blob: d17e33e66a1e5228cead65463a3255adb383fa71 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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
Mark Fashehccd979b2005-12-15 14:31:24 -080071static kmem_cache_t *ocfs2_inode_cachep = NULL;
72
73kmem_cache_t *ocfs2_lock_cache = NULL;
74
75/* OCFS2 needs to schedule several differnt types of work which
76 * require cluster locking, disk I/O, recovery waits, etc. Since these
77 * types of work tend to be heavy we avoid using the kernel events
78 * workqueue and schedule on our own. */
79struct workqueue_struct *ocfs2_wq = NULL;
80
81static struct dentry *ocfs2_debugfs_root = NULL;
82
83MODULE_AUTHOR("Oracle");
84MODULE_LICENSE("GPL");
85
86static int ocfs2_parse_options(struct super_block *sb, char *options,
87 unsigned long *mount_opt, int is_remount);
88static void ocfs2_put_super(struct super_block *sb);
89static int ocfs2_mount_volume(struct super_block *sb);
90static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
91static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
92static int ocfs2_initialize_mem_caches(void);
93static void ocfs2_free_mem_caches(void);
94static void ocfs2_delete_osb(struct ocfs2_super *osb);
95
David Howells726c3342006-06-23 02:02:58 -070096static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
Mark Fashehccd979b2005-12-15 14:31:24 -080097
98static int ocfs2_sync_fs(struct super_block *sb, int wait);
99
100static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
101static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
102static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
103static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
104static int ocfs2_check_volume(struct ocfs2_super *osb);
105static int ocfs2_verify_volume(struct ocfs2_dinode *di,
106 struct buffer_head *bh,
107 u32 sectsize);
108static int ocfs2_initialize_super(struct super_block *sb,
109 struct buffer_head *bh,
110 int sector_size);
111static int ocfs2_get_sector(struct super_block *sb,
112 struct buffer_head **bh,
113 int block,
114 int sect_size);
115static void ocfs2_write_super(struct super_block *sb);
116static struct inode *ocfs2_alloc_inode(struct super_block *sb);
117static void ocfs2_destroy_inode(struct inode *inode);
118
119static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
120
121static struct super_operations ocfs2_sops = {
122 .statfs = ocfs2_statfs,
123 .alloc_inode = ocfs2_alloc_inode,
124 .destroy_inode = ocfs2_destroy_inode,
125 .drop_inode = ocfs2_drop_inode,
126 .clear_inode = ocfs2_clear_inode,
127 .delete_inode = ocfs2_delete_inode,
128 .sync_fs = ocfs2_sync_fs,
129 .write_super = ocfs2_write_super,
130 .put_super = ocfs2_put_super,
131 .remount_fs = ocfs2_remount,
132};
133
134enum {
135 Opt_barrier,
136 Opt_err_panic,
137 Opt_err_ro,
138 Opt_intr,
139 Opt_nointr,
140 Opt_hb_none,
141 Opt_hb_local,
142 Opt_data_ordered,
143 Opt_data_writeback,
144 Opt_err,
145};
146
147static match_table_t tokens = {
148 {Opt_barrier, "barrier=%u"},
149 {Opt_err_panic, "errors=panic"},
150 {Opt_err_ro, "errors=remount-ro"},
151 {Opt_intr, "intr"},
152 {Opt_nointr, "nointr"},
153 {Opt_hb_none, OCFS2_HB_NONE},
154 {Opt_hb_local, OCFS2_HB_LOCAL},
155 {Opt_data_ordered, "data=ordered"},
156 {Opt_data_writeback, "data=writeback"},
157 {Opt_err, NULL}
158};
159
160/*
161 * write_super and sync_fs ripped right out of ext3.
162 */
163static void ocfs2_write_super(struct super_block *sb)
164{
Ingo Molnar7892f2f2006-01-09 15:59:25 -0800165 if (mutex_trylock(&sb->s_lock) != 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800166 BUG();
167 sb->s_dirt = 0;
168}
169
170static int ocfs2_sync_fs(struct super_block *sb, int wait)
171{
172 int status = 0;
173 tid_t target;
174 struct ocfs2_super *osb = OCFS2_SB(sb);
175
176 sb->s_dirt = 0;
177
178 if (ocfs2_is_hard_readonly(osb))
179 return -EROFS;
180
181 if (wait) {
182 status = ocfs2_flush_truncate_log(osb);
183 if (status < 0)
184 mlog_errno(status);
185 } else {
186 ocfs2_schedule_truncate_log_flush(osb, 0);
187 }
188
189 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
190 if (wait)
191 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
192 target);
193 }
194 return 0;
195}
196
197static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
198{
199 struct inode *new = NULL;
200 int status = 0;
201 int i;
202
203 mlog_entry_void();
204
205 new = ocfs2_iget(osb, osb->root_blkno);
206 if (IS_ERR(new)) {
207 status = PTR_ERR(new);
208 mlog_errno(status);
209 goto bail;
210 }
211 osb->root_inode = new;
212
213 new = ocfs2_iget(osb, osb->system_dir_blkno);
214 if (IS_ERR(new)) {
215 status = PTR_ERR(new);
216 mlog_errno(status);
217 goto bail;
218 }
219 osb->sys_root_inode = new;
220
221 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
222 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
223 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
224 if (!new) {
225 ocfs2_release_system_inodes(osb);
226 status = -EINVAL;
227 mlog_errno(status);
228 /* FIXME: Should ERROR_RO_FS */
229 mlog(ML_ERROR, "Unable to load system inode %d, "
230 "possibly corrupt fs?", i);
231 goto bail;
232 }
233 // the array now has one ref, so drop this one
234 iput(new);
235 }
236
237bail:
238 mlog_exit(status);
239 return status;
240}
241
242static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
243{
244 struct inode *new = NULL;
245 int status = 0;
246 int i;
247
248 mlog_entry_void();
249
250 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
251 i < NUM_SYSTEM_INODES;
252 i++) {
253 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
254 if (!new) {
255 ocfs2_release_system_inodes(osb);
256 status = -EINVAL;
257 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
258 status, i, osb->slot_num);
259 goto bail;
260 }
261 /* the array now has one ref, so drop this one */
262 iput(new);
263 }
264
265bail:
266 mlog_exit(status);
267 return status;
268}
269
270static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
271{
272 int status = 0, i;
273 struct inode *inode;
274
275 mlog_entry_void();
276
277 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
278 inode = osb->system_inodes[i];
279 if (inode) {
280 iput(inode);
281 osb->system_inodes[i] = NULL;
282 }
283 }
284
285 inode = osb->sys_root_inode;
286 if (inode) {
287 iput(inode);
288 osb->sys_root_inode = NULL;
289 }
290
291 inode = osb->root_inode;
292 if (inode) {
293 iput(inode);
294 osb->root_inode = NULL;
295 }
296
297 mlog_exit(status);
298 return status;
299}
300
301/* We're allocating fs objects, use GFP_NOFS */
302static struct inode *ocfs2_alloc_inode(struct super_block *sb)
303{
304 struct ocfs2_inode_info *oi;
305
306 oi = kmem_cache_alloc(ocfs2_inode_cachep, SLAB_NOFS);
307 if (!oi)
308 return NULL;
309
310 return &oi->vfs_inode;
311}
312
313static void ocfs2_destroy_inode(struct inode *inode)
314{
315 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
316}
317
318/* From xfs_super.c:xfs_max_file_offset
319 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
320 */
321static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
322{
323 unsigned int pagefactor = 1;
324 unsigned int bitshift = BITS_PER_LONG - 1;
325
326 /* Figure out maximum filesize, on Linux this can depend on
327 * the filesystem blocksize (on 32 bit platforms).
