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");
86 unsigned long mount_opt;
87 unsigned int atime_quantum;
91 static int ocfs2_parse_options(struct super_block *sb, char *options,
92 struct mount_options *mopt,
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);
102 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
104 static int ocfs2_sync_fs(struct super_block *sb, int wait);
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,
114 static int ocfs2_initialize_super(struct super_block *sb,
115 struct buffer_head *bh,
117 static int ocfs2_get_sector(struct super_block *sb,
118 struct buffer_head **bh,
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);
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,
153 static match_table_t tokens = {
154 {Opt_barrier, "barrier=%u"},
155 {Opt_err_panic, "errors=panic"},
156 {Opt_err_ro, "errors=remount-ro"},
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"},
169 * write_super and sync_fs ripped right out of ext3.
171 static void ocfs2_write_super(struct super_block *sb)
173 if (mutex_trylock(&sb->s_lock) != 0)
178 static int ocfs2_sync_fs(struct super_block *sb, int wait)
182 struct ocfs2_super *osb = OCFS2_SB(sb);
186 if (ocfs2_is_hard_readonly(osb))
190 status = ocfs2_flush_truncate_log(osb);
194 ocfs2_schedule_truncate_log_flush(osb, 0);
197 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
199 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
205 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
207 struct inode *new = NULL;
213 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
215 status = PTR_ERR(new);
219 osb->root_inode = new;
221 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
223 status = PTR_ERR(new);
227 osb->sys_root_inode = new;
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);
233 ocfs2_release_system_inodes(osb);
236 /* FIXME: Should ERROR_RO_FS */
237 mlog(ML_ERROR, "Unable to load system inode %d, "
238 "possibly corrupt fs?", i);
241 // the array now has one ref, so drop this one
250 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
252 struct inode *new = NULL;
258 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
259 i < NUM_SYSTEM_INODES;
261 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
263 ocfs2_release_system_inodes(osb);
265 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
266 status, i, osb->slot_num);
269 /* the array now has one ref, so drop this one */
278 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
285 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
286 inode = osb->system_inodes[i];
289 osb->system_inodes[i] = NULL;
293 inode = osb->sys_root_inode;
296 osb->sys_root_inode = NULL;
299 inode = osb->root_inode;
302 osb->root_inode = NULL;
309 /* We're allocating fs objects, use GFP_NOFS */
310 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
312 struct ocfs2_inode_info *oi;
314 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
318 return &oi->vfs_inode;
321 static void ocfs2_destroy_inode(struct inode *inode)
323 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
326 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
329 unsigned int bytes = 1 << cbits;
330 unsigned int trim = bytes;
331 unsigned int bitshift = 32;
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.
339 #if BITS_PER_LONG == 32
340 # if defined(CONFIG_LBD)
341 BUILD_BUG_ON(sizeof(sector_t) != 8);
343 * We might be limited by page cache size.
345 if (bytes > PAGE_CACHE_SIZE) {
346 bytes = PAGE_CACHE_SIZE;
349 * Shift by 31 here so that we don't get larger than
356 * We are limited by the size of sector_t. Use block size, as
357 * that's what we expose to the VFS.
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.
370 return (((unsigned long long)bytes) << bitshift) - trim;
373 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
375 int incompat_features;
377 struct mount_options parsed_options;
378 struct ocfs2_super *osb = OCFS2_SB(sb);
380 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
385 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
386 (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
388 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
392 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
393 (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
395 mlog(ML_ERROR, "Cannot change data mode on remount\n");
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");
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;
415 mlog(0, "Making ro filesystem writeable.\n");
417 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
418 mlog(ML_ERROR, "Cannot remount RDWR "
419 "filesystem due to previous errors.\n");
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);
431 sb->s_flags &= ~MS_RDONLY;
432 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
435 spin_unlock(&osb->osb_lock);
439 if (!ocfs2_is_hard_readonly(osb))
440 ocfs2_set_journal_params(osb);
442 /* Only save off the new mount options in case of a successful
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;
452 static int ocfs2_sb_probe(struct super_block *sb,
453 struct buffer_head **bh,
456 int status = 0, tmpstat;
457 struct ocfs1_vol_disk_hdr *hdr;
458 struct ocfs2_dinode *di;
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);
472 /* Can this really happen? */
473 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
474 *sector_size = OCFS2_MIN_BLOCKSIZE;
476 /* check block zero for old format */
477 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
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);
488 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
489 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
490 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
497 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
498 "upgraded before mounting with ocfs v2\n");
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.
508 for (blksize = *sector_size;
509 blksize <= OCFS2_MAX_BLOCKSIZE;
511 tmpstat = ocfs2_get_sector(sb, bh,
512 OCFS2_SUPER_BLOCK_BLKNO,
519 di = (struct ocfs2_dinode *) (*bh)->b_data;
520 status = ocfs2_verify_volume(di, *bh, blksize);
525 if (status != -EAGAIN)
533 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
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");
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");
554 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
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;
564 mlog_entry("%p, %p, %i", sb, data, silent);
566 if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
568 goto read_super_error;
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()) {
576 goto read_super_error;
580 /* probe for superblock */
581 status = ocfs2_sb_probe(sb, &bh, §or_size);
583 mlog(ML_ERROR, "superblock probe failed!\n");
584 goto read_super_error;
587 status = ocfs2_initialize_super(sb, bh, sector_size);
591 goto read_super_error;
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;
599 sb->s_magic = OCFS2_SUPER_MAGIC;
601 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
603 if (bdev_read_only(sb->s_bdev)) {
604 if (!(sb->s_flags & MS_RDONLY)) {
606 mlog(ML_ERROR, "Readonly device detected but readonly "
607 "mount was not specified.\n");
608 goto read_super_error;
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) {
615 mlog(ML_ERROR, "Local heartbeat specified on readonly "
617 goto read_super_error;
620 status = ocfs2_check_journals_nolocks(osb);
622 if (status == -EROFS)
623 mlog(ML_ERROR, "Recovery required on readonly "
624 "file system, but write access is "
628 goto read_super_error;
631 ocfs2_set_ro_flag(osb, 1);
633 printk(KERN_NOTICE "Readonly device detected. No cluster "
634 "services will be utilized for this mount. Recovery "
635 "will be skipped.\n");
638 if (!ocfs2_is_hard_readonly(osb)) {
639 if (sb->s_flags & MS_RDONLY)
640 ocfs2_set_ro_flag(osb, 0);
643 status = ocfs2_verify_heartbeat(osb);
646 goto read_super_error;
649 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
651 if (!osb->osb_debug_root) {
653 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
654 goto read_super_error;
657 status = ocfs2_mount_volume(sb);
659 inode = igrab(osb->root_inode);
662 goto read_super_error;
667 goto read_super_error;
670 root = d_alloc_root(inode);
674 goto read_super_error;
679 ocfs2_complete_mount_recovery(osb);
681 if (ocfs2_mount_local(osb))
682 snprintf(nodestr, sizeof(nodestr), "local");
684 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
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" :
692 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
693 wake_up(&osb->osb_mount_event);
706 atomic_set(&osb->vol_state, VOLUME_DISABLED);
707 wake_up(&osb->osb_mount_event);
708 ocfs2_dismount_volume(sb, 1);
715 static int ocfs2_get_sb(struct file_system_type *fs_type,
717 const char *dev_name,
719 struct vfsmount *mnt)
721 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
725 static struct file_system_type ocfs2_fs_type = {
726 .owner = THIS_MODULE,
728 .get_sb = ocfs2_get_sb, /* is this called when we mount
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,
737 static int ocfs2_parse_options(struct super_block *sb,
739 struct mount_options *mopt,
745 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
746 options ? options : "(none)");
749 mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
750 mopt->slot = OCFS2_INVALID_SLOT;
757 while ((p = strsep(&options, ",")) != NULL) {
759 substring_t args[MAX_OPT_ARGS];
764 token = match_token(p, tokens, args);
767 mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
770 mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
773 if (match_int(&args[0], &option)) {
778 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
780 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
783 mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
786 mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
789 mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
792 mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
794 case Opt_data_ordered:
795 mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
797 case Opt_data_writeback:
798 mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
800 case Opt_atime_quantum:
801 if (match_int(&args[0], &option)) {
806 mopt->atime_quantum = option;
810 if (match_int(&args[0], &option)) {
815 mopt->slot = (s16)option;
819 "Unrecognized mount option \"%s\" "
820 "or missing value\n", p);
833 static int __init ocfs2_init(void)
839 ocfs2_print_version();
841 status = init_ocfs2_uptodate_cache();
847 status = ocfs2_initialize_mem_caches();
853 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
859 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
860 if (!ocfs2_debugfs_root) {
862 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
867 ocfs2_free_mem_caches();
868 exit_ocfs2_uptodate_cache();
874 return register_filesystem(&ocfs2_fs_type);
879 static void __exit ocfs2_exit(void)
884 flush_workqueue(ocfs2_wq);
885 destroy_workqueue(ocfs2_wq);
888 debugfs_remove(ocfs2_debugfs_root);
890 ocfs2_free_mem_caches();
892 unregister_filesystem(&ocfs2_fs_type);
894 exit_ocfs2_uptodate_cache();
899 static void ocfs2_put_super(struct super_block *sb)
901 mlog_entry("(0x%p)\n", sb);
903 ocfs2_sync_blockdev(sb);
904 ocfs2_dismount_volume(sb, 0);
909 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
911 struct ocfs2_super *osb;
912 u32 numbits, freebits;
914 struct ocfs2_dinode *bm_lock;
915 struct buffer_head *bh = NULL;
916 struct inode *inode = NULL;
918 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
920 osb = OCFS2_SB(dentry->d_sb);
922 inode = ocfs2_get_system_file_inode(osb,
923 GLOBAL_BITMAP_SYSTEM_INODE,
926 mlog(ML_ERROR, "failed to get bitmap inode\n");
931 status = ocfs2_meta_lock(inode, &bh, 0);
937 bm_lock = (struct ocfs2_dinode *) bh->b_data;
939 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
940 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
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;
955 ocfs2_meta_unlock(inode, 0);
966 static void ocfs2_inode_init_once(void *data,
967 struct kmem_cache *cachep,
970 struct ocfs2_inode_info *oi = data;
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;
981 init_rwsem(&oi->ip_alloc_sem);
982 mutex_init(&oi->ip_io_mutex);
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);
992 ocfs2_metadata_cache_init(&oi->vfs_inode);
994 inode_init_once(&oi->vfs_inode);
997 static int ocfs2_initialize_mem_caches(void)
999 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1000 sizeof(struct ocfs2_inode_info),
1002 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1004 ocfs2_inode_init_once);
1005 if (!ocfs2_inode_cachep)
1011 static void ocfs2_free_mem_caches(void)
1013 if (ocfs2_inode_cachep)
1014 kmem_cache_destroy(ocfs2_inode_cachep);
1016 ocfs2_inode_cachep = NULL;
1019 static int ocfs2_get_sector(struct super_block *sb,
1020 struct buffer_head **bh,
1024 if (!sb_set_blocksize(sb, sect_size)) {
1025 mlog(ML_ERROR, "unable to set blocksize\n");
1029 *bh = sb_getblk(sb, block);
1035 if (!buffer_dirty(*bh))
1036 clear_buffer_uptodate(*bh);
1038 ll_rw_block(READ, 1, bh);
1039 wait_on_buffer(*bh);
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)
1048 /* XXX hold a ref on the node while mounte? easy enough, if
1050 if (ocfs2_mount_local(osb))
1053 osb->node_num = o2nm_this_node();
1055 if (osb->node_num == O2NM_MAX_NODES) {
1056 mlog(ML_ERROR, "could not find this host's node number\n");
1061 mlog(0, "I am node %d\n", osb->node_num);
1068 static int ocfs2_mount_volume(struct super_block *sb)
1071 int unlock_super = 0;
1072 struct ocfs2_super *osb = OCFS2_SB(sb);
1076 if (ocfs2_is_hard_readonly(osb))
1079 status = ocfs2_fill_local_node_info(osb);
1085 status = ocfs2_register_hb_callbacks(osb);
1091 status = ocfs2_dlm_init(osb);
1097 /* requires vote_thread to be running. */
1098 status = ocfs2_register_net_handlers(osb);
1104 status = ocfs2_super_lock(osb, 1);
1111 /* This will load up the node map and add ourselves to it. */
1112 status = ocfs2_find_slot(osb);
1118 ocfs2_populate_mounted_map(osb);
1120 /* load all node-local system inodes */
1121 status = ocfs2_init_local_system_inodes(osb);
1127 status = ocfs2_check_volume(osb);
1133 status = ocfs2_truncate_log_init(osb);
1139 if (ocfs2_mount_local(osb))
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);
1153 ocfs2_super_unlock(osb, 1);
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
1162 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1165 return osb->recovery_thread_task != NULL;
1168 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1171 struct ocfs2_super *osb = NULL;
1174 mlog_entry("(0x%p)\n", sb);
1180 ocfs2_shutdown_local_alloc(osb);
1182 ocfs2_truncate_log_shutdown(osb);
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));
1191 /* At this point, we know that no more recovery threads can be
1192 * launched, so wait for any recovery completion work to
1194 flush_workqueue(ocfs2_wq);
1196 ocfs2_journal_shutdown(osb);
1198 ocfs2_sync_blockdev(sb);
1200 /* No dlm means we've failed during mount, so skip all the
1201 * steps which depended on that to complete. */
1203 tmp = ocfs2_super_lock(osb, 1);
1209 tmp = ocfs2_request_umount_vote(osb);
1213 if (osb->slot_num != OCFS2_INVALID_SLOT)
1214 ocfs2_put_slot(osb);
1216 ocfs2_super_unlock(osb, 1);
1219 ocfs2_release_system_inodes(osb);
1222 ocfs2_unregister_net_handlers(osb);
1224 ocfs2_dlm_shutdown(osb);
1227 ocfs2_clear_hb_callbacks(osb);
1229 debugfs_remove(osb->osb_debug_root);
1232 ocfs2_stop_heartbeat(osb);
1234 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1236 if (ocfs2_mount_local(osb))
1237 snprintf(nodestr, sizeof(nodestr), "local");
1239 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1241 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1242 osb->dev_str, nodestr);
1244 ocfs2_delete_osb(osb);
1247 sb->s_fs_info = NULL;
1250 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1251 unsigned uuid_bytes)
1256 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1258 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1259 if (osb->uuid_str == NULL)
1262 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1263 /* print with null */
1264 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1265 if (ret != 2) /* drop super cleans up */
1267 /* then only advance past the last char */
1274 static int ocfs2_initialize_super(struct super_block *sb,
1275 struct buffer_head *bh,
1279 int i, cbits, bbits;
1280 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1281 struct inode *inode = NULL;
1282 struct buffer_head *bitmap_bh = NULL;
1283 struct ocfs2_journal *journal;
1284 __le32 uuid_net_key;
1285 struct ocfs2_super *osb;
1289 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1296 sb->s_fs_info = osb;
1297 sb->s_op = &ocfs2_sops;
1298 sb->s_export_op = &ocfs2_export_ops;
1299 sb->s_time_gran = 1;
1300 sb->s_flags |= MS_NOATIME;
1301 /* this is needed to support O_LARGEFILE */
1302 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1303 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1304 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1307 /* Save off for ocfs2_rw_direct */
1308 osb->s_sectsize_bits = blksize_bits(sector_size);
1309 BUG_ON(!osb->s_sectsize_bits);
1311 osb->net_response_ids = 0;
1312 spin_lock_init(&osb->net_response_lock);
1313 INIT_LIST_HEAD(&osb->net_response_list);
1315 INIT_LIST_HEAD(&osb->osb_net_handlers);
1316 init_waitqueue_head(&osb->recovery_event);
1317 spin_lock_init(&osb->vote_task_lock);
1318 init_waitqueue_head(&osb->vote_event);
1319 osb->vote_work_sequence = 0;
1320 osb->vote_wake_sequence = 0;
1321 INIT_LIST_HEAD(&osb->blocked_lock_list);
1322 osb->blocked_lock_count = 0;
1323 INIT_LIST_HEAD(&osb->vote_list);
1324 spin_lock_init(&osb->osb_lock);
1326 atomic_set(&osb->alloc_stats.moves, 0);
1327 atomic_set(&osb->alloc_stats.local_data, 0);
1328 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1329 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1330 atomic_set(&osb->alloc_stats.bg_extends, 0);
1332 ocfs2_init_node_maps(osb);
1334 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1335 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1337 mutex_init(&osb->recovery_lock);
1339 osb->disable_recovery = 0;
1340 osb->recovery_thread_task = NULL;
1342 init_waitqueue_head(&osb->checkpoint_event);
1343 atomic_set(&osb->needs_checkpoint, 0);
1345 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1347 osb->node_num = O2NM_INVALID_NODE_NUM;
1348 osb->slot_num = OCFS2_INVALID_SLOT;
1350 osb->local_alloc_state = OCFS2_LA_UNUSED;
1351 osb->local_alloc_bh = NULL;
1353 ocfs2_setup_hb_callbacks(osb);
1355 init_waitqueue_head(&osb->osb_mount_event);
1357 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1358 if (!osb->vol_label) {
1359 mlog(ML_ERROR, "unable to alloc vol label\n");
1364 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1365 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1366 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1371 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1373 init_waitqueue_head(&osb->osb_wipe_event);
1374 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1375 sizeof(*osb->osb_orphan_wipes),
1377 if (!osb->osb_orphan_wipes) {
1383 osb->s_feature_compat =
1384 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1385 osb->s_feature_ro_compat =
1386 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1387 osb->s_feature_incompat =
1388 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1390 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1391 mlog(ML_ERROR, "couldn't mount because of unsupported "
1392 "optional features (%x).\n", i);
1396 if (!(osb->sb->s_flags & MS_RDONLY) &&
1397 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1398 mlog(ML_ERROR, "couldn't mount RDWR because of "
1399 "unsupported optional features (%x).\n", i);
1404 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1407 * This should be done in ocfs2_journal_init(), but unknown
1408 * ordering issues will cause the filesystem to crash.
1409 * If anyone wants to figure out what part of the code
1410 * refers to osb->journal before ocfs2_journal_init() is run,
1413 /* initialize our journal structure */
1415 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1417 mlog(ML_ERROR, "unable to alloc journal\n");
1421 osb->journal = journal;
1422 journal->j_osb = osb;
1424 atomic_set(&journal->j_num_trans, 0);
1425 init_rwsem(&journal->j_trans_barrier);
1426 init_waitqueue_head(&journal->j_checkpointed);
1427 spin_lock_init(&journal->j_lock);
1428 journal->j_trans_id = (unsigned long) 1;
1429 INIT_LIST_HEAD(&journal->j_la_cleanups);
1430 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1431 journal->j_state = OCFS2_JOURNAL_FREE;
1433 /* get some pseudo constants for clustersize bits */
1434 osb->s_clustersize_bits =
1435 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1436 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1437 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1439 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1440 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1441 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1442 osb->s_clustersize);
1447 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1449 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1450 "what jbd can address in 32 bits.\n");
1455 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1456 sizeof(di->id2.i_super.s_uuid))) {
1457 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1462 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1463 osb->net_key = le32_to_cpu(uuid_net_key);
1465 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1466 osb->vol_label[63] = '\0';
1467 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1468 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1469 osb->first_cluster_group_blkno =
1470 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1471 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1472 mlog(0, "vol_label: %s\n", osb->vol_label);
1473 mlog(0, "uuid: %s\n", osb->uuid_str);
1474 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1475 (unsigned long long)osb->root_blkno,
1476 (unsigned long long)osb->system_dir_blkno);
1478 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1479 if (!osb->osb_dlm_debug) {
1485 atomic_set(&osb->vol_state, VOLUME_INIT);
1487 /* load root, system_dir, and all global system inodes */
1488 status = ocfs2_init_global_system_inodes(osb);
1497 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1498 OCFS2_INVALID_SLOT);
1505 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1507 /* We don't have a cluster lock on the bitmap here because
1508 * we're only interested in static information and the extra
1509 * complexity at mount time isn't worht it. Don't pass the
1510 * inode in to the read function though as we don't want it to
1511 * be put in the cache. */
1512 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1520 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1521 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1523 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1524 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1526 status = ocfs2_init_slot_info(osb);
1538 * will return: -EAGAIN if it is ok to keep searching for superblocks
1539 * -EINVAL if there is a bad superblock
1542 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1543 struct buffer_head *bh,
1546 int status = -EAGAIN;
1550 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1551 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1553 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1554 mlog(ML_ERROR, "found superblock with incorrect block "
1555 "size: found %u, should be %u\n",
1556 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1558 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1559 OCFS2_MAJOR_REV_LEVEL ||
1560 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1561 OCFS2_MINOR_REV_LEVEL) {
1562 mlog(ML_ERROR, "found superblock with bad version: "
1563 "found %u.%u, should be %u.%u\n",
1564 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1565 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1566 OCFS2_MAJOR_REV_LEVEL,
1567 OCFS2_MINOR_REV_LEVEL);
1568 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1569 mlog(ML_ERROR, "bad block number on superblock: "
1570 "found %llu, should be %llu\n",
1571 (unsigned long long)le64_to_cpu(di->i_blkno),
1572 (unsigned long long)bh->b_blocknr);
1573 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1574 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1575 mlog(ML_ERROR, "bad cluster size found: %u\n",
1576 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1577 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1578 mlog(ML_ERROR, "bad root_blkno: 0\n");
1579 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1580 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1581 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1583 "Superblock slots found greater than file system "
1584 "maximum: found %u, max %u\n",
1585 le16_to_cpu(di->id2.i_super.s_max_slots),
1597 static int ocfs2_check_volume(struct ocfs2_super *osb)
1602 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1608 /* Init our journal object. */
1609 status = ocfs2_journal_init(osb->journal, &dirty);
1611 mlog(ML_ERROR, "Could not initialize journal!\n");
1615 /* If the journal was unmounted cleanly then we don't want to
1616 * recover anything. Otherwise, journal_load will do that
1617 * dirty work for us :) */
1619 status = ocfs2_journal_wipe(osb->journal, 0);
1625 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1626 "recovering volume.\n");
1629 local = ocfs2_mount_local(osb);
1631 /* will play back anything left in the journal. */
1632 ocfs2_journal_load(osb->journal, local);
1635 /* recover my local alloc if we didn't unmount cleanly. */
1636 status = ocfs2_begin_local_alloc_recovery(osb,
1643 /* we complete the recovery process after we've marked
1644 * ourselves as mounted. */
1647 mlog(0, "Journal loaded.\n");
1649 status = ocfs2_load_local_alloc(osb);
1656 /* Recovery will be completed after we've mounted the
1657 * rest of the volume. */
1659 osb->local_alloc_copy = local_alloc;
1663 /* go through each journal, trylock it and if you get the
1664 * lock, and it's marked as dirty, set the bit in the recover
1665 * map and launch a recovery thread for it. */
1666 status = ocfs2_mark_dead_nodes(osb);
1679 * The routine gets called from dismount or close whenever a dismount on
1680 * volume is requested and the osb open count becomes 1.
1681 * It will remove the osb from the global list and also free up all the
1682 * initialized resources and fileobject.
1684 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1688 /* This function assumes that the caller has the main osb resource */
1691 ocfs2_free_slot_info(osb->slot_info);
1693 kfree(osb->osb_orphan_wipes);
1695 * This belongs in journal shutdown, but because we have to
1696 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1699 kfree(osb->journal);
1700 if (osb->local_alloc_copy)
1701 kfree(osb->local_alloc_copy);
1702 kfree(osb->uuid_str);
1703 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1704 memset(osb, 0, sizeof(struct ocfs2_super));
1709 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1710 * panic(). We do not support continue-on-error operation. */
1711 static void ocfs2_handle_error(struct super_block *sb)
1713 struct ocfs2_super *osb = OCFS2_SB(sb);
1715 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1716 panic("OCFS2: (device %s): panic forced after error\n",
1719 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1721 if (sb->s_flags & MS_RDONLY &&
1722 (ocfs2_is_soft_readonly(osb) ||
1723 ocfs2_is_hard_readonly(osb)))
1726 printk(KERN_CRIT "File system is now read-only due to the potential "
1727 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1728 "system is unmounted.\n");
1729 sb->s_flags |= MS_RDONLY;
1730 ocfs2_set_ro_flag(osb, 0);
1733 static char error_buf[1024];
1735 void __ocfs2_error(struct super_block *sb,
1736 const char *function,
1737 const char *fmt, ...)
1741 va_start(args, fmt);
1742 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1745 /* Not using mlog here because we want to show the actual
1746 * function the error came from. */
1747 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1748 sb->s_id, function, error_buf);
1750 ocfs2_handle_error(sb);
1753 /* Handle critical errors. This is intentionally more drastic than
1754 * ocfs2_handle_error, so we only use for things like journal errors,
1756 void __ocfs2_abort(struct super_block* sb,
1757 const char *function,
1758 const char *fmt, ...)
1762 va_start(args, fmt);
1763 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1766 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1767 sb->s_id, function, error_buf);
1769 /* We don't have the cluster support yet to go straight to
1770 * hard readonly in here. Until then, we want to keep
1771 * ocfs2_abort() so that we can at least mark critical
1774 * TODO: This should abort the journal and alert other nodes
1775 * that our slot needs recovery. */
1777 /* Force a panic(). This stinks, but it's better than letting
1778 * things continue without having a proper hard readonly
1780 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1781 ocfs2_handle_error(sb);
1784 module_init(ocfs2_init);
1785 module_exit(ocfs2_exit);