocfs2_clear_inode,
133 	.delete_inode	= ocfs2_delete_inode,
134 	.sync_fs	= ocfs2_sync_fs,
135 	.write_super	= ocfs2_write_super,
136 	.put_super	= ocfs2_put_super,
137 	.remount_fs	= ocfs2_remount,
138 	.show_options   = ocfs2_show_options,
139 };
140 
141 enum {
142 	Opt_barrier,
143 	Opt_err_panic,
144 	Opt_err_ro,
145 	Opt_intr,
146 	Opt_nointr,
147 	Opt_hb_none,
148 	Opt_hb_local,
149 	Opt_data_ordered,
150 	Opt_data_writeback,
151 	Opt_atime_quantum,
152 	Opt_slot,
153 	Opt_err,
154 };
155 
156 static match_table_t tokens = {
157 	{Opt_barrier, "barrier=%u"},
158 	{Opt_err_panic, "errors=panic"},
159 	{Opt_err_ro, "errors=remount-ro"},
160 	{Opt_intr, "intr"},
161 	{Opt_nointr, "nointr"},
162 	{Opt_hb_none, OCFS2_HB_NONE},
163 	{Opt_hb_local, OCFS2_HB_LOCAL},
164 	{Opt_data_ordered, "data=ordered"},
165 	{Opt_data_writeback, "data=writeback"},
166 	{Opt_atime_quantum, "atime_quantum=%u"},
167 	{Opt_slot, "preferred_slot=%u"},
168 	{Opt_err, NULL}
169 };
170 
171 /*
172  * write_super and sync_fs ripped right out of ext3.
173  */
174 static void ocfs2_write_super(struct super_block *sb)
175 {
176 	if (mutex_trylock(&sb->s_lock) != 0)
177 		BUG();
178 	sb->s_dirt = 0;
179 }
180 
181 static int ocfs2_sync_fs(struct super_block *sb, int wait)
182 {
183 	int status;
184 	tid_t target;
185 	struct ocfs2_super *osb = OCFS2_SB(sb);
186 
187 	sb->s_dirt = 0;
188 
189 	if (ocfs2_is_hard_readonly(osb))
190 		return -EROFS;
191 
192 	if (wait) {
193 		status = ocfs2_flush_truncate_log(osb);
194 		if (status < 0)
195 			mlog_errno(status);
196 	} else {
197 		ocfs2_schedule_truncate_log_flush(osb, 0);
198 	}
199 
200 	if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
201 		if (wait)
202 			log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
203 					target);
204 	}
205 	return 0;
206 }
207 
208 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
209 {
210 	struct inode *new = NULL;
211 	int status = 0;
212 	int i;
213 
214 	mlog_entry_void();
215 
216 	new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
217 	if (IS_ERR(new)) {
218 		status = PTR_ERR(new);
219 		mlog_errno(status);
220 		goto bail;
221 	}
222 	osb->root_inode = new;
223 
224 	new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
225 	if (IS_ERR(new)) {
226 		status = PTR_ERR(new);
227 		mlog_errno(status);
228 		goto bail;
229 	}
230 	osb->sys_root_inode = new;
231 
232 	for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
233 	     i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
234 		new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
235 		if (!new) {
236 			ocfs2_release_system_inodes(osb);
237 			status = -EINVAL;
238 			mlog_errno(status);
239 			/* FIXME: Should ERROR_RO_FS */
240 			mlog(ML_ERROR, "Unable to load system inode %d, "
241 			     "possibly corrupt fs?", i);
242 			goto bail;
243 		}
244 		// the array now has one ref, so drop this one
245 		iput(new);
246 	}
247 
248 bail:
249 	mlog_exit(status);
250 	return status;
251 }
252 
253 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
254 {
255 	struct inode *new = NULL;
256 	int status = 0;
257 	int i;
258 
259 	mlog_entry_void();
260 
261 	for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
262 	     i < NUM_SYSTEM_INODES;
263 	     i++) {
264 		new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
265 		if (!new) {
266 			ocfs2_release_system_inodes(osb);
267 			status = -EINVAL;
268 			mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
269 			     status, i, osb->slot_num);
270 			goto bail;
271 		}
272 		/* the array now has one ref, so drop this one */
273 		iput(new);
274 	}
275 
276 bail:
277 	mlog_exit(status);
278 	return status;
279 }
280 
281 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
282 {
283 	int i;
284 	struct inode *inode;
285 
286 	mlog_entry_void();
287 
288 	for (i = 0; i < NUM_SYSTEM_INODES; i++) {
289 		inode = osb->system_inodes[i];
290 		if (inode) {
291 			iput(inode);
292 			osb->system_inodes[i] = NULL;
293 		}
294 	}
295 
296 	inode = osb->sys_root_inode;
297 	if (inode) {
298 		iput(inode);
299 		osb->sys_root_inode = NULL;
300 	}
301 
302 	inode = osb->root_inode;
303 	if (inode) {
304 		iput(inode);
305 		osb->root_inode = NULL;
306 	}
307 
308 	mlog_exit(0);
309 }
310 
311 /* We're allocating fs objects, use GFP_NOFS */
312 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
313 {
314 	struct ocfs2_inode_info *oi;
315 
316 	oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
317 	if (!oi)
318 		return NULL;
319 
320 	return &oi->vfs_inode;
321 }
322 
323 static void ocfs2_destroy_inode(struct inode *inode)
324 {
325 	kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
326 }
327 
328 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
329 						unsigned int cbits)
330 {
331 	unsigned int bytes = 1 << cbits;
332 	unsigned int trim = bytes;
333 	unsigned int bitshift = 32;
334 
335 	/*
336 	 * i_size and all block offsets in ocfs2 are always 64 bits
337 	 * wide. i_clusters is 32 bits, in cluster-sized units. So on
338 	 * 64 bit platforms, cluster size will be the limiting factor.
339 	 */
340 
341 #if BITS_PER_LONG == 32
342 # if defined(CONFIG_LBD)
343 	BUILD_BUG_ON(sizeof(sector_t) != 8);
344 	/*
345 	 * We might be limited by page cache size.
346 	 */
347 	if (bytes > PAGE_CACHE_SIZE) {
348 		bytes = PAGE_CACHE_SIZE;
349 		trim = 1;
350 		/*
351 		 * Shift by 31 here so that we don't get larger than
352 		 * MAX_LFS_FILESIZE
353 		 */
354 		bitshift = 31;
355 	}
356 # else
357 	/*
358 	 * We are limited by the size of sector_t. Use block size, as
359 	 * that's what we expose to the VFS.