328 * __block_prepare_write does this in an [unsigned] long...
329 * page->index << (PAGE_CACHE_SHIFT - bbits)
330 * So, for page sized blocks (4K on 32 bit platforms),
331 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
332 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
333 * but for smaller blocksizes it is less (bbits = log2 bsize).
334 * Note1: get_block_t takes a long (implicit cast from above)
335 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
336 * can optionally convert the [unsigned] long from above into
337 * an [unsigned] long long.
338 */
339
340#if BITS_PER_LONG == 32
341# if defined(CONFIG_LBD)
342 BUG_ON(sizeof(sector_t) != 8);
343 pagefactor = PAGE_CACHE_SIZE;
344 bitshift = BITS_PER_LONG;
345# else
346 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
347# endif
348#endif
349
350 return (((unsigned long long)pagefactor) << bitshift) - 1;
351}
352
353static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
354{
355 int incompat_features;
356 int ret = 0;
357 unsigned long parsed_options;
358 struct ocfs2_super *osb = OCFS2_SB(sb);
359
360 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
361 ret = -EINVAL;
362 goto out;
363 }
364
365 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
366 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
367 ret = -EINVAL;
368 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
369 goto out;
370 }
371
372 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
373 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
374 ret = -EINVAL;
375 mlog(ML_ERROR, "Cannot change data mode on remount\n");
376 goto out;
377 }
378
379 /* We're going to/from readonly mode. */
380 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
381 /* Lock here so the check of HARD_RO and the potential
382 * setting of SOFT_RO is atomic. */
383 spin_lock(&osb->osb_lock);
384 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
385 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
386 ret = -EROFS;
387 goto unlock_osb;
388 }
389
390 if (*flags & MS_RDONLY) {
391 mlog(0, "Going to ro mode.\n");
392 sb->s_flags |= MS_RDONLY;
393 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
394 } else {
395 mlog(0, "Making ro filesystem writeable.\n");
396
397 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
398 mlog(ML_ERROR, "Cannot remount RDWR "
399 "filesystem due to previous errors.\n");
400 ret = -EROFS;
401 goto unlock_osb;
402 }
403 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
404 if (incompat_features) {
405 mlog(ML_ERROR, "Cannot remount RDWR because "
406 "of unsupported optional features "
407 "(%x).\n", incompat_features);
408 ret = -EINVAL;
409 goto unlock_osb;
410 }
411 sb->s_flags &= ~MS_RDONLY;
412 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
413 }
414unlock_osb:
415 spin_unlock(&osb->osb_lock);
416 }
417
418 if (!ret) {
419 if (!ocfs2_is_hard_readonly(osb))
420 ocfs2_set_journal_params(osb);
421
422 /* Only save off the new mount options in case of a successful
423 * remount. */
424 osb->s_mount_opt = parsed_options;
425 }
426out:
427 return ret;
428}
429
430static int ocfs2_sb_probe(struct super_block *sb,
431 struct buffer_head **bh,
432 int *sector_size)
433{
434 int status = 0, tmpstat;
435 struct ocfs1_vol_disk_hdr *hdr;
436 struct ocfs2_dinode *di;
437 int blksize;
438
439 *bh = NULL;
440
441 /* may be > 512 */
442 *sector_size = bdev_hardsect_size(sb->s_bdev);
443 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
444 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
445 *sector_size, OCFS2_MAX_BLOCKSIZE);
446 status = -EINVAL;
447 goto bail;
448 }
449
450 /* Can this really happen? */
451 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
452 *sector_size = OCFS2_MIN_BLOCKSIZE;
453
454 /* check block zero for old format */
455 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
456 if (status < 0) {
457 mlog_errno(status);
458 goto bail;
459 }
460 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
461 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
462 mlog(ML_ERROR, "incompatible version: %u.%u\n",
463 hdr->major_version, hdr->minor_version);
464 status = -EINVAL;
465 }
466 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
467 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
468 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
469 hdr->signature);
470 status = -EINVAL;
471 }
472 brelse(*bh);
473 *bh = NULL;
474 if (status < 0) {
475 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
476 "upgraded before mounting with ocfs v2\n");
477 goto bail;
478 }
479
480 /*
481 * Now check at magic offset for 512, 1024, 2048, 4096
482 * blocksizes. 4096 is the maximum blocksize because it is
483 * the minimum clustersize.
