1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * load/unload driver, mount/dismount volumes
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
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.
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.
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.
26 #include <linux/module.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>
43 #include <cluster/nodemanager.h>
45 #define MLOG_MASK_PREFIX ML_SUPER
46 #include <cluster/masklog.h>
50 /* this should be the only file to include a version 1 header */
51 #include "ocfs1_fs_compat.h"
56 #include "extent_map.h"
57 #include "heartbeat.h"
60 #include "localalloc.h"
69 #include "buffer_head_io.h"
74 static spinlock_t ocfs2_globals_lock = SPIN_LOCK_UNLOCKED;
76 static u32 osb_id; /* Keeps track of next available OSB Id */
78 static kmem_cache_t *ocfs2_inode_cachep = NULL;
80 kmem_cache_t *ocfs2_lock_cache = NULL;
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. */
86 struct workqueue_struct *ocfs2_wq = NULL;
88 static struct dentry *ocfs2_debugfs_root = NULL;
90 MODULE_AUTHOR("Oracle");
91 MODULE_LICENSE("GPL");
93 static int ocfs2_parse_options(struct super_block *sb, char *options,
94 unsigned long *mount_opt, int is_remount);
95 static void ocfs2_put_super(struct super_block *sb);
96 static int ocfs2_mount_volume(struct super_block *sb);
97 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
98 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
99 static int ocfs2_initialize_mem_caches(void);
100 static void ocfs2_free_mem_caches(void);
101 static void ocfs2_delete_osb(struct ocfs2_super *osb);
103 static int ocfs2_statfs(struct super_block *sb, struct kstatfs *buf);
105 static int ocfs2_sync_fs(struct super_block *sb, int wait);
107 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
108 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
109 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
110 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
111 static int ocfs2_check_volume(struct ocfs2_super *osb);
112 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
113 struct buffer_head *bh,
115 static int ocfs2_initialize_super(struct super_block *sb,
116 struct buffer_head *bh,
118 static int ocfs2_get_sector(struct super_block *sb,
119 struct buffer_head **bh,
122 static void ocfs2_write_super(struct super_block *sb);
123 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
124 static void ocfs2_destroy_inode(struct inode *inode);
126 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
128 static 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,
154 static match_table_t tokens = {
155 {Opt_barrier, "barrier=%u"},
156 {Opt_err_panic, "errors=panic"},
157 {Opt_err_ro, "errors=remount-ro"},
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"},
168 * write_super and sync_fs ripped right out of ext3.
170 static void ocfs2_write_super(struct super_block *sb)
172 if (mutex_trylock(&sb->s_lock) != 0)
177 static int ocfs2_sync_fs(struct super_block *sb, int wait)
181 struct ocfs2_super *osb = OCFS2_SB(sb);
185 if (ocfs2_is_hard_readonly(osb))
189 status = ocfs2_flush_truncate_log(osb);
193 ocfs2_schedule_truncate_log_flush(osb, 0);
196 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
198 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
204 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
206 struct inode *new = NULL;
212 new = ocfs2_iget(osb, osb->root_blkno);
214 status = PTR_ERR(new);
218 osb->root_inode = new;
220 new = ocfs2_iget(osb, osb->system_dir_blkno);
222 status = PTR_ERR(new);
226 osb->sys_root_inode = new;
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);
232 ocfs2_release_system_inodes(osb);
235 /* FIXME: Should ERROR_RO_FS */
236 mlog(ML_ERROR, "Unable to load system inode %d, "
237 "possibly corrupt fs?", i);
240 // the array now has one ref, so drop this one
249 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
251 struct inode *new = NULL;
257 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
258 i < NUM_SYSTEM_INODES;
260 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
262 ocfs2_release_system_inodes(osb);
264 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
265 status, i, osb->slot_num);
268 /* the array now has one ref, so drop this one */
277 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
284 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
285 inode = osb->system_inodes[i];
288 osb->system_inodes[i] = NULL;
292 inode = osb->sys_root_inode;
295 osb->sys_root_inode = NULL;
298 inode = osb->root_inode;
301 osb->root_inode = NULL;
308 /* We're allocating fs objects, use GFP_NOFS */
309 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
311 struct ocfs2_inode_info *oi;
313 oi = kmem_cache_alloc(ocfs2_inode_cachep, SLAB_NOFS);
317 return &oi->vfs_inode;
320 static void ocfs2_destroy_inode(struct inode *inode)
322 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
325 /* From xfs_super.c:xfs_max_file_offset
326 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
328 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
330 unsigned int pagefactor = 1;
331 unsigned int bitshift = BITS_PER_LONG - 1;
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.
