projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ocfs2: Clear slot map when umounting a local volume
[deliverable/linux.git] / fs / ocfs2 / super.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * super.c
5 *
6 * load/unload driver, mount/dismount volumes
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 */
25
26 #include <linux/module.h>
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
42
43 #include <cluster/nodemanager.h>
44
45 #define MLOG_MASK_PREFIX ML_SUPER
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 /* this should be the only file to include a version 1 header */
51 #include "ocfs1_fs_compat.h"
52
53 #include "alloc.h"
54 #include "dlmglue.h"
55 #include "export.h"
56 #include "extent_map.h"
57 #include "heartbeat.h"
58 #include "inode.h"
59 #include "journal.h"
60 #include "localalloc.h"
61 #include "namei.h"
62 #include "slot_map.h"
63 #include "super.h"
64 #include "sysfile.h"
65 #include "uptodate.h"
66 #include "ver.h"
67 #include "vote.h"
68
69 #include "buffer_head_io.h"
70
71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
72
73 /* OCFS2 needs to schedule several differnt types of work which
74 * require cluster locking, disk I/O, recovery waits, etc. Since these
75 * types of work tend to be heavy we avoid using the kernel events
76 * workqueue and schedule on our own. */
77 struct workqueue_struct *ocfs2_wq = NULL;
78
79 static struct dentry *ocfs2_debugfs_root = NULL;
80
81 MODULE_AUTHOR("Oracle");
82 MODULE_LICENSE("GPL");
83
84 struct mount_options
85 {
86 unsigned long mount_opt;
87 unsigned int atime_quantum;
88 signed short slot;
89 };
90
91 static int ocfs2_parse_options(struct super_block *sb, char *options,
92 struct mount_options *mopt,
93 int is_remount);
94 static void ocfs2_put_super(struct super_block *sb);
95 static int ocfs2_mount_volume(struct super_block *sb);
96 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
97 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
98 static int ocfs2_initialize_mem_caches(void);
99 static void ocfs2_free_mem_caches(void);
100 static void ocfs2_delete_osb(struct ocfs2_super *osb);
101
102 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
103
104 static int ocfs2_sync_fs(struct super_block *sb, int wait);
105
106 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
107 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
108 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
109 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
110 static int ocfs2_check_volume(struct ocfs2_super *osb);
111 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
112 struct buffer_head *bh,
113 u32 sectsize);
114 static int ocfs2_initialize_super(struct super_block *sb,
115 struct buffer_head *bh,
116 int sector_size);
117 static int ocfs2_get_sector(struct super_block *sb,
118 struct buffer_head **bh,
119 int block,
120 int sect_size);
121 static void ocfs2_write_super(struct super_block *sb);
122 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
123 static void ocfs2_destroy_inode(struct inode *inode);
124
125 static const struct super_operations ocfs2_sops = {
126 .statfs = ocfs2_statfs,
127 .alloc_inode = ocfs2_alloc_inode,
128 .destroy_inode = ocfs2_destroy_inode,
129 .drop_inode = ocfs2_drop_inode,
130 .clear_inode = ocfs2_clear_inode,
131 .delete_inode = ocfs2_delete_inode,
132 .sync_fs = ocfs2_sync_fs,
133 .write_super = ocfs2_write_super,
134 .put_super = ocfs2_put_super,
135 .remount_fs = ocfs2_remount,
136 };
137
138 enum {
139 Opt_barrier,
140 Opt_err_panic,
141 Opt_err_ro,
142 Opt_intr,
143 Opt_nointr,
144 Opt_hb_none,
145 Opt_hb_local,
146 Opt_data_ordered,
147 Opt_data_writeback,
148 Opt_atime_quantum,
149 Opt_slot,
150 Opt_err,
151 };
152
153 static match_table_t tokens = {
154 {Opt_barrier, "barrier=%u"},
155 {Opt_err_panic, "errors=panic"},
156 {Opt_err_ro, "errors=remount-ro"},
157 {Opt_intr, "intr"},
158 {Opt_nointr, "nointr"},
159 {Opt_hb_none, OCFS2_HB_NONE},
160 {Opt_hb_local, OCFS2_HB_LOCAL},
161 {Opt_data_ordered, "data=ordered"},
162 {Opt_data_writeback, "data=writeback"},
163 {Opt_atime_quantum, "atime_quantum=%u"},
164 {Opt_slot, "preferred_slot=%u"},
165 {Opt_err, NULL}
166 };
167
168 /*
169 * write_super and sync_fs ripped right out of ext3.