360 	 */
361 	bytes = 1 << bbits;
362 	trim = 1;
363 	bitshift = 31;
364 # endif
365 #endif
366 
367 	/*
368 	 * Trim by a whole cluster when we can actually approach the
369 	 * on-disk limits. Otherwise we can overflow i_clusters when
370 	 * an extent start is at the max offset.
371 	 */
372 	return (((unsigned long long)bytes) << bitshift) - trim;
373 }
374 
375 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
376 {
377 	int incompat_features;
378 	int ret = 0;
379 	struct mount_options parsed_options;
380 	struct ocfs2_super *osb = OCFS2_SB(sb);
381 
382 	if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
383 		ret = -EINVAL;
384 		goto out;
385 	}
386 
387 	if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
388 	    (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
389 		ret = -EINVAL;
390 		mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
391 		goto out;
392 	}
393 
394 	if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
395 	    (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
396 		ret = -EINVAL;
397 		mlog(ML_ERROR, "Cannot change data mode on remount\n");
398 		goto out;
399 	}
400 
401 	/* We're going to/from readonly mode. */
402 	if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
403 		/* Lock here so the check of HARD_RO and the potential
404 		 * setting of SOFT_RO is atomic. */
405 		spin_lock(&osb->osb_lock);
406 		if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
407 			mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
408 			ret = -EROFS;
409 			goto unlock_osb;
410 		}
411 
412 		if (*flags & MS_RDONLY) {
413 			mlog(0, "Going to ro mode.\n");
414 			sb->s_flags |= MS_RDONLY;
415 			osb->osb_flags |= OCFS2_OSB_SOFT_RO;
416 		} else {
417 			mlog(0, "Making ro filesystem writeable.\n");
418 
419 			if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
420 				mlog(ML_ERROR, "Cannot remount RDWR "
421 				     "filesystem due to previous errors.\n");
422 				ret = -EROFS;
423 				goto unlock_osb;
424 			}
425 			incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
426 			if (incompat_features) {
427 				mlog(ML_ERROR, "Cannot remount RDWR because "
428 				     "of unsupported optional features "
429 				     "(%x).\n", incompat_features);
430 				ret = -EINVAL;
431 				goto unlock_osb;
432 			}
433 			sb->s_flags &= ~MS_RDONLY;
434 			osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
435 		}
436 unlock_osb:
437 		spin_unlock(&osb->osb_lock);
438 	}
439 
440 	if (!ret) {
441 		if (!ocfs2_is_hard_readonly(osb))
442 			ocfs2_set_journal_params(osb);
443 
444 		/* Only save off the new mount options in case of a successful
445 		 * remount. */
446 		osb->s_mount_opt = parsed_options.mount_opt;
447 		osb->s_atime_quantum = parsed_options.atime_quantum;
448 		osb->preferred_slot = parsed_options.slot;
449 	}
450 out:
451 	return ret;
452 }
453 
454 static int ocfs2_sb_probe(struct super_block *sb,
455 			  struct buffer_head **bh,
456 			  int *sector_size)
457 {
458 	int status, tmpstat;
459 	struct ocfs1_vol_disk_hdr *hdr;
460 	struct ocfs2_dinode *di;
461 	int blksize;
462 
463 	*bh = NULL;
464 
465 	/* may be > 512 */
466 	*sector_size = bdev_hardsect_size(sb->s_bdev);
467 	if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
468 		mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
469 		     *sector_size, OCFS2_MAX_BLOCKSIZE);
470 		status = -EINVAL;
471 		goto bail;
472 	}
473 
474 	/* Can this really happen? */
475 	if (*sector_size < OCFS2_MIN_BLOCKSIZE)
476 		*sector_size = OCFS2_MIN_BLOCKSIZE;
477 
478 	/* check block zero for old format */
479 	status = ocfs2_get_sector(sb, bh, 0, *sector_size);
480 	if (status < 0) {
481 		mlog_errno(status);
482 		goto bail;
483 	}
484 	hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
485 	if (hdr->major_version == OCFS1_MAJOR_VERSION) {
486 		mlog(ML_ERROR, "incompatible version: %u.%u\n",
487 		     hdr->major_version, hdr->minor_version);
488 		status = -EINVAL;
489 	}
490 	if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
491 		   strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
492 		mlog(ML_ERROR, "incompatible volume signature: %8s\n",
493 		     hdr->signature);
494 		status = -EINVAL;
495 	}
496 	brelse(*bh);
497 	*bh = NULL;
498 	if (status < 0) {
499 		mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
500 		     "upgraded before mounting with ocfs v2\n");
501 		goto bail;
502 	}
503 
504 	/*
505 	 * Now check at magic offset for 512, 1024, 2048, 4096
506 	 * blocksizes.  4096 is the maximum blocksize because it is
507 	 * the minimum clustersize.
508 	 */
509 	status = -EINVAL;
510 	for (blksize = *sector_size;
511 	     blksize <= OCFS2_MAX_BLOCKSIZE;
512 	     blksize <<= 1) {
513 		tmpstat = ocfs2_get_sector(sb, bh,
514 					   OCFS2_SUPER_BLOCK_BLKNO,
515 					   blksize);
516 		if (tmpstat < 0) {
517 			status = tmpstat;
518 			mlog_errno(status);
519 			goto bail;
520 		}
521 		di = (struct ocfs2_dinode *) (*bh)->b_data;
522 		status = ocfs2_verify_volume(di, *bh, blksize);
523 		if (status >= 0)
524 			goto bail;
525 		brelse(*bh);
526 		*bh = NULL;
527 		if (status != -EAGAIN)
528 			break;
529 	}
530 
531 bail:
532 	return status;
533 }
534 
535 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
536 {
537 	if (ocfs2_mount_local(osb)) {
538 		if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
539 			mlog(ML_ERROR, "Cannot heartbeat on a locally "
540 			     "mounted device.\n");
541 			return -EINVAL;
542 		}
543 	}
544 
545 	if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
546 		if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
547 			mlog(ML_ERROR, "Heartbeat has to be started to mount "
548 			     "a read-write clustered device.\n");
549 			return -EINVAL;
550 		}
551 	}
552 
553 	return 0;
554 }
555 
556 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
557 {
558 	struct dentry *root;
559 	int status, sector_size;
560 	struct mount_options parsed_options;
561 	struct inode *inode = NULL;
562 	struct ocfs2_super *osb = NULL;