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;
353 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
357 return (((unsigned long long)pagefactor) << bitshift) - 1;
360 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
362 int incompat_features;
364 unsigned long parsed_options;
365 struct ocfs2_super *osb = OCFS2_SB(sb);
367 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
372 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
373 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
375 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
379 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
380 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
382 mlog(ML_ERROR, "Cannot change data mode on remount\n");
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");
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;
402 mlog(0, "Making ro filesystem writeable.\n");
404 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
405 mlog(ML_ERROR, "Cannot remount RDWR "
406 "filesystem due to previous errors.\n");
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);
418 sb->s_flags &= ~MS_RDONLY;
419 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
422 spin_unlock(&osb->osb_lock);
426 if (!ocfs2_is_hard_readonly(osb))
427 ocfs2_set_journal_params(osb);
429 /* Only save off the new mount options in case of a successful
431 osb->s_mount_opt = parsed_options;
437 static int ocfs2_sb_probe(struct super_block *sb,
438 struct buffer_head **bh,
441 int status = 0, tmpstat;
442 struct ocfs1_vol_disk_hdr *hdr;
443 struct ocfs2_dinode *di;
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);
457 /* Can this really happen? */
458 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
459 *sector_size = OCFS2_MIN_BLOCKSIZE;
461 /* check block zero for old format */
462 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
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);
473 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
474 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
475 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
482 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
483 "upgraded before mounting with ocfs v2\n");
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.
493 for (blksize = *sector_size;
494 blksize <= OCFS2_MAX_BLOCKSIZE;
496 tmpstat = ocfs2_get_sector(sb, bh,
497 OCFS2_SUPER_BLOCK_BLKNO,
504 di = (struct ocfs2_dinode *) (*bh)->b_data;
505 status = ocfs2_verify_volume(di, *bh, blksize);
510 if (status != -EAGAIN)
518 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
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;
527 mlog_entry("%p, %p, %i", sb, data, silent);
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()) {
533 goto read_super_error;
536 /* probe for superblock */
537 status = ocfs2_sb_probe(sb, &bh, §or_size);
539 mlog(ML_ERROR, "superblock probe failed!\n");
540 goto read_super_error;
543 status = ocfs2_initialize_super(sb, bh, sector_size);
547 goto read_super_error;
552 if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
554 goto read_super_error;
556 osb->s_mount_opt = parsed_opt;
558 sb->s_magic = OCFS2_SUPER_MAGIC;
560 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
562 if (bdev_read_only(sb->s_bdev)) {
563 if (!(sb->s_flags & MS_RDONLY)) {
565 mlog(ML_ERROR, "Readonly device detected but readonly "
566 "mount was not specified.\n");
567 goto read_super_error;
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) {
574 mlog(ML_ERROR, "Local heartbeat specified on readonly "
576 goto read_super_error;
579 status = ocfs2_check_journals_nolocks(osb);
581 if (status == -EROFS)
582 mlog(ML_ERROR, "Recovery required on readonly "
583 "file system, but write access is "
587 goto read_super_error;
590 ocfs2_set_ro_flag(osb, 1);
592 printk(KERN_NOTICE "Readonly device detected. No cluster "
593 "services will be utilized for this mount. Recovery "
594 "will be skipped.\n");
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
602 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