170 */
171 static void ocfs2_write_super(struct super_block *sb)
172 {
173 if (mutex_trylock(&sb->s_lock) != 0)
174 BUG();
175 sb->s_dirt = 0;
176 }
177
178 static int ocfs2_sync_fs(struct super_block *sb, int wait)
179 {
180 int status = 0;
181 tid_t target;
182 struct ocfs2_super *osb = OCFS2_SB(sb);
183
184 sb->s_dirt = 0;
185
186 if (ocfs2_is_hard_readonly(osb))
187 return -EROFS;
188
189 if (wait) {
190 status = ocfs2_flush_truncate_log(osb);
191 if (status < 0)
192 mlog_errno(status);
193 } else {
194 ocfs2_schedule_truncate_log_flush(osb, 0);
195 }
196
197 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
198 if (wait)
199 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
200 target);
201 }
202 return 0;
203 }
204
205 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
206 {
207 struct inode *new = NULL;
208 int status = 0;
209 int i;
210
211 mlog_entry_void();
212
213 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
214 if (IS_ERR(new)) {
215 status = PTR_ERR(new);
216 mlog_errno(status);
217 goto bail;
218 }
219 osb->root_inode = new;
220
221 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
222 if (IS_ERR(new)) {
223 status = PTR_ERR(new);
224 mlog_errno(status);
225 goto bail;
226 }
227 osb->sys_root_inode = new;
228
229 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
230 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
231 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
232 if (!new) {
233 ocfs2_release_system_inodes(osb);
234 status = -EINVAL;
235 mlog_errno(status);
236 /* FIXME: Should ERROR_RO_FS */
237 mlog(ML_ERROR, "Unable to load system inode %d, "
238 "possibly corrupt fs?", i);
239 goto bail;
240 }
241 // the array now has one ref, so drop this one
242 iput(new);
243 }
244
245 bail:
246 mlog_exit(status);
247 return status;
248 }
249
250 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
251 {
252 struct inode *new = NULL;
253 int status = 0;
254 int i;
255
256 mlog_entry_void();
257
258 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
259 i < NUM_SYSTEM_INODES;
260 i++) {
261 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
262 if (!new) {
263 ocfs2_release_system_inodes(osb);
264 status = -EINVAL;
265 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
266 status, i, osb->slot_num);
267 goto bail;
268 }
269 /* the array now has one ref, so drop this one */
270 iput(new);
271 }
272
273 bail:
274 mlog_exit(status);
275 return status;
276 }
277
278 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
279 {
280 int status = 0, i;
281 struct inode *inode;
282
283 mlog_entry_void();
284
285 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
286 inode = osb->system_inodes[i];
287 if (inode) {
288 iput(inode);
289 osb->system_inodes[i] = NULL;
290 }
291 }
292
293 inode = osb->sys_root_inode;
294 if (inode) {
295 iput(inode);
296 osb->sys_root_inode = NULL;
297 }
298
299 inode = osb->root_inode;
300 if (inode) {
301 iput(inode);
302 osb->root_inode = NULL;
303 }
304
305 mlog_exit(status);
306 return status;
307 }
308
309 /* We're allocating fs objects, use GFP_NOFS */
310 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
311 {
312 struct ocfs2_inode_info *oi;
313
314 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
315 if (!oi)
316 return NULL;
317
318 return &oi->vfs_inode;
319 }
320
321 static void ocfs2_destroy_inode(struct inode *inode)
322 {
323 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
324 }
325
326 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
327 unsigned int cbits)
328 {
329 unsigned int bytes = 1 << cbits;
330 unsigned int trim = bytes;
331 unsigned int bitshift = 32;
332
333 /*
334 * i_size and all block offsets in ocfs2 are always 64 bits
335 * wide. i_clusters is 32 bits, in cluster-sized units. So on
336 * 64 bit platforms, cluster size will be the limiting factor.
337 */
338
339 #if BITS_PER_LONG == 32
340 # if defined(CONFIG_LBD)
341 BUILD_BUG_ON(sizeof(sector_t) != 8);
342 /*
343 * We might be limited by page cache size.
344 */
345 if (bytes > PAGE_CACHE_SIZE) {
346 bytes = PAGE_CACHE_SIZE;
347 trim = 1;
348 /*
349 * Shift by 31 here so that we don't get larger than
350 * MAX_LFS_FILESIZE
351 */
352 bitshift = 31;
353 }
354 # else
355 /*
356 * We are limited by the size of sector_t. Use block size, as
357 * that's what we expose to the VFS.
358 */
359 bytes = 1 << bbits;
360 trim = 1;
361 bitshift = 31;
362 # endif
363 #endif
364
365 /*
366 * Trim by a whole cluster when we can actually approach the
367 * on-disk limits. Otherwise we can overflow i_clusters when
368 * an extent start is at the max offset.
369 */
370 return (((unsigned long long)bytes) << bitshift) - trim;
371 }
372
373 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
374 {
375 int incompat_features;
376 int ret = 0;
377 struct mount_options parsed_options;
378 struct ocfs2_super *osb = OCFS2_SB(sb);
379
380 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
381 ret = -EINVAL;
382 goto out;
383 }
384
385 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
386 (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
387 ret = -EINVAL;
388 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
389 goto out;
390 }
391
392 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
393 (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
394 ret = -EINVAL;
395 mlog(ML_ERROR, "Cannot change data mode on remount\n");
396 goto out;
397 }
398
399 /* We're going to/from readonly mode. */
400 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
401 /* Lock here so the check of HARD_RO and the potential
402 * setting of SOFT_RO is atomic. */
403 spin_lock(&osb->osb_lock);
404 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
405 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
406 ret = -EROFS;
407 goto unlock_osb;
408 }
409
410 if (*flags & MS_RDONLY) {
411 mlog(0, "Going to ro mode.\n");
412 sb->s_flags |= MS_RDONLY;
413 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
414 } else {
415 mlog(0, "Making ro filesystem writeable.\n");
416
417 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
418 mlog(ML_ERROR, "Cannot remount RDWR "
419 "filesystem due to previous errors.\n");
420 ret = -EROFS;
421 goto unlock_osb;
422 }
423 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
424 if (incompat_features) {
425 mlog(ML_ERROR, "Cannot remount RDWR because "
426 "of unsupported optional features "
427 "(%x).\n", incompat_features);
428 ret = -EINVAL;
429 goto unlock_osb;
430 }
431 sb->s_flags &= ~MS_RDONLY;
432 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
433 }
434 unlock_osb:
435 spin_unlock(&osb->osb_lock);
436 }
437
438 if (!ret) {
439 if (!ocfs2_is_hard_readonly(osb))
440 ocfs2_set_journal_params(osb);
441
442 /* Only save off the new mount options in case of a successful
443 * remount. */
444 osb->s_mount_opt = parsed_options.mount_opt;
445 osb->s_atime_quantum = parsed_options.atime_quantum;
446 osb->preferred_slot = parsed_options.slot;
447 }
448 out:
449 return ret;
450 }
451
452 static int ocfs2_sb_probe(struct super_block *sb,
453 struct buffer_head **bh,
454 int *sector_size)
455 {
456 int status = 0, tmpstat;
457 struct ocfs1_vol_disk_hdr *hdr;
458 struct ocfs2_dinode *di;
459 int blksize;
460
461 *bh = NULL;
462
463 /* may be > 512 */
464 *sector_size = bdev_hardsect_size(sb->s_bdev);
465 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
466 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
467 *sector_size, OCFS2_MAX_BLOCKSIZE);
468 status = -EINVAL;
469 goto bail;
470 }
471
472 /* Can this really happen? */
473 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
474 *sector_size = OCFS2_MIN_BLOCKSIZE;
475
476 /* check block zero for old format */
477 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
478 if (status < 0) {
479 mlog_errno(status);
480 goto bail;
481 }
482 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
483 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
484 mlog(ML_ERROR, "incompatible version: %u.%u\n",
485 hdr->major_version, hdr->minor_version);
486 status = -EINVAL;
487 }
488 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
489 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
490 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
491 hdr->signature);
492 status = -EINVAL;
493 }
494 brelse(*bh);
495 *bh = NULL;
496 if (status < 0) {
497 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
498 "upgraded before mounting with ocfs v2\n");
499 goto bail;
500 }
501
502 /*
503 * Now check at magic offset for 512, 1024, 2048, 4096
504 * blocksizes. 4096 is the maximum blocksize because it is
505 * the minimum clustersize.