484 */
485 status = -EINVAL;
486 for (blksize = *sector_size;
487 blksize <= OCFS2_MAX_BLOCKSIZE;
488 blksize <<= 1) {
489 tmpstat = ocfs2_get_sector(sb, bh,
490 OCFS2_SUPER_BLOCK_BLKNO,
491 blksize);
492 if (tmpstat < 0) {
493 status = tmpstat;
494 mlog_errno(status);
495 goto bail;
496 }
497 di = (struct ocfs2_dinode *) (*bh)->b_data;
498 status = ocfs2_verify_volume(di, *bh, blksize);
499 if (status >= 0)
500 goto bail;
501 brelse(*bh);
502 *bh = NULL;
503 if (status != -EAGAIN)
504 break;
505 }
506
507bail:
508 return status;
509}
510
511static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
512{
513 struct dentry *root;
514 int status, sector_size;
515 unsigned long parsed_opt;
516 struct inode *inode = NULL;
517 struct ocfs2_super *osb = NULL;
518 struct buffer_head *bh = NULL;
519
520 mlog_entry("%p, %p, %i", sb, data, silent);
521
522 /* for now we only have one cluster/node, make sure we see it
523 * in the heartbeat universe */
524 if (!o2hb_check_local_node_heartbeating()) {
525 status = -EINVAL;
526 goto read_super_error;
527 }
528
529 /* probe for superblock */
530 status = ocfs2_sb_probe(sb, &bh, &sector_size);
531 if (status < 0) {
532 mlog(ML_ERROR, "superblock probe failed!\n");
533 goto read_super_error;
534 }
535
536 status = ocfs2_initialize_super(sb, bh, sector_size);
537 osb = OCFS2_SB(sb);
538 if (status < 0) {
539 mlog_errno(status);
540 goto read_super_error;
541 }
542 brelse(bh);
543 bh = NULL;
544
545 if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
546 status = -EINVAL;
547 goto read_super_error;
548 }
549 osb->s_mount_opt = parsed_opt;
550
551 sb->s_magic = OCFS2_SUPER_MAGIC;
552
553 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
554 * heartbeat=none */
555 if (bdev_read_only(sb->s_bdev)) {
556 if (!(sb->s_flags & MS_RDONLY)) {
557 status = -EACCES;
558 mlog(ML_ERROR, "Readonly device detected but readonly "
559 "mount was not specified.\n");
560 goto read_super_error;
561 }
562
563 /* You should not be able to start a local heartbeat
564 * on a readonly device. */
565 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
566 status = -EROFS;
567 mlog(ML_ERROR, "Local heartbeat specified on readonly "
568 "device.\n");
569 goto read_super_error;
570 }
571
572 status = ocfs2_check_journals_nolocks(osb);
573 if (status < 0) {
574 if (status == -EROFS)
575 mlog(ML_ERROR, "Recovery required on readonly "
576 "file system, but write access is "
577 "unavailable.\n");
578 else
579 mlog_errno(status);
580 goto read_super_error;
581 }
582
583 ocfs2_set_ro_flag(osb, 1);
584
585 printk(KERN_NOTICE "Readonly device detected. No cluster "
586 "services will be utilized for this mount. Recovery "
587 "will be skipped.\n");
588 }
589
590 if (!ocfs2_is_hard_readonly(osb)) {
591 /* If this isn't a hard readonly mount, then we need
592 * to make sure that heartbeat is in a valid state,
593 * and that we mark ourselves soft readonly is -oro
594 * was specified. */
595 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
596 mlog(ML_ERROR, "No heartbeat for device (%s)\n",
597 sb->s_id);
598 status = -EINVAL;
599 goto read_super_error;
600 }
601
602 if (sb->s_flags & MS_RDONLY)
603 ocfs2_set_ro_flag(osb, 0);
604 }
605
606 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
607 ocfs2_debugfs_root);
608 if (!osb->osb_debug_root) {
609 status = -EINVAL;
610 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
611 goto read_super_error;
612 }
613
614 status = ocfs2_mount_volume(sb);
615 if (osb->root_inode)
616 inode = igrab(osb->root_inode);
617
618 if (status < 0)
619 goto read_super_error;
620
621 if (!inode) {
622 status = -EIO;
623 mlog_errno(status);
624 goto read_super_error;
625 }
626
627 root = d_alloc_root(inode);
628 if (!root) {
629 status = -ENOMEM;
630 mlog_errno(status);
631 goto read_super_error;
632 }
633
634 sb->s_root = root;
635
636 ocfs2_complete_mount_recovery(osb);
637
Sunil Mushran781ee3e2006-04-27 16:41:31 -0700638 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %d, slot %d) "
639 "with %s data mode.\n",
640 osb->dev_str, osb->node_num, osb->slot_num,
Mark Fashehccd979b2005-12-15 14:31:24 -0800641 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
642 "ordered");
643
644 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
645 wake_up(&osb->osb_mount_event);
646
647 mlog_exit(status);
648 return status;
649
650read_super_error:
651 if (bh != NULL)
652 brelse(bh);
653
654 if (inode)
655 iput(inode);
656
657 if (osb) {
658 atomic_set(&osb->vol_state, VOLUME_DISABLED);
659 wake_up(&osb->osb_mount_event);
660 ocfs2_dismount_volume(sb, 1);
661 }
662
663 mlog_exit(status);
664 return status;
665}
666
David Howells454e2392006-06-23 02:02:57 -0700667static int ocfs2_get_sb(struct file_system_type *fs_type,
668 int flags,
669 const char *dev_name,
670 void *data,
671 struct vfsmount *mnt)
Mark Fashehccd979b2005-12-15 14:31:24 -0800672{
David Howells454e2392006-06-23 02:02:57 -0700673 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
674 mnt);
Mark Fashehccd979b2005-12-15 14:31:24 -0800675}
676
677static struct file_system_type ocfs2_fs_type = {
678 .owner = THIS_MODULE,
679 .name = "ocfs2",
680 .get_sb = ocfs2_get_sb, /* is this called when we mount
681 * the fs? */
682 .kill_sb = kill_block_super, /* set to the generic one
683 * right now, but do we
684 * need to change that? */
685 .fs_flags = FS_REQUIRES_DEV,
686 .next = NULL
687};
688
689static int ocfs2_parse_options(struct super_block *sb,
690 char *options,
691 unsigned long *mount_opt,
692 int is_remount)
693{
694 int status;
695 char *p;
696
697 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
698 options ? options : "(none)");
699
700 *mount_opt = 0;
701
702 if (!options) {
703 status = 1;
704 goto bail;
705 }
706
707 while ((p = strsep(&options, ",")) != NULL) {
708 int token, option;
709 substring_t args[MAX_OPT_ARGS];
710
711 if (!*p)
712 continue;
