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"
71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
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;
79 static struct dentry *ocfs2_debugfs_root = NULL;
81 MODULE_AUTHOR("Oracle");
82 MODULE_LICENSE("GPL");
84 static int ocfs2_parse_options(struct super_block *sb, char *options,
85 unsigned long *mount_opt, s16 *slot,
87 static void ocfs2_put_super(struct super_block *sb);
88 static int ocfs2_mount_volume(struct super_block *sb);
89 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
90 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
91 static int ocfs2_initialize_mem_caches(void);
92 static void ocfs2_free_mem_caches(void);
93 static void ocfs2_delete_osb(struct ocfs2_super *osb);
95 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
97 static int ocfs2_sync_fs(struct super_block *sb, int wait);
99 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
100 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
101 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
102 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
103 static int ocfs2_check_volume(struct ocfs2_super *osb);
104 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
105 struct buffer_head *bh,
107 static int ocfs2_initialize_super(struct super_block *sb,
108 struct buffer_head *bh,
110 static int ocfs2_get_sector(struct super_block *sb,
111 struct buffer_head **bh,
114 static void ocfs2_write_super(struct super_block *sb);
115 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
116 static void ocfs2_destroy_inode(struct inode *inode);
118 static const struct super_operations ocfs2_sops = {
119 .statfs = ocfs2_statfs,
120 .alloc_inode = ocfs2_alloc_inode,
121 .destroy_inode = ocfs2_destroy_inode,
122 .drop_inode = ocfs2_drop_inode,
123 .clear_inode = ocfs2_clear_inode,
124 .delete_inode = ocfs2_delete_inode,
125 .sync_fs = ocfs2_sync_fs,
126 .write_super = ocfs2_write_super,
127 .put_super = ocfs2_put_super,
128 .remount_fs = ocfs2_remount,
146 static match_table_t tokens = {
147 {Opt_barrier, "barrier=%u"},
148 {Opt_err_panic, "errors=panic"},
149 {Opt_err_ro, "errors=remount-ro"},
151 {Opt_nointr, "nointr"},
152 {Opt_hb_none, OCFS2_HB_NONE},
153 {Opt_hb_local, OCFS2_HB_LOCAL},
154 {Opt_data_ordered, "data=ordered"},
155 {Opt_data_writeback, "data=writeback"},
156 {Opt_atime_quantum, "atime_quantum=%u"},
157 {Opt_slot, "preferred_slot=%u"},
162 * write_super and sync_fs ripped right out of ext3.
164 static void ocfs2_write_super(struct super_block *sb)
166 if (mutex_trylock(&sb->s_lock) != 0)
171 static int ocfs2_sync_fs(struct super_block *sb, int wait)
175 struct ocfs2_super *osb = OCFS2_SB(sb);
179 if (ocfs2_is_hard_readonly(osb))
183 status = ocfs2_flush_truncate_log(osb);
187 ocfs2_schedule_truncate_log_flush(osb, 0);
190 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
192 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
198 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
200 struct inode *new = NULL;
206 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
208 status = PTR_ERR(new);
212 osb->root_inode = new;
214 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
216 status = PTR_ERR(new);
220 osb->sys_root_inode = new;
222 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
223 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
224 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
226 ocfs2_release_system_inodes(osb);
229 /* FIXME: Should ERROR_RO_FS */
230 mlog(ML_ERROR, "Unable to load system inode %d, "
231 "possibly corrupt fs?", i);
234 // the array now has one ref, so drop this one
243 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
245 struct inode *new = NULL;
251 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
252 i < NUM_SYSTEM_INODES;
254 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
256 ocfs2_release_system_inodes(osb);
258 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
259 status, i, osb->slot_num);
262 /* the array now has one ref, so drop this one */
271 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
278 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
279 inode = osb->system_inodes[i];
282 osb->system_inodes[i] = NULL;
286 inode = osb->sys_root_inode;
289 osb->sys_root_inode = NULL;
292 inode = osb->root_inode;
295 osb->root_inode = NULL;
302 /* We're allocating fs objects, use GFP_NOFS */
303 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
305 struct ocfs2_inode_info *oi;
307 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
311 return &oi->vfs_inode;
314 static void ocfs2_destroy_inode(struct inode *inode)
316 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
319 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
322 unsigned int bytes = 1 << cbits;
323 unsigned int trim = bytes;
324 unsigned int bitshift = 32;
327 * i_size and all block offsets in ocfs2 are always 64 bits
328 * wide. i_clusters is 32 bits, in cluster-sized units. So on
329 * 64 bit platforms, cluster size will be the limiting factor.