506 */
507 status = -EINVAL;
508 for (blksize = *sector_size;
509 blksize <= OCFS2_MAX_BLOCKSIZE;
510 blksize <<= 1) {
511 tmpstat = ocfs2_get_sector(sb, bh,
512 OCFS2_SUPER_BLOCK_BLKNO,
513 blksize);
514 if (tmpstat < 0) {
515 status = tmpstat;
516 mlog_errno(status);
517 goto bail;
518 }
519 di = (struct ocfs2_dinode *) (*bh)->b_data;
520 status = ocfs2_verify_volume(di, *bh, blksize);
521 if (status >= 0)
522 goto bail;
523 brelse(*bh);
524 *bh = NULL;
525 if (status != -EAGAIN)
526 break;
527 }
528
529 bail:
530 return status;
531 }
532
533 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
534 {
535 if (ocfs2_mount_local(osb)) {
536 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
537 mlog(ML_ERROR, "Cannot heartbeat on a locally "
538 "mounted device.\n");
539 return -EINVAL;
540 }
541 }
542
543 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
544 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
545 mlog(ML_ERROR, "Heartbeat has to be started to mount "
546 "a read-write clustered device.\n");
547 return -EINVAL;
548 }
549 }
550
551 return 0;
552 }
553
554 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
555 {
556 struct dentry *root;
557 int status, sector_size;
558 struct mount_options parsed_options;
559 struct inode *inode = NULL;
560 struct ocfs2_super *osb = NULL;
561 struct buffer_head *bh = NULL;
562 char nodestr[8];
563
564 mlog_entry("%p, %p, %i", sb, data, silent);
565
566 if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
567 status = -EINVAL;
568 goto read_super_error;
569 }
570
571 /* for now we only have one cluster/node, make sure we see it
572 * in the heartbeat universe */
573 if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
574 if (!o2hb_check_local_node_heartbeating()) {
575 status = -EINVAL;
576 goto read_super_error;
577 }
578 }
579
580 /* probe for superblock */
581 status = ocfs2_sb_probe(sb, &bh, &sector_size);
582 if (status < 0) {
583 mlog(ML_ERROR, "superblock probe failed!\n");
584 goto read_super_error;
585 }
586
587 status = ocfs2_initialize_super(sb, bh, sector_size);
588 osb = OCFS2_SB(sb);
589 if (status < 0) {
590 mlog_errno(status);
591 goto read_super_error;
592 }
593 brelse(bh);
594 bh = NULL;
595 osb->s_mount_opt = parsed_options.mount_opt;
596 osb->s_atime_quantum = parsed_options.atime_quantum;
597 osb->preferred_slot = parsed_options.slot;
598
599 sb->s_magic = OCFS2_SUPER_MAGIC;
600
601 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
602 * heartbeat=none */
603 if (bdev_read_only(sb->s_bdev)) {
604 if (!(sb->s_flags & MS_RDONLY)) {
605 status = -EACCES;
606 mlog(ML_ERROR, "Readonly device detected but readonly "
607 "mount was not specified.\n");
608 goto read_super_error;
609 }
610
611 /* You should not be able to start a local heartbeat
612 * on a readonly device. */
613 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
614 status = -EROFS;
615 mlog(ML_ERROR, "Local heartbeat specified on readonly "
616 "device.\n");
617 goto read_super_error;
618 }
619
620 status = ocfs2_check_journals_nolocks(osb);
621 if (status < 0) {
622 if (status == -EROFS)
623 mlog(ML_ERROR, "Recovery required on readonly "
624 "file system, but write access is "
625 "unavailable.\n");
626 else
627 mlog_errno(status);
628 goto read_super_error;
629 }
630
631 ocfs2_set_ro_flag(osb, 1);
632
633 printk(KERN_NOTICE "Readonly device detected. No cluster "
634 "services will be utilized for this mount. Recovery "
635 "will be skipped.\n");
636 }
637
638 if (!ocfs2_is_hard_readonly(osb)) {
639 if (sb->s_flags & MS_RDONLY)
640 ocfs2_set_ro_flag(osb, 0);
641 }
642
643 status = ocfs2_verify_heartbeat(osb);
644 if (status < 0) {
645 mlog_errno(status);
646 goto read_super_error;
647 }
648
649 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
650 ocfs2_debugfs_root);
651 if (!osb->osb_debug_root) {
652 status = -EINVAL;
653 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
654 goto read_super_error;
655 }
656
657 status = ocfs2_mount_volume(sb);
658 if (osb->root_inode)
659 inode = igrab(osb->root_inode);
660
661 if (status < 0)
662 goto read_super_error;
663
664 if (!inode) {
665 status = -EIO;
666 mlog_errno(status);
667 goto read_super_error;
668 }
669
670 root = d_alloc_root(inode);
671 if (!root) {
672 status = -ENOMEM;
673 mlog_errno(status);
674 goto read_super_error;
675 }
676
677 sb->s_root = root;
678
679 ocfs2_complete_mount_recovery(osb);
680
681 if (ocfs2_mount_local(osb))
682 snprintf(nodestr, sizeof(nodestr), "local");
683 else
684 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
685
686 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
687 "with %s data mode.\n",
688 osb->dev_str, nodestr, osb->slot_num,
689 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
690 "ordered");
691
692 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
693 wake_up(&osb->osb_mount_event);
694
695 mlog_exit(status);
696 return status;
697
698 read_super_error:
699 if (bh != NULL)
700 brelse(bh);
701
702 if (inode)
703 iput(inode);
704
705 if (osb) {
706 atomic_set(&osb->vol_state, VOLUME_DISABLED);
707 wake_up(&osb->osb_mount_event);
708 ocfs2_dismount_volume(sb, 1);
709 }
710
711 mlog_exit(status);
712 return status;
713 }
714
715 static int ocfs2_get_sb(struct file_system_type *fs_type,
716 int flags,
717 const char *dev_name,
718 void *data,
719 struct vfsmount *mnt)
720 {
721 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
722 mnt);
723 }
724
725 static struct file_system_type ocfs2_fs_type = {
726 .owner = THIS_MODULE,
727 .name = "ocfs2",
728 .get_sb = ocfs2_get_sb, /* is this called when we mount
729 * the fs? */
730 .kill_sb = kill_block_super, /* set to the generic one
731 * right now, but do we
732 * need to change that? */