563 	struct buffer_head *bh = NULL;
564 	char nodestr[8];
565 
566 	mlog_entry("%p, %p, %i", sb, data, silent);
567 
568 	if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
569 		status = -EINVAL;
570 		goto read_super_error;
571 	}
572 
573 	/* for now we only have one cluster/node, make sure we see it
574 	 * in the heartbeat universe */
575 	if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
576 		if (!o2hb_check_local_node_heartbeating()) {
577 			status = -EINVAL;
578 			goto read_super_error;
579 		}
580 	}
581 
582 	/* probe for superblock */
583 	status = ocfs2_sb_probe(sb, &bh, &sector_size);
584 	if (status < 0) {
585 		mlog(ML_ERROR, "superblock probe failed!\n");
586 		goto read_super_error;
587 	}
588 
589 	status = ocfs2_initialize_super(sb, bh, sector_size);
590 	osb = OCFS2_SB(sb);
591 	if (status < 0) {
592 		mlog_errno(status);
593 		goto read_super_error;
594 	}
595 	brelse(bh);
596 	bh = NULL;
597 	osb->s_mount_opt = parsed_options.mount_opt;
598 	osb->s_atime_quantum = parsed_options.atime_quantum;
599 	osb->preferred_slot = parsed_options.slot;
600 
601 	sb->s_magic = OCFS2_SUPER_MAGIC;
602 
603 	/* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
604 	 * heartbeat=none */
605 	if (bdev_read_only(sb->s_bdev)) {
606 		if (!(sb->s_flags & MS_RDONLY)) {
607 			status = -EACCES;
608 			mlog(ML_ERROR, "Readonly device detected but readonly "
609 			     "mount was not specified.\n");
610 			goto read_super_error;
611 		}
612 
613 		/* You should not be able to start a local heartbeat
614 		 * on a readonly device. */
615 		if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
616 			status = -EROFS;
617 			mlog(ML_ERROR, "Local heartbeat specified on readonly "
618 			     "device.\n");
619 			goto read_super_error;
620 		}
621 
622 		status = ocfs2_check_journals_nolocks(osb);
623 		if (status < 0) {
624 			if (status == -EROFS)
625 				mlog(ML_ERROR, "Recovery required on readonly "
626 				     "file system, but write access is "
627 				     "unavailable.\n");
628 			else
629 				mlog_errno(status);
630 			goto read_super_error;
631 		}
632 
633 		ocfs2_set_ro_flag(osb, 1);
634 
635 		printk(KERN_NOTICE "Readonly device detected. No cluster "
636 		       "services will be utilized for this mount. Recovery "
637 		       "will be skipped.\n");
638 	}
639 
640 	if (!ocfs2_is_hard_readonly(osb)) {
641 		if (sb->s_flags & MS_RDONLY)
642 			ocfs2_set_ro_flag(osb, 0);
643 	}
644 
645 	status = ocfs2_verify_heartbeat(osb);
646 	if (status < 0) {
647 		mlog_errno(status);
648 		goto read_super_error;
649 	}
650 
651 	osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
652 						 ocfs2_debugfs_root);
653 	if (!osb->osb_debug_root) {
654 		status = -EINVAL;
655 		mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
656 		goto read_super_error;
657 	}
658 
659 	status = ocfs2_mount_volume(sb);
660 	if (osb->root_inode)
661 		inode = igrab(osb->root_inode);
662 
663 	if (status < 0)
664 		goto read_super_error;
665 
666 	if (!inode) {
667 		status = -EIO;
668 		mlog_errno(status);
669 		goto read_super_error;
670 	}
671 
672 	root = d_alloc_root(inode);
673 	if (!root) {
674 		status = -ENOMEM;
675 		mlog_errno(status);
676 		goto read_super_error;
677 	}
678 
679 	sb->s_root = root;
680 
681 	ocfs2_complete_mount_recovery(osb);
682 
683 	if (ocfs2_mount_local(osb))
684 		snprintf(nodestr, sizeof(nodestr), "local");
685 	else
686 		snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
687 
688 	printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
689 	       "with %s data mode.\n",
690 	       osb->dev_str, nodestr, osb->slot_num,
691 	       osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
692 	       "ordered");
693 
694 	atomic_set(&osb->vol_state, VOLUME_MOUNTED);
695 	wake_up(&osb->osb_mount_event);
696 
697 	mlog_exit(status);
698 	return status;
699 
700 read_super_error:
701 	if (bh != NULL)
702 		brelse(bh);
703 
704 	if (inode)
705 		iput(inode);
706 
707 	if (osb) {
708 		atomic_set(&osb->vol_state, VOLUME_DISABLED);
709 		wake_up(&osb->osb_mount_event);
710 		ocfs2_dismount_volume(sb, 1);
711 	}
712 
713 	mlog_exit(status);
714 	return status;
715 }
716 
717 static int ocfs2_get_sb(struct file_system_type *fs_type,
718 			int flags,
719 			const char *dev_name,
720 			void *data,
721 			struct vfsmount *mnt)
722 {
723 	return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
724 			   mnt);
725 }
726 
727 static struct file_system_type ocfs2_fs_type = {
728 	.owner          = THIS_MODULE,
729 	.name           = "ocfs2",
730 	.get_sb         = ocfs2_get_sb, /* is this called when we mount
731 					* the fs? */
732 	.kill_sb        = kill_block_super, /* set to the generic one
733 					     * right now, but do we
734 					     * need to change that? */
735 	.fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
736 	.next           = NULL
737 };
738 
739 static int ocfs2_parse_options(struct super_block *sb,
740 			       char *options,
741 			       struct mount_options *mopt,
742 			       int is_remount)
743 {
744 	int status;
745 	char *p;
746 
747 	mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
748 		   options ? options : "(none)");
749 
750 	mopt->mount_opt = 0;
751 	mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
752 	mopt->slot = OCFS2_INVALID_SLOT;
753 
754 	if (!options) {
755 		status = 1;
756 		goto bail;
757 	}
758 
759 	while ((p = strsep(&options, ",")) != NULL) {
760 		int token, option;
761 		substring_t args[MAX_OPT_ARGS];
762 
763 		if (!*p)
764 			continue;
765 
766 		token = match_token(p, tokens, args);
767 		switch (token) {
768 		case Opt_hb_local:
769 			mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
770 			break;
771 		case Opt_hb_none:
772 			mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
773 			break;
774 		case Opt_barrier:
775 			if (match_int(&args[0], &option)) {
776 				status = 0;
777 				goto bail;
778 			}
779 			if (option)
780 				mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
781 			else
