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_check_volume(struct ocfs2_super *osb);
112 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
113 struct buffer_head *bh,
115 static int ocfs2_initialize_super(struct super_block *sb,
116 struct buffer_head *bh,
118 static int ocfs2_get_sector(struct super_block *sb,
119 struct buffer_head **bh,
122 static void ocfs2_write_super(struct super_block *sb);
123 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
124 static void ocfs2_destroy_inode(struct inode *inode);
126 static const struct super_operations ocfs2_sops = {
127 .statfs = ocfs2_statfs,
128 .alloc_inode = ocfs2_alloc_inode,
129 .destroy_inode = ocfs2_destroy_inode,
130 .drop_inode = ocfs2_drop_inode,
131 .clear_inode = ocfs2_clear_inode,
132 .delete_inode = ocfs2_delete_inode,
133 .sync_fs = ocfs2_sync_fs,
134 .write_super = ocfs2_write_super,
135 .put_super = ocfs2_put_super,
136 .remount_fs = ocfs2_remount,
137 .show_options = ocfs2_show_options,
158 static match_table_t tokens = {
159 {Opt_barrier, "barrier=%u"},
160 {Opt_err_panic, "errors=panic"},
161 {Opt_err_ro, "errors=remount-ro"},
163 {Opt_nointr, "nointr"},
164 {Opt_hb_none, OCFS2_HB_NONE},
165 {Opt_hb_local, OCFS2_HB_LOCAL},
166 {Opt_data_ordered, "data=ordered"},
167 {Opt_data_writeback, "data=writeback"},
168 {Opt_atime_quantum, "atime_quantum=%u"},
169 {Opt_slot, "preferred_slot=%u"},
170 {Opt_commit, "commit=%u"},
171 {Opt_localalloc, "localalloc=%d"},
172 {Opt_localflocks, "localflocks"},
177 * write_super and sync_fs ripped right out of ext3.
179 static void ocfs2_write_super(struct super_block *sb)
181 if (mutex_trylock(&sb->s_lock) != 0)
186 static int ocfs2_sync_fs(struct super_block *sb, int wait)
190 struct ocfs2_super *osb = OCFS2_SB(sb);
194 if (ocfs2_is_hard_readonly(osb))
198 status = ocfs2_flush_truncate_log(osb);
202 ocfs2_schedule_truncate_log_flush(osb, 0);
205 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
207 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
213 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
215 struct inode *new = NULL;
221 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
223 status = PTR_ERR(new);
227 osb->root_inode = new;
229 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
231 status = PTR_ERR(new);
235 osb->sys_root_inode = new;
237 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
238 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
239 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
241 ocfs2_release_system_inodes(osb);
244 /* FIXME: Should ERROR_RO_FS */
245 mlog(ML_ERROR, "Unable to load system inode %d, "
246 "possibly corrupt fs?", i);
249 // the array now has one ref, so drop this one
258 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
260 struct inode *new = NULL;
266 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
267 i < NUM_SYSTEM_INODES;
269 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
271 ocfs2_release_system_inodes(osb);
273 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
274 status, i, osb->slot_num);
277 /* the array now has one ref, so drop this one */
286 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
293 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
294 inode = osb->system_inodes[i];
297 osb->system_inodes[i] = NULL;
301 inode = osb->sys_root_inode;
304 osb->sys_root_inode = NULL;
307 inode = osb->root_inode;
310 osb->root_inode = NULL;
316 /* We're allocating fs objects, use GFP_NOFS */
317 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
319 struct ocfs2_inode_info *oi;
321 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
325 return &oi->vfs_inode;
328 static void ocfs2_destroy_inode(struct inode *inode)
330 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
333 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
336 unsigned int bytes = 1 << cbits;
337 unsigned int trim = bytes;
338 unsigned int bitshift = 32;
341 * i_size and all block offsets in ocfs2 are always 64 bits
342 * wide. i_clusters is 32 bits, in cluster-sized units. So on
343 * 64 bit platforms, cluster size will be the limiting factor.
346 #if BITS_PER_LONG == 32
347 # if defined(CONFIG_LBD)
348 BUILD_BUG_ON(sizeof(sector_t) != 8);
350 * We might be limited by page cache size.
352 if (bytes > PAGE_CACHE_SIZE) {
353 bytes = PAGE_CACHE_SIZE;
356 * Shift by 31 here so that we don't get larger than
363 * We are limited by the size of sector_t. Use block size, as
364 * that's what we expose to the VFS.
373 * Trim by a whole cluster when we can actually approach the
374 * on-disk limits. Otherwise we can overflow i_clusters when
375 * an extent start is at the max offset.
377 return (((unsigned long long)bytes) << bitshift) - trim;
380 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
382 int incompat_features;
384 struct mount_options parsed_options;
385 struct ocfs2_super *osb = OCFS2_SB(sb);
387 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
392 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
393 (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
395 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
399 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
400 (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
402 mlog(ML_ERROR, "Cannot change data mode on remount\n");
406 /* We're going to/from readonly mode. */
407 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
408 /* Lock here so the check of HARD_RO and the potential
409 * setting of SOFT_RO is atomic. */
410 spin_lock(&osb->osb_lock);
411 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
412 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
417 if (*flags & MS_RDONLY) {
418 mlog(0, "Going to ro mode.\n");
419 sb->s_flags |= MS_RDONLY;
420 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
422 mlog(0, "Making ro filesystem writeable.\n");
424 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
425 mlog(ML_ERROR, "Cannot remount RDWR "
426 "filesystem due to previous errors.\n");
430 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
431 if (incompat_features) {
432 mlog(ML_ERROR, "Cannot remount RDWR because "
433 "of unsupported optional features "
434 "(%x).\n", incompat_features);
438 sb->s_flags &= ~MS_RDONLY;
439 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
442 spin_unlock(&osb->osb_lock);
446 /* Only save off the new mount options in case of a successful
448 osb->s_mount_opt = parsed_options.mount_opt;
449 osb->s_atime_quantum = parsed_options.atime_quantum;
450 osb->preferred_slot = parsed_options.slot;
451 if (parsed_options.commit_interval)
452 osb->osb_commit_interval = parsed_options.commit_interval;
454 if (!ocfs2_is_hard_readonly(osb))
455 ocfs2_set_journal_params(osb);
461 static int ocfs2_sb_probe(struct super_block *sb,
462 struct buffer_head **bh,
466 struct ocfs1_vol_disk_hdr *hdr;
467 struct ocfs2_dinode *di;
473 *sector_size = bdev_hardsect_size(sb->s_bdev);
474 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
475 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
476 *sector_size, OCFS2_MAX_BLOCKSIZE);
481 /* Can this really happen? */
482 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
483 *sector_size = OCFS2_MIN_BLOCKSIZE;
485 /* check block zero for old format */
486 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
491 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
492 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
493 mlog(ML_ERROR, "incompatible version: %u.%u\n",
494 hdr->major_version, hdr->minor_version);
497 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
498 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
499 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
506 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
507 "upgraded before mounting with ocfs v2\n");
512 * Now check at magic offset for 512, 1024, 2048, 4096
513 * blocksizes. 4096 is the maximum blocksize because it is
514 * the minimum clustersize.
517 for (blksize = *sector_size;
518 blksize <= OCFS2_MAX_BLOCKSIZE;
520 tmpstat = ocfs2_get_sector(sb, bh,
521 OCFS2_SUPER_BLOCK_BLKNO,
528 di = (struct ocfs2_dinode *) (*bh)->b_data;
529 status = ocfs2_verify_volume(di, *bh, blksize);
534 if (status != -EAGAIN)
542 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
544 if (ocfs2_mount_local(osb)) {
545 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
546 mlog(ML_ERROR, "Cannot heartbeat on a locally "
547 "mounted device.\n");
552 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
553 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
554 mlog(ML_ERROR, "Heartbeat has to be started to mount "
555 "a read-write clustered device.\n");
563 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
566 int status, sector_size;
567 struct mount_options parsed_options;
568 struct inode *inode = NULL;
569 struct ocfs2_super *osb = NULL;
570 struct buffer_head *bh = NULL;
573 mlog_entry("%p, %p, %i", sb, data, silent);
575 if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
577 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;
598 osb->osb_commit_interval = parsed_options.commit_interval;
599 osb->local_alloc_size = parsed_options.localalloc_opt;
601 sb->s_magic = OCFS2_SUPER_MAGIC;
603 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
605 if (bdev_read_only(sb->s_bdev)) {
606 if (!(sb->s_flags & MS_RDONLY)) {
608 mlog(ML_ERROR, "Readonly device detected but readonly "
609 "mount was not specified.\n");
610 goto read_super_error;
613 /* You should not be able to start a local heartbeat
614 * on a readonly device. */
615 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
617 mlog(ML_ERROR, "