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>
42 #include <linux/mount.h>
43 #include <linux/seq_file.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"
68 #include "buffer_head_io.h"
70 static struct kmem_cache *ocfs2_inode_cachep = NULL;
72 /* OCFS2 needs to schedule several differnt types of work which
73 * require cluster locking, disk I/O, recovery waits, etc. Since these
74 * types of work tend to be heavy we avoid using the kernel events
75 * workqueue and schedule on our own. */
76 struct workqueue_struct *ocfs2_wq = NULL;
78 static struct dentry *ocfs2_debugfs_root = NULL;
80 MODULE_AUTHOR("Oracle");
81 MODULE_LICENSE("GPL");
85 unsigned long commit_interval;
86 unsigned long mount_opt;
87 unsigned int atime_quantum;
89 unsigned int localalloc_opt;
92 static int ocfs2_parse_options(struct super_block *sb, char *options,
93 struct mount_options *mopt,
95 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
96 static void ocfs2_put_super(struct super_block *sb);
97 static int ocfs2_mount_volume(struct super_block *sb);
98 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
99 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
100 static int ocfs2_initialize_mem_caches(void);
101 static void ocfs2_free_mem_caches(void);
102 static void ocfs2_delete_osb(struct ocfs2_super *osb);
104 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
106 static int ocfs2_sync_fs(struct super_block *sb, int wait);
108 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
109 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
110 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
111 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
112 static int ocfs2_check_volume(struct ocfs2_super *osb);
113 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
114 struct buffer_head *bh,
116 static int ocfs2_initialize_super(struct super_block *sb,
117 struct buffer_head *bh,
119 static int ocfs2_get_sector(struct super_block *sb,
120 struct buffer_head **bh,
123 static void ocfs2_write_super(struct super_block *sb);
124 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
125 static void ocfs2_destroy_inode(struct inode *inode);
127 static const struct super_operations ocfs2_sops = {
128 .statfs = ocfs2_statfs,
129 .alloc_inode = ocfs2_alloc_inode,
130 .destroy_inode = ocfs2_destroy_inode,
131 .drop_inode = ocfs2_drop_inode,
132 .clear_inode = ocfs2_clear_inode,
133 .delete_inode = ocfs2_delete_inode,
134 .sync_fs = ocfs2_sync_fs,
135 .write_super = ocfs2_write_super,
136 .put_super = ocfs2_put_super,
137 .remount_fs = ocfs2_remount,
138 .show_options = ocfs2_show_options,
159 static match_table_t tokens = {
160 {Opt_barrier, "barrier=%u"},
161 {Opt_err_panic, "errors=panic"},
162 {Opt_err_ro, "errors=remount-ro"},
164 {Opt_nointr, "nointr"},
165 {Opt_hb_none, OCFS2_HB_NONE},
166 {Opt_hb_local, OCFS2_HB_LOCAL},
167 {Opt_data_ordered, "data=ordered"},
168 {Opt_data_writeback, "data=writeback"},
169 {Opt_atime_quantum, "atime_quantum=%u"},
170 {Opt_slot, "preferred_slot=%u"},
171 {Opt_commit, "commit=%u"},
172 {Opt_localalloc, "localalloc=%d"},
173 {Opt_localflocks, "localflocks"},
178 * write_super and sync_fs ripped right out of ext3.
180 static void ocfs2_write_super(struct super_block *sb)
182 if (mutex_trylock(&sb->s_lock) != 0)
187 static int ocfs2_sync_fs(struct super_block *sb, int wait)
191 struct ocfs2_super *osb = OCFS2_SB(sb);
195 if (ocfs2_is_hard_readonly(osb))
199 status = ocfs2_flush_truncate_log(osb);
203 ocfs2_schedule_truncate_log_flush(osb, 0);
206 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
208 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
214 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
216 struct inode *new = NULL;
222 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
224 status = PTR_ERR(new);
228 osb->root_inode = new;
230 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
232 status = PTR_ERR(new);
236 osb->sys_root_inode = new;
238 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
239 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
240 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
242 ocfs2_release_system_inodes(osb);
245 /* FIXME: Should ERROR_RO_FS */
246 mlog(ML_ERROR, "Unable to load system inode %d, "
247 "possibly corrupt fs?", i);
250 // the array now has one ref, so drop this one
259 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
261 struct inode *new = NULL;
267 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
268 i < NUM_SYSTEM_INODES;
270 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
272 ocfs2_release_system_inodes(osb);
274 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
275 status, i, osb->slot_num);
278 /* the array now has one ref, so drop this one */
287 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
294 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
295 inode = osb->system_inodes[i];
298 osb->system_inodes[i] = NULL;
302 inode = osb->sys_root_inode;
305 osb->sys_root_inode = NULL;
308 inode = osb->root_inode;
311 osb->root_inode = NULL;
317 /* We're allocating fs objects, use GFP_NOFS */
318 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
320 struct ocfs2_inode_info *oi;
322 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
326 return &oi->vfs_inode;
329 static void ocfs2_destroy_inode(struct inode *inode)
331 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
334 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
337 unsigned int bytes = 1 << cbits;
338 unsigned int trim = bytes;
339 unsigned int bitshift = 32;
342 * i_size and all block offsets in ocfs2 are always 64 bits
343 * wide. i_clusters is 32 bits, in cluster-sized units. So on
344 * 64 bit platforms, cluster size will be the limiting factor.
347 #if BITS_PER_LONG == 32
348 # if defined(CONFIG_LBD)
349 BUILD_BUG_ON(sizeof(sector_t) != 8);
351 * We might be limited by page cache size.
353 if (bytes > PAGE_CACHE_SIZE) {
354 bytes = PAGE_CACHE_SIZE;
357 * Shift by 31 here so that we don't get larger than
364 * We are limited by the size of sector_t. Use block size, as
365 * that's what we expose to the VFS.
374 * Trim by a whole cluster when we can actually approach the
375 * on-disk limits. Otherwise we can overflow i_clusters when
376 * an extent start is at the max offset.
378 return (((unsigned long long)bytes) << bitshift) - trim;
381 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
383 int incompat_features;
385 struct mount_options parsed_options;
386 struct ocfs2_super *osb = OCFS2_SB(sb);
388 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
393 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
394 (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
396 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
400 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
401 (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
403 mlog(ML_ERROR, "Cannot change data mode on remount\n");
407 /* We're going to/from readonly mode. */
408 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
409 /* Lock here so the check of HARD_RO and the potential
410 * setting of SOFT_RO is atomic. */
411 spin_lock(&osb->osb_lock);
412 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
413 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
418 if (*flags & MS_RDONLY) {
419 mlog(0, "Going to ro mode.\n");
420 sb->s_flags |= MS_RDONLY;
421 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
423 mlog(0, "Making ro filesystem writeable.\n");
425 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
426 mlog(ML_ERROR, "Cannot remount RDWR "
427 "filesystem due to previous errors.\n");
431 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
432 if (incompat_features) {
433 mlog(ML_ERROR, "Cannot remount RDWR because "
434 "of unsupported optional features "
435 "(%x).\n", incompat_features);
439 sb->s_flags &= ~MS_RDONLY;
440 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
443 spin_unlock(&osb->osb_lock);
447 /* Only save off the new mount options in case of a successful
449 osb->s_mount_opt = parsed_options.mount_opt;
450 osb->s_atime_quantum = parsed_options.atime_quantum;
451 osb->preferred_slot = parsed_options.slot;
452 if (parsed_options.commit_interval)
453 osb->osb_commit_interval = parsed_options.commit_interval;
455 if (!ocfs2_is_hard_readonly(osb))
456 ocfs2_set_journal_params(osb);
462 static int ocfs2_sb_probe(struct super_block *sb,
463 struct buffer_head **bh,
467 struct ocfs1_vol_disk_hdr *hdr;
468 struct ocfs2_dinode *di;
474 *sector_size = bdev_hardsect_size(sb->s_bdev);
475 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
476 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
477 *sector_size, OCFS2_MAX_BLOCKSIZE);
482 /* Can this really happen? */
483 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
484 *sector_size = OCFS2_MIN_BLOCKSIZE;
486 /* check block zero for old format */
487 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
492 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
493 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
494 mlog(ML_ERROR, "incompatible version: %u.%u\n",
495 hdr->major_version, hdr->minor_version);
498 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
499 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
500 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
507 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
508 "upgraded before mounting with ocfs v2\n");
513 * Now check at magic offset for 512, 1024, 2048, 4096
514 * blocksizes. 4096 is the maximum blocksize because it is
515 * the minimum clustersize.
518 for (blksize = *sector_size;
519 blksize <= OCFS2_MAX_BLOCKSIZE;
521 tmpstat = ocfs2_get_sector(sb, bh,
522 OCFS2_SUPER_BLOCK_BLKNO,
529 di = (struct ocfs2_dinode *) (*bh)->b_data;
530 status = ocfs2_verify_volume(di, *bh, blksize);
535 if (status != -EAGAIN)
543 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
545 if (ocfs2_mount_local(osb)) {
546 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
547 mlog(ML_ERROR, "Cannot heartbeat on a locally "
548 "mounted device.\n");
553 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
554 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
555 mlog(ML_ERROR, "Heartbeat has to be started to mount "
556 "a read-write clustered device.\n");
564 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
567 int status, sector_size;
568 struct mount_options parsed_options;
569 struct inode *inode = NULL;
570 struct ocfs2_super *osb = NULL;
571 struct buffer_head *bh = NULL;
574 mlog_entry("%p, %p, %i", sb, data, silent);
576 if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
578 goto read_super_error;
581 /* probe for superblock */
582 status = ocfs2_sb_probe(sb, &bh, §or_size);
584 mlog(ML_ERROR, "superblock probe failed!\n");
585 goto read_super_error;
588 status = ocfs2_initialize_super(sb, bh, sector_size);
592 goto read_super_error;
596 osb->s_mount_opt = parsed_options.mount_opt;
597 osb->s_atime_quantum = parsed_options.atime_quantum;
598 osb->preferred_slot = parsed_options.slot;
599 osb->osb_commit_interval = parsed_options.commit_interval;
600 osb->local_alloc_size = parsed_options.localalloc_opt;
602 sb->s_magic = OCFS2_SUPER_MAGIC;
604 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
606 if (bdev_read_only(sb->s_bdev)) {
607 if (!(sb->s_flags & MS_RDONLY)) {
609 mlog(ML_ERROR, "Readonly device detected but readonly "
610 "mount was not specified.\n");
611 goto read_super_error;
614 /* You should not be able to start a local heartbeat
615 * on a readonly device. */
616 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
618 mlog(ML_ERROR, "Local heartbeat specified on readonly "
620 goto read_super_error;
623 status = ocfs2_check_journals_nolocks(osb);
625 if (status == -EROFS)
626 mlog(ML_ERROR, "Recovery required on readonly "
627 "file system, but write access is "
631 goto read_super_error;
634 ocfs2_set_ro_flag(osb, 1);
636 printk(KERN_NOTICE "Readonly device detected. No cluster "
637 "services will be utilized for this mount. Recovery "
638 "will be skipped.\n");
641 if (!ocfs2_is_hard_readonly(osb)) {
642 if (sb->s_flags & MS_RDONLY)
643 ocfs2_set_ro_flag(osb, 0);
646 status = ocfs2_verify_heartbeat(osb);
649 goto read_super_error;
652 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
654 if (!osb->osb_debug_root) {
656 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
657 goto read_super_error;
660 status = ocfs2_mount_volume(sb);
662 inode = igrab(osb->root_inode);
665 goto read_super_error;
670 goto read_super_error;
673 root = d_alloc_root(inode);
677 goto read_super_error;
682 ocfs2_complete_mount_recovery(osb);
684 if (ocfs2_mount_local(osb))
685 snprintf(nodestr, sizeof(nodestr), "local");
687 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
689 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
690 "with %s data mode.\n",
691 osb->dev_str, nodestr, osb->slot_num,
692 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
695 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
696 wake_up(&osb->osb_mount_event);
709 atomic_set(&osb->vol_state, VOLUME_DISABLED);
710 wake_up(&osb->osb_mount_event);
711 ocfs2_dismount_volume(sb, 1);
718 static int ocfs2_get_sb(struct file_system_type *fs_type,
720 const char *dev_name,
722 struct vfsmount *mnt)
724 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
728 static struct file_system_type ocfs2_fs_type = {
729 .owner = THIS_MODULE,
731 .get_sb = ocfs2_get_sb, /* is this called when we mount
733 .kill_sb = kill_block_super, /* set to the generic one
734 * right now, but do we
735 * need to change that? */
736 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
740 static int ocfs2_parse_options(struct super_block *sb,
742 struct mount_options *mopt,
748 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
749 options ? options : "(none)");
751 mopt->commit_interval = 0;
753 mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
754 mopt->slot = OCFS2_INVALID_SLOT;
755 mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
762 while ((p = strsep(&options, ",")) != NULL) {
764 substring_t args[MAX_OPT_ARGS];
769 token = match_token(p, tokens, args);
772 mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
775 mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
778 if (match_int(&args[0], &option)) {
783 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
785 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
788 mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
791 mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
794 mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
797 mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
799 case Opt_data_ordered:
800 mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
802 case Opt_data_writeback:
803 mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
805 case Opt_atime_quantum:
806 if (match_int(&args[0], &option)) {
811 mopt->atime_quantum = option;
815 if (match_int(&args[0], &option)) {
820 mopt->slot = (s16)option;
824 if (match_int(&args[0], &option)) {
831 option = JBD_DEFAULT_MAX_COMMIT_AGE;
832 mopt->commit_interval = HZ * option;
836 if (match_int(&args[0], &option)) {
840 if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
841 mopt->localalloc_opt = option;
843 case Opt_localflocks:
845 * Changing this during remount could race
846 * flock() requests, or "unbalance" existing
847 * ones (e.g., a lock is taken in one mode but
848 * dropped in the other). If users care enough
849 * to flip locking modes during remount, we
850 * could add a "local" flag to individual
851 * flock structures for proper tracking of
855 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
859 "Unrecognized mount option \"%s\" "
860 "or missing value\n", p);
873 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
875 struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
876 unsigned long opts = osb->s_mount_opt;
878 if (opts & OCFS2_MOUNT_HB_LOCAL)
879 seq_printf(s, ",_netdev,heartbeat=local");
881 seq_printf(s, ",heartbeat=none");
883 if (opts & OCFS2_MOUNT_NOINTR)
884 seq_printf(s, ",nointr");
886 if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
887 seq_printf(s, ",data=writeback");
889 seq_printf(s, ",data=ordered");
891 if (opts & OCFS2_MOUNT_BARRIER)
892 seq_printf(s, ",barrier=1");
894 if (opts & OCFS2_MOUNT_ERRORS_PANIC)
895 seq_printf(s, ",errors=panic");
897 seq_printf(s, ",errors=remount-ro");
899 if (osb->preferred_slot != OCFS2_INVALID_SLOT)
900 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
902 if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
903 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
905 if (osb->osb_commit_interval)
906 seq_printf(s, ",commit=%u",
907 (unsigned) (osb->osb_commit_interval / HZ));
909 if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
910 seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
912 if (opts & OCFS2_MOUNT_LOCALFLOCKS)
913 seq_printf(s, ",localflocks,");
918 static int __init ocfs2_init(void)
924 ocfs2_print_version();
926 dlmglue_init_stack();
928 status = init_ocfs2_uptodate_cache();
934 status = ocfs2_initialize_mem_caches();
940 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
946 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
947 if (!ocfs2_debugfs_root) {
949 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
954 ocfs2_free_mem_caches();
955 exit_ocfs2_uptodate_cache();
961 return register_filesystem(&ocfs2_fs_type);
966 static void __exit ocfs2_exit(void)
971 flush_workqueue(ocfs2_wq);
972 destroy_workqueue(ocfs2_wq);
975 debugfs_remove(ocfs2_debugfs_root);
977 ocfs2_free_mem_caches();
979 unregister_filesystem(&ocfs2_fs_type);
981 exit_ocfs2_uptodate_cache();
983 dlmglue_exit_stack();
988 static void ocfs2_put_super(struct super_block *sb)
990 mlog_entry("(0x%p)\n", sb);
992 ocfs2_sync_blockdev(sb);
993 ocfs2_dismount_volume(sb, 0);
998 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1000 struct ocfs2_super *osb;
1001 u32 numbits, freebits;
1003 struct ocfs2_dinode *bm_lock;
1004 struct buffer_head *bh = NULL;
1005 struct inode *inode = NULL;
1007 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1009 osb = OCFS2_SB(dentry->d_sb);
1011 inode = ocfs2_get_system_file_inode(osb,
1012 GLOBAL_BITMAP_SYSTEM_INODE,
1013 OCFS2_INVALID_SLOT);
1015 mlog(ML_ERROR, "failed to get bitmap inode\n");
1020 status = ocfs2_inode_lock(inode, &bh, 0);
1026 bm_lock = (struct ocfs2_dinode *) bh->b_data;
1028 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1029 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1031 buf->f_type = OCFS2_SUPER_MAGIC;
1032 buf->f_bsize = dentry->d_sb->s_blocksize;
1033 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1034 buf->f_blocks = ((sector_t) numbits) *
1035 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1036 buf->f_bfree = ((sector_t) freebits) *
1037 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1038 buf->f_bavail = buf->f_bfree;
1039 buf->f_files = numbits;
1040 buf->f_ffree = freebits;
1044 ocfs2_inode_unlock(inode, 0);
1055 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1057 struct ocfs2_inode_info *oi = data;
1060 oi->ip_open_count = 0;
1061 spin_lock_init(&oi->ip_lock);
1062 ocfs2_extent_map_init(&oi->vfs_inode);
1063 INIT_LIST_HEAD(&oi->ip_io_markers);
1064 oi->ip_created_trans = 0;
1065 oi->ip_last_trans = 0;
1066 oi->ip_dir_start_lookup = 0;
1068 init_rwsem(&oi->ip_alloc_sem);
1069 mutex_init(&oi->ip_io_mutex);
1071 oi->ip_blkno = 0ULL;
1072 oi->ip_clusters = 0;
1074 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1075 ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1076 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1078 ocfs2_metadata_cache_init(&oi->vfs_inode);
1080 inode_init_once(&oi->vfs_inode);
1083 static int ocfs2_initialize_mem_caches(void)
1085 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1086 sizeof(struct ocfs2_inode_info),
1088 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1090 ocfs2_inode_init_once);
1091 if (!ocfs2_inode_cachep)
1097 static void ocfs2_free_mem_caches(void)
1099 if (ocfs2_inode_cachep)
1100 kmem_cache_destroy(ocfs2_inode_cachep);
1102 ocfs2_inode_cachep = NULL;
1105 static int ocfs2_get_sector(struct super_block *sb,
1106 struct buffer_head **bh,
1110 if (!sb_set_blocksize(sb, sect_size)) {
1111 mlog(ML_ERROR, "unable to set blocksize\n");
1115 *bh = sb_getblk(sb, block);
1121 if (!buffer_dirty(*bh))
1122 clear_buffer_uptodate(*bh);
1124 ll_rw_block(READ, 1, bh);
1125 wait_on_buffer(*bh);
1129 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1130 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1134 /* XXX hold a ref on the node while mounte? easy enough, if
1136 if (ocfs2_mount_local(osb))
1139 status = ocfs2_cluster_this_node(&osb->node_num);
1143 "could not find this host's node number\n");
1148 mlog(0, "I am node %u\n", osb->node_num);
1155 static int ocfs2_mount_volume(struct super_block *sb)
1158 int unlock_super = 0;
1159 struct ocfs2_super *osb = OCFS2_SB(sb);
1163 if (ocfs2_is_hard_readonly(osb))
1166 status = ocfs2_fill_local_node_info(osb);
1172 status = ocfs2_dlm_init(osb);
1178 status = ocfs2_super_lock(osb, 1);
1185 /* This will load up the node map and add ourselves to it. */
1186 status = ocfs2_find_slot(osb);
1192 /* load all node-local system inodes */
1193 status = ocfs2_init_local_system_inodes(osb);
1199 status = ocfs2_check_volume(osb);
1205 status = ocfs2_truncate_log_init(osb);
1211 if (ocfs2_mount_local(osb))
1216 ocfs2_super_unlock(osb, 1);
1222 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1225 struct ocfs2_super *osb = NULL;
1228 mlog_entry("(0x%p)\n", sb);
1234 ocfs2_shutdown_local_alloc(osb);
1236 ocfs2_truncate_log_shutdown(osb);
1238 /* This will disable recovery and flush any recovery work. */
1239 ocfs2_recovery_exit(osb);
1241 ocfs2_journal_shutdown(osb);
1243 ocfs2_sync_blockdev(sb);
1245 /* No cluster connection means we've failed during mount, so skip
1246 * all the steps which depended on that to complete. */
1248 tmp = ocfs2_super_lock(osb, 1);
1255 if (osb->slot_num != OCFS2_INVALID_SLOT)
1256 ocfs2_put_slot(osb);
1259 ocfs2_super_unlock(osb, 1);
1261 ocfs2_release_system_inodes(osb);
1264 ocfs2_dlm_shutdown(osb);
1266 debugfs_remove(osb->osb_debug_root);
1269 * This is a small hack to move ocfs2_hb_ctl into stackglue.
1270 * If we're dismounting due to mount error, mount.ocfs2 will clean
1271 * up heartbeat. If we're a local mount, there is no heartbeat.
1272 * If we failed before we got a uuid_str yet, we can't stop
1273 * heartbeat. Otherwise, do it.
1275 if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1276 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1278 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1280 if (ocfs2_mount_local(osb))
1281 snprintf(nodestr, sizeof(nodestr), "local");
1283 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1285 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1286 osb->dev_str, nodestr);
1288 ocfs2_delete_osb(osb);
1291 sb->s_fs_info = NULL;
1294 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1295 unsigned uuid_bytes)
1300 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1302 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1303 if (osb->uuid_str == NULL)
1306 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1307 /* print with null */
1308 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1309 if (ret != 2) /* drop super cleans up */
1311 /* then only advance past the last char */
1318 static int ocfs2_initialize_super(struct super_block *sb,
1319 struct buffer_head *bh,
1323 int i, cbits, bbits;
1324 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1325 struct inode *inode = NULL;
1326 struct ocfs2_journal *journal;
1327 __le32 uuid_net_key;
1328 struct ocfs2_super *osb;
1332 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1339 sb->s_fs_info = osb;
1340 sb->s_op = &ocfs2_sops;
1341 sb->s_export_op = &ocfs2_export_ops;
1342 sb->s_time_gran = 1;
1343 sb->s_flags |= MS_NOATIME;
1344 /* this is needed to support O_LARGEFILE */
1345 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1346 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1347 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1350 /* Save off for ocfs2_rw_direct */
1351 osb->s_sectsize_bits = blksize_bits(sector_size);
1352 BUG_ON(!osb->s_sectsize_bits);
1354 spin_lock_init(&osb->dc_task_lock);
1355 init_waitqueue_head(&osb->dc_event);
1356 osb->dc_work_sequence = 0;
1357 osb->dc_wake_sequence = 0;
1358 INIT_LIST_HEAD(&osb->blocked_lock_list);
1359 osb->blocked_lock_count = 0;
1360 spin_lock_init(&osb->osb_lock);
1362 atomic_set(&osb->alloc_stats.moves, 0);
1363 atomic_set(&osb->alloc_stats.local_data, 0);
1364 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1365 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1366 atomic_set(&osb->alloc_stats.bg_extends, 0);
1368 ocfs2_init_node_maps(osb);
1370 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1371 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1373 status = ocfs2_recovery_init(osb);
1375 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1380 init_waitqueue_head(&osb->checkpoint_event);
1381 atomic_set(&osb->needs_checkpoint, 0);
1383 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1385 osb->slot_num = OCFS2_INVALID_SLOT;
1387 osb->local_alloc_state = OCFS2_LA_UNUSED;
1388 osb->local_alloc_bh = NULL;
1390 init_waitqueue_head(&osb->osb_mount_event);
1392 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1393 if (!osb->vol_label) {
1394 mlog(ML_ERROR, "unable to alloc vol label\n");
1399 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1400 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1401 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1406 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1408 init_waitqueue_head(&osb->osb_wipe_event);
1409 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1410 sizeof(*osb->osb_orphan_wipes),
1412 if (!osb->osb_orphan_wipes) {
1418 osb->s_feature_compat =
1419 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1420 osb->s_feature_ro_compat =
1421 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1422 osb->s_feature_incompat =
1423 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1425 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1426 mlog(ML_ERROR, "couldn't mount because of unsupported "
1427 "optional features (%x).\n", i);
1431 if (!(osb->sb->s_flags & MS_RDONLY) &&
1432 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1433 mlog(ML_ERROR, "couldn't mount RDWR because of "
1434 "unsupported optional features (%x).\n", i);
1439 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1442 * This should be done in ocfs2_journal_init(), but unknown
1443 * ordering issues will cause the filesystem to crash.
1444 * If anyone wants to figure out what part of the code
1445 * refers to osb->journal before ocfs2_journal_init() is run,
1448 /* initialize our journal structure */
1450 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1452 mlog(ML_ERROR, "unable to alloc journal\n");
1456 osb->journal = journal;
1457 journal->j_osb = osb;
1459 atomic_set(&journal->j_num_trans, 0);
1460 init_rwsem(&journal->j_trans_barrier);
1461 init_waitqueue_head(&journal->j_checkpointed);
1462 spin_lock_init(&journal->j_lock);
1463 journal->j_trans_id = (unsigned long) 1;
1464 INIT_LIST_HEAD(&journal->j_la_cleanups);
1465 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1466 journal->j_state = OCFS2_JOURNAL_FREE;
1468 /* get some pseudo constants for clustersize bits */
1469 osb->s_clustersize_bits =
1470 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1471 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1472 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1474 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1475 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1476 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1477 osb->s_clustersize);
1482 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1484 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1485 "what jbd can address in 32 bits.\n");
1490 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1491 sizeof(di->id2.i_super.s_uuid))) {
1492 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1497 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1499 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1500 osb->vol_label[63] = '\0';
1501 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1502 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1503 osb->first_cluster_group_blkno =
1504 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1505 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1506 mlog(0, "vol_label: %s\n", osb->vol_label);
1507 mlog(0, "uuid: %s\n", osb->uuid_str);
1508 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1509 (unsigned long long)osb->root_blkno,
1510 (unsigned long long)osb->system_dir_blkno);
1512 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1513 if (!osb->osb_dlm_debug) {
1519 atomic_set(&osb->vol_state, VOLUME_INIT);
1521 /* load root, system_dir, and all global system inodes */
1522 status = ocfs2_init_global_system_inodes(osb);
1531 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1532 OCFS2_INVALID_SLOT);
1539 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1542 osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1544 status = ocfs2_init_slot_info(osb);
1556 * will return: -EAGAIN if it is ok to keep searching for superblocks
1557 * -EINVAL if there is a bad superblock
1560 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1561 struct buffer_head *bh,
1564 int status = -EAGAIN;
1568 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1569 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1571 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1572 mlog(ML_ERROR, "found superblock with incorrect block "
1573 "size: found %u, should be %u\n",
1574 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1576 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1577 OCFS2_MAJOR_REV_LEVEL ||
1578 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1579 OCFS2_MINOR_REV_LEVEL) {
1580 mlog(ML_ERROR, "found superblock with bad version: "
1581 "found %u.%u, should be %u.%u\n",
1582 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1583 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1584 OCFS2_MAJOR_REV_LEVEL,
1585 OCFS2_MINOR_REV_LEVEL);
1586 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1587 mlog(ML_ERROR, "bad block number on superblock: "
1588 "found %llu, should be %llu\n",
1589 (unsigned long long)le64_to_cpu(di->i_blkno),
1590 (unsigned long long)bh->b_blocknr);
1591 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1592 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1593 mlog(ML_ERROR, "bad cluster size found: %u\n",
1594 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1595 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1596 mlog(ML_ERROR, "bad root_blkno: 0\n");
1597 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1598 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1599 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1601 "Superblock slots found greater than file system "
1602 "maximum: found %u, max %u\n",
1603 le16_to_cpu(di->id2.i_super.s_max_slots),
1615 static int ocfs2_check_volume(struct ocfs2_super *osb)
1620 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1626 /* Init our journal object. */
1627 status = ocfs2_journal_init(osb->journal, &dirty);
1629 mlog(ML_ERROR, "Could not initialize journal!\n");
1633 /* If the journal was unmounted cleanly then we don't want to
1634 * recover anything. Otherwise, journal_load will do that
1635 * dirty work for us :) */
1637 status = ocfs2_journal_wipe(osb->journal, 0);
1643 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1644 "recovering volume.\n");
1647 local = ocfs2_mount_local(osb);
1649 /* will play back anything left in the journal. */
1650 ocfs2_journal_load(osb->journal, local);
1653 /* recover my local alloc if we didn't unmount cleanly. */
1654 status = ocfs2_begin_local_alloc_recovery(osb,
1661 /* we complete the recovery process after we've marked
1662 * ourselves as mounted. */
1665 mlog(0, "Journal loaded.\n");
1667 status = ocfs2_load_local_alloc(osb);
1674 /* Recovery will be completed after we've mounted the
1675 * rest of the volume. */
1677 osb->local_alloc_copy = local_alloc;
1681 /* go through each journal, trylock it and if you get the
1682 * lock, and it's marked as dirty, set the bit in the recover
1683 * map and launch a recovery thread for it. */
1684 status = ocfs2_mark_dead_nodes(osb);
1697 * The routine gets called from dismount or close whenever a dismount on
1698 * volume is requested and the osb open count becomes 1.
1699 * It will remove the osb from the global list and also free up all the
1700 * initialized resources and fileobject.
1702 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1706 /* This function assumes that the caller has the main osb resource */
1708 ocfs2_free_slot_info(osb);
1710 kfree(osb->osb_orphan_wipes);
1712 * This belongs in journal shutdown, but because we have to
1713 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1716 kfree(osb->journal);
1717 if (osb->local_alloc_copy)
1718 kfree(osb->local_alloc_copy);
1719 kfree(osb->uuid_str);
1720 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1721 memset(osb, 0, sizeof(struct ocfs2_super));
1726 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1727 * panic(). We do not support continue-on-error operation. */
1728 static void ocfs2_handle_error(struct super_block *sb)
1730 struct ocfs2_super *osb = OCFS2_SB(sb);
1732 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1733 panic("OCFS2: (device %s): panic forced after error\n",
1736 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1738 if (sb->s_flags & MS_RDONLY &&
1739 (ocfs2_is_soft_readonly(osb) ||
1740 ocfs2_is_hard_readonly(osb)))
1743 printk(KERN_CRIT "File system is now read-only due to the potential "
1744 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1745 "system is unmounted.\n");
1746 sb->s_flags |= MS_RDONLY;
1747 ocfs2_set_ro_flag(osb, 0);
1750 static char error_buf[1024];
1752 void __ocfs2_error(struct super_block *sb,
1753 const char *function,
1754 const char *fmt, ...)
1758 va_start(args, fmt);
1759 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1762 /* Not using mlog here because we want to show the actual
1763 * function the error came from. */
1764 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1765 sb->s_id, function, error_buf);
1767 ocfs2_handle_error(sb);
1770 /* Handle critical errors. This is intentionally more drastic than
1771 * ocfs2_handle_error, so we only use for things like journal errors,
1773 void __ocfs2_abort(struct super_block* sb,
1774 const char *function,
1775 const char *fmt, ...)
1779 va_start(args, fmt);
1780 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1783 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1784 sb->s_id, function, error_buf);
1786 /* We don't have the cluster support yet to go straight to
1787 * hard readonly in here. Until then, we want to keep
1788 * ocfs2_abort() so that we can at least mark critical
1791 * TODO: This should abort the journal and alert other nodes
1792 * that our slot needs recovery. */
1794 /* Force a panic(). This stinks, but it's better than letting
1795 * things continue without having a proper hard readonly
1797 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1798 ocfs2_handle_error(sb);
1801 module_init(ocfs2_init);
1802 module_exit(ocfs2_exit);