332 #if BITS_PER_LONG == 32
333 # if defined(CONFIG_LBD)
334 BUILD_BUG_ON(sizeof(sector_t) != 8);
336 * We might be limited by page cache size.
338 if (bytes > PAGE_CACHE_SIZE) {
339 bytes = PAGE_CACHE_SIZE;
342 * Shift by 31 here so that we don't get larger than
349 * We are limited by the size of sector_t. Use block size, as
350 * that's what we expose to the VFS.
359 * Trim by a whole cluster when we can actually approach the
360 * on-disk limits. Otherwise we can overflow i_clusters when
361 * an extent start is at the max offset.
363 return (((unsigned long long)bytes) << bitshift) - trim;
366 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
368 int incompat_features;
370 unsigned long parsed_options;
372 struct ocfs2_super *osb = OCFS2_SB(sb);
374 if (!ocfs2_parse_options(sb, data, &parsed_options, &slot, 1)) {
379 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
380 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
382 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
386 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
387 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
389 mlog(ML_ERROR, "Cannot change data mode on remount\n");
393 /* We're going to/from readonly mode. */
394 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
395 /* Lock here so the check of HARD_RO and the potential
396 * setting of SOFT_RO is atomic. */
397 spin_lock(&osb->osb_lock);
398 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
399 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
404 if (*flags & MS_RDONLY) {
405 mlog(0, "Going to ro mode.\n");
406 sb->s_flags |= MS_RDONLY;
407 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
409 mlog(0, "Making ro filesystem writeable.\n");
411 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
412 mlog(ML_ERROR, "Cannot remount RDWR "
413 "filesystem due to previous errors.\n");
417 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
418 if (incompat_features) {
419 mlog(ML_ERROR, "Cannot remount RDWR because "
420 "of unsupported optional features "
421 "(%x).\n", incompat_features);
425 sb->s_flags &= ~MS_RDONLY;
426 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
429 spin_unlock(&osb->osb_lock);
433 if (!ocfs2_is_hard_readonly(osb))
434 ocfs2_set_journal_params(osb);
436 /* Only save off the new mount options in case of a successful
438 osb->s_mount_opt = parsed_options;
444 static int ocfs2_sb_probe(struct super_block *sb,
445 struct buffer_head **bh,
448 int status = 0, tmpstat;
449 struct ocfs1_vol_disk_hdr *hdr;
450 struct ocfs2_dinode *di;
456 *sector_size = bdev_hardsect_size(sb->s_bdev);
457 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
458 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
459 *sector_size, OCFS2_MAX_BLOCKSIZE);
464 /* Can this really happen? */
465 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
466 *sector_size = OCFS2_MIN_BLOCKSIZE;
468 /* check block zero for old format */
469 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
474 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
475 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
476 mlog(ML_ERROR, "incompatible version: %u.%u\n",
477 hdr->major_version, hdr->minor_version);
480 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
481 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
482 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
489 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
490 "upgraded before mounting with ocfs v2\n");
495 * Now check at magic offset for 512, 1024, 2048, 4096
496 * blocksizes. 4096 is the maximum blocksize because it is
497 * the minimum clustersize.