733 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
734 .next = NULL
735 };
736
737 static int ocfs2_parse_options(struct super_block *sb,
738 char *options,
739 struct mount_options *mopt,
740 int is_remount)
741 {
742 int status;
743 char *p;
744
745 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
746 options ? options : "(none)");
747
748 mopt->mount_opt = 0;
749 mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
750 mopt->slot = OCFS2_INVALID_SLOT;
751
752 if (!options) {
753 status = 1;
754 goto bail;
755 }
756
757 while ((p = strsep(&options, ",")) != NULL) {
758 int token, option;
759 substring_t args[MAX_OPT_ARGS];
760
761 if (!*p)
762 continue;
763
764 token = match_token(p, tokens, args);
765 switch (token) {
766 case Opt_hb_local:
767 mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
768 break;
769 case Opt_hb_none:
770 mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
771 break;
772 case Opt_barrier:
773 if (match_int(&args[0], &option)) {
774 status = 0;
775 goto bail;
776 }
777 if (option)
778 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
779 else
780 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
781 break;
782 case Opt_intr:
783 mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
784 break;
785 case Opt_nointr:
786 mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
787 break;
788 case Opt_err_panic:
789 mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
790 break;
791 case Opt_err_ro:
792 mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
793 break;
794 case Opt_data_ordered:
795 mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
796 break;
797 case Opt_data_writeback:
798 mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
799 break;
800 case Opt_atime_quantum:
801 if (match_int(&args[0], &option)) {
802 status = 0;
803 goto bail;
804 }
805 if (option >= 0)
806 mopt->atime_quantum = option;
807 break;
808 case Opt_slot:
809 option = 0;
810 if (match_int(&args[0], &option)) {
811 status = 0;
812 goto bail;
813 }
814 if (option)
815 mopt->slot = (s16)option;
816 break;
817 default:
818 mlog(ML_ERROR,
819 "Unrecognized mount option \"%s\" "
820 "or missing value\n", p);
821 status = 0;
822 goto bail;
823 }
824 }
825
826 status = 1;
827
828 bail:
829 mlog_exit(status);
830 return status;
831 }
832
833 static int __init ocfs2_init(void)
834 {
835 int status;
836
837 mlog_entry_void();
838
839 ocfs2_print_version();
840
841 status = init_ocfs2_uptodate_cache();
842 if (status < 0) {
843 mlog_errno(status);
844 goto leave;
845 }
846
847 status = ocfs2_initialize_mem_caches();
848 if (status < 0) {
849 mlog_errno(status);
850 goto leave;
851 }
852
853 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
854 if (!ocfs2_wq) {
855 status = -ENOMEM;
856 goto leave;
857 }
858
859 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
860 if (!ocfs2_debugfs_root) {
861 status = -EFAULT;
862 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
863 }
864
865 leave:
866 if (status < 0) {
867 ocfs2_free_mem_caches();
868 exit_ocfs2_uptodate_cache();
869 }
870
871 mlog_exit(status);
872
873 if (status >= 0) {
874 return register_filesystem(&ocfs2_fs_type);
875 } else
876 return -1;
877 }
878
879 static void __exit ocfs2_exit(void)
880 {
881 mlog_entry_void();
882
883 if (ocfs2_wq) {
884 flush_workqueue(ocfs2_wq);
885 destroy_workqueue(ocfs2_wq);
886 }
887
888 debugfs_remove(ocfs2_debugfs_root);
889
890 ocfs2_free_mem_caches();
891
892 unregister_filesystem(&ocfs2_fs_type);
893
894 exit_ocfs2_uptodate_cache();
895
896 mlog_exit_void();
897 }
898
899 static void ocfs2_put_super(struct super_block *sb)
900 {
901 mlog_entry("(0x%p)\n", sb);
902
903 ocfs2_sync_blockdev(sb);
904 ocfs2_dismount_volume(sb, 0);
905
906 mlog_exit_void();
907 }
908
909 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
910 {
911 struct ocfs2_super *osb;
912 u32 numbits, freebits;
913 int status;
914 struct ocfs2_dinode *bm_lock;
915 struct buffer_head *bh = NULL;
916 struct inode *inode = NULL;
917
918 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
919
920 osb = OCFS2_SB(dentry->d_sb);
921
922 inode = ocfs2_get_system_file_inode(osb,
923 GLOBAL_BITMAP_SYSTEM_INODE,
924 OCFS2_INVALID_SLOT);
925 if (!inode) {
926 mlog(ML_ERROR, "failed to get bitmap inode\n");
927 status = -EIO;
928 goto bail;
929 }
930
931 status = ocfs2_meta_lock(inode, &bh, 0);
932 if (status < 0) {
933 mlog_errno(status);
934 goto bail;
935 }
936
937 bm_lock = (struct ocfs2_dinode *) bh->b_data;
938
939 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
940 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
941
942 buf->f_type = OCFS2_SUPER_MAGIC;
943 buf->f_bsize = dentry->d_sb->s_blocksize;
944 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
945 buf->f_blocks = ((sector_t) numbits) *
946 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
947 buf->f_bfree = ((sector_t) freebits) *
948 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
949 buf->f_bavail = buf->f_bfree;
950 buf->f_files = numbits;
951 buf->f_ffree = freebits;
952
953 brelse(bh);
954
955 ocfs2_meta_unlock(inode, 0);
956 status = 0;
957 bail:
958 if (inode)
959 iput(inode);
960
961 mlog_exit(status);
962
963 return status;
964 }
965
966 static void ocfs2_inode_init_once(void *data,
967 struct kmem_cache *cachep,
968 unsigned long flags)
969 {
970 struct ocfs2_inode_info *oi = data;
971
972 oi->ip_flags = 0;
973 oi->ip_open_count = 0;
974 spin_lock_init(&oi->ip_lock);
975 ocfs2_extent_map_init(&oi->vfs_inode);
976 INIT_LIST_HEAD(&oi->ip_io_markers);
977 oi->ip_created_trans = 0;
978 oi->ip_last_trans = 0;
979 oi->ip_dir_start_lookup = 0;
980
981 init_rwsem(&oi->ip_alloc_sem);
982 mutex_init(&oi->ip_io_mutex);
983
984 oi->ip_blkno = 0ULL;
985 oi->ip_clusters = 0;
986