603 mlog(ML_ERROR, "No heartbeat for device (%s)\n",
606 goto read_super_error;
609 if (sb->s_flags & MS_RDONLY)
610 ocfs2_set_ro_flag(osb, 0);
613 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
615 if (!osb->osb_debug_root) {
617 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
618 goto read_super_error;
621 status = ocfs2_mount_volume(sb);
623 inode = igrab(osb->root_inode);
626 goto read_super_error;
631 goto read_super_error;
634 root = d_alloc_root(inode);
638 goto read_super_error;
643 ocfs2_complete_mount_recovery(osb);
645 printk("ocfs2: Mounting device (%u,%u) on (node %d, slot %d) with %s "
647 MAJOR(sb->s_dev), MINOR(sb->s_dev), osb->node_num,
649 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
652 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
653 wake_up(&osb->osb_mount_event);
666 atomic_set(&osb->vol_state, VOLUME_DISABLED);
667 wake_up(&osb->osb_mount_event);
668 ocfs2_dismount_volume(sb, 1);
675 static struct super_block *ocfs2_get_sb(struct file_system_type *fs_type,
677 const char *dev_name,
680 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
683 static struct file_system_type ocfs2_fs_type = {
684 .owner = THIS_MODULE,
686 .get_sb = ocfs2_get_sb, /* is this called when we mount
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,
695 static int ocfs2_parse_options(struct super_block *sb,
697 unsigned long *mount_opt,
703 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
704 options ? options : "(none)");
713 while ((p = strsep(&options, ",")) != NULL) {
715 substring_t args[MAX_OPT_ARGS];
720 token = match_token(p, tokens, args);
723 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
726 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
729 if (match_int(&args[0], &option)) {
734 *mount_opt |= OCFS2_MOUNT_BARRIER;
736 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
739 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
742 *mount_opt |= OCFS2_MOUNT_NOINTR;
745 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
748 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
750 case Opt_data_ordered:
751 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
753 case Opt_data_writeback:
754 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
758 "Unrecognized mount option \"%s\" "
759 "or missing value\n", p);
772 static int __init ocfs2_init(void)
778 ocfs2_print_version();
780 if (init_ocfs2_extent_maps())
783 status = init_ocfs2_uptodate_cache();
789 status = ocfs2_initialize_mem_caches();
795 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
801 spin_lock(&ocfs2_globals_lock);
803 spin_unlock(&ocfs2_globals_lock);
805 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
806 if (!ocfs2_debugfs_root) {
808 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
813 ocfs2_free_mem_caches();
814 exit_ocfs2_uptodate_cache();
815 exit_ocfs2_extent_maps();
821 return register_filesystem(&ocfs2_fs_type);
826 static void __exit ocfs2_exit(void)
831 flush_workqueue(ocfs2_wq);
832 destroy_workqueue(ocfs2_wq);
835 debugfs_remove(ocfs2_debugfs_root);
837 ocfs2_free_mem_caches();
839 unregister_filesystem(&ocfs2_fs_type);
841 exit_ocfs2_extent_maps();
843 exit_ocfs2_uptodate_cache();
848 static void ocfs2_put_super(struct super_block *sb)
850 mlog_entry("(0x%p)\n", sb);
852 ocfs2_sync_blockdev(sb);
853 ocfs2_dismount_volume(sb, 0);
858 static int ocfs2_statfs(struct super_block *sb, struct kstatfs *buf)
860 struct ocfs2_super *osb;
861 u32 numbits, freebits;
863 struct ocfs2_dinode *bm_lock;
864 struct buffer_head *bh = NULL;
865 struct inode *inode = NULL;
867 mlog_entry("(%p, %p)\n", sb, buf);
871 inode = ocfs2_get_system_file_inode(osb,
872 GLOBAL_BITMAP_SYSTEM_INODE,
875 mlog(ML_ERROR, "failed to get bitmap inode\n");
880 status = ocfs2_meta_lock(inode, NULL, &bh, 0);
886 bm_lock = (struct ocfs2_dinode *) bh->b_data;
888 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
889 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
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;