500 for (blksize = *sector_size;
501 blksize <= OCFS2_MAX_BLOCKSIZE;
503 tmpstat = ocfs2_get_sector(sb, bh,
504 OCFS2_SUPER_BLOCK_BLKNO,
511 di = (struct ocfs2_dinode *) (*bh)->b_data;
512 status = ocfs2_verify_volume(di, *bh, blksize);
517 if (status != -EAGAIN)
525 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
527 if (ocfs2_mount_local(osb)) {
528 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
529 mlog(ML_ERROR, "Cannot heartbeat on a locally "
530 "mounted device.\n");
535 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
536 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
537 mlog(ML_ERROR, "Heartbeat has to be started to mount "
538 "a read-write clustered device.\n");
546 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
549 int status, sector_size;
550 unsigned long parsed_opt;
552 struct inode *inode = NULL;
553 struct ocfs2_super *osb = NULL;
554 struct buffer_head *bh = NULL;
557 mlog_entry("%p, %p, %i", sb, data, silent);
559 if (!ocfs2_parse_options(sb, data, &parsed_opt, &slot, 0)) {
561 goto read_super_error;
564 /* for now we only have one cluster/node, make sure we see it
565 * in the heartbeat universe */
566 if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
567 if (!o2hb_check_local_node_heartbeating()) {
569 goto read_super_error;
573 /* probe for superblock */
574 status = ocfs2_sb_probe(sb, &bh, §or_size);
576 mlog(ML_ERROR, "superblock probe failed!\n");
577 goto read_super_error;
580 status = ocfs2_initialize_super(sb, bh, sector_size);
584 goto read_super_error;
588 osb->s_mount_opt = parsed_opt;
589 osb->preferred_slot = slot;
591 sb->s_magic = OCFS2_SUPER_MAGIC;
593 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
595 if (bdev_read_only(sb->s_bdev)) {
596 if (!(sb->s_flags & MS_RDONLY)) {
598 mlog(ML_ERROR, "Readonly device detected but readonly "
599 "mount was not specified.\n");
600 goto read_super_error;
603 /* You should not be able to start a local heartbeat
604 * on a readonly device. */
605 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
607 mlog(ML_ERROR, "Local heartbeat specified on readonly "
609 goto read_super_error;
612 status = ocfs2_check_journals_nolocks(osb);
614 if (status == -EROFS)
615 mlog(ML_ERROR, "Recovery required on readonly "
616 "file system, but write access is "
620 goto read_super_error;
623 ocfs2_set_ro_flag(osb, 1);
625 printk(KERN_NOTICE "Readonly device detected. No cluster "
626 "services will be utilized for this mount. Recovery "
627 "will be skipped.\n");
630 if (!ocfs2_is_hard_readonly(osb)) {
631 if (sb->s_flags & MS_RDONLY)
632 ocfs2_set_ro_flag(osb, 0);
635 status = ocfs2_verify_heartbeat(osb);
638 goto read_super_error;
641 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
643 if (!osb->osb_debug_root) {
645 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
646 goto read_super_error;
649 status = ocfs2_mount_volume(sb);
651 inode = igrab(osb->root_inode);
654 goto read_super_error;
659 goto read_super_error;
662 root = d_alloc_root(inode);
666 goto read_super_error;
671 ocfs2_complete_mount_recovery(osb);
673 if (ocfs2_mount_local(osb))
674 snprintf(nodestr, sizeof(nodestr), "local");
676 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
678 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
679 "with %s data mode.\n",
680 osb->dev_str, nodestr, osb->slot_num,
681 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
684 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
685 wake_up(&osb->osb_mount_event);
698 atomic_set(&osb->vol_state, VOLUME_DISABLED);
699 wake_up(&osb->osb_mount_event);
700 ocfs2_dismount_volume(sb, 1);
707 static int ocfs2_get_sb(struct file_system_type *fs_type,
709 const char *dev_name,
711 struct vfsmount *mnt)
713 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
717 static struct file_system_type ocfs2_fs_type = {
718 .owner = THIS_MODULE,
720 .get_sb = ocfs2_get_sb, /* is this called when we mount
722 .kill_sb = kill_block_super, /* set to the generic one
723 * right now, but do we
724 * need to change that? */
725 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