987 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
988 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
989 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
990 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
991
992 ocfs2_metadata_cache_init(&oi->vfs_inode);
993
994 inode_init_once(&oi->vfs_inode);
995 }
996
997 static int ocfs2_initialize_mem_caches(void)
998 {
999 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1000 sizeof(struct ocfs2_inode_info),
1001 0,
1002 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1003 SLAB_MEM_SPREAD),
1004 ocfs2_inode_init_once);
1005 if (!ocfs2_inode_cachep)
1006 return -ENOMEM;
1007
1008 return 0;
1009 }
1010
1011 static void ocfs2_free_mem_caches(void)
1012 {
1013 if (ocfs2_inode_cachep)
1014 kmem_cache_destroy(ocfs2_inode_cachep);
1015
1016 ocfs2_inode_cachep = NULL;
1017 }
1018
1019 static int ocfs2_get_sector(struct super_block *sb,
1020 struct buffer_head **bh,
1021 int block,
1022 int sect_size)
1023 {
1024 if (!sb_set_blocksize(sb, sect_size)) {
1025 mlog(ML_ERROR, "unable to set blocksize\n");
1026 return -EIO;
1027 }
1028
1029 *bh = sb_getblk(sb, block);
1030 if (!*bh) {
1031 mlog_errno(-EIO);
1032 return -EIO;
1033 }
1034 lock_buffer(*bh);
1035 if (!buffer_dirty(*bh))
1036 clear_buffer_uptodate(*bh);
1037 unlock_buffer(*bh);
1038 ll_rw_block(READ, 1, bh);
1039 wait_on_buffer(*bh);
1040 return 0;
1041 }
1042
1043 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1044 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1045 {
1046 int status;
1047
1048 /* XXX hold a ref on the node while mounte? easy enough, if
1049 * desirable. */
1050 if (ocfs2_mount_local(osb))
1051 osb->node_num = 0;
1052 else
1053 osb->node_num = o2nm_this_node();
1054
1055 if (osb->node_num == O2NM_MAX_NODES) {
1056 mlog(ML_ERROR, "could not find this host's node number\n");
1057 status = -ENOENT;
1058 goto bail;
1059 }
1060
1061 mlog(0, "I am node %d\n", osb->node_num);
1062
1063 status = 0;
1064 bail:
1065 return status;
1066 }
1067
1068 static int ocfs2_mount_volume(struct super_block *sb)
1069 {
1070 int status = 0;
1071 int unlock_super = 0;
1072 struct ocfs2_super *osb = OCFS2_SB(sb);
1073
1074 mlog_entry_void();
1075
1076 if (ocfs2_is_hard_readonly(osb))
1077 goto leave;
1078
1079 status = ocfs2_fill_local_node_info(osb);
1080 if (status < 0) {
1081 mlog_errno(status);
1082 goto leave;
1083 }
1084
1085 status = ocfs2_register_hb_callbacks(osb);
1086 if (status < 0) {
1087 mlog_errno(status);
1088 goto leave;
1089 }
1090
1091 status = ocfs2_dlm_init(osb);
1092 if (status < 0) {
1093 mlog_errno(status);
1094 goto leave;
1095 }
1096
1097 /* requires vote_thread to be running. */
1098 status = ocfs2_register_net_handlers(osb);
1099 if (status < 0) {
1100 mlog_errno(status);
1101 goto leave;
1102 }
1103
1104 status = ocfs2_super_lock(osb, 1);
1105 if (status < 0) {
1106 mlog_errno(status);
1107 goto leave;
1108 }
1109 unlock_super = 1;
1110
1111 /* This will load up the node map and add ourselves to it. */
1112 status = ocfs2_find_slot(osb);
1113 if (status < 0) {
1114 mlog_errno(status);
1115 goto leave;
1116 }
1117
1118 ocfs2_populate_mounted_map(osb);
1119
1120 /* load all node-local system inodes */
1121 status = ocfs2_init_local_system_inodes(osb);
1122 if (status < 0) {
1123 mlog_errno(status);
1124 goto leave;
1125 }
1126
1127 status = ocfs2_check_volume(osb);
1128 if (status < 0) {
1129 mlog_errno(status);
1130 goto leave;
1131 }
1132
1133 status = ocfs2_truncate_log_init(osb);
1134 if (status < 0) {
1135 mlog_errno(status);
1136 goto leave;
1137 }
1138
1139 if (ocfs2_mount_local(osb))
1140 goto leave;
1141
1142 /* This should be sent *after* we recovered our journal as it
1143 * will cause other nodes to unmark us as needing
1144 * recovery. However, we need to send it *before* dropping the
1145 * super block lock as otherwise their recovery threads might
1146 * try to clean us up while we're live! */
1147 status = ocfs2_request_mount_vote(osb);
1148 if (status < 0)
1149 mlog_errno(status);
1150
1151 leave:
1152 if (unlock_super)
1153 ocfs2_super_unlock(osb, 1);
1154
1155 mlog_exit(status);
1156 return status;
1157 }
1158
1159 /* we can't grab the goofy sem lock from inside wait_event, so we use
1160 * memory barriers to make sure that we'll see the null task before
1161 * being woken up */
1162 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1163 {
1164 mb();
1165 return osb->recovery_thread_task != NULL;
1166 }
1167
1168 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1169 {
1170 int tmp;
1171 struct ocfs2_super *osb = NULL;
1172 char nodestr[8];
1173
1174 mlog_entry("(0x%p)\n", sb);
1175
1176 BUG_ON(!sb);
1177 osb = OCFS2_SB(sb);
1178 BUG_ON(!osb);
1179
1180 ocfs2_shutdown_local_alloc(osb);
1181
1182 ocfs2_truncate_log_shutdown(osb);
1183
1184 /* disable any new recovery threads and wait for any currently
1185 * running ones to exit. Do this before setting the vol_state. */
1186 mutex_lock(&osb->recovery_lock);
1187 osb->disable_recovery = 1;
1188 mutex_unlock(&osb->recovery_lock);
1189 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1190
1191 /* At this point, we know that no more recovery threads can be
1192 * launched, so wait for any recovery completion work to
1193 * complete. */
1194 flush_workqueue(ocfs2_wq);
1195
1196 ocfs2_journal_shutdown(osb);
1197
1198 ocfs2_sync_blockdev(sb);
1199
1200 /* No dlm means we've failed during mount, so skip all the
1201 * steps which depended on that to complete. */
1202 if (osb->dlm) {
1203 tmp = ocfs2_super_lock(osb, 1);
1204 if (tmp < 0) {
1205 mlog_errno(tmp);
1206 return;
1207 }
1208
1209 tmp = ocfs2_request_umount_vote(osb);
1210 if (tmp < 0)
1211 mlog_errno(tmp);
1212 }
1213
1214 if (osb->slot_num != OCFS2_INVALID_SLOT)