904 ocfs2_meta_unlock(inode, 0);
915 static void ocfs2_inode_init_once(void *data,
916 kmem_cache_t *cachep,
919 struct ocfs2_inode_info *oi = data;
921 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
922 SLAB_CTOR_CONSTRUCTOR) {
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;
934 init_rwsem(&oi->ip_alloc_sem);
935 mutex_init(&oi->ip_io_mutex);
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);
944 ocfs2_metadata_cache_init(&oi->vfs_inode);
946 inode_init_once(&oi->vfs_inode);
950 static int ocfs2_initialize_mem_caches(void)
952 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
953 sizeof(struct ocfs2_inode_info),
955 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
957 ocfs2_inode_init_once, NULL);
958 if (!ocfs2_inode_cachep)
961 ocfs2_lock_cache = kmem_cache_create("ocfs2_lock",
962 sizeof(struct ocfs2_journal_lock),
966 if (!ocfs2_lock_cache)
972 static void ocfs2_free_mem_caches(void)
974 if (ocfs2_inode_cachep)
975 kmem_cache_destroy(ocfs2_inode_cachep);
976 if (ocfs2_lock_cache)
977 kmem_cache_destroy(ocfs2_lock_cache);
979 ocfs2_inode_cachep = NULL;
980 ocfs2_lock_cache = NULL;
983 static int ocfs2_get_sector(struct super_block *sb,
984 struct buffer_head **bh,
988 if (!sb_set_blocksize(sb, sect_size)) {
989 mlog(ML_ERROR, "unable to set blocksize\n");
993 *bh = sb_getblk(sb, block);
999 if (!buffer_dirty(*bh))
1000 clear_buffer_uptodate(*bh);
1002 ll_rw_block(READ, 1, bh);
1003 wait_on_buffer(*bh);
1007 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1008 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1012 /* XXX hold a ref on the node while mounte? easy enough, if
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");
1021 mlog(ML_NOTICE, "I am node %d\n", osb->node_num);
1028 static int ocfs2_mount_volume(struct super_block *sb)
1031 int unlock_super = 0;
1032 struct ocfs2_super *osb = OCFS2_SB(sb);
1036 if (ocfs2_is_hard_readonly(osb))
1039 status = ocfs2_fill_local_node_info(osb);
1045 status = ocfs2_register_hb_callbacks(osb);
1051 status = ocfs2_dlm_init(osb);
1057 /* requires vote_thread to be running. */
1058 status = ocfs2_register_net_handlers(osb);
1064 status = ocfs2_super_lock(osb, 1);
1071 /* This will load up the node map and add ourselves to it. */
1072 status = ocfs2_find_slot(osb);
1078 ocfs2_populate_mounted_map(osb);
1080 /* load all node-local system inodes */
1081 status = ocfs2_init_local_system_inodes(osb);
1087 status = ocfs2_check_volume(osb);
1093 status = ocfs2_truncate_log_init(osb);
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);
1110 ocfs2_super_unlock(osb, 1);
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
1119 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1122 return osb->recovery_thread_task != NULL;
1125 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1128 struct ocfs2_super *osb = NULL;
1130 mlog_entry("(0x%p)\n", sb);
1136 ocfs2_shutdown_local_alloc(osb);
1138 ocfs2_truncate_log_shutdown(osb);
1140 /* disable any new recovery threads and wait for any currently
1141 * running ones to exit. Do this before setting the vol_state. */
1142 mutex_lock(&osb->recovery_lock);
1143 osb->disable_recovery = 1;
1144 mutex_unlock(&osb->recovery_lock);
1145 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1147 /* At this point, we know that no more recovery threads can be
1148 * launched, so wait for any recovery completion work to
1150 flush_workqueue(ocfs2_wq);
1152 ocfs2_journal_shutdown(osb);
1154 ocfs2_sync_blockdev(sb);
1156 /* No dlm means we've failed during mount, so skip all the
1157 * steps which depended on that to complete. */
1159 tmp = ocfs2_super_lock(osb, 1);
1165 tmp = ocfs2_request_umount_vote(osb);
1169 if (osb->slot_num != OCFS2_INVALID_SLOT)
1170 ocfs2_put_slot(osb);
1172 ocfs2_super_unlock(osb, 1);
1175 ocfs2_release_system_inodes(osb);
1178 ocfs2_unregister_net_handlers(osb);
1180 ocfs2_dlm_shutdown(osb);
1183 ocfs2_clear_hb_callbacks(osb);
1185 debugfs_remove(osb->osb_debug_root);
1188 ocfs2_stop_heartbeat(osb);
1190 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