782 				mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
783 			break;
784 		case Opt_intr:
785 			mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
786 			break;
787 		case Opt_nointr:
788 			mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
789 			break;
790 		case Opt_err_panic:
791 			mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
792 			break;
793 		case Opt_err_ro:
794 			mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
795 			break;
796 		case Opt_data_ordered:
797 			mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
798 			break;
799 		case Opt_data_writeback:
800 			mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
801 			break;
802 		case Opt_atime_quantum:
803 			if (match_int(&args[0], &option)) {
804 				status = 0;
805 				goto bail;
806 			}
807 			if (option >= 0)
808 				mopt->atime_quantum = option;
809 			break;
810 		case Opt_slot:
811 			option = 0;
812 			if (match_int(&args[0], &option)) {
813 				status = 0;
814 				goto bail;
815 			}
816 			if (option)
817 				mopt->slot = (s16)option;
818 			break;
819 		default:
820 			mlog(ML_ERROR,
821 			     "Unrecognized mount option \"%s\" "
822 			     "or missing value\n", p);
823 			status = 0;
824 			goto bail;
825 		}
826 	}
827 
828 	status = 1;
829 
830 bail:
831 	mlog_exit(status);
832 	return status;
833 }
834 
835 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
836 {
837 	struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
838 	unsigned long opts = osb->s_mount_opt;
839 
840 	if (opts & OCFS2_MOUNT_HB_LOCAL)
841 		seq_printf(s, ",_netdev,heartbeat=local");
842 	else
843 		seq_printf(s, ",heartbeat=none");
844 
845 	if (opts & OCFS2_MOUNT_NOINTR)
846 		seq_printf(s, ",nointr");
847 
848 	if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
849 		seq_printf(s, ",data=writeback");
850 	else
851 		seq_printf(s, ",data=ordered");
852 
853 	if (opts & OCFS2_MOUNT_BARRIER)
854 		seq_printf(s, ",barrier=1");
855 
856 	if (opts & OCFS2_MOUNT_ERRORS_PANIC)
857 		seq_printf(s, ",errors=panic");
858 	else
859 		seq_printf(s, ",errors=remount-ro");
860 
861 	if (osb->preferred_slot != OCFS2_INVALID_SLOT)
862 		seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
863 
864 	if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
865 		seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
866 
867 	return 0;
868 }
869 
870 static int __init ocfs2_init(void)
871 {
872 	int status;
873 
874 	mlog_entry_void();
875 
876 	ocfs2_print_version();
877 
878 	status = init_ocfs2_uptodate_cache();
879 	if (status < 0) {
880 		mlog_errno(status);
881 		goto leave;
882 	}
883 
884 	status = ocfs2_initialize_mem_caches();
885 	if (status < 0) {
886 		mlog_errno(status);
887 		goto leave;
888 	}
889 
890 	ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
891 	if (!ocfs2_wq) {
892 		status = -ENOMEM;
893 		goto leave;
894 	}
895 
896 	ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
897 	if (!ocfs2_debugfs_root) {
898 		status = -EFAULT;
899 		mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
900 	}
901 
902 leave:
903 	if (status < 0) {
904 		ocfs2_free_mem_caches();
905 		exit_ocfs2_uptodate_cache();
906 	}
907 
908 	mlog_exit(status);
909 
910 	if (status >= 0) {
911 		return register_filesystem(&ocfs2_fs_type);
912 	} else
913 		return -1;
914 }
915 
916 static void __exit ocfs2_exit(void)
917 {
918 	mlog_entry_void();
919 
920 	if (ocfs2_wq) {
921 		flush_workqueue(ocfs2_wq);
922 		destroy_workqueue(ocfs2_wq);
923 	}
924 
925 	debugfs_remove(ocfs2_debugfs_root);
926 
927 	ocfs2_free_mem_caches();
928 
929 	unregister_filesystem(&ocfs2_fs_type);
930 
931 	exit_ocfs2_uptodate_cache();
932 
933 	mlog_exit_void();
934 }
935 
936 static void ocfs2_put_super(struct super_block *sb)
937 {
938 	mlog_entry("(0x%p)\n", sb);
939 
940 	ocfs2_sync_blockdev(sb);
941 	ocfs2_dismount_volume(sb, 0);
942 
943 	mlog_exit_void();
944 }
945 
946 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
947 {
948 	struct ocfs2_super *osb;
949 	u32 numbits, freebits;
950 	int status;
951 	struct ocfs2_dinode *bm_lock;
952 	struct buffer_head *bh = NULL;
953 	struct inode *inode = NULL;
954 
955 	mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
956 
957 	osb = OCFS2_SB(dentry->d_sb);
958 
959 	inode = ocfs2_get_system_file_inode(osb,
960 					    GLOBAL_BITMAP_SYSTEM_INODE,
961 					    OCFS2_INVALID_SLOT);
962 	if (!inode) {
963 		mlog(ML_ERROR, "failed to get bitmap inode\n");
964 		status = -EIO;
965 		goto bail;
966 	}
967 
968 	status = ocfs2_meta_lock(inode, &bh, 0);
969 	if (status < 0) {
970 		mlog_errno(status);
971 		goto bail;
972 	}
973 
974 	bm_lock = (struct ocfs2_dinode *) bh->b_data;
975 
976 	numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
977 	freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
978 
979 	buf->f_type = OCFS2_SUPER_MAGIC;
980 	buf->f_bsize = dentry->d_sb->s_blocksize;
981 	buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
982 	buf->f_blocks = ((sector_t) numbits) *
983 			(osb->s_clustersize >> osb->sb->s_blocksize_bits);
984 	buf->f_bfree = ((sector_t) freebits) *
985 		       (osb->s_clustersize >> osb->sb->s_blocksize_bits);
986 	buf->f_bavail = buf->f_bfree;
987 	buf->f_files = numbits;
988 	buf->f_ffree = freebits;
989 
990 	brelse(bh);
991 
992 	ocfs2_meta_unlock(inode, 0);
993 	status = 0;
994 bail:
995 	if (inode)
996 		iput(inode);
997 
998 	mlog_exit(status);
999 
1000 	return status;
1001 }
1002 
1003 static void ocfs2_inode_init_once(void *data,
1004 				  struct kmem_cache *cachep,
1005 				  unsigned long flags)
1006 {
1007 	struct ocfs2_inode_info *oi = data;
1008 
1009 	oi->ip_flags = 0;
1010 	oi->ip_open_count = 0;
1011 	spin_lock_init(&oi->ip_lock);
1012 	ocfs2_extent_map_init(&oi->vfs_inode);
1013 	INIT_LIST_HEAD(&oi->ip_io_markers);
1014 	oi->ip_created_trans = 0;
1015 	oi->ip_last_trans = 0;
1016 	oi->ip_dir_start_lookup = 0;
1017 
1018 	init_rwsem(&oi->ip_alloc_sem);
1019 	mutex_init(&oi->ip_io_mutex);
1020 
1021 	oi->ip_blkno = 0ULL;
1022 	oi->ip_clusters = 0;
1023 