729 static int ocfs2_parse_options(struct super_block *sb,
731 unsigned long *mount_opt,
738 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
739 options ? options : "(none)");
742 *slot = OCFS2_INVALID_SLOT;
749 while ((p = strsep(&options, ",")) != NULL) {
751 substring_t args[MAX_OPT_ARGS];
752 struct ocfs2_super * osb = OCFS2_SB(sb);
757 token = match_token(p, tokens, args);
760 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
763 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
766 if (match_int(&args[0], &option)) {
771 *mount_opt |= OCFS2_MOUNT_BARRIER;
773 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
776 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
779 *mount_opt |= OCFS2_MOUNT_NOINTR;
782 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
785 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
787 case Opt_data_ordered:
788 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
790 case Opt_data_writeback:
791 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
793 case Opt_atime_quantum:
794 if (match_int(&args[0], &option)) {
799 osb->s_atime_quantum = option;
801 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
805 if (match_int(&args[0], &option)) {
814 "Unrecognized mount option \"%s\" "
815 "or missing value\n", p);
828 static int __init ocfs2_init(void)
834 ocfs2_print_version();
836 status = init_ocfs2_uptodate_cache();
842 status = ocfs2_initialize_mem_caches();
848 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
854 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
855 if (!ocfs2_debugfs_root) {
857 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
862 ocfs2_free_mem_caches();
863 exit_ocfs2_uptodate_cache();
869 return register_filesystem(&ocfs2_fs_type);
874 static void __exit ocfs2_exit(void)
879 flush_workqueue(ocfs2_wq);
880 destroy_workqueue(ocfs2_wq);
883 debugfs_remove(ocfs2_debugfs_root);
885 ocfs2_free_mem_caches();
887 unregister_filesystem(&ocfs2_fs_type);
889 exit_ocfs2_uptodate_cache();
894 static void ocfs2_put_super(struct super_block *sb)
896 mlog_entry("(0x%p)\n", sb);
898 ocfs2_sync_blockdev(sb);
899 ocfs2_dismount_volume(sb, 0);
904 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
906 struct ocfs2_super *osb;
907 u32 numbits, freebits;
909 struct ocfs2_dinode *bm_lock;
910 struct buffer_head *bh = NULL;
911 struct inode *inode = NULL;
913 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
915 osb = OCFS2_SB(dentry->d_sb);
917 inode = ocfs2_get_system_file_inode(osb,
918 GLOBAL_BITMAP_SYSTEM_INODE,
921 mlog(ML_ERROR, "failed to get bitmap inode\n");
926 status = ocfs2_meta_lock(inode, &bh, 0);
932 bm_lock = (struct ocfs2_dinode *) bh->b_data;
934 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
935 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
937 buf->f_type = OCFS2_SUPER_MAGIC;
938 buf->f_bsize = dentry->d_sb->s_blocksize;
939 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
940 buf->f_blocks = ((sector_t) numbits) *
941 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
942 buf->f_bfree = ((sector_t) freebits) *
943 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
944 buf->f_bavail = buf->f_bfree;
945 buf->f_files = numbits;
946 buf->f_ffree = freebits;
950 ocfs2_meta_unlock(inode, 0);
961 static void ocfs2_inode_init_once(void *data,
962 struct kmem_cache *cachep,
965 struct ocfs2_inode_info *oi = data;
968 oi->ip_open_count = 0;
969 spin_lock_init(&oi->ip_lock);
970 ocfs2_extent_map_init(&oi->vfs_inode);
971 INIT_LIST_HEAD(&oi->ip_io_markers);
972 oi->ip_created_trans = 0;
973 oi->ip_last_trans = 0;
974 oi->ip_dir_start_lookup = 0;
976 init_rwsem(&oi->ip_alloc_sem);
977 mutex_init(&oi->ip_io_mutex);
982 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
983 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
984 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
985 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
987 ocfs2_metadata_cache_init(&oi->vfs_inode);
989 inode_init_once(&oi->vfs_inode);
992 static int ocfs2_initialize_mem_caches(void)
994 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