1215 ocfs2_put_slot(osb);
1216
1217 if (osb->dlm)
1218 ocfs2_super_unlock(osb, 1);
1219
1220 ocfs2_release_system_inodes(osb);
1221
1222 if (osb->dlm) {
1223 ocfs2_unregister_net_handlers(osb);
1224
1225 ocfs2_dlm_shutdown(osb);
1226 }
1227
1228 ocfs2_clear_hb_callbacks(osb);
1229
1230 debugfs_remove(osb->osb_debug_root);
1231
1232 if (!mnt_err)
1233 ocfs2_stop_heartbeat(osb);
1234
1235 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1236
1237 if (ocfs2_mount_local(osb))
1238 snprintf(nodestr, sizeof(nodestr), "local");
1239 else
1240 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1241
1242 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1243 osb->dev_str, nodestr);
1244
1245 ocfs2_delete_osb(osb);
1246 kfree(osb);
1247 sb->s_dev = 0;
1248 sb->s_fs_info = NULL;
1249 }
1250
1251 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1252 unsigned uuid_bytes)
1253 {
1254 int i, ret;
1255 char *ptr;
1256
1257 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1258
1259 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1260 if (osb->uuid_str == NULL)
1261 return -ENOMEM;
1262
1263 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1264 /* print with null */
1265 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1266 if (ret != 2) /* drop super cleans up */
1267 return -EINVAL;
1268 /* then only advance past the last char */
1269 ptr += 2;
1270 }
1271
1272 return 0;
1273 }
1274
1275 static int ocfs2_initialize_super(struct super_block *sb,
1276 struct buffer_head *bh,
1277 int sector_size)
1278 {
1279 int status = 0;
1280 int i, cbits, bbits;
1281 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1282 struct inode *inode = NULL;
1283 struct buffer_head *bitmap_bh = NULL;
1284 struct ocfs2_journal *journal;
1285 __le32 uuid_net_key;
1286 struct ocfs2_super *osb;
1287
1288 mlog_entry_void();
1289
1290 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1291 if (!osb) {
1292 status = -ENOMEM;
1293 mlog_errno(status);
1294 goto bail;
1295 }
1296
1297 sb->s_fs_info = osb;
1298 sb->s_op = &ocfs2_sops;
1299 sb->s_export_op = &ocfs2_export_ops;
1300 sb->s_time_gran = 1;
1301 sb->s_flags |= MS_NOATIME;
1302 /* this is needed to support O_LARGEFILE */
1303 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1304 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1305 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1306
1307 osb->sb = sb;
1308 /* Save off for ocfs2_rw_direct */
1309 osb->s_sectsize_bits = blksize_bits(sector_size);
1310 BUG_ON(!osb->s_sectsize_bits);
1311
1312 osb->net_response_ids = 0;
1313 spin_lock_init(&osb->net_response_lock);
1314 INIT_LIST_HEAD(&osb->net_response_list);
1315
1316 INIT_LIST_HEAD(&osb->osb_net_handlers);
1317 init_waitqueue_head(&osb->recovery_event);
1318 spin_lock_init(&osb->vote_task_lock);
1319 init_waitqueue_head(&osb->vote_event);
1320 osb->vote_work_sequence = 0;
1321 osb->vote_wake_sequence = 0;
1322 INIT_LIST_HEAD(&osb->blocked_lock_list);
1323 osb->blocked_lock_count = 0;
1324 INIT_LIST_HEAD(&osb->vote_list);
1325 spin_lock_init(&osb->osb_lock);
1326
1327 atomic_set(&osb->alloc_stats.moves, 0);
1328 atomic_set(&osb->alloc_stats.local_data, 0);
1329 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1330 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1331 atomic_set(&osb->alloc_stats.bg_extends, 0);
1332
1333 ocfs2_init_node_maps(osb);
1334
1335 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1336 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1337
1338 mutex_init(&osb->recovery_lock);
1339
1340 osb->disable_recovery = 0;
1341 osb->recovery_thread_task = NULL;
1342
1343 init_waitqueue_head(&osb->checkpoint_event);
1344 atomic_set(&osb->needs_checkpoint, 0);
1345
1346 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1347
1348 osb->node_num = O2NM_INVALID_NODE_NUM;
1349 osb->slot_num = OCFS2_INVALID_SLOT;
1350
1351 osb->local_alloc_state = OCFS2_LA_UNUSED;
1352 osb->local_alloc_bh = NULL;
1353
1354 ocfs2_setup_hb_callbacks(osb);
1355
1356 init_waitqueue_head(&osb->osb_mount_event);
1357
1358 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1359 if (!osb->vol_label) {
1360 mlog(ML_ERROR, "unable to alloc vol label\n");
1361 status = -ENOMEM;
1362 goto bail;
1363 }
1364
1365 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1366 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1367 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1368 osb->max_slots);
1369 status = -EINVAL;
1370 goto bail;
1371 }
1372 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1373
1374 init_waitqueue_head(&osb->osb_wipe_event);
1375 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1376 sizeof(*osb->osb_orphan_wipes),
1377 GFP_KERNEL);
1378 if (!osb->osb_orphan_wipes) {
1379 status = -ENOMEM;
1380 mlog_errno(status);
1381 goto bail;
1382 }
1383
1384 osb->s_feature_compat =
1385 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1386 osb->s_feature_ro_compat =
1387 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1388 osb->s_feature_incompat =
1389 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1390
1391 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1392 mlog(ML_ERROR, "couldn't mount because of unsupported "
1393 "optional features (%x).\n", i);
1394 status = -EINVAL;
1395 goto bail;
1396 }
1397 if (!(osb->sb->s_flags & MS_RDONLY) &&
1398 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1399 mlog(ML_ERROR, "couldn't mount RDWR because of "
1400 "unsupported optional features (%x).\n", i);
1401 status = -EINVAL;
1402 goto bail;
1403 }
1404
1405 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1406
1407 /* FIXME
1408 * This should be done in ocfs2_journal_init(), but unknown
1409 * ordering issues will cause the filesystem to crash.