1024 	ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1025 	ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
1026 	ocfs2_lock_res_init_once(&oi->ip_data_lockres);
1027 	ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1028 
1029 	ocfs2_metadata_cache_init(&oi->vfs_inode);
1030 
1031 	inode_init_once(&oi->vfs_inode);
1032 }
1033 
1034 static int ocfs2_initialize_mem_caches(void)
1035 {
1036 	ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1037 				       sizeof(struct ocfs2_inode_info),
1038 				       0,
1039 				       (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1040 						SLAB_MEM_SPREAD),
1041 				       ocfs2_inode_init_once);
1042 	if (!ocfs2_inode_cachep)
1043 		return -ENOMEM;
1044 
1045 	return 0;
1046 }
1047 
1048 static void ocfs2_free_mem_caches(void)
1049 {
1050 	if (ocfs2_inode_cachep)
1051 		kmem_cache_destroy(ocfs2_inode_cachep);
1052 
1053 	ocfs2_inode_cachep = NULL;
1054 }
1055 
1056 static int ocfs2_get_sector(struct super_block *sb,
1057 			    struct buffer_head **bh,
1058 			    int block,
1059 			    int sect_size)
1060 {
1061 	if (!sb_set_blocksize(sb, sect_size)) {
1062 		mlog(ML_ERROR, "unable to set blocksize\n");
1063 		return -EIO;
1064 	}
1065 
1066 	*bh = sb_getblk(sb, block);
1067 	if (!*bh) {
1068 		mlog_errno(-EIO);
1069 		return -EIO;
1070 	}
1071 	lock_buffer(*bh);
1072 	if (!buffer_dirty(*bh))
1073 		clear_buffer_uptodate(*bh);
1074 	unlock_buffer(*bh);
1075 	ll_rw_block(READ, 1, bh);
1076 	wait_on_buffer(*bh);
1077 	return 0;
1078 }
1079 
1080 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1081 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1082 {
1083 	int status;
1084 
1085 	/* XXX hold a ref on the node while mounte?  easy enough, if
1086 	 * desirable. */
1087 	if (ocfs2_mount_local(osb))
1088 		osb->node_num = 0;
1089 	else
1090 		osb->node_num = o2nm_this_node();
1091 
1092 	if (osb->node_num == O2NM_MAX_NODES) {
1093 		mlog(ML_ERROR, "could not find this host's node number\n");
1094 		status = -ENOENT;
1095 		goto bail;
1096 	}
1097 
1098 	mlog(0, "I am node %d\n", osb->node_num);
1099 
1100 	status = 0;
1101 bail:
1102 	return status;
1103 }
1104 
1105 static int ocfs2_mount_volume(struct super_block *sb)
1106 {
1107 	int status = 0;
1108 	int unlock_super = 0;
1109 	struct ocfs2_super *osb = OCFS2_SB(sb);
1110 
1111 	mlog_entry_void();
1112 
1113 	if (ocfs2_is_hard_readonly(osb))
1114 		goto leave;
1115 
1116 	status = ocfs2_fill_local_node_info(osb);
1117 	if (status < 0) {
1118 		mlog_errno(status);
1119 		goto leave;
1120 	}
1121 
1122 	status = ocfs2_register_hb_callbacks(osb);
1123 	if (status < 0) {
1124 		mlog_errno(status);
1125 		goto leave;
1126 	}
1127 
1128 	status = ocfs2_dlm_init(osb);
1129 	if (status < 0) {
1130 		mlog_errno(status);
1131 		goto leave;
1132 	}
1133 
1134 	/* requires vote_thread to be running. */
1135 	status = ocfs2_register_net_handlers(osb);
1136 	if (status < 0) {
1137 		mlog_errno(status);
1138 		goto leave;
1139 	}
1140 
1141 	status = ocfs2_super_lock(osb, 1);
1142 	if (status < 0) {
1143 		mlog_errno(status);
1144 		goto leave;
1145 	}
1146 	unlock_super = 1;
1147 
1148 	/* This will load up the node map and add ourselves to it. */
1149 	status = ocfs2_find_slot(osb);
1150 	if (status < 0) {
1151 		mlog_errno(status);
1152 		goto leave;
1153 	}
1154 
1155 	ocfs2_populate_mounted_map(osb);
1156 
1157 	/* load all node-local system inodes */
1158 	status = ocfs2_init_local_system_inodes(osb);
1159 	if (status < 0) {
1160 		mlog_errno(status);
1161 		goto leave;
1162 	}
1163 
1164 	status = ocfs2_check_volume(osb);
1165 	if (status < 0) {
1166 		mlog_errno(status);
1167 		goto leave;
1168 	}
1169 
1170 	status = ocfs2_truncate_log_init(osb);
1171 	if (status < 0) {
1172 		mlog_errno(status);
1173 		goto leave;
1174 	}
1175 
1176 	if (ocfs2_mount_local(osb))
1177 		goto leave;
1178 
1179 	/* This should be sent *after* we recovered our journal as it
1180 	 * will cause other nodes to unmark us as needing
1181 	 * recovery. However, we need to send it *before* dropping the
1182 	 * super block lock as otherwise their recovery threads might
1183 	 * try to clean us up while we're live! */
1184 	status = ocfs2_request_mount_vote(osb);
1185 	if (status < 0)
1186 		mlog_errno(status);
1187 
1188 leave:
1189 	if (unlock_super)
1190 		ocfs2_super_unlock(osb, 1);
1191 
1192 	mlog_exit(status);
1193 	return status;
1194 }
1195 
1196 /* we can't grab the goofy sem lock from inside wait_event, so we use
1197  * memory barriers to make sure that we'll see the null task before
1198  * being woken up */
1199 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1200 {
1201 	mb();
1202 	return osb->recovery_thread_task != NULL;
1203 }
1204 
1205 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1206 {
1207 	int tmp;
1208 	struct ocfs2_super *osb = NULL;
1209 	char nodestr[8];
1210 
1211 	mlog_entry("(0x%p)\n", sb);
1212 
1213 	BUG_ON(!sb);
1214 	osb = OCFS2_SB(sb);
1215 	BUG_ON(!osb);
1216 
1217 	ocfs2_shutdown_local_alloc(osb);
1218 
1219 	ocfs2_truncate_log_shutdown(osb);
1220 
1221 	/* disable any new recovery threads and wait for any currently
1222 	 * running ones to exit. Do this before setting the vol_state. */
1223 	mutex_lock(&osb->recovery_lock);
1224 	osb->disable_recovery = 1;
1225 	mutex_unlock(&osb->recovery_lock);
1226 	wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1227 
1228 	/* At this point, we know that no more recovery threads can be
1229 	 * launched, so wait for any recovery completion work to
1230 	 * complete. */
1231 	flush_workqueue(ocfs2_wq);
1232 
1233 	ocfs2_journal_shutdown(osb);
1234 
1235 	ocfs2_sync_blockdev(sb);
1236 
1237 	/* No dlm means we've failed during mount, so skip all the
1238 	 * steps which depended on that to complete. */
1239 	if (osb->dlm) {
1240 		tmp = ocfs2_super_lock(osb, 1);
1241 		if (tmp < 0) {
1242 			mlog_errno(tmp);
1243 			return;
1244 		}
1245 
1246 		tmp = ocfs2_request_umount_vote(osb);
1247 		if (tmp < 0)
1248 			mlog_errno(tmp);
1249 	}
1250 
1251 	if (osb->slot_num != OCFS2_INVALID_SLOT)