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);
1195 ocfs2_delete_osb(osb);
1198 sb->s_fs_info = NULL;
1201 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1202 unsigned uuid_bytes)
1207 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1209 osb->uuid_str = kcalloc(1, OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1210 if (osb->uuid_str == NULL)
1213 memcpy(osb->uuid, uuid, OCFS2_VOL_UUID_LEN);
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 */
1220 /* then only advance past the last char */
1227 static int ocfs2_initialize_super(struct super_block *sb,
1228 struct buffer_head *bh,
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;
1242 osb = kcalloc(1, sizeof(struct ocfs2_super), GFP_KERNEL);
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);
1257 /* Save off for ocfs2_rw_direct */
1258 osb->s_sectsize_bits = blksize_bits(sector_size);
1259 BUG_ON(!osb->s_sectsize_bits);
1261 osb->net_response_ids = 0;
1262 spin_lock_init(&osb->net_response_lock);
1263 INIT_LIST_HEAD(&osb->net_response_list);
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);
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);
1282 ocfs2_init_node_maps(osb);
1284 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1285 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1287 mutex_init(&osb->recovery_lock);
1289 osb->disable_recovery = 0;
1290 osb->recovery_thread_task = NULL;
1292 init_waitqueue_head(&osb->checkpoint_event);
1293 atomic_set(&osb->needs_checkpoint, 0);
1295 osb->node_num = O2NM_INVALID_NODE_NUM;
1296 osb->slot_num = OCFS2_INVALID_SLOT;
1298 osb->local_alloc_state = OCFS2_LA_UNUSED;
1299 osb->local_alloc_bh = NULL;
1301 ocfs2_setup_hb_callbacks(osb);
1303 init_waitqueue_head(&osb->osb_mount_event);
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");
1312 osb->uuid = kmalloc(OCFS2_VOL_UUID_LEN, GFP_KERNEL);
1314 mlog(ML_ERROR, "unable to alloc uuid\n");
1319 di = (struct ocfs2_dinode *)bh->b_data;
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",
1328 mlog(ML_NOTICE, "max_slots for this device: %u\n", osb->max_slots);
1330 init_waitqueue_head(&osb->osb_wipe_event);
1331 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1332 sizeof(*osb->osb_orphan_wipes),
1334 if (!osb->osb_orphan_wipes) {
1340 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);
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);
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);
1361 get_random_bytes(&osb->s_next_generation, sizeof(u32));
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,
1370 /* initialize our journal structure */
1372 journal = kcalloc(1, sizeof(struct ocfs2_journal), GFP_KERNEL);
1374 mlog(ML_ERROR, "unable to alloc journal\n");
1378 osb->journal = journal;
1379 journal->j_osb = osb;
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;
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);
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);
1404 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1406 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1407 "what jbd can address in 32 bits.\n");
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");
1419 memcpy(&uuid_net_key, &osb->uuid[i], sizeof(osb->net_key));
1420 osb->net_key = le32_to_cpu(uuid_net_key);
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);
1431 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);
1435 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1436 if (!osb->osb_dlm_debug) {
1442 atomic_set(&osb->vol_state, VOLUME_INIT);
1444 /* load root, system_dir, and all global system inodes */
1445 status = ocfs2_init_global_system_inodes(osb);
1454 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1455 OCFS2_INVALID_SLOT);
1462 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1464 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
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);
1476 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1477 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1479 status = ocfs2_init_slot_info(osb);