1410 * If anyone wants to figure out what part of the code
1411 * refers to osb->journal before ocfs2_journal_init() is run,
1412 * be my guest.
1413 */
1414 /* initialize our journal structure */
1415
1416 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1417 if (!journal) {
1418 mlog(ML_ERROR, "unable to alloc journal\n");
1419 status = -ENOMEM;
1420 goto bail;
1421 }
1422 osb->journal = journal;
1423 journal->j_osb = osb;
1424
1425 atomic_set(&journal->j_num_trans, 0);
1426 init_rwsem(&journal->j_trans_barrier);
1427 init_waitqueue_head(&journal->j_checkpointed);
1428 spin_lock_init(&journal->j_lock);
1429 journal->j_trans_id = (unsigned long) 1;
1430 INIT_LIST_HEAD(&journal->j_la_cleanups);
1431 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1432 journal->j_state = OCFS2_JOURNAL_FREE;
1433
1434 /* get some pseudo constants for clustersize bits */
1435 osb->s_clustersize_bits =
1436 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1437 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1438 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1439
1440 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1441 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1442 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1443 osb->s_clustersize);
1444 status = -EINVAL;
1445 goto bail;
1446 }
1447
1448 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1449 > (u32)~0UL) {
1450 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1451 "what jbd can address in 32 bits.\n");
1452 status = -EINVAL;
1453 goto bail;
1454 }
1455
1456 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1457 sizeof(di->id2.i_super.s_uuid))) {
1458 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1459 status = -ENOMEM;
1460 goto bail;
1461 }
1462
1463 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1464 osb->net_key = le32_to_cpu(uuid_net_key);
1465
1466 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1467 osb->vol_label[63] = '\0';
1468 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1469 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1470 osb->first_cluster_group_blkno =
1471 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1472 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1473 mlog(0, "vol_label: %s\n", osb->vol_label);
1474 mlog(0, "uuid: %s\n", osb->uuid_str);
1475 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1476 (unsigned long long)osb->root_blkno,
1477 (unsigned long long)osb->system_dir_blkno);
1478
1479 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1480 if (!osb->osb_dlm_debug) {
1481 status = -ENOMEM;
1482 mlog_errno(status);
1483 goto bail;
1484 }
1485
1486 atomic_set(&osb->vol_state, VOLUME_INIT);
1487
1488 /* load root, system_dir, and all global system inodes */
1489 status = ocfs2_init_global_system_inodes(osb);
1490 if (status < 0) {
1491 mlog_errno(status);
1492 goto bail;
1493 }
1494
1495 /*
1496 * global bitmap
1497 */
1498 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1499 OCFS2_INVALID_SLOT);
1500 if (!inode) {
1501 status = -EINVAL;
1502 mlog_errno(status);
1503 goto bail;
1504 }
1505
1506 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1507
1508 /* We don't have a cluster lock on the bitmap here because
1509 * we're only interested in static information and the extra
1510 * complexity at mount time isn't worht it. Don't pass the
1511 * inode in to the read function though as we don't want it to
1512 * be put in the cache. */
1513 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1514 NULL);
1515 iput(inode);
1516 if (status < 0) {
1517 mlog_errno(status);
1518 goto bail;
1519 }
1520
1521 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1522 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1523 brelse(bitmap_bh);
1524 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1525 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1526
1527 status = ocfs2_init_slot_info(osb);
1528 if (status < 0) {
1529 mlog_errno(status);
1530 goto bail;
1531 }
1532
1533 bail:
1534 mlog_exit(status);
1535 return status;
1536 }
1537
1538 /*
1539 * will return: -EAGAIN if it is ok to keep searching for superblocks
1540 * -EINVAL if there is a bad superblock
1541 * 0 on success
1542 */
1543 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1544 struct buffer_head *bh,
1545 u32 blksz)
1546 {
1547 int status = -EAGAIN;
1548
1549 mlog_entry_void();
1550
1551 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1552 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1553 status = -EINVAL;
1554 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1555 mlog(ML_ERROR, "found superblock with incorrect block "
1556 "size: found %u, should be %u\n",
1557 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1558 blksz);
1559 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1560 OCFS2_MAJOR_REV_LEVEL ||
1561 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1562 OCFS2_MINOR_REV_LEVEL) {
1563 mlog(ML_ERROR, "found superblock with bad version: "
1564 "found %u.%u, should be %u.%u\n",
1565 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1566 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1567 OCFS2_MAJOR_REV_LEVEL,
1568 OCFS2_MINOR_REV_LEVEL);
1569 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1570 mlog(ML_ERROR, "bad block number on superblock: "
1571 "found %llu, should be %llu\n",
1572 (unsigned long long)le64_to_cpu(di->i_blkno),
1573 (unsigned long long)bh->b_blocknr);
1574 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1575 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1576 mlog(ML_ERROR, "bad cluster size found: %u\n",
1577 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1578 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1579 mlog(ML_ERROR, "bad root_blkno: 0\n");
1580 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1581 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1582 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1583 mlog(ML_ERROR,