1252 		ocfs2_put_slot(osb);
1253 
1254 	if (osb->dlm)
1255 		ocfs2_super_unlock(osb, 1);
1256 
1257 	ocfs2_release_system_inodes(osb);
1258 
1259 	if (osb->dlm) {
1260 		ocfs2_unregister_net_handlers(osb);
1261 
1262 		ocfs2_dlm_shutdown(osb);
1263 	}
1264 
1265 	ocfs2_clear_hb_callbacks(osb);
1266 
1267 	debugfs_remove(osb->osb_debug_root);
1268 
1269 	if (!mnt_err)
1270 		ocfs2_stop_heartbeat(osb);
1271 
1272 	atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1273 
1274 	if (ocfs2_mount_local(osb))
1275 		snprintf(nodestr, sizeof(nodestr), "local");
1276 	else
1277 		snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1278 
1279 	printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1280 	       osb->dev_str, nodestr);
1281 
1282 	ocfs2_delete_osb(osb);
1283 	kfree(osb);
1284 	sb->s_dev = 0;
1285 	sb->s_fs_info = NULL;
1286 }
1287 
1288 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1289 				unsigned uuid_bytes)
1290 {
1291 	int i, ret;
1292 	char *ptr;
1293 
1294 	BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1295 
1296 	osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1297 	if (osb->uuid_str == NULL)
1298 		return -ENOMEM;
1299 
1300 	for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1301 		/* print with null */
1302 		ret = snprintf(ptr, 3, "%02X", uuid[i]);
1303 		if (ret != 2) /* drop super cleans up */
1304 			return -EINVAL;
1305 		/* then only advance past the last char */
1306 		ptr += 2;
1307 	}
1308 
1309 	return 0;
1310 }
1311 
1312 static int ocfs2_initialize_super(struct super_block *sb,
1313 				  struct buffer_head *bh,
1314 				  int sector_size)
1315 {
1316 	int status;
1317 	int i, cbits, bbits;
1318 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1319 	struct inode *inode = NULL;
1320 	struct buffer_head *bitmap_bh = NULL;
1321 	struct ocfs2_journal *journal;
1322 	__le32 uuid_net_key;
1323 	struct ocfs2_super *osb;
1324 
1325 	mlog_entry_void();
1326 
1327 	osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1328 	if (!osb) {
1329 		status = -ENOMEM;
1330 		mlog_errno(status);
1331 		goto bail;
1332 	}
1333 
1334 	sb->s_fs_info = osb;
1335 	sb->s_op = &ocfs2_sops;
1336 	sb->s_export_op = &ocfs2_export_ops;
1337 	sb->s_time_gran = 1;
1338 	sb->s_flags |= MS_NOATIME;
1339 	/* this is needed to support O_LARGEFILE */
1340 	cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1341 	bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1342 	sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1343 
1344 	osb->sb = sb;
1345 	/* Save off for ocfs2_rw_direct */
1346 	osb->s_sectsize_bits = blksize_bits(sector_size);
1347 	BUG_ON(!osb->s_sectsize_bits);
1348 
1349 	osb->net_response_ids = 0;
1350 	spin_lock_init(&osb->net_response_lock);
1351 	INIT_LIST_HEAD(&osb->net_response_list);
1352 
1353 	INIT_LIST_HEAD(&osb->osb_net_handlers);
1354 	init_waitqueue_head(&osb->recovery_event);
1355 	spin_lock_init(&osb->vote_task_lock);
1356 	init_waitqueue_head(&osb->vote_event);
1357 	osb->vote_work_sequence = 0;
1358 	osb->vote_wake_sequence = 0;
1359 	INIT_LIST_HEAD(&osb->blocked_lock_list);
1360 	osb->blocked_lock_count = 0;
1361 	INIT_LIST_HEAD(&osb->vote_list);
1362 	spin_lock_init(&osb->osb_lock);
1363 
1364 	atomic_set(&osb->alloc_stats.moves, 0);
1365 	atomic_set(&osb->alloc_stats.local_data, 0);
1366 	atomic_set(&osb->alloc_stats.bitmap_data, 0);
1367 	atomic_set(&osb->alloc_stats.bg_allocs, 0);
1368 	atomic_set(&osb->alloc_stats.bg_extends, 0);
1369 
1370 	ocfs2_init_node_maps(osb);
1371 
1372 	snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1373 		 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1374 
1375 	mutex_init(&osb->recovery_lock);
1376 
1377 	osb->disable_recovery = 0;
1378 	osb->recovery_thread_task = NULL;
1379 
1380 	init_waitqueue_head(&osb->checkpoint_event);
1381 	atomic_set(&osb->needs_checkpoint, 0);
1382 
1383 	osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1384 
1385 	osb->node_num = O2NM_INVALID_NODE_NUM;
1386 	osb->slot_num = OCFS2_INVALID_SLOT;
1387 
1388 	osb->local_alloc_state = OCFS2_LA_UNUSED;
1389 	osb->local_alloc_bh = NULL;
1390 
1391 	ocfs2_setup_hb_callbacks(osb);
1392 
1393 	init_waitqueue_head(&osb->osb_mount_event);
1394 
1395 	osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1396 	if (!osb->vol_label) {
1397 		mlog(ML_ERROR, "unable to alloc vol label\n");
1398 		status = -ENOMEM;
1399 		goto bail;
1400 	}
1401 
1402 	osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1403 	if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1404 		mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1405 		     osb->max_slots);
1406 		status = -EINVAL;
1407 		goto bail;
1408 	}
1409 	mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1410 
1411 	init_waitqueue_head(&osb->osb_wipe_event);
1412 	osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1413 					sizeof(*osb->osb_orphan_wipes),
1414 					GFP_KERNEL);
1415 	if (!osb->osb_orphan_wipes) {
1416 		status = -ENOMEM;
1417 		mlog_errno(status);
1418 		goto bail;
1419 	}
1420 
1421 	osb->s_feature_compat =
1422 		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1423 	osb->s_feature_ro_compat =
1424 		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1425 	osb->s_feature_incompat =
1426 		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1427 
1428 	if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1429 		mlog(ML_ERROR, "couldn't mount because of unsupported "
1430 		     "optional features (%x).\n", i);
1431 		status = -EINVAL;
1432 		goto bail;
1433 	}
1434 	if (!(osb->sb->s_flags & MS_RDONLY) &&
1435 	    (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1436 		mlog(ML_ERROR, "couldn't mount RDWR because of "
1437 		     "unsupported optional features (%x).\n", i);
1438 		status = -EINVAL;
1439 		goto bail;
1440 	}
1441 
1442 	get_random_bytes(&osb->s_next_generation, sizeof(u32));
1443 
1444 	/* FIXME
1445 	 * This should be done in ocfs2_journal_init(), but unknown
1446 	 * ordering issues will cause the filesystem to crash.