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)
1492 mlog(ML_ERROR, "Too many volumes mounted\n");
1495 spin_unlock(&ocfs2_globals_lock);
1503 * will return: -EAGAIN if it is ok to keep searching for superblocks
1504 * -EINVAL if there is a bad superblock
1507 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1508 struct buffer_head *bh,
1511 int status = -EAGAIN;
1515 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1516 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
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),
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: "
1535 "found %llu, should be %llu\n",
1536 (unsigned long long)di->i_blkno,
1537 (unsigned long long)bh->b_blocknr);
1538 } 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) {
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),
1562 static int ocfs2_check_volume(struct ocfs2_super *osb)
1566 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1572 /* Init our journal object. */
1573 status = ocfs2_journal_init(osb->journal, &dirty);
1575 mlog(ML_ERROR, "Could not initialize journal!\n");
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 :) */
1583 status = ocfs2_journal_wipe(osb->journal, 0);
1589 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1590 "recovering volume.\n");
1593 /* will play back anything left in the journal. */
1594 ocfs2_journal_load(osb->journal);
1597 /* recover my local alloc if we didn't unmount cleanly. */
1598 status = ocfs2_begin_local_alloc_recovery(osb,
1605 /* we complete the recovery process after we've marked
1606 * ourselves as mounted. */
1609 mlog(0, "Journal loaded.\n");
1611 status = ocfs2_load_local_alloc(osb);
1618 /* Recovery will be completed after we've mounted the
1619 * rest of the volume. */
1621 osb->local_alloc_copy = local_alloc;
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);
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.
1646 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1650 /* This function assumes that the caller has the main osb resource */
1653 ocfs2_free_slot_info(osb->slot_info);
1655 kfree(osb->osb_orphan_wipes);
1657 * This belongs in journal shutdown, but because we have to
1658 * allocate osb->journal at the start of ocfs2_initalize_osb(),
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));
1671 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1672 * panic(). We do not support continue-on-error operation. */
1673 static void ocfs2_handle_error(struct super_block *sb)
1675 struct ocfs2_super *osb = OCFS2_SB(sb);
1677 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1678 panic("OCFS2: (device %s): panic forced after error\n",
1681 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1683 if (sb->s_flags & MS_RDONLY &&
1684 (ocfs2_is_soft_readonly(osb) ||
1685 ocfs2_is_hard_readonly(osb)))
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);
1695 static char error_buf[1024];
1697 void __ocfs2_error(struct super_block *sb,
1698 const char *function,
1699 const char *fmt, ...)
1703 va_start(args, fmt);
1704 vsprintf(error_buf, fmt, args);
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);
1712 ocfs2_handle_error(sb);
1715 /* Handle critical errors. This is intentionally more drastic than
1716 * ocfs2_handle_error, so we only use for things like journal errors,
1718 void __ocfs2_abort(struct super_block* sb,
1719 const char *function,
1720 const char *fmt, ...)
1724 va_start(args, fmt);
1725 vsprintf(error_buf, fmt, args);
1728 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1729 sb->s_id, function, error_buf);
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
1736 * TODO: This should abort the journal and alert other nodes
1737 * that our slot needs recovery. */
1739 /* Force a panic(). This stinks, but it's better than letting
1740 * things continue without having a proper hard readonly
1742 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1743 ocfs2_handle_error(sb);
1746 module_init(ocfs2_init);
1747 module_exit(ocfs2_exit);