1584 "Superblock slots found greater than file system "
1585 "maximum: found %u, max %u\n",
1586 le16_to_cpu(di->id2.i_super.s_max_slots),
1587 OCFS2_MAX_SLOTS);
1588 } else {
1589 /* found it! */
1590 status = 0;
1591 }
1592 }
1593
1594 mlog_exit(status);
1595 return status;
1596 }
1597
1598 static int ocfs2_check_volume(struct ocfs2_super *osb)
1599 {
1600 int status = 0;
1601 int dirty;
1602 int local;
1603 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1604 * recover
1605 * ourselves. */
1606
1607 mlog_entry_void();
1608
1609 /* Init our journal object. */
1610 status = ocfs2_journal_init(osb->journal, &dirty);
1611 if (status < 0) {
1612 mlog(ML_ERROR, "Could not initialize journal!\n");
1613 goto finally;
1614 }
1615
1616 /* If the journal was unmounted cleanly then we don't want to
1617 * recover anything. Otherwise, journal_load will do that
1618 * dirty work for us :) */
1619 if (!dirty) {
1620 status = ocfs2_journal_wipe(osb->journal, 0);
1621 if (status < 0) {
1622 mlog_errno(status);
1623 goto finally;
1624 }
1625 } else {
1626 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1627 "recovering volume.\n");
1628 }
1629
1630 local = ocfs2_mount_local(osb);
1631
1632 /* will play back anything left in the journal. */
1633 ocfs2_journal_load(osb->journal, local);
1634
1635 if (dirty) {
1636 /* recover my local alloc if we didn't unmount cleanly. */
1637 status = ocfs2_begin_local_alloc_recovery(osb,
1638 osb->slot_num,
1639 &local_alloc);
1640 if (status < 0) {
1641 mlog_errno(status);
1642 goto finally;
1643 }
1644 /* we complete the recovery process after we've marked
1645 * ourselves as mounted. */
1646 }
1647
1648 mlog(0, "Journal loaded.\n");
1649
1650 status = ocfs2_load_local_alloc(osb);
1651 if (status < 0) {
1652 mlog_errno(status);
1653 goto finally;
1654 }
1655
1656 if (dirty) {
1657 /* Recovery will be completed after we've mounted the
1658 * rest of the volume. */
1659 osb->dirty = 1;
1660 osb->local_alloc_copy = local_alloc;
1661 local_alloc = NULL;
1662 }
1663
1664 /* go through each journal, trylock it and if you get the
1665 * lock, and it's marked as dirty, set the bit in the recover
1666 * map and launch a recovery thread for it. */
1667 status = ocfs2_mark_dead_nodes(osb);
1668 if (status < 0)
1669 mlog_errno(status);
1670
1671 finally:
1672 if (local_alloc)
1673 kfree(local_alloc);
1674
1675 mlog_exit(status);
1676 return status;
1677 }
1678
1679 /*
1680 * The routine gets called from dismount or close whenever a dismount on
1681 * volume is requested and the osb open count becomes 1.
1682 * It will remove the osb from the global list and also free up all the
1683 * initialized resources and fileobject.
1684 */
1685 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1686 {
1687 mlog_entry_void();
1688
1689 /* This function assumes that the caller has the main osb resource */
1690
1691 if (osb->slot_info)
1692 ocfs2_free_slot_info(osb->slot_info);
1693
1694 kfree(osb->osb_orphan_wipes);
1695 /* FIXME
1696 * This belongs in journal shutdown, but because we have to
1697 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1698 * we free it here.
1699 */
1700 kfree(osb->journal);
1701 if (osb->local_alloc_copy)
1702 kfree(osb->local_alloc_copy);
1703 kfree(osb->uuid_str);
1704 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1705 memset(osb, 0, sizeof(struct ocfs2_super));
1706
1707 mlog_exit_void();
1708 }
1709
1710 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1711 * panic(). We do not support continue-on-error operation. */
1712 static void ocfs2_handle_error(struct super_block *sb)
1713 {
1714 struct ocfs2_super *osb = OCFS2_SB(sb);
1715
1716 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1717 panic("OCFS2: (device %s): panic forced after error\n",
1718 sb->s_id);
1719
1720 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1721
1722 if (sb->s_flags & MS_RDONLY &&
1723 (ocfs2_is_soft_readonly(osb) ||
1724 ocfs2_is_hard_readonly(osb)))
1725 return;
1726
1727 printk(KERN_CRIT "File system is now read-only due to the potential "
1728 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1729 "system is unmounted.\n");
1730 sb->s_flags |= MS_RDONLY;
1731 ocfs2_set_ro_flag(osb, 0);
1732 }
1733
1734 static char error_buf[1024];
1735
1736 void __ocfs2_error(struct super_block *sb,
1737 const char *function,
1738 const char *fmt, ...)
1739 {
1740 va_list args;
1741
1742 va_start(args, fmt);
1743 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1744 va_end(args);
1745
1746 /* Not using mlog here because we want to show the actual
1747 * function the error came from. */
1748 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1749 sb->s_id, function, error_buf);
1750
1751 ocfs2_handle_error(sb);
1752 }
1753
1754 /* Handle critical errors. This is intentionally more drastic than
1755 * ocfs2_handle_error, so we only use for things like journal errors,
1756 * etc. */
1757 void __ocfs2_abort(struct super_block* sb,
1758 const char *function,
1759 const char *fmt, ...)
1760 {
1761 va_list args;
1762
1763 va_start(args, fmt);
1764 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1765 va_end(args);
1766
1767 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1768 sb->s_id, function, error_buf);
1769
1770 /* We don't have the cluster support yet to go straight to
1771 * hard readonly in here. Until then, we want to keep
1772 * ocfs2_abort() so that we can at least mark critical
1773 * errors.
1774 *
1775 * TODO: This should abort the journal and alert other nodes
1776 * that our slot needs recovery. */
1777
1778 /* Force a panic(). This stinks, but it's better than letting
1779 * things continue without having a proper hard readonly
1780 * here. */
1781 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1782 ocfs2_handle_error(sb);
1783 }
1784
1785 module_init(ocfs2_init);
1786 module_exit(ocfs2_exit);
Linux kernel - Deliverables
This page took 0.07547 seconds and 6 git commands to generate.