1447 	 * If anyone wants to figure out what part of the code
1448 	 * refers to osb->journal before ocfs2_journal_init() is run,
1449 	 * be my guest.
1450 	 */
1451 	/* initialize our journal structure */
1452 
1453 	journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1454 	if (!journal) {
1455 		mlog(ML_ERROR, "unable to alloc journal\n");
1456 		status = -ENOMEM;
1457 		goto bail;
1458 	}
1459 	osb->journal = journal;
1460 	journal->j_osb = osb;
1461 
1462 	atomic_set(&journal->j_num_trans, 0);
1463 	init_rwsem(&journal->j_trans_barrier);
1464 	init_waitqueue_head(&journal->j_checkpointed);
1465 	spin_lock_init(&journal->j_lock);
1466 	journal->j_trans_id = (unsigned long) 1;
1467 	INIT_LIST_HEAD(&journal->j_la_cleanups);
1468 	INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1469 	journal->j_state = OCFS2_JOURNAL_FREE;
1470 
1471 	/* get some pseudo constants for clustersize bits */
1472 	osb->s_clustersize_bits =
1473 		le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1474 	osb->s_clustersize = 1 << osb->s_clustersize_bits;
1475 	mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1476 
1477 	if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1478 	    osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1479 		mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1480 		     osb->s_clustersize);
1481 		status = -EINVAL;
1482 		goto bail;
1483 	}
1484 
1485 	if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1486 	    > (u32)~0UL) {
1487 		mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1488 		     "what jbd can address in 32 bits.\n");
1489 		status = -EINVAL;
1490 		goto bail;
1491 	}
1492 
1493 	if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1494 				 sizeof(di->id2.i_super.s_uuid))) {
1495 		mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1496 		status = -ENOMEM;
1497 		goto bail;
1498 	}
1499 
1500 	memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1501 	osb->net_key = le32_to_cpu(uuid_net_key);
1502 
1503 	strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1504 	osb->vol_label[63] = '\0';
1505 	osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1506 	osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1507 	osb->first_cluster_group_blkno =
1508 		le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1509 	osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1510 	mlog(0, "vol_label: %s\n", osb->vol_label);
1511 	mlog(0, "uuid: %s\n", osb->uuid_str);
1512 	mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1513 	     (unsigned long long)osb->root_blkno,
1514 	     (unsigned long long)osb->system_dir_blkno);
1515 
1516 	osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1517 	if (!osb->osb_dlm_debug) {
1518 		status = -ENOMEM;
1519 		mlog_errno(status);
1520 		goto bail;
1521 	}
1522 
1523 	atomic_set(&osb->vol_state, VOLUME_INIT);
1524 
1525 	/* load root, system_dir, and all global system inodes */
1526 	status = ocfs2_init_global_system_inodes(osb);
1527 	if (status < 0) {
1528 		mlog_errno(status);
1529 		goto bail;
1530 	}
1531 
1532 	/*
1533 	 * global bitmap
1534 	 */
1535 	inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1536 					    OCFS2_INVALID_SLOT);
1537 	if (!inode) {
1538 		status = -EINVAL;
1539 		mlog_errno(status);
1540 		goto bail;
1541 	}
1542 
1543 	osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1544 
1545 	/* We don't have a cluster lock on the bitmap here because
1546 	 * we're only interested in static information and the extra
1547 	 * complexity at mount time isn't worht it. Don't pass the
1548 	 * inode in to the read function though as we don't want it to
1549 	 * be put in the cache. */
1550 	status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1551 				  NULL);
1552 	iput(inode);
1553 	if (status < 0) {
1554 		mlog_errno(status);
1555 		goto bail;
1556 	}
1557 
1558 	di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1559 	osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1560 	brelse(bitmap_bh);
1561 	mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1562 	     (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1563 
1564 	status = ocfs2_init_slot_info(osb);
1565 	if (status < 0) {
1566 		mlog_errno(status);
1567 		goto bail;
1568 	}
1569 
1570 bail:
1571 	mlog_exit(status);
1572 	return status;
1573 }
1574 
1575 /*
1576  * will return: -EAGAIN if it is ok to keep searching for superblocks
1577  *              -EINVAL if there is a bad superblock
1578  *              0 on success
1579  */
1580 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1581 			       struct buffer_head *bh,
1582 			       u32 blksz)
1583 {
1584 	int status = -EAGAIN;
1585 
1586 	mlog_entry_void();
1587 
1588 	if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1589 		   strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1590 		status = -EINVAL;
1591 		if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1592 			mlog(ML_ERROR, "found superblock with incorrect block "
1593 			     "size: found %u, should be %u\n",
1594 			     1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1595 			       blksz);
1596 		} else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1597 			   OCFS2_MAJOR_REV_LEVEL ||
1598 			   le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1599 			   OCFS2_MINOR_REV_LEVEL) {
1600 			mlog(ML_ERROR, "found superblock with bad version: "
1601 			     "found %u.%u, should be %u.%u\n",
1602 			     le16_to_cpu(di->id2.i_super.s_major_rev_level),
1603 			     le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1604 			     OCFS2_MAJOR_REV_LEVEL,
1605 			     OCFS2_MINOR_REV_LEVEL);
1606 		} else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1607 			mlog(ML_ERROR, "bad block number on superblock: "
1608 			     "found %llu, should be %llu\n",
1609 			     (unsigned long long)le64_to_cpu(di->i_blkno),
1610 			     (unsigned long long)bh->b_blocknr);
1611 		} else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1612 			    le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1613 			mlog(ML_ERROR, "bad cluster size found: %u\n",
1614 			     1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1615 		} else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1616 			mlog(ML_ERROR, "bad root_blkno: 0\n");