995 sizeof(struct ocfs2_inode_info),
997 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
999 ocfs2_inode_init_once);
1000 if (!ocfs2_inode_cachep)
1006 static void ocfs2_free_mem_caches(void)
1008 if (ocfs2_inode_cachep)
1009 kmem_cache_destroy(ocfs2_inode_cachep);
1011 ocfs2_inode_cachep = NULL;
1014 static int ocfs2_get_sector(struct super_block *sb,
1015 struct buffer_head **bh,
1019 if (!sb_set_blocksize(sb, sect_size)) {
1020 mlog(ML_ERROR, "unable to set blocksize\n");
1024 *bh = sb_getblk(sb, block);
1030 if (!buffer_dirty(*bh))
1031 clear_buffer_uptodate(*bh);
1033 ll_rw_block(READ, 1, bh);
1034 wait_on_buffer(*bh);
1038 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1039 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1043 /* XXX hold a ref on the node while mounte? easy enough, if
1045 if (ocfs2_mount_local(osb))
1048 osb->node_num = o2nm_this_node();
1050 if (osb->node_num == O2NM_MAX_NODES) {
1051 mlog(ML_ERROR, "could not find this host's node number\n");
1056 mlog(0, "I am node %d\n", osb->node_num);
1063 static int ocfs2_mount_volume(struct super_block *sb)
1066 int unlock_super = 0;
1067 struct ocfs2_super *osb = OCFS2_SB(sb);
1071 if (ocfs2_is_hard_readonly(osb))
1074 status = ocfs2_fill_local_node_info(osb);
1080 status = ocfs2_register_hb_callbacks(osb);
1086 status = ocfs2_dlm_init(osb);
1092 /* requires vote_thread to be running. */
1093 status = ocfs2_register_net_handlers(osb);
1099 status = ocfs2_super_lock(osb, 1);
1106 /* This will load up the node map and add ourselves to it. */
1107 status = ocfs2_find_slot(osb);
1113 ocfs2_populate_mounted_map(osb);
1115 /* load all node-local system inodes */
1116 status = ocfs2_init_local_system_inodes(osb);
1122 status = ocfs2_check_volume(osb);
1128 status = ocfs2_truncate_log_init(osb);
1134 if (ocfs2_mount_local(osb))
1137 /* This should be sent *after* we recovered our journal as it
1138 * will cause other nodes to unmark us as needing
1139 * recovery. However, we need to send it *before* dropping the
1140 * super block lock as otherwise their recovery threads might
1141 * try to clean us up while we're live! */
1142 status = ocfs2_request_mount_vote(osb);
1148 ocfs2_super_unlock(osb, 1);
1154 /* we can't grab the goofy sem lock from inside wait_event, so we use
1155 * memory barriers to make sure that we'll see the null task before
1157 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1160 return osb->recovery_thread_task != NULL;
1163 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1166 struct ocfs2_super *osb = NULL;
1169 mlog_entry("(0x%p)\n", sb);
1175 ocfs2_shutdown_local_alloc(osb);
1177 ocfs2_truncate_log_shutdown(osb);
1179 /* disable any new recovery threads and wait for any currently
1180 * running ones to exit. Do this before setting the vol_state. */
1181 mutex_lock(&osb->recovery_lock);
1182 osb->disable_recovery = 1;
1183 mutex_unlock(&osb->recovery_lock);
1184 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1186 /* At this point, we know that no more recovery threads can be
1187 * launched, so wait for any recovery completion work to
1189 flush_workqueue(ocfs2_wq);
1191 ocfs2_journal_shutdown(osb);
1193 ocfs2_sync_blockdev(sb);
1195 /* No dlm means we've failed during mount, so skip all the
1196 * steps which depended on that to complete. */
1198 tmp = ocfs2_super_lock(osb, 1);
1204 tmp = ocfs2_request_umount_vote(osb);
1208 if (osb->slot_num != OCFS2_INVALID_SLOT)
1209 ocfs2_put_slot(osb);
1211 ocfs2_super_unlock(osb, 1);
1214 ocfs2_release_system_inodes(osb);
1217 ocfs2_unregister_net_handlers(osb);
1219 ocfs2_dlm_shutdown(osb);
1222 ocfs2_clear_hb_callbacks(osb);
1224 debugfs_remove(osb->osb_debug_root);
1227 ocfs2_stop_heartbeat(osb);
1229 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1231 if (ocfs2_mount_local(osb))
1232 snprintf(nodestr, sizeof(nodestr), "local");
1234 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1236 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1237 osb->dev_str, nodestr);
1239 ocfs2_delete_osb(osb);
1242 sb->s_fs_info = NULL;
1245 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1246 unsigned uuid_bytes)