713
714 token = match_token(p, tokens, args);
715 switch (token) {
716 case Opt_hb_local:
717 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
718 break;
719 case Opt_hb_none:
720 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
721 break;
722 case Opt_barrier:
723 if (match_int(&args[0], &option)) {
724 status = 0;
725 goto bail;
726 }
727 if (option)
728 *mount_opt |= OCFS2_MOUNT_BARRIER;
729 else
730 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
731 break;
732 case Opt_intr:
733 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
734 break;
735 case Opt_nointr:
736 *mount_opt |= OCFS2_MOUNT_NOINTR;
737 break;
738 case Opt_err_panic:
739 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
740 break;
741 case Opt_err_ro:
742 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
743 break;
744 case Opt_data_ordered:
745 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
746 break;
747 case Opt_data_writeback:
748 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
749 break;
750 default:
751 mlog(ML_ERROR,
752 "Unrecognized mount option \"%s\" "
753 "or missing value\n", p);
754 status = 0;
755 goto bail;
756 }
757 }
758
759 status = 1;
760
761bail:
762 mlog_exit(status);
763 return status;
764}
765
766static int __init ocfs2_init(void)
767{
768 int status;
769
770 mlog_entry_void();
771
772 ocfs2_print_version();
773
774 if (init_ocfs2_extent_maps())
775 return -ENOMEM;
776
777 status = init_ocfs2_uptodate_cache();
778 if (status < 0) {
779 mlog_errno(status);
780 goto leave;
781 }
782
783 status = ocfs2_initialize_mem_caches();
784 if (status < 0) {
785 mlog_errno(status);
786 goto leave;
787 }
788
789 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
790 if (!ocfs2_wq) {
791 status = -ENOMEM;
792 goto leave;
793 }
794
Mark Fashehccd979b2005-12-15 14:31:24 -0800795 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
796 if (!ocfs2_debugfs_root) {
797 status = -EFAULT;
798 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
799 }
800
801leave:
802 if (status < 0) {
803 ocfs2_free_mem_caches();
804 exit_ocfs2_uptodate_cache();
805 exit_ocfs2_extent_maps();
806 }
807
808 mlog_exit(status);
809
810 if (status >= 0) {
811 return register_filesystem(&ocfs2_fs_type);
812 } else
813 return -1;
814}
815
816static void __exit ocfs2_exit(void)
817{
818 mlog_entry_void();
819
820 if (ocfs2_wq) {
821 flush_workqueue(ocfs2_wq);
822 destroy_workqueue(ocfs2_wq);
823 }
824
825 debugfs_remove(ocfs2_debugfs_root);
826
827 ocfs2_free_mem_caches();
828
829 unregister_filesystem(&ocfs2_fs_type);
830
831 exit_ocfs2_extent_maps();
832
833 exit_ocfs2_uptodate_cache();
834
835 mlog_exit_void();
836}
837
838static void ocfs2_put_super(struct super_block *sb)
839{
840 mlog_entry("(0x%p)\n", sb);
841
842 ocfs2_sync_blockdev(sb);
843 ocfs2_dismount_volume(sb, 0);
844
845 mlog_exit_void();
846}
847
David Howells726c3342006-06-23 02:02:58 -0700848static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
Mark Fashehccd979b2005-12-15 14:31:24 -0800849{
850 struct ocfs2_super *osb;
851 u32 numbits, freebits;
852 int status;
853 struct ocfs2_dinode *bm_lock;
854 struct buffer_head *bh = NULL;
855 struct inode *inode = NULL;
856
David Howells726c3342006-06-23 02:02:58 -0700857 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
Mark Fashehccd979b2005-12-15 14:31:24 -0800858
David Howells726c3342006-06-23 02:02:58 -0700859 osb = OCFS2_SB(dentry->d_sb);
Mark Fashehccd979b2005-12-15 14:31:24 -0800860
861 inode = ocfs2_get_system_file_inode(osb,
862 GLOBAL_BITMAP_SYSTEM_INODE,
863 OCFS2_INVALID_SLOT);
864 if (!inode) {
865 mlog(ML_ERROR, "failed to get bitmap inode\n");
866 status = -EIO;
867 goto bail;
868 }
869
870 status = ocfs2_meta_lock(inode, NULL, &bh, 0);
871 if (status < 0) {
872 mlog_errno(status);
873 goto bail;
874 }
875
876 bm_lock = (struct ocfs2_dinode *) bh->b_data;
877
878 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
879 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
880
881 buf->f_type = OCFS2_SUPER_MAGIC;
David Howells726c3342006-06-23 02:02:58 -0700882 buf->f_bsize = dentry->d_sb->s_blocksize;
Mark Fashehccd979b2005-12-15 14:31:24 -0800883 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
884 buf->f_blocks = ((sector_t) numbits) *
885 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
886 buf->f_bfree = ((sector_t) freebits) *
887 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
888 buf->f_bavail = buf->f_bfree;
889 buf->f_files = numbits;
890 buf->f_ffree = freebits;
891
892 brelse(bh);
893
894 ocfs2_meta_unlock(inode, 0);
895 status = 0;
896bail:
897 if (inode)
898 iput(inode);
899
900 mlog_exit(status);
901
902 return status;
903}
904
905static void ocfs2_inode_init_once(void *data,
906 kmem_cache_t *cachep,
907 unsigned long flags)
908{
909 struct ocfs2_inode_info *oi = data;
910
911 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
912 SLAB_CTOR_CONSTRUCTOR) {
913 oi->ip_flags = 0;
914 oi->ip_open_count = 0;
915 spin_lock_init(&oi->ip_lock);
916 ocfs2_extent_map_init(&oi->vfs_inode);
917 INIT_LIST_HEAD(&oi->ip_handle_list);
918 INIT_LIST_HEAD(&oi->ip_io_markers);
919 oi->ip_handle = NULL;
920 oi->ip_created_trans = 0;
921 oi->ip_last_trans = 0;
922 oi->ip_dir_start_lookup = 0;
923
924 init_rwsem(&oi->ip_alloc_sem);
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800925 mutex_init(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800926
927 oi->ip_blkno = 0ULL;
928 oi->ip_clusters = 0;
929
930 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
931 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
932 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
933
934 ocfs2_metadata_cache_init(&oi->vfs_inode);
935
936 inode_init_once(&oi->vfs_inode);
937 }
938}
939
940static int ocfs2_initialize_mem_caches(void)
941{
942 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800943 sizeof(struct ocfs2_inode_info),
944 0,
945 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
946 SLAB_MEM_SPREAD),
947 ocfs2_inode_init_once, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800948 if (!ocfs2_inode_cachep)
949 return -ENOMEM;
950
951 ocfs2_lock_cache = kmem_cache_create("ocfs2_lock",
952 sizeof(struct ocfs2_journal_lock),
953 0,
Christoph Lameterac2b8982006-03-22 00:08:15 -0800954 SLAB_HWCACHE_ALIGN,
Mark Fashehccd979b2005-12-15 14:31:24 -0800955 NULL, NULL);
956 if (!ocfs2_lock_cache)
957 return -ENOMEM;
958
959 return 0;
960}
961
962static void ocfs2_free_mem_caches(void)
963{
964 if (ocfs2_inode_cachep)
965 kmem_cache_destroy(ocfs2_inode_cachep);
966 if (ocfs2_lock_cache)
967 kmem_cache_destroy(ocfs2_lock_cache);
968
969 ocfs2_inode_cachep = NULL;
970 ocfs2_lock_cache = NULL;
971}
972
973static int ocfs2_get_sector(struct super_block *sb,
974 struct buffer_head **bh,
975 int block,
976 int sect_size)
977{
978 if (!sb_set_blocksize(sb, sect_size)) {
979 mlog(ML_ERROR, "unable to set blocksize\n");
980 return -EIO;
981 }
982
983 *bh = sb_getblk(sb, block);
984 if (!*bh) {
985 mlog_errno(-EIO);
986 return -EIO;
987 }
988 lock_buffer(*bh);
989 if (!buffer_dirty(*bh))
990 clear_buffer_uptodate(*bh);
991 unlock_buffer(*bh);
992 ll_rw_block(READ, 1, bh);
993 wait_on_buffer(*bh);
994 return 0;
995}
996
997/* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
998static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
999{
1000 int status;
1001
1002 /* XXX hold a ref on the node while mounte? easy enough, if
1003 * desirable. */
1004 osb->node_num = o2nm_this_node();
1005 if (osb->node_num == O2NM_MAX_NODES) {
1006 mlog(ML_ERROR, "could not find this host's node number\n");
1007 status = -ENOENT;
1008 goto bail;
1009 }
1010
Sunil Mushran781ee3e2006-04-27 16:41:31 -07001011 mlog(0, "I am node %d\n", osb->node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001012
1013 status = 0;
1014bail:
1015 return status;
1016}
1017
1018static int ocfs2_mount_volume(struct super_block *sb)
1019{
1020 int status = 0;
1021 int unlock_super = 0;
1022 struct ocfs2_super *osb = OCFS2_SB(sb);
1023
1024 mlog_entry_void();
1025
1026 if (ocfs2_is_hard_readonly(osb))
1027 goto leave;
1028
1029 status = ocfs2_fill_local_node_info(osb);
1030 if (status < 0) {
1031 mlog_errno(status);
1032 goto leave;
1033 }
1034
1035 status = ocfs2_register_hb_callbacks(osb);
1036 if (status < 0) {
1037 mlog_errno(status);
1038 goto leave;
1039 }
1040
1041 status = ocfs2_dlm_init(osb);
1042 if (status < 0) {
1043 mlog_errno(status);
1044 goto leave;
1045 }
1046
1047 /* requires vote_thread to be running. */
1048 status = ocfs2_register_net_handlers(osb);
1049 if (status < 0) {
1050 mlog_errno(status);
1051 goto leave;
1052 }
1053
1054 status = ocfs2_super_lock(osb, 1);
1055 if (status < 0) {
1056 mlog_errno(status);
1057 goto leave;
1058 }
1059 unlock_super = 1;
1060
1061 /* This will load up the node map and add ourselves to it. */
1062 status = ocfs2_find_slot(osb);
1063 if (status < 0) {
1064 mlog_errno(status);
1065 goto leave;
1066 }
1067
1068 ocfs2_populate_mounted_map(osb);
1069
1070 /* load all node-local system inodes */
1071 status = ocfs2_init_local_system_inodes(osb);
1072 if (status < 0) {
1073 mlog_errno(status);
1074 goto leave;
1075 }
1076
1077 status = ocfs2_check_volume(osb);
1078 if (status < 0) {
1079 mlog_errno(status);
1080 goto leave;
1081 }
1082
1083 status = ocfs2_truncate_log_init(osb);
1084 if (status < 0) {
1085 mlog_errno(status);
1086 goto leave;
1087 }
1088
1089 /* This should be sent *after* we recovered our journal as it
1090 * will cause other nodes to unmark us as needing
1091 * recovery. However, we need to send it *before* dropping the
1092 * super block lock as otherwise their recovery threads might
1093 * try to clean us up while we're live! */
1094 status = ocfs2_request_mount_vote(osb);
1095 if (status < 0)
1096 mlog_errno(status);
1097
1098leave:
1099 if (unlock_super)
1100 ocfs2_super_unlock(osb, 1);
1101
1102 mlog_exit(status);
1103 return status;
1104}
1105
1106/* we can't grab the goofy sem lock from inside wait_event, so we use
1107 * memory barriers to make sure that we'll see the null task before
1108 * being woken up */
1109static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1110{
1111 mb();
1112 return osb->recovery_thread_task != NULL;
1113}
1114
1115static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1116{
1117 int tmp;
1118 struct ocfs2_super *osb = NULL;
1119
1120 mlog_entry("(0x%p)\n", sb);
1121
1122 BUG_ON(!sb);
1123 osb = OCFS2_SB(sb);
1124 BUG_ON(!osb);
1125
1126 ocfs2_shutdown_local_alloc(osb);
1127
1128 ocfs2_truncate_log_shutdown(osb);
1129
1130 /* disable any new recovery threads and wait for any currently
1131 * running ones to exit. Do this before setting the vol_state. */
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001132 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001133 osb->disable_recovery = 1;
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001134 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001135 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1136
1137 /* At this point, we know that no more recovery threads can be
1138 * launched, so wait for any recovery completion work to
1139 * complete. */
1140 flush_workqueue(ocfs2_wq);
1141
1142 ocfs2_journal_shutdown(osb);
1143
1144 ocfs2_sync_blockdev(sb);
1145
1146 /* No dlm means we've failed during mount, so skip all the
1147 * steps which depended on that to complete. */
1148 if (osb->dlm) {
1149 tmp = ocfs2_super_lock(osb, 1);
1150 if (tmp < 0) {
1151 mlog_errno(tmp);
1152 return;
1153 }
1154
1155 tmp = ocfs2_request_umount_vote(osb);
1156 if (tmp < 0)
1157 mlog_errno(tmp);
1158
1159 if (osb->slot_num != OCFS2_INVALID_SLOT)
1160 ocfs2_put_slot(osb);
1161
1162 ocfs2_super_unlock(osb, 1);
1163 }
1164
1165 ocfs2_release_system_inodes(osb);
1166
1167 if (osb->dlm) {
1168 ocfs2_unregister_net_handlers(osb);
1169
1170 ocfs2_dlm_shutdown(osb);
1171 }
1172
1173 ocfs2_clear_hb_callbacks(osb);
1174
1175 debugfs_remove(osb->osb_debug_root);
1176
1177 if (!mnt_err)
1178 ocfs2_stop_heartbeat(osb);
1179
1180 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1181
Sunil Mushran781ee3e2006-04-27 16:41:31 -07001182 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %d)\n",
1183 osb->dev_str, osb->node_num);
Mark Fashehccd979b2005-12-15 14:31:24 -08001184
1185 ocfs2_delete_osb(osb);
1186 kfree(osb);
1187 sb->s_dev = 0;
1188 sb->s_fs_info = NULL;
1189}
1190
1191static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1192 unsigned uuid_bytes)
1193{
1194 int i, ret;
1195 char *ptr;
1196
1197 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1198
1199 osb->uuid_str = kcalloc(1, OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1200 if (osb->uuid_str == NULL)
1201 return -ENOMEM;
1202
Mark Fashehccd979b2005-12-15 14:31:24 -08001203 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1204 /* print with null */
1205 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1206 if (ret != 2) /* drop super cleans up */
1207 return -EINVAL;
1208 /* then only advance past the last char */
1209 ptr += 2;
1210 }
1211
1212 return 0;
1213}
1214
1215static int ocfs2_initialize_super(struct super_block *sb,
1216 struct buffer_head *bh,
1217 int sector_size)
1218{
1219 int status = 0;
1220 int i;
1221 struct ocfs2_dinode *di = NULL;
1222 struct inode *inode = NULL;
1223 struct buffer_head *bitmap_bh = NULL;
1224 struct ocfs2_journal *journal;
1225 __le32 uuid_net_key;
1226 struct ocfs2_super *osb;
1227
1228 mlog_entry_void();
1229
1230 osb = kcalloc(1, sizeof(struct ocfs2_super), GFP_KERNEL);
1231 if (!osb) {
1232 status = -ENOMEM;
1233 mlog_errno(status);
1234 goto bail;
1235 }
1236
1237 sb->s_fs_info = osb;
1238 sb->s_op = &ocfs2_sops;
1239 sb->s_export_op = &ocfs2_export_ops;
1240 sb->s_flags |= MS_NOATIME;
1241 /* this is needed to support O_LARGEFILE */
1242 sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1243
1244 osb->sb = sb;
1245 /* Save off for ocfs2_rw_direct */
1246 osb->s_sectsize_bits = blksize_bits(sector_size);
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001247 BUG_ON(!osb->s_sectsize_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001248
1249 osb->net_response_ids = 0;
1250 spin_lock_init(&osb->net_response_lock);
1251 INIT_LIST_HEAD(&osb->net_response_list);
1252
1253 INIT_LIST_HEAD(&osb->osb_net_handlers);
1254 init_waitqueue_head(&osb->recovery_event);
1255 spin_lock_init(&osb->vote_task_lock);
1256 init_waitqueue_head(&osb->vote_event);
1257 osb->vote_work_sequence = 0;
1258 osb->vote_wake_sequence = 0;
1259 INIT_LIST_HEAD(&osb->blocked_lock_list);
1260 osb->blocked_lock_count = 0;
1261 INIT_LIST_HEAD(&osb->vote_list);
1262 spin_lock_init(&osb->osb_lock);
1263
1264 atomic_set(&osb->alloc_stats.moves, 0);
1265 atomic_set(&osb->alloc_stats.local_data, 0);
1266 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1267 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1268 atomic_set(&osb->alloc_stats.bg_extends, 0);
1269
1270 ocfs2_init_node_maps(osb);
1271
1272 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1273 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1274
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001275 mutex_init(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001276
1277 osb->disable_recovery = 0;
1278 osb->recovery_thread_task = NULL;
1279
1280 init_waitqueue_head(&osb->checkpoint_event);
1281 atomic_set(&osb->needs_checkpoint, 0);
1282
1283 osb->node_num = O2NM_INVALID_NODE_NUM;
1284 osb->slot_num = OCFS2_INVALID_SLOT;
1285
1286 osb->local_alloc_state = OCFS2_LA_UNUSED;
1287 osb->local_alloc_bh = NULL;
1288
1289 ocfs2_setup_hb_callbacks(osb);
1290
1291 init_waitqueue_head(&osb->osb_mount_event);
1292
1293 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1294 if (!osb->vol_label) {
1295 mlog(ML_ERROR, "unable to alloc vol label\n");
1296 status = -ENOMEM;
1297 goto bail;
1298 }
1299
Mark Fashehccd979b2005-12-15 14:31:24 -08001300 di = (struct ocfs2_dinode *)bh->b_data;
1301
1302 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1303 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1304 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1305 osb->max_slots);
1306 status = -EINVAL;
1307 goto bail;
1308 }
Sunil Mushran781ee3e2006-04-27 16:41:31 -07001309 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
Mark Fashehccd979b2005-12-15 14:31:24 -08001310
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001311 init_waitqueue_head(&osb->osb_wipe_event);
1312 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1313 sizeof(*osb->osb_orphan_wipes),
1314 GFP_KERNEL);
1315 if (!osb->osb_orphan_wipes) {
1316 status = -ENOMEM;
1317 mlog_errno(status);
1318 goto bail;
1319 }
1320
Mark Fashehccd979b2005-12-15 14:31:24 -08001321 osb->s_feature_compat =
1322 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1323 osb->s_feature_ro_compat =
1324 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1325 osb->s_feature_incompat =
1326 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1327
1328 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1329 mlog(ML_ERROR, "couldn't mount because of unsupported "
1330 "optional features (%x).\n", i);
1331 status = -EINVAL;
1332 goto bail;
1333 }
1334 if (!(osb->sb->s_flags & MS_RDONLY) &&
1335 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1336 mlog(ML_ERROR, "couldn't mount RDWR because of "
1337 "unsupported optional features (%x).\n", i);
1338 status = -EINVAL;
1339 goto bail;
1340 }
1341
1342 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1343
1344 /* FIXME
1345 * This should be done in ocfs2_journal_init(), but unknown
1346 * ordering issues will cause the filesystem to crash.
1347 * If anyone wants to figure out what part of the code
1348 * refers to osb->journal before ocfs2_journal_init() is run,
1349 * be my guest.
1350 */
1351 /* initialize our journal structure */
1352
1353 journal = kcalloc(1, sizeof(struct ocfs2_journal), GFP_KERNEL);
1354 if (!journal) {
1355 mlog(ML_ERROR, "unable to alloc journal\n");
1356 status = -ENOMEM;
1357 goto bail;
1358 }
1359 osb->journal = journal;
1360 journal->j_osb = osb;
1361
1362 atomic_set(&journal->j_num_trans, 0);
1363 init_rwsem(&journal->j_trans_barrier);
1364 init_waitqueue_head(&journal->j_checkpointed);
1365 spin_lock_init(&journal->j_lock);
1366 journal->j_trans_id = (unsigned long) 1;
1367 INIT_LIST_HEAD(&journal->j_la_cleanups);
1368 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery, osb);
1369 journal->j_state = OCFS2_JOURNAL_FREE;
1370
1371 /* get some pseudo constants for clustersize bits */
1372 osb->s_clustersize_bits =
1373 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1374 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1375 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1376
1377 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1378 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1379 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1380 osb->s_clustersize);
1381 status = -EINVAL;
1382 goto bail;
1383 }
1384
1385 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1386 > (u32)~0UL) {
1387 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1388 "what jbd can address in 32 bits.\n");
1389 status = -EINVAL;
1390 goto bail;
1391 }
1392
1393 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1394 sizeof(di->id2.i_super.s_uuid))) {
1395 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1396 status = -ENOMEM;
1397 goto bail;
1398 }
1399
Mark Fasheh78427042006-05-04 12:03:26 -07001400 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
Mark Fashehccd979b2005-12-15 14:31:24 -08001401 osb->net_key = le32_to_cpu(uuid_net_key);
1402
1403 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1404 osb->vol_label[63] = '\0';
1405 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1406 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1407 osb->first_cluster_group_blkno =
1408 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1409 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1410 mlog(0, "vol_label: %s\n", osb->vol_label);
1411 mlog(0, "uuid: %s\n", osb->uuid_str);
Mark Fashehb0697052006-03-03 10:24:33 -08001412 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1413 (unsigned long long)osb->root_blkno,
1414 (unsigned long long)osb->system_dir_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001415
1416 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1417 if (!osb->osb_dlm_debug) {
1418 status = -ENOMEM;
1419 mlog_errno(status);
1420 goto bail;
1421 }
1422
1423 atomic_set(&osb->vol_state, VOLUME_INIT);
1424
1425 /* load root, system_dir, and all global system inodes */
1426 status = ocfs2_init_global_system_inodes(osb);
1427 if (status < 0) {
1428 mlog_errno(status);
1429 goto bail;
1430 }
1431
1432 /*
1433 * global bitmap
1434 */
1435 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1436 OCFS2_INVALID_SLOT);
1437 if (!inode) {
1438 status = -EINVAL;
1439 mlog_errno(status);
1440 goto bail;
1441 }
1442
1443 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1444
Mark Fasheh101ebf22006-05-02 17:54:45 -07001445 /* We don't have a cluster lock on the bitmap here because
1446 * we're only interested in static information and the extra
1447 * complexity at mount time isn't worht it. Don't pass the
1448 * inode in to the read function though as we don't want it to
1449 * be put in the cache. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001450 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
Mark Fasheh101ebf22006-05-02 17:54:45 -07001451 NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001452 iput(inode);
1453 if (status < 0) {
1454 mlog_errno(status);
1455 goto bail;
1456 }
1457
1458 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1459 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
Mark Fashehccd979b2005-12-15 14:31:24 -08001460 brelse(bitmap_bh);
Mark Fashehb0697052006-03-03 10:24:33 -08001461 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1462 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
Mark Fashehccd979b2005-12-15 14:31:24 -08001463
1464 status = ocfs2_init_slot_info(osb);
1465 if (status < 0) {
1466 mlog_errno(status);
1467 goto bail;
1468 }
1469
Mark Fashehccd979b2005-12-15 14:31:24 -08001470bail:
1471 mlog_exit(status);
1472 return status;
1473}
1474
1475/*
1476 * will return: -EAGAIN if it is ok to keep searching for superblocks
1477 * -EINVAL if there is a bad superblock
1478 * 0 on success
1479 */
1480static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1481 struct buffer_head *bh,
1482 u32 blksz)
1483{
1484 int status = -EAGAIN;
1485
1486 mlog_entry_void();
1487
1488 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1489 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1490 status = -EINVAL;
1491 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1492 mlog(ML_ERROR, "found superblock with incorrect block "
1493 "size: found %u, should be %u\n",
1494 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1495 blksz);
1496 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1497 OCFS2_MAJOR_REV_LEVEL ||
1498 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1499 OCFS2_MINOR_REV_LEVEL) {
1500 mlog(ML_ERROR, "found superblock with bad version: "
1501 "found %u.%u, should be %u.%u\n",
1502 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1503 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1504 OCFS2_MAJOR_REV_LEVEL,
1505 OCFS2_MINOR_REV_LEVEL);
1506 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1507 mlog(ML_ERROR, "bad block number on superblock: "
Mark Fashehb0697052006-03-03 10:24:33 -08001508 "found %llu, should be %llu\n",
1509 (unsigned long long)di->i_blkno,
1510 (unsigned long long)bh->b_blocknr);
Mark Fashehccd979b2005-12-15 14:31:24 -08001511 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1512 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1513 mlog(ML_ERROR, "bad cluster size found: %u\n",
1514 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1515 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1516 mlog(ML_ERROR, "bad root_blkno: 0\n");
1517 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1518 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1519 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1520 mlog(ML_ERROR,
1521 "Superblock slots found greater than file system "
1522 "maximum: found %u, max %u\n",
1523 le16_to_cpu(di->id2.i_super.s_max_slots),
1524 OCFS2_MAX_SLOTS);
1525 } else {
1526 /* found it! */
1527 status = 0;
1528 }
1529 }
1530
1531 mlog_exit(status);
1532 return status;
1533}
1534
1535static int ocfs2_check_volume(struct ocfs2_super *osb)
1536{
1537 int status = 0;
1538 int dirty;
1539 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1540 * recover
1541 * ourselves. */
1542
1543 mlog_entry_void();
1544
1545 /* Init our journal object. */
1546 status = ocfs2_journal_init(osb->journal, &dirty);
1547 if (status < 0) {
1548 mlog(ML_ERROR, "Could not initialize journal!\n");
1549 goto finally;
1550 }
1551
1552 /* If the journal was unmounted cleanly then we don't want to
1553 * recover anything. Otherwise, journal_load will do that
1554 * dirty work for us :) */
1555 if (!dirty) {
1556 status = ocfs2_journal_wipe(osb->journal, 0);
1557 if (status < 0) {
1558 mlog_errno(status);
1559 goto finally;
1560 }
1561 } else {
1562 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1563 "recovering volume.\n");
1564 }
1565
1566 /* will play back anything left in the journal. */
1567 ocfs2_journal_load(osb->journal);
1568
1569 if (dirty) {
1570 /* recover my local alloc if we didn't unmount cleanly. */
1571 status = ocfs2_begin_local_alloc_recovery(osb,
1572 osb->slot_num,
1573 &local_alloc);
1574 if (status < 0) {
1575 mlog_errno(status);
1576 goto finally;
1577 }
1578 /* we complete the recovery process after we've marked
1579 * ourselves as mounted. */
1580 }
1581
1582 mlog(0, "Journal loaded.\n");
1583
1584 status = ocfs2_load_local_alloc(osb);
1585 if (status < 0) {
1586 mlog_errno(status);
1587 goto finally;
1588 }
1589
1590 if (dirty) {
1591 /* Recovery will be completed after we've mounted the
1592 * rest of the volume. */
1593 osb->dirty = 1;
1594 osb->local_alloc_copy = local_alloc;
1595 local_alloc = NULL;
1596 }
1597
1598 /* go through each journal, trylock it and if you get the
1599 * lock, and it's marked as dirty, set the bit in the recover
1600 * map and launch a recovery thread for it. */
1601 status = ocfs2_mark_dead_nodes(osb);
1602 if (status < 0)
1603 mlog_errno(status);
1604
1605finally:
1606 if (local_alloc)
1607 kfree(local_alloc);
1608
1609 mlog_exit(status);
1610 return status;
1611}
1612
1613/*
1614 * The routine gets called from dismount or close whenever a dismount on
1615 * volume is requested and the osb open count becomes 1.
1616 * It will remove the osb from the global list and also free up all the
1617 * initialized resources and fileobject.
1618 */
1619static void ocfs2_delete_osb(struct ocfs2_super *osb)
1620{
1621 mlog_entry_void();
1622
1623 /* This function assumes that the caller has the main osb resource */
1624
1625 if (osb->slot_info)
1626 ocfs2_free_slot_info(osb->slot_info);
1627
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001628 kfree(osb->osb_orphan_wipes);
Mark Fashehccd979b2005-12-15 14:31:24 -08001629 /* FIXME
1630 * This belongs in journal shutdown, but because we have to
1631 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1632 * we free it here.
1633 */
1634 kfree(osb->journal);
1635 if (osb->local_alloc_copy)
1636 kfree(osb->local_alloc_copy);
1637 kfree(osb->uuid_str);
1638 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1639 memset(osb, 0, sizeof(struct ocfs2_super));
1640
1641 mlog_exit_void();
1642}
1643
1644/* Put OCFS2 into a readonly state, or (if the user specifies it),
1645 * panic(). We do not support continue-on-error operation. */
1646static void ocfs2_handle_error(struct super_block *sb)
1647{
1648 struct ocfs2_super *osb = OCFS2_SB(sb);
1649
1650 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1651 panic("OCFS2: (device %s): panic forced after error\n",
1652 sb->s_id);
1653
1654 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1655
1656 if (sb->s_flags & MS_RDONLY &&
1657 (ocfs2_is_soft_readonly(osb) ||
1658 ocfs2_is_hard_readonly(osb)))
1659 return;
1660
1661 printk(KERN_CRIT "File system is now read-only due to the potential "
1662 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1663 "system is unmounted.\n");
1664 sb->s_flags |= MS_RDONLY;
1665 ocfs2_set_ro_flag(osb, 0);
1666}
1667
1668static char error_buf[1024];
1669
1670void __ocfs2_error(struct super_block *sb,
1671 const char *function,
1672 const char *fmt, ...)
1673{
1674 va_list args;
1675
1676 va_start(args, fmt);
1677 vsprintf(error_buf, fmt, args);
1678 va_end(args);
1679
1680 /* Not using mlog here because we want to show the actual
1681 * function the error came from. */
1682 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1683 sb->s_id, function, error_buf);
1684
1685 ocfs2_handle_error(sb);
1686}
1687
1688/* Handle critical errors. This is intentionally more drastic than
1689 * ocfs2_handle_error, so we only use for things like journal errors,
1690 * etc. */
1691void __ocfs2_abort(struct super_block* sb,
1692 const char *function,
1693 const char *fmt, ...)
1694{
1695 va_list args;
1696
1697 va_start(args, fmt);
1698 vsprintf(error_buf, fmt, args);
1699 va_end(args);
1700
1701 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1702 sb->s_id, function, error_buf);
1703
1704 /* We don't have the cluster support yet to go straight to
1705 * hard readonly in here. Until then, we want to keep
1706 * ocfs2_abort() so that we can at least mark critical
1707 * errors.
1708 *
1709 * TODO: This should abort the journal and alert other nodes
1710 * that our slot needs recovery. */
1711
1712 /* Force a panic(). This stinks, but it's better than letting
1713 * things continue without having a proper hard readonly
1714 * here. */
1715 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1716 ocfs2_handle_error(sb);
1717}
1718
1719module_init(ocfs2_init);
1720module_exit(ocfs2_exit);