1617 		} else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1618 			mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1619 		} else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1620 			mlog(ML_ERROR,
1621 			     "Superblock slots found greater than file system "
1622 			     "maximum: found %u, max %u\n",
1623 			     le16_to_cpu(di->id2.i_super.s_max_slots),
1624 			     OCFS2_MAX_SLOTS);
1625 		} else {
1626 			/* found it! */
1627 			status = 0;
1628 		}
1629 	}
1630 
1631 	mlog_exit(status);
1632 	return status;
1633 }
1634 
1635 static int ocfs2_check_volume(struct ocfs2_super *osb)
1636 {
1637 	int status;
1638 	int dirty;
1639 	int local;
1640 	struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1641 						  * recover
1642 						  * ourselves. */
1643 
1644 	mlog_entry_void();
1645 
1646 	/* Init our journal object. */
1647 	status = ocfs2_journal_init(osb->journal, &dirty);
1648 	if (status < 0) {
1649 		mlog(ML_ERROR, "Could not initialize journal!\n");
1650 		goto finally;
1651 	}
1652 
1653 	/* If the journal was unmounted cleanly then we don't want to
1654 	 * recover anything. Otherwise, journal_load will do that
1655 	 * dirty work for us :) */
1656 	if (!dirty) {
1657 		status = ocfs2_journal_wipe(osb->journal, 0);
1658 		if (status < 0) {
1659 			mlog_errno(status);
1660 			goto finally;
1661 		}
1662 	} else {
1663 		mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1664 		     "recovering volume.\n");
1665 	}
1666 
1667 	local = ocfs2_mount_local(osb);
1668 
1669 	/* will play back anything left in the journal. */
1670 	ocfs2_journal_load(osb->journal, local);
1671 
1672 	if (dirty) {
1673 		/* recover my local alloc if we didn't unmount cleanly. */
1674 		status = ocfs2_begin_local_alloc_recovery(osb,
1675 							  osb->slot_num,
1676 							  &local_alloc);
1677 		if (status < 0) {
1678 			mlog_errno(status);
1679 			goto finally;
1680 		}
1681 		/* we complete the recovery process after we've marked
1682 		 * ourselves as mounted. */
1683 	}
1684 
1685 	mlog(0, "Journal loaded.\n");
1686 
1687 	status = ocfs2_load_local_alloc(osb);
1688 	if (status < 0) {
1689 		mlog_errno(status);
1690 		goto finally;
1691 	}
1692 
1693 	if (dirty) {
1694 		/* Recovery will be completed after we've mounted the
1695 		 * rest of the volume. */
1696 		osb->dirty = 1;
1697 		osb->local_alloc_copy = local_alloc;
1698 		local_alloc = NULL;
1699 	}
1700 
1701 	/* go through each journal, trylock it and if you get the
1702 	 * lock, and it's marked as dirty, set the bit in the recover
1703 	 * map and launch a recovery thread for it. */
1704 	status = ocfs2_mark_dead_nodes(osb);
1705 	if (status < 0)
1706 		mlog_errno(status);
1707 
1708 finally:
1709 	if (local_alloc)
1710 		kfree(local_alloc);
1711 
1712 	mlog_exit(status);
1713 	return status;
1714 }
1715 
1716 /*
1717  * The routine gets called from dismount or close whenever a dismount on
1718  * volume is requested and the osb open count becomes 1.
1719  * It will remove the osb from the global list and also free up all the
1720  * initialized resources and fileobject.
1721  */
1722 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1723 {
1724 	mlog_entry_void();
1725 
1726 	/* This function assumes that the caller has the main osb resource */
1727 
1728 	if (osb->slot_info)
1729 		ocfs2_free_slot_info(osb->slot_info);
1730 
1731 	kfree(osb->osb_orphan_wipes);
1732 	/* FIXME
1733 	 * This belongs in journal shutdown, but because we have to
1734 	 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1735 	 * we free it here.
1736 	 */
1737 	kfree(osb->journal);
1738 	if (osb->local_alloc_copy)
1739 		kfree(osb->local_alloc_copy);
1740 	kfree(osb->uuid_str);
1741 	ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1742 	memset(osb, 0, sizeof(struct ocfs2_super));
1743 
1744 	mlog_exit_void();
1745 }
1746 
1747 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1748  * panic(). We do not support continue-on-error operation. */
1749 static void ocfs2_handle_error(struct super_block *sb)
1750 {
1751 	struct ocfs2_super *osb = OCFS2_SB(sb);
1752 
1753 	if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1754 		panic("OCFS2: (device %s): panic forced after error\n",
1755 		      sb->s_id);
1756 
1757 	ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1758 
1759 	if (sb->s_flags & MS_RDONLY &&
1760 	    (ocfs2_is_soft_readonly(osb) ||
1761 	     ocfs2_is_hard_readonly(osb)))
1762 		return;
1763 
1764 	printk(KERN_CRIT "File system is now read-only due to the potential "
1765 	       "of on-disk corruption. Please run fsck.ocfs2 once the file "
1766 	       "system is unmounted.\n");
1767 	sb->s_flags |= MS_RDONLY;
1768 	ocfs2_set_ro_flag(osb, 0);
1769 }
1770 
1771 static char error_buf[1024];
1772 
1773 void __ocfs2_error(struct super_block *sb,
1774 		   const char *function,
1775 		   const char *fmt, ...)
1776 {
1777 	va_list args;
1778 
1779 	va_start(args, fmt);
1780 	vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1781 	va_end(args);
1782 
1783 	/* Not using mlog here because we want to show the actual
1784 	 * function the error came from. */
1785 	printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1786 	       sb->s_id, function, error_buf);
1787 
1788 	ocfs2_handle_error(sb);
1789 }
1790 
1791 /* Handle critical errors. This is intentionally more drastic than
1792  * ocfs2_handle_error, so we only use for things like journal errors,
1793  * etc. */
1794 void __ocfs2_abort(struct super_block* sb,
1795 		   const char *function,
1796 		   const char *fmt, ...)
1797 {
1798 	va_list args;
1799 
1800 	va_start(args, fmt);
1801 	vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1802 	va_end(args);
1803 
1804 	printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1805 	       sb->s_id, function, error_buf);
1806 
1807 	/* We don't have the cluster support yet to go straight to
1808 	 * hard readonly in here. Until then, we want to keep
1809 	 * ocfs2_abort() so that we can at least mark critical
1810 	 * errors.
1811 	 *
1812 	 * TODO: This should abort the journal and alert other nodes
1813 	 * that our slot needs recovery. */
1814 
1815 	/* Force a panic(). This stinks, but it's better than letting
1816 	 * things continue without having a proper hard readonly
1817 	 * here. */
1818 	OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1819 	ocfs2_handle_error(sb);
1820 }
1821 
1822 module_init(ocfs2_init);
1823 module_exit(ocfs2_exit);
1824