1251 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1253 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1254 if (osb->uuid_str == NULL)
1257 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1258 /* print with null */
1259 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1260 if (ret != 2) /* drop super cleans up */
1262 /* then only advance past the last char */
1269 static int ocfs2_initialize_super(struct super_block *sb,
1270 struct buffer_head *bh,
1274 int i, cbits, bbits;
1275 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1276 struct inode *inode = NULL;
1277 struct buffer_head *bitmap_bh = NULL;
1278 struct ocfs2_journal *journal;
1279 __le32 uuid_net_key;
1280 struct ocfs2_super *osb;
1284 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1291 sb->s_fs_info = osb;
1292 sb->s_op = &ocfs2_sops;
1293 sb->s_export_op = &ocfs2_export_ops;
1294 sb->s_time_gran = 1;
1295 sb->s_flags |= MS_NOATIME;
1296 /* this is needed to support O_LARGEFILE */
1297 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1298 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1299 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1302 /* Save off for ocfs2_rw_direct */
1303 osb->s_sectsize_bits = blksize_bits(sector_size);
1304 BUG_ON(!osb->s_sectsize_bits);
1306 osb->net_response_ids = 0;
1307 spin_lock_init(&osb->net_response_lock);
1308 INIT_LIST_HEAD(&osb->net_response_list);
1310 INIT_LIST_HEAD(&osb->osb_net_handlers);
1311 init_waitqueue_head(&osb->recovery_event);
1312 spin_lock_init(&osb->vote_task_lock);
1313 init_waitqueue_head(&osb->vote_event);
1314 osb->vote_work_sequence = 0;
1315 osb->vote_wake_sequence = 0;
1316 INIT_LIST_HEAD(&osb->blocked_lock_list);
1317 osb->blocked_lock_count = 0;
1318 INIT_LIST_HEAD(&osb->vote_list);
1319 spin_lock_init(&osb->osb_lock);
1321 atomic_set(&osb->alloc_stats.moves, 0);
1322 atomic_set(&osb->alloc_stats.local_data, 0);
1323 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1324 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1325 atomic_set(&osb->alloc_stats.bg_extends, 0);
1327 ocfs2_init_node_maps(osb);
1329 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1330 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1332 mutex_init(&osb->recovery_lock);
1334 osb->disable_recovery = 0;
1335 osb->recovery_thread_task = NULL;
1337 init_waitqueue_head(&osb->checkpoint_event);
1338 atomic_set(&osb->needs_checkpoint, 0);
1340 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1342 osb->node_num = O2NM_INVALID_NODE_NUM;
1343 osb->slot_num = OCFS2_INVALID_SLOT;
1345 osb->local_alloc_state = OCFS2_LA_UNUSED;
1346 osb->local_alloc_bh = NULL;
1348 ocfs2_setup_hb_callbacks(osb);
1350 init_waitqueue_head(&osb->osb_mount_event);
1352 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1353 if (!osb->vol_label) {
1354 mlog(ML_ERROR, "unable to alloc vol label\n");
1359 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1360 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1361 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1366 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1368 init_waitqueue_head(&osb->osb_wipe_event);
1369 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1370 sizeof(*osb->osb_orphan_wipes),
1372 if (!osb->osb_orphan_wipes) {
1378 osb->s_feature_compat =
1379 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1380 osb->s_feature_ro_compat =
1381 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1382 osb->s_feature_incompat =
1383 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1385 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1386 mlog(ML_ERROR, "couldn't mount because of unsupported "
1387 "optional features (%x).\n", i);
1391 if (!(osb->sb->s_flags & MS_RDONLY) &&
1392 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1393 mlog(ML_ERROR, "couldn't mount RDWR because of "
1394 "unsupported optional features (%x).\n", i);
1399 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1402 * This should be done in ocfs2_journal_init(), but unknown
1403 * ordering issues will cause the filesystem to crash.
1404 * If anyone wants to figure out what part of the code
1405 * refers to osb->journal before ocfs2_journal_init() is run,
1408 /* initialize our journal structure */
1410 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1412 mlog(ML_ERROR, "unable to alloc journal\n");
1416 osb->journal = journal;
1417 journal->j_osb = osb;
1419 atomic_set(&journal->j_num_trans, 0);
1420 init_rwsem(&journal->j_trans_barrier);
1421 init_waitqueue_head(&journal->j_checkpointed);
1422 spin_lock_init(&journal->j_lock);
1423 journal->j_trans_id = (unsigned long) 1;
1424 INIT_LIST_HEAD(&journal->j_la_cleanups);
1425 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1426 journal->j_state = OCFS2_JOURNAL_FREE;
1428 /* get some pseudo constants for clustersize bits */
1429 osb->s_clustersize_bits =
1430 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1431 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1432 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1434 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1435 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1436 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1437 osb->s_clustersize);
1442 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1444 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1445 "what jbd can address in 32 bits.\n");
1450 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1451 sizeof(di->id2.i_super.s_uuid))) {
1452 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1457 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1458 osb->net_key = le32_to_cpu(uuid_net_key);
1460 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1461 osb->vol_label[63] = '\0';
1462 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1463 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1464 osb->first_cluster_group_blkno =
1465 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1466 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1467 mlog(0, "vol_label: %s\n", osb->vol_label);
1468 mlog(0, "uuid: %s\n", osb->uuid_str);
1469 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1470 (unsigned long long)osb->root_blkno,
1471 (unsigned long long)osb->system_dir_blkno);
1473 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1474 if (!osb->osb_dlm_debug) {
1480 atomic_set(&osb->vol_state, VOLUME_INIT);
1482 /* load root, system_dir, and all global system inodes */
1483 status = ocfs2_init_global_system_inodes(osb);
1492 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1493 OCFS2_INVALID_SLOT);
1500 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1502 /* We don't have a cluster lock on the bitmap here because
1503 * we're only interested in static information and the extra
1504 * complexity at mount time isn't worht it. Don't pass the
1505 * inode in to the read function though as we don't want it to
1506 * be put in the cache. */
1507 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1515 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1516 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1518 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1519 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1521 status = ocfs2_init_slot_info(osb);
1533 * will return: -EAGAIN if it is ok to keep searching for superblocks
1534 * -EINVAL if there is a bad superblock
1537 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1538 struct buffer_head *bh,
1541 int status = -EAGAIN;
1545 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1546 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1548 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1549 mlog(ML_ERROR, "found superblock with incorrect block "
1550 "size: found %u, should be %u\n",
1551 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1553 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1554 OCFS2_MAJOR_REV_LEVEL ||
1555 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1556 OCFS2_MINOR_REV_LEVEL) {
1557 mlog(ML_ERROR, "found superblock with bad version: "
1558 "found %u.%u, should be %u.%u\n",
1559 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1560 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1561 OCFS2_MAJOR_REV_LEVEL,
1562 OCFS2_MINOR_REV_LEVEL);
1563 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1564 mlog(ML_ERROR, "bad block number on superblock: "
1565 "found %llu, should be %llu\n",
1566 (unsigned long long)le64_to_cpu(di->i_blkno),
1567 (unsigned long long)bh->b_blocknr);
1568 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1569 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1570 mlog(ML_ERROR, "bad cluster size found: %u\n",
1571 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1572 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1573 mlog(ML_ERROR, "bad root_blkno: 0\n");
1574 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1575 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1576 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1578 "Superblock slots found greater than file system "
1579 "maximum: found %u, max %u\n",
1580 le16_to_cpu(di->id2.i_super.s_max_slots),
1592 static int ocfs2_check_volume(struct ocfs2_super *osb)
1597 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1603 /* Init our journal object. */
1604 status = ocfs2_journal_init(osb->journal, &dirty);
1606 mlog(ML_ERROR, "Could not initialize journal!\n");
1610 /* If the journal was unmounted cleanly then we don't want to
1611 * recover anything. Otherwise, journal_load will do that
1612 * dirty work for us :) */
1614 status = ocfs2_journal_wipe(osb->journal, 0);
1620 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1621 "recovering volume.\n");
1624 local = ocfs2_mount_local(osb);
1626 /* will play back anything left in the journal. */
1627 ocfs2_journal_load(osb->journal, local);
1630 /* recover my local alloc if we didn't unmount cleanly. */
1631 status = ocfs2_begin_local_alloc_recovery(osb,
1638 /* we complete the recovery process after we've marked
1639 * ourselves as mounted. */
1642 mlog(0, "Journal loaded.\n");
1644 status = ocfs2_load_local_alloc(osb);
1651 /* Recovery will be completed after we've mounted the
1652 * rest of the volume. */
1654 osb->local_alloc_copy = local_alloc;
1658 /* go through each journal, trylock it and if you get the
1659 * lock, and it's marked as dirty, set the bit in the recover
1660 * map and launch a recovery thread for it. */
1661 status = ocfs2_mark_dead_nodes(osb);
1674 * The routine gets called from dismount or close whenever a dismount on
1675 * volume is requested and the osb open count becomes 1.
1676 * It will remove the osb from the global list and also free up all the
1677 * initialized resources and fileobject.
1679 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1683 /* This function assumes that the caller has the main osb resource */
1686 ocfs2_free_slot_info(osb->slot_info);
1688 kfree(osb->osb_orphan_wipes);
1690 * This belongs in journal shutdown, but because we have to
1691 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1694 kfree(osb->journal);
1695 if (osb->local_alloc_copy)
1696 kfree(osb->local_alloc_copy);
1697 kfree(osb->uuid_str);
1698 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1699 memset(osb, 0, sizeof(struct ocfs2_super));
1704 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1705 * panic(). We do not support continue-on-error operation. */
1706 static void ocfs2_handle_error(struct super_block *sb)
1708 struct ocfs2_super *osb = OCFS2_SB(sb);
1710 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1711 panic("OCFS2: (device %s): panic forced after error\n",
1714 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1716 if (sb->s_flags & MS_RDONLY &&
1717 (ocfs2_is_soft_readonly(osb) ||
1718 ocfs2_is_hard_readonly(osb)))
1721 printk(KERN_CRIT "File system is now read-only due to the potential "
1722 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1723 "system is unmounted.\n");
1724 sb->s_flags |= MS_RDONLY;
1725 ocfs2_set_ro_flag(osb, 0);
1728 static char error_buf[1024];
1730 void __ocfs2_error(struct super_block *sb,
1731 const char *function,
1732 const char *fmt, ...)
1736 va_start(args, fmt);
1737 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1740 /* Not using mlog here because we want to show the actual
1741 * function the error came from. */
1742 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1743 sb->s_id, function, error_buf);
1745 ocfs2_handle_error(sb);
1748 /* Handle critical errors. This is intentionally more drastic than
1749 * ocfs2_handle_error, so we only use for things like journal errors,
1751 void __ocfs2_abort(struct super_block* sb,
1752 const char *function,
1753 const char *fmt, ...)
1757 va_start(args, fmt);
1758 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1761 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1762 sb->s_id, function, error_buf);
1764 /* We don't have the cluster support yet to go straight to
1765 * hard readonly in here. Until then, we want to keep
1766 * ocfs2_abort() so that we can at least mark critical
1769 * TODO: This should abort the journal and alert other nodes
1770 * that our slot needs recovery. */
1772 /* Force a panic(). This stinks, but it's better than letting
1773 * things continue without having a proper hard readonly
1775 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1776 ocfs2_handle_error(sb);
1779 module_init(ocfs2_init);
1780 module_exit(ocfs2_exit);