blob: 949b3dac30f142127cefd25226d149c4c8bf465a [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
71/*
72 * Globals
73 */
74static spinlock_t ocfs2_globals_lock = SPIN_LOCK_UNLOCKED;
75
76static u32 osb_id; /* Keeps track of next available OSB Id */
77
78static kmem_cache_t *ocfs2_inode_cachep = NULL;
79
80kmem_cache_t *ocfs2_lock_cache = NULL;
81
82/* OCFS2 needs to schedule several differnt types of work which
83 * require cluster locking, disk I/O, recovery waits, etc. Since these
84 * types of work tend to be heavy we avoid using the kernel events
85 * workqueue and schedule on our own. */
86struct workqueue_struct *ocfs2_wq = NULL;
87
88static struct dentry *ocfs2_debugfs_root = NULL;
89
90MODULE_AUTHOR("Oracle");
91MODULE_LICENSE("GPL");
92
93static int ocfs2_parse_options(struct super_block *sb, char *options,
94 unsigned long *mount_opt, int is_remount);
95static void ocfs2_put_super(struct super_block *sb);
96static int ocfs2_mount_volume(struct super_block *sb);
97static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
98static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
99static int ocfs2_initialize_mem_caches(void);
100static void ocfs2_free_mem_caches(void);
101static void ocfs2_delete_osb(struct ocfs2_super *osb);
102
103static int ocfs2_statfs(struct super_block *sb, struct kstatfs *buf);
104
105static int ocfs2_sync_fs(struct super_block *sb, int wait);
106
107static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
108static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
109static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
110static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
111static int ocfs2_check_volume(struct ocfs2_super *osb);
112static int ocfs2_verify_volume(struct ocfs2_dinode *di,
113 struct buffer_head *bh,
114 u32 sectsize);
115static int ocfs2_initialize_super(struct super_block *sb,
116 struct buffer_head *bh,
117 int sector_size);
118static int ocfs2_get_sector(struct super_block *sb,
119 struct buffer_head **bh,
120 int block,
121 int sect_size);
122static void ocfs2_write_super(struct super_block *sb);
123static struct inode *ocfs2_alloc_inode(struct super_block *sb);
124static void ocfs2_destroy_inode(struct inode *inode);
125
126static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
127
128static struct super_operations ocfs2_sops = {
129 .statfs = ocfs2_statfs,
130 .alloc_inode = ocfs2_alloc_inode,
131 .destroy_inode = ocfs2_destroy_inode,
132 .drop_inode = ocfs2_drop_inode,
133 .clear_inode = ocfs2_clear_inode,
134 .delete_inode = ocfs2_delete_inode,
135 .sync_fs = ocfs2_sync_fs,
136 .write_super = ocfs2_write_super,
137 .put_super = ocfs2_put_super,
138 .remount_fs = ocfs2_remount,
139};
140
141enum {
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_err,
152};
153
154static match_table_t tokens = {
155 {Opt_barrier, "barrier=%u"},
156 {Opt_err_panic, "errors=panic"},
157 {Opt_err_ro, "errors=remount-ro"},
158 {Opt_intr, "intr"},
159 {Opt_nointr, "nointr"},
160 {Opt_hb_none, OCFS2_HB_NONE},
161 {Opt_hb_local, OCFS2_HB_LOCAL},
162 {Opt_data_ordered, "data=ordered"},
163 {Opt_data_writeback, "data=writeback"},
164 {Opt_err, NULL}
165};
166
167/*
168 * write_super and sync_fs ripped right out of ext3.
169 */
170static void ocfs2_write_super(struct super_block *sb)
171{
Ingo Molnar7892f2f2006-01-09 15:59:25 -0800172 if (mutex_trylock(&sb->s_lock) != 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800173 BUG();
174 sb->s_dirt = 0;
175}
176
177static int ocfs2_sync_fs(struct super_block *sb, int wait)
178{
179 int status = 0;
180 tid_t target;
181 struct ocfs2_super *osb = OCFS2_SB(sb);
182
183 sb->s_dirt = 0;
184
185 if (ocfs2_is_hard_readonly(osb))
186 return -EROFS;
187
188 if (wait) {
189 status = ocfs2_flush_truncate_log(osb);
190 if (status < 0)
191 mlog_errno(status);
192 } else {
193 ocfs2_schedule_truncate_log_flush(osb, 0);
194 }
195
196 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
197 if (wait)
198 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
199 target);
200 }
201 return 0;
202}
203
204static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
205{
206 struct inode *new = NULL;
207 int status = 0;
208 int i;
209
210 mlog_entry_void();
211
212 new = ocfs2_iget(osb, osb->root_blkno);
213 if (IS_ERR(new)) {
214 status = PTR_ERR(new);
215 mlog_errno(status);
216 goto bail;
217 }
218 osb->root_inode = new;
219
220 new = ocfs2_iget(osb, osb->system_dir_blkno);
221 if (IS_ERR(new)) {
222 status = PTR_ERR(new);
223 mlog_errno(status);
224 goto bail;
225 }
226 osb->sys_root_inode = new;
227
228 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
229 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
230 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
231 if (!new) {
232 ocfs2_release_system_inodes(osb);
233 status = -EINVAL;
234 mlog_errno(status);
235 /* FIXME: Should ERROR_RO_FS */
236 mlog(ML_ERROR, "Unable to load system inode %d, "
237 "possibly corrupt fs?", i);
238 goto bail;
239 }
240 // the array now has one ref, so drop this one
241 iput(new);
242 }
243
244bail:
245 mlog_exit(status);
246 return status;
247}
248
249static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
250{
251 struct inode *new = NULL;
252 int status = 0;
253 int i;
254
255 mlog_entry_void();
256
257 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
258 i < NUM_SYSTEM_INODES;
259 i++) {
260 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
261 if (!new) {
262 ocfs2_release_system_inodes(osb);
263 status = -EINVAL;
264 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
265 status, i, osb->slot_num);
266 goto bail;
267 }
268 /* the array now has one ref, so drop this one */
269 iput(new);
270 }
271
272bail:
273 mlog_exit(status);
274 return status;
275}
276
277static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
278{
279 int status = 0, i;
280 struct inode *inode;
281
282 mlog_entry_void();
283
284 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
285 inode = osb->system_inodes[i];
286 if (inode) {
287 iput(inode);
288 osb->system_inodes[i] = NULL;
289 }
290 }
291
292 inode = osb->sys_root_inode;
293 if (inode) {
294 iput(inode);
295 osb->sys_root_inode = NULL;
296 }
297
298 inode = osb->root_inode;
299 if (inode) {
300 iput(inode);
301 osb->root_inode = NULL;
302 }
303
304 mlog_exit(status);
305 return status;
306}
307
308/* We're allocating fs objects, use GFP_NOFS */
309static struct inode *ocfs2_alloc_inode(struct super_block *sb)
310{
311 struct ocfs2_inode_info *oi;
312
313 oi = kmem_cache_alloc(ocfs2_inode_cachep, SLAB_NOFS);
314 if (!oi)
315 return NULL;
316
317 return &oi->vfs_inode;
318}
319
320static void ocfs2_destroy_inode(struct inode *inode)
321{
322 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
323}
324
325/* From xfs_super.c:xfs_max_file_offset
326 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
327 */
328static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
329{
330 unsigned int pagefactor = 1;
331 unsigned int bitshift = BITS_PER_LONG - 1;
332
333 /* Figure out maximum filesize, on Linux this can depend on
334 * the filesystem blocksize (on 32 bit platforms).
335 * __block_prepare_write does this in an [unsigned] long...
336 * page->index << (PAGE_CACHE_SHIFT - bbits)
337 * So, for page sized blocks (4K on 32 bit platforms),
338 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
339 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
340 * but for smaller blocksizes it is less (bbits = log2 bsize).
341 * Note1: get_block_t takes a long (implicit cast from above)
342 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
343 * can optionally convert the [unsigned] long from above into
344 * an [unsigned] long long.
345 */
346
347#if BITS_PER_LONG == 32
348# if defined(CONFIG_LBD)
349 BUG_ON(sizeof(sector_t) != 8);
350 pagefactor = PAGE_CACHE_SIZE;
351 bitshift = BITS_PER_LONG;
352# else
353 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
354# endif
355#endif
356
357 return (((unsigned long long)pagefactor) << bitshift) - 1;
358}
359
360static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
361{
362 int incompat_features;
363 int ret = 0;
364 unsigned long parsed_options;
365 struct ocfs2_super *osb = OCFS2_SB(sb);
366
367 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
368 ret = -EINVAL;
369 goto out;
370 }
371
372 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
373 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
374 ret = -EINVAL;
375 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
376 goto out;
377 }
378
379 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
380 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
381 ret = -EINVAL;
382 mlog(ML_ERROR, "Cannot change data mode on remount\n");
383 goto out;
384 }
385
386 /* We're going to/from readonly mode. */
387 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
388 /* Lock here so the check of HARD_RO and the potential
389 * setting of SOFT_RO is atomic. */
390 spin_lock(&osb->osb_lock);
391 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
392 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
393 ret = -EROFS;
394 goto unlock_osb;
395 }
396
397 if (*flags & MS_RDONLY) {
398 mlog(0, "Going to ro mode.\n");
399 sb->s_flags |= MS_RDONLY;
400 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
401 } else {
402 mlog(0, "Making ro filesystem writeable.\n");
403
404 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
405 mlog(ML_ERROR, "Cannot remount RDWR "
406 "filesystem due to previous errors.\n");
407 ret = -EROFS;
408 goto unlock_osb;
409 }
410 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
411 if (incompat_features) {
412 mlog(ML_ERROR, "Cannot remount RDWR because "
413 "of unsupported optional features "
414 "(%x).\n", incompat_features);
415 ret = -EINVAL;
416 goto unlock_osb;
417 }
418 sb->s_flags &= ~MS_RDONLY;
419 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
420 }
421unlock_osb:
422 spin_unlock(&osb->osb_lock);
423 }
424
425 if (!ret) {
426 if (!ocfs2_is_hard_readonly(osb))
427 ocfs2_set_journal_params(osb);
428
429 /* Only save off the new mount options in case of a successful
430 * remount. */
431 osb->s_mount_opt = parsed_options;
432 }
433out:
434 return ret;
435}
436
437static int ocfs2_sb_probe(struct super_block *sb,
438 struct buffer_head **bh,
439 int *sector_size)
440{
441 int status = 0, tmpstat;
442 struct ocfs1_vol_disk_hdr *hdr;
443 struct ocfs2_dinode *di;
444 int blksize;
445
446 *bh = NULL;
447
448 /* may be > 512 */
449 *sector_size = bdev_hardsect_size(sb->s_bdev);
450 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
451 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
452 *sector_size, OCFS2_MAX_BLOCKSIZE);
453 status = -EINVAL;
454 goto bail;
455 }
456
457 /* Can this really happen? */
458 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
459 *sector_size = OCFS2_MIN_BLOCKSIZE;
460
461 /* check block zero for old format */
462 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
463 if (status < 0) {
464 mlog_errno(status);
465 goto bail;
466 }
467 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
468 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
469 mlog(ML_ERROR, "incompatible version: %u.%u\n",
470 hdr->major_version, hdr->minor_version);
471 status = -EINVAL;
472 }
473 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
474 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
475 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
476 hdr->signature);
477 status = -EINVAL;
478 }
479 brelse(*bh);
480 *bh = NULL;
481 if (status < 0) {
482 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
483 "upgraded before mounting with ocfs v2\n");
484 goto bail;
485 }
486
487 /*
488 * Now check at magic offset for 512, 1024, 2048, 4096
489 * blocksizes. 4096 is the maximum blocksize because it is
490 * the minimum clustersize.
491 */
492 status = -EINVAL;
493 for (blksize = *sector_size;
494 blksize <= OCFS2_MAX_BLOCKSIZE;
495 blksize <<= 1) {
496 tmpstat = ocfs2_get_sector(sb, bh,
497 OCFS2_SUPER_BLOCK_BLKNO,
498 blksize);
499 if (tmpstat < 0) {
500 status = tmpstat;
501 mlog_errno(status);
502 goto bail;
503 }
504 di = (struct ocfs2_dinode *) (*bh)->b_data;
505 status = ocfs2_verify_volume(di, *bh, blksize);
506 if (status >= 0)
507 goto bail;
508 brelse(*bh);
509 *bh = NULL;
510 if (status != -EAGAIN)
511 break;
512 }
513
514bail:
515 return status;
516}
517
518static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
519{
520 struct dentry *root;
521 int status, sector_size;
522 unsigned long parsed_opt;
523 struct inode *inode = NULL;
524 struct ocfs2_super *osb = NULL;
525 struct buffer_head *bh = NULL;
526
527 mlog_entry("%p, %p, %i", sb, data, silent);
528
529 /* for now we only have one cluster/node, make sure we see it
530 * in the heartbeat universe */
531 if (!o2hb_check_local_node_heartbeating()) {
532 status = -EINVAL;
533 goto read_super_error;
534 }
535
536 /* probe for superblock */
537 status = ocfs2_sb_probe(sb, &bh, &sector_size);
538 if (status < 0) {
539 mlog(ML_ERROR, "superblock probe failed!\n");
540 goto read_super_error;
541 }
542
543 status = ocfs2_initialize_super(sb, bh, sector_size);
544 osb = OCFS2_SB(sb);
545 if (status < 0) {
546 mlog_errno(status);
547 goto read_super_error;
548 }
549 brelse(bh);
550 bh = NULL;
551
552 if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
553 status = -EINVAL;
554 goto read_super_error;
555 }
556 osb->s_mount_opt = parsed_opt;
557
558 sb->s_magic = OCFS2_SUPER_MAGIC;
559
560 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
561 * heartbeat=none */
562 if (bdev_read_only(sb->s_bdev)) {
563 if (!(sb->s_flags & MS_RDONLY)) {
564 status = -EACCES;
565 mlog(ML_ERROR, "Readonly device detected but readonly "
566 "mount was not specified.\n");
567 goto read_super_error;
568 }
569
570 /* You should not be able to start a local heartbeat
571 * on a readonly device. */
572 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
573 status = -EROFS;
574 mlog(ML_ERROR, "Local heartbeat specified on readonly "
575 "device.\n");
576 goto read_super_error;
577 }
578
579 status = ocfs2_check_journals_nolocks(osb);
580 if (status < 0) {
581 if (status == -EROFS)
582 mlog(ML_ERROR, "Recovery required on readonly "
583 "file system, but write access is "
584 "unavailable.\n");
585 else
586 mlog_errno(status);
587 goto read_super_error;
588 }
589
590 ocfs2_set_ro_flag(osb, 1);
591
592 printk(KERN_NOTICE "Readonly device detected. No cluster "
593 "services will be utilized for this mount. Recovery "
594 "will be skipped.\n");
595 }
596
597 if (!ocfs2_is_hard_readonly(osb)) {
598 /* If this isn't a hard readonly mount, then we need
599 * to make sure that heartbeat is in a valid state,
600 * and that we mark ourselves soft readonly is -oro
601 * was specified. */
602 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
603 mlog(ML_ERROR, "No heartbeat for device (%s)\n",
604 sb->s_id);
605 status = -EINVAL;
606 goto read_super_error;
607 }
608
609 if (sb->s_flags & MS_RDONLY)
610 ocfs2_set_ro_flag(osb, 0);
611 }
612
613 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
614 ocfs2_debugfs_root);
615 if (!osb->osb_debug_root) {
616 status = -EINVAL;
617 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
618 goto read_super_error;
619 }
620
621 status = ocfs2_mount_volume(sb);
622 if (osb->root_inode)
623 inode = igrab(osb->root_inode);
624
625 if (status < 0)
626 goto read_super_error;
627
628 if (!inode) {
629 status = -EIO;
630 mlog_errno(status);
631 goto read_super_error;
632 }
633
634 root = d_alloc_root(inode);
635 if (!root) {
636 status = -ENOMEM;
637 mlog_errno(status);
638 goto read_super_error;
639 }
640
641 sb->s_root = root;
642
643 ocfs2_complete_mount_recovery(osb);
644
645 printk("ocfs2: Mounting device (%u,%u) on (node %d, slot %d) with %s "
646 "data mode.\n",
647 MAJOR(sb->s_dev), MINOR(sb->s_dev), osb->node_num,
648 osb->slot_num,
649 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
650 "ordered");
651
652 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
653 wake_up(&osb->osb_mount_event);
654
655 mlog_exit(status);
656 return status;
657
658read_super_error:
659 if (bh != NULL)
660 brelse(bh);
661
662 if (inode)
663 iput(inode);
664
665 if (osb) {
666 atomic_set(&osb->vol_state, VOLUME_DISABLED);
667 wake_up(&osb->osb_mount_event);
668 ocfs2_dismount_volume(sb, 1);
669 }
670
671 mlog_exit(status);
672 return status;
673}
674
675static struct super_block *ocfs2_get_sb(struct file_system_type *fs_type,
676 int flags,
677 const char *dev_name,
678 void *data)
679{
680 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
681}
682
683static struct file_system_type ocfs2_fs_type = {
684 .owner = THIS_MODULE,
685 .name = "ocfs2",
686 .get_sb = ocfs2_get_sb, /* is this called when we mount
687 * the fs? */
688 .kill_sb = kill_block_super, /* set to the generic one
689 * right now, but do we
690 * need to change that? */
691 .fs_flags = FS_REQUIRES_DEV,
692 .next = NULL
693};
694
695static int ocfs2_parse_options(struct super_block *sb,
696 char *options,
697 unsigned long *mount_opt,
698 int is_remount)
699{
700 int status;
701 char *p;
702
703 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
704 options ? options : "(none)");
705
706 *mount_opt = 0;
707
708 if (!options) {
709 status = 1;
710 goto bail;
711 }
712
713 while ((p = strsep(&options, ",")) != NULL) {
714 int token, option;
715 substring_t args[MAX_OPT_ARGS];
716
717 if (!*p)
718 continue;
719
720 token = match_token(p, tokens, args);
721 switch (token) {
722 case Opt_hb_local:
723 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
724 break;
725 case Opt_hb_none:
726 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
727 break;
728 case Opt_barrier:
729 if (match_int(&args[0], &option)) {
730 status = 0;
731 goto bail;
732 }
733 if (option)
734 *mount_opt |= OCFS2_MOUNT_BARRIER;
735 else
736 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
737 break;
738 case Opt_intr:
739 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
740 break;
741 case Opt_nointr:
742 *mount_opt |= OCFS2_MOUNT_NOINTR;
743 break;
744 case Opt_err_panic:
745 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
746 break;
747 case Opt_err_ro:
748 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
749 break;
750 case Opt_data_ordered:
751 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
752 break;
753 case Opt_data_writeback:
754 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
755 break;
756 default:
757 mlog(ML_ERROR,
758 "Unrecognized mount option \"%s\" "
759 "or missing value\n", p);
760 status = 0;
761 goto bail;
762 }
763 }
764
765 status = 1;
766
767bail:
768 mlog_exit(status);
769 return status;
770}
771
772static int __init ocfs2_init(void)
773{
774 int status;
775
776 mlog_entry_void();
777
778 ocfs2_print_version();
779
780 if (init_ocfs2_extent_maps())
781 return -ENOMEM;
782
783 status = init_ocfs2_uptodate_cache();
784 if (status < 0) {
785 mlog_errno(status);
786 goto leave;
787 }
788
789 status = ocfs2_initialize_mem_caches();
790 if (status < 0) {
791 mlog_errno(status);
792 goto leave;
793 }
794
795 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
796 if (!ocfs2_wq) {
797 status = -ENOMEM;
798 goto leave;
799 }
800
801 spin_lock(&ocfs2_globals_lock);
802 osb_id = 0;
803 spin_unlock(&ocfs2_globals_lock);
804
805 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
806 if (!ocfs2_debugfs_root) {
807 status = -EFAULT;
808 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
809 }
810
811leave:
812 if (status < 0) {
813 ocfs2_free_mem_caches();
814 exit_ocfs2_uptodate_cache();
815 exit_ocfs2_extent_maps();
816 }
817
818 mlog_exit(status);
819
820 if (status >= 0) {
821 return register_filesystem(&ocfs2_fs_type);
822 } else
823 return -1;
824}
825
826static void __exit ocfs2_exit(void)
827{
828 mlog_entry_void();
829
830 if (ocfs2_wq) {
831 flush_workqueue(ocfs2_wq);
832 destroy_workqueue(ocfs2_wq);
833 }
834
835 debugfs_remove(ocfs2_debugfs_root);
836
837 ocfs2_free_mem_caches();
838
839 unregister_filesystem(&ocfs2_fs_type);
840
841 exit_ocfs2_extent_maps();
842
843 exit_ocfs2_uptodate_cache();
844
845 mlog_exit_void();
846}
847
848static void ocfs2_put_super(struct super_block *sb)
849{
850 mlog_entry("(0x%p)\n", sb);
851
852 ocfs2_sync_blockdev(sb);
853 ocfs2_dismount_volume(sb, 0);
854
855 mlog_exit_void();
856}
857
858static int ocfs2_statfs(struct super_block *sb, struct kstatfs *buf)
859{
860 struct ocfs2_super *osb;
861 u32 numbits, freebits;
862 int status;
863 struct ocfs2_dinode *bm_lock;
864 struct buffer_head *bh = NULL;
865 struct inode *inode = NULL;
866
867 mlog_entry("(%p, %p)\n", sb, buf);
868
869 osb = OCFS2_SB(sb);
870
871 inode = ocfs2_get_system_file_inode(osb,
872 GLOBAL_BITMAP_SYSTEM_INODE,
873 OCFS2_INVALID_SLOT);
874 if (!inode) {
875 mlog(ML_ERROR, "failed to get bitmap inode\n");
876 status = -EIO;
877 goto bail;
878 }
879
880 status = ocfs2_meta_lock(inode, NULL, &bh, 0);
881 if (status < 0) {
882 mlog_errno(status);
883 goto bail;
884 }
885
886 bm_lock = (struct ocfs2_dinode *) bh->b_data;
887
888 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
889 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
890
891 buf->f_type = OCFS2_SUPER_MAGIC;
892 buf->f_bsize = sb->s_blocksize;
893 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
894 buf->f_blocks = ((sector_t) numbits) *
895 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
896 buf->f_bfree = ((sector_t) freebits) *
897 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
898 buf->f_bavail = buf->f_bfree;
899 buf->f_files = numbits;
900 buf->f_ffree = freebits;
901
902 brelse(bh);
903
904 ocfs2_meta_unlock(inode, 0);
905 status = 0;
906bail:
907 if (inode)
908 iput(inode);
909
910 mlog_exit(status);
911
912 return status;
913}
914
915static void ocfs2_inode_init_once(void *data,
916 kmem_cache_t *cachep,
917 unsigned long flags)
918{
919 struct ocfs2_inode_info *oi = data;
920
921 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
922 SLAB_CTOR_CONSTRUCTOR) {
923 oi->ip_flags = 0;
924 oi->ip_open_count = 0;
925 spin_lock_init(&oi->ip_lock);
926 ocfs2_extent_map_init(&oi->vfs_inode);
927 INIT_LIST_HEAD(&oi->ip_handle_list);
928 INIT_LIST_HEAD(&oi->ip_io_markers);
929 oi->ip_handle = NULL;
930 oi->ip_created_trans = 0;
931 oi->ip_last_trans = 0;
932 oi->ip_dir_start_lookup = 0;
933
934 init_rwsem(&oi->ip_alloc_sem);
Mark Fasheh251b6ec2006-01-10 15:41:43 -0800935 mutex_init(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800936
937 oi->ip_blkno = 0ULL;
938 oi->ip_clusters = 0;
939
940 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
941 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
942 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
943
944 ocfs2_metadata_cache_init(&oi->vfs_inode);
945
946 inode_init_once(&oi->vfs_inode);
947 }
948}
949
950static int ocfs2_initialize_mem_caches(void)
951{
952 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800953 sizeof(struct ocfs2_inode_info),
954 0,
955 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
956 SLAB_MEM_SPREAD),
957 ocfs2_inode_init_once, NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800958 if (!ocfs2_inode_cachep)
959 return -ENOMEM;
960
961 ocfs2_lock_cache = kmem_cache_create("ocfs2_lock",
962 sizeof(struct ocfs2_journal_lock),
963 0,
Christoph Lameterac2b8982006-03-22 00:08:15 -0800964 SLAB_HWCACHE_ALIGN,
Mark Fashehccd979b2005-12-15 14:31:24 -0800965 NULL, NULL);
966 if (!ocfs2_lock_cache)
967 return -ENOMEM;
968
969 return 0;
970}
971
972static void ocfs2_free_mem_caches(void)
973{
974 if (ocfs2_inode_cachep)
975 kmem_cache_destroy(ocfs2_inode_cachep);
976 if (ocfs2_lock_cache)
977 kmem_cache_destroy(ocfs2_lock_cache);
978
979 ocfs2_inode_cachep = NULL;
980 ocfs2_lock_cache = NULL;
981}
982
983static int ocfs2_get_sector(struct super_block *sb,
984 struct buffer_head **bh,
985 int block,
986 int sect_size)
987{
988 if (!sb_set_blocksize(sb, sect_size)) {
989 mlog(ML_ERROR, "unable to set blocksize\n");
990 return -EIO;
991 }
992
993 *bh = sb_getblk(sb, block);
994 if (!*bh) {
995 mlog_errno(-EIO);
996 return -EIO;
997 }
998 lock_buffer(*bh);
999 if (!buffer_dirty(*bh))
1000 clear_buffer_uptodate(*bh);
1001 unlock_buffer(*bh);
1002 ll_rw_block(READ, 1, bh);
1003 wait_on_buffer(*bh);
1004 return 0;
1005}
1006
1007/* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1008static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1009{
1010 int status;
1011
1012 /* XXX hold a ref on the node while mounte? easy enough, if
1013 * desirable. */
1014 osb->node_num = o2nm_this_node();
1015 if (osb->node_num == O2NM_MAX_NODES) {
1016 mlog(ML_ERROR, "could not find this host's node number\n");
1017 status = -ENOENT;
1018 goto bail;
1019 }
1020
1021 mlog(ML_NOTICE, "I am node %d\n", osb->node_num);
1022
1023 status = 0;
1024bail:
1025 return status;
1026}
1027
1028static int ocfs2_mount_volume(struct super_block *sb)
1029{
1030 int status = 0;
1031 int unlock_super = 0;
1032 struct ocfs2_super *osb = OCFS2_SB(sb);
1033
1034 mlog_entry_void();
1035
1036 if (ocfs2_is_hard_readonly(osb))
1037 goto leave;
1038
1039 status = ocfs2_fill_local_node_info(osb);
1040 if (status < 0) {
1041 mlog_errno(status);
1042 goto leave;
1043 }
1044
1045 status = ocfs2_register_hb_callbacks(osb);
1046 if (status < 0) {
1047 mlog_errno(status);
1048 goto leave;
1049 }
1050
1051 status = ocfs2_dlm_init(osb);
1052 if (status < 0) {
1053 mlog_errno(status);
1054 goto leave;
1055 }
1056
1057 /* requires vote_thread to be running. */
1058 status = ocfs2_register_net_handlers(osb);
1059 if (status < 0) {
1060 mlog_errno(status);
1061 goto leave;
1062 }
1063
1064 status = ocfs2_super_lock(osb, 1);
1065 if (status < 0) {
1066 mlog_errno(status);
1067 goto leave;
1068 }
1069 unlock_super = 1;
1070
1071 /* This will load up the node map and add ourselves to it. */
1072 status = ocfs2_find_slot(osb);
1073 if (status < 0) {
1074 mlog_errno(status);
1075 goto leave;
1076 }
1077
1078 ocfs2_populate_mounted_map(osb);
1079
1080 /* load all node-local system inodes */
1081 status = ocfs2_init_local_system_inodes(osb);
1082 if (status < 0) {
1083 mlog_errno(status);
1084 goto leave;
1085 }
1086
1087 status = ocfs2_check_volume(osb);
1088 if (status < 0) {
1089 mlog_errno(status);
1090 goto leave;
1091 }
1092
1093 status = ocfs2_truncate_log_init(osb);
1094 if (status < 0) {
1095 mlog_errno(status);
1096 goto leave;
1097 }
1098
1099 /* This should be sent *after* we recovered our journal as it
1100 * will cause other nodes to unmark us as needing
1101 * recovery. However, we need to send it *before* dropping the
1102 * super block lock as otherwise their recovery threads might
1103 * try to clean us up while we're live! */
1104 status = ocfs2_request_mount_vote(osb);
1105 if (status < 0)
1106 mlog_errno(status);
1107
1108leave:
1109 if (unlock_super)
1110 ocfs2_super_unlock(osb, 1);
1111
1112 mlog_exit(status);
1113 return status;
1114}
1115
1116/* we can't grab the goofy sem lock from inside wait_event, so we use
1117 * memory barriers to make sure that we'll see the null task before
1118 * being woken up */
1119static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1120{
1121 mb();
1122 return osb->recovery_thread_task != NULL;
1123}
1124
1125static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1126{
1127 int tmp;
1128 struct ocfs2_super *osb = NULL;
1129
1130 mlog_entry("(0x%p)\n", sb);
1131
1132 BUG_ON(!sb);
1133 osb = OCFS2_SB(sb);
1134 BUG_ON(!osb);
1135
1136 ocfs2_shutdown_local_alloc(osb);
1137
1138 ocfs2_truncate_log_shutdown(osb);
1139
1140 /* disable any new recovery threads and wait for any currently
1141 * running ones to exit. Do this before setting the vol_state. */
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001142 mutex_lock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001143 osb->disable_recovery = 1;
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001144 mutex_unlock(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001145 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1146
1147 /* At this point, we know that no more recovery threads can be
1148 * launched, so wait for any recovery completion work to
1149 * complete. */
1150 flush_workqueue(ocfs2_wq);
1151
1152 ocfs2_journal_shutdown(osb);
1153
1154 ocfs2_sync_blockdev(sb);
1155
1156 /* No dlm means we've failed during mount, so skip all the
1157 * steps which depended on that to complete. */
1158 if (osb->dlm) {
1159 tmp = ocfs2_super_lock(osb, 1);
1160 if (tmp < 0) {
1161 mlog_errno(tmp);
1162 return;
1163 }
1164
1165 tmp = ocfs2_request_umount_vote(osb);
1166 if (tmp < 0)
1167 mlog_errno(tmp);
1168
1169 if (osb->slot_num != OCFS2_INVALID_SLOT)
1170 ocfs2_put_slot(osb);
1171
1172 ocfs2_super_unlock(osb, 1);
1173 }
1174
1175 ocfs2_release_system_inodes(osb);
1176
1177 if (osb->dlm) {
1178 ocfs2_unregister_net_handlers(osb);
1179
1180 ocfs2_dlm_shutdown(osb);
1181 }
1182
1183 ocfs2_clear_hb_callbacks(osb);
1184
1185 debugfs_remove(osb->osb_debug_root);
1186
1187 if (!mnt_err)
1188 ocfs2_stop_heartbeat(osb);
1189
1190 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1191
1192 printk("ocfs2: Unmounting device (%u,%u) on (node %d)\n",
1193 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev), osb->node_num);
1194
1195 ocfs2_delete_osb(osb);
1196 kfree(osb);
1197 sb->s_dev = 0;
1198 sb->s_fs_info = NULL;
1199}
1200
1201static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1202 unsigned uuid_bytes)
1203{
1204 int i, ret;
1205 char *ptr;
1206
1207 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1208
1209 osb->uuid_str = kcalloc(1, OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1210 if (osb->uuid_str == NULL)
1211 return -ENOMEM;
1212
1213 memcpy(osb->uuid, uuid, OCFS2_VOL_UUID_LEN);
1214
1215 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1216 /* print with null */
1217 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1218 if (ret != 2) /* drop super cleans up */
1219 return -EINVAL;
1220 /* then only advance past the last char */
1221 ptr += 2;
1222 }
1223
1224 return 0;
1225}
1226
1227static int ocfs2_initialize_super(struct super_block *sb,
1228 struct buffer_head *bh,
1229 int sector_size)
1230{
1231 int status = 0;
1232 int i;
1233 struct ocfs2_dinode *di = NULL;
1234 struct inode *inode = NULL;
1235 struct buffer_head *bitmap_bh = NULL;
1236 struct ocfs2_journal *journal;
1237 __le32 uuid_net_key;
1238 struct ocfs2_super *osb;
1239
1240 mlog_entry_void();
1241
1242 osb = kcalloc(1, sizeof(struct ocfs2_super), GFP_KERNEL);
1243 if (!osb) {
1244 status = -ENOMEM;
1245 mlog_errno(status);
1246 goto bail;
1247 }
1248
1249 sb->s_fs_info = osb;
1250 sb->s_op = &ocfs2_sops;
1251 sb->s_export_op = &ocfs2_export_ops;
1252 sb->s_flags |= MS_NOATIME;
1253 /* this is needed to support O_LARGEFILE */
1254 sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1255
1256 osb->sb = sb;
1257 /* Save off for ocfs2_rw_direct */
1258 osb->s_sectsize_bits = blksize_bits(sector_size);
Eric Sesterhenn / snakebyteebdec832006-01-27 10:32:52 +01001259 BUG_ON(!osb->s_sectsize_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -08001260
1261 osb->net_response_ids = 0;
1262 spin_lock_init(&osb->net_response_lock);
1263 INIT_LIST_HEAD(&osb->net_response_list);
1264
1265 INIT_LIST_HEAD(&osb->osb_net_handlers);
1266 init_waitqueue_head(&osb->recovery_event);
1267 spin_lock_init(&osb->vote_task_lock);
1268 init_waitqueue_head(&osb->vote_event);
1269 osb->vote_work_sequence = 0;
1270 osb->vote_wake_sequence = 0;
1271 INIT_LIST_HEAD(&osb->blocked_lock_list);
1272 osb->blocked_lock_count = 0;
1273 INIT_LIST_HEAD(&osb->vote_list);
1274 spin_lock_init(&osb->osb_lock);
1275
1276 atomic_set(&osb->alloc_stats.moves, 0);
1277 atomic_set(&osb->alloc_stats.local_data, 0);
1278 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1279 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1280 atomic_set(&osb->alloc_stats.bg_extends, 0);
1281
1282 ocfs2_init_node_maps(osb);
1283
1284 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1285 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1286
Arjan van de Venc74ec2f2006-01-13 21:54:23 -08001287 mutex_init(&osb->recovery_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001288
1289 osb->disable_recovery = 0;
1290 osb->recovery_thread_task = NULL;
1291
1292 init_waitqueue_head(&osb->checkpoint_event);
1293 atomic_set(&osb->needs_checkpoint, 0);
1294
1295 osb->node_num = O2NM_INVALID_NODE_NUM;
1296 osb->slot_num = OCFS2_INVALID_SLOT;
1297
1298 osb->local_alloc_state = OCFS2_LA_UNUSED;
1299 osb->local_alloc_bh = NULL;
1300
1301 ocfs2_setup_hb_callbacks(osb);
1302
1303 init_waitqueue_head(&osb->osb_mount_event);
1304
1305 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1306 if (!osb->vol_label) {
1307 mlog(ML_ERROR, "unable to alloc vol label\n");
1308 status = -ENOMEM;
1309 goto bail;
1310 }
1311
1312 osb->uuid = kmalloc(OCFS2_VOL_UUID_LEN, GFP_KERNEL);
1313 if (!osb->uuid) {
1314 mlog(ML_ERROR, "unable to alloc uuid\n");
1315 status = -ENOMEM;
1316 goto bail;
1317 }
1318
1319 di = (struct ocfs2_dinode *)bh->b_data;
1320
1321 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1322 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1323 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1324 osb->max_slots);
1325 status = -EINVAL;
1326 goto bail;
1327 }
1328 mlog(ML_NOTICE, "max_slots for this device: %u\n", osb->max_slots);
1329
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001330 init_waitqueue_head(&osb->osb_wipe_event);
1331 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1332 sizeof(*osb->osb_orphan_wipes),
1333 GFP_KERNEL);
1334 if (!osb->osb_orphan_wipes) {
1335 status = -ENOMEM;
1336 mlog_errno(status);
1337 goto bail;
1338 }
1339
Mark Fashehccd979b2005-12-15 14:31:24 -08001340 osb->s_feature_compat =
1341 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1342 osb->s_feature_ro_compat =
1343 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1344 osb->s_feature_incompat =
1345 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1346
1347 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1348 mlog(ML_ERROR, "couldn't mount because of unsupported "
1349 "optional features (%x).\n", i);
1350 status = -EINVAL;
1351 goto bail;
1352 }
1353 if (!(osb->sb->s_flags & MS_RDONLY) &&
1354 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1355 mlog(ML_ERROR, "couldn't mount RDWR because of "
1356 "unsupported optional features (%x).\n", i);
1357 status = -EINVAL;
1358 goto bail;
1359 }
1360
1361 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1362
1363 /* FIXME
1364 * This should be done in ocfs2_journal_init(), but unknown
1365 * ordering issues will cause the filesystem to crash.
1366 * If anyone wants to figure out what part of the code
1367 * refers to osb->journal before ocfs2_journal_init() is run,
1368 * be my guest.
1369 */
1370 /* initialize our journal structure */
1371
1372 journal = kcalloc(1, sizeof(struct ocfs2_journal), GFP_KERNEL);
1373 if (!journal) {
1374 mlog(ML_ERROR, "unable to alloc journal\n");
1375 status = -ENOMEM;
1376 goto bail;
1377 }
1378 osb->journal = journal;
1379 journal->j_osb = osb;
1380
1381 atomic_set(&journal->j_num_trans, 0);
1382 init_rwsem(&journal->j_trans_barrier);
1383 init_waitqueue_head(&journal->j_checkpointed);
1384 spin_lock_init(&journal->j_lock);
1385 journal->j_trans_id = (unsigned long) 1;
1386 INIT_LIST_HEAD(&journal->j_la_cleanups);
1387 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery, osb);
1388 journal->j_state = OCFS2_JOURNAL_FREE;
1389
1390 /* get some pseudo constants for clustersize bits */
1391 osb->s_clustersize_bits =
1392 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1393 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1394 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1395
1396 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1397 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1398 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1399 osb->s_clustersize);
1400 status = -EINVAL;
1401 goto bail;
1402 }
1403
1404 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1405 > (u32)~0UL) {
1406 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1407 "what jbd can address in 32 bits.\n");
1408 status = -EINVAL;
1409 goto bail;
1410 }
1411
1412 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1413 sizeof(di->id2.i_super.s_uuid))) {
1414 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1415 status = -ENOMEM;
1416 goto bail;
1417 }
1418
1419 memcpy(&uuid_net_key, &osb->uuid[i], sizeof(osb->net_key));
1420 osb->net_key = le32_to_cpu(uuid_net_key);
1421
1422 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1423 osb->vol_label[63] = '\0';
1424 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1425 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1426 osb->first_cluster_group_blkno =
1427 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1428 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1429 mlog(0, "vol_label: %s\n", osb->vol_label);
1430 mlog(0, "uuid: %s\n", osb->uuid_str);
Mark Fashehb0697052006-03-03 10:24:33 -08001431 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1432 (unsigned long long)osb->root_blkno,
1433 (unsigned long long)osb->system_dir_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001434
1435 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1436 if (!osb->osb_dlm_debug) {
1437 status = -ENOMEM;
1438 mlog_errno(status);
1439 goto bail;
1440 }
1441
1442 atomic_set(&osb->vol_state, VOLUME_INIT);
1443
1444 /* load root, system_dir, and all global system inodes */
1445 status = ocfs2_init_global_system_inodes(osb);
1446 if (status < 0) {
1447 mlog_errno(status);
1448 goto bail;
1449 }
1450
1451 /*
1452 * global bitmap
1453 */
1454 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1455 OCFS2_INVALID_SLOT);
1456 if (!inode) {
1457 status = -EINVAL;
1458 mlog_errno(status);
1459 goto bail;
1460 }
1461
1462 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1463
1464 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1465 inode);
1466 iput(inode);
1467 if (status < 0) {
1468 mlog_errno(status);
1469 goto bail;
1470 }
1471
1472 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1473 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1474 osb->num_clusters = le32_to_cpu(di->id1.bitmap1.i_total);
1475 brelse(bitmap_bh);
Mark Fashehb0697052006-03-03 10:24:33 -08001476 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1477 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
Mark Fashehccd979b2005-12-15 14:31:24 -08001478
1479 status = ocfs2_init_slot_info(osb);
1480 if (status < 0) {
1481 mlog_errno(status);
1482 goto bail;
1483 }
1484
1485 /* Link this osb onto the global linked list of all osb structures. */
1486 /* The Global Link List is mainted for the whole driver . */
1487 spin_lock(&ocfs2_globals_lock);
1488 osb->osb_id = osb_id;
1489 if (osb_id < OCFS2_MAX_OSB_ID)
1490 osb_id++;
1491 else {
1492 mlog(ML_ERROR, "Too many volumes mounted\n");
1493 status = -ENOMEM;
1494 }
1495 spin_unlock(&ocfs2_globals_lock);
1496
1497bail:
1498 mlog_exit(status);
1499 return status;
1500}
1501
1502/*
1503 * will return: -EAGAIN if it is ok to keep searching for superblocks
1504 * -EINVAL if there is a bad superblock
1505 * 0 on success
1506 */
1507static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1508 struct buffer_head *bh,
1509 u32 blksz)
1510{
1511 int status = -EAGAIN;
1512
1513 mlog_entry_void();
1514
1515 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1516 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1517 status = -EINVAL;
1518 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1519 mlog(ML_ERROR, "found superblock with incorrect block "
1520 "size: found %u, should be %u\n",
1521 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1522 blksz);
1523 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1524 OCFS2_MAJOR_REV_LEVEL ||
1525 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1526 OCFS2_MINOR_REV_LEVEL) {
1527 mlog(ML_ERROR, "found superblock with bad version: "
1528 "found %u.%u, should be %u.%u\n",
1529 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1530 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1531 OCFS2_MAJOR_REV_LEVEL,
1532 OCFS2_MINOR_REV_LEVEL);
1533 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1534 mlog(ML_ERROR, "bad block number on superblock: "
Mark Fashehb0697052006-03-03 10:24:33 -08001535 "found %llu, should be %llu\n",
1536 (unsigned long long)di->i_blkno,
1537 (unsigned long long)bh->b_blocknr);
Mark Fashehccd979b2005-12-15 14:31:24 -08001538 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1539 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1540 mlog(ML_ERROR, "bad cluster size found: %u\n",
1541 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1542 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1543 mlog(ML_ERROR, "bad root_blkno: 0\n");
1544 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1545 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1546 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1547 mlog(ML_ERROR,
1548 "Superblock slots found greater than file system "
1549 "maximum: found %u, max %u\n",
1550 le16_to_cpu(di->id2.i_super.s_max_slots),
1551 OCFS2_MAX_SLOTS);
1552 } else {
1553 /* found it! */
1554 status = 0;
1555 }
1556 }
1557
1558 mlog_exit(status);
1559 return status;
1560}
1561
1562static int ocfs2_check_volume(struct ocfs2_super *osb)
1563{
1564 int status = 0;
1565 int dirty;
1566 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1567 * recover
1568 * ourselves. */
1569
1570 mlog_entry_void();
1571
1572 /* Init our journal object. */
1573 status = ocfs2_journal_init(osb->journal, &dirty);
1574 if (status < 0) {
1575 mlog(ML_ERROR, "Could not initialize journal!\n");
1576 goto finally;
1577 }
1578
1579 /* If the journal was unmounted cleanly then we don't want to
1580 * recover anything. Otherwise, journal_load will do that
1581 * dirty work for us :) */
1582 if (!dirty) {
1583 status = ocfs2_journal_wipe(osb->journal, 0);
1584 if (status < 0) {
1585 mlog_errno(status);
1586 goto finally;
1587 }
1588 } else {
1589 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1590 "recovering volume.\n");
1591 }
1592
1593 /* will play back anything left in the journal. */
1594 ocfs2_journal_load(osb->journal);
1595
1596 if (dirty) {
1597 /* recover my local alloc if we didn't unmount cleanly. */
1598 status = ocfs2_begin_local_alloc_recovery(osb,
1599 osb->slot_num,
1600 &local_alloc);
1601 if (status < 0) {
1602 mlog_errno(status);
1603 goto finally;
1604 }
1605 /* we complete the recovery process after we've marked
1606 * ourselves as mounted. */
1607 }
1608
1609 mlog(0, "Journal loaded.\n");
1610
1611 status = ocfs2_load_local_alloc(osb);
1612 if (status < 0) {
1613 mlog_errno(status);
1614 goto finally;
1615 }
1616
1617 if (dirty) {
1618 /* Recovery will be completed after we've mounted the
1619 * rest of the volume. */
1620 osb->dirty = 1;
1621 osb->local_alloc_copy = local_alloc;
1622 local_alloc = NULL;
1623 }
1624
1625 /* go through each journal, trylock it and if you get the
1626 * lock, and it's marked as dirty, set the bit in the recover
1627 * map and launch a recovery thread for it. */
1628 status = ocfs2_mark_dead_nodes(osb);
1629 if (status < 0)
1630 mlog_errno(status);
1631
1632finally:
1633 if (local_alloc)
1634 kfree(local_alloc);
1635
1636 mlog_exit(status);
1637 return status;
1638}
1639
1640/*
1641 * The routine gets called from dismount or close whenever a dismount on
1642 * volume is requested and the osb open count becomes 1.
1643 * It will remove the osb from the global list and also free up all the
1644 * initialized resources and fileobject.
1645 */
1646static void ocfs2_delete_osb(struct ocfs2_super *osb)
1647{
1648 mlog_entry_void();
1649
1650 /* This function assumes that the caller has the main osb resource */
1651
1652 if (osb->slot_info)
1653 ocfs2_free_slot_info(osb->slot_info);
1654
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001655 kfree(osb->osb_orphan_wipes);
Mark Fashehccd979b2005-12-15 14:31:24 -08001656 /* FIXME
1657 * This belongs in journal shutdown, but because we have to
1658 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1659 * we free it here.
1660 */
1661 kfree(osb->journal);
1662 if (osb->local_alloc_copy)
1663 kfree(osb->local_alloc_copy);
1664 kfree(osb->uuid_str);
1665 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1666 memset(osb, 0, sizeof(struct ocfs2_super));
1667
1668 mlog_exit_void();
1669}
1670
1671/* Put OCFS2 into a readonly state, or (if the user specifies it),
1672 * panic(). We do not support continue-on-error operation. */
1673static void ocfs2_handle_error(struct super_block *sb)
1674{
1675 struct ocfs2_super *osb = OCFS2_SB(sb);
1676
1677 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1678 panic("OCFS2: (device %s): panic forced after error\n",
1679 sb->s_id);
1680
1681 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1682
1683 if (sb->s_flags & MS_RDONLY &&
1684 (ocfs2_is_soft_readonly(osb) ||
1685 ocfs2_is_hard_readonly(osb)))
1686 return;
1687
1688 printk(KERN_CRIT "File system is now read-only due to the potential "
1689 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1690 "system is unmounted.\n");
1691 sb->s_flags |= MS_RDONLY;
1692 ocfs2_set_ro_flag(osb, 0);
1693}
1694
1695static char error_buf[1024];
1696
1697void __ocfs2_error(struct super_block *sb,
1698 const char *function,
1699 const char *fmt, ...)
1700{
1701 va_list args;
1702
1703 va_start(args, fmt);
1704 vsprintf(error_buf, fmt, args);
1705 va_end(args);
1706
1707 /* Not using mlog here because we want to show the actual
1708 * function the error came from. */
1709 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1710 sb->s_id, function, error_buf);
1711
1712 ocfs2_handle_error(sb);
1713}
1714
1715/* Handle critical errors. This is intentionally more drastic than
1716 * ocfs2_handle_error, so we only use for things like journal errors,
1717 * etc. */
1718void __ocfs2_abort(struct super_block* sb,
1719 const char *function,
1720 const char *fmt, ...)
1721{
1722 va_list args;
1723
1724 va_start(args, fmt);
1725 vsprintf(error_buf, fmt, args);
1726 va_end(args);
1727
1728 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1729 sb->s_id, function, error_buf);
1730
1731 /* We don't have the cluster support yet to go straight to
1732 * hard readonly in here. Until then, we want to keep
1733 * ocfs2_abort() so that we can at least mark critical
1734 * errors.
1735 *
1736 * TODO: This should abort the journal and alert other nodes
1737 * that our slot needs recovery. */
1738
1739 /* Force a panic(). This stinks, but it's better than letting
1740 * things continue without having a proper hard readonly
1741 * here. */
1742 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1743 ocfs2_handle_error(sb);
1744}
1745
1746module_init(ocfs2_init);
1747module_exit(ocfs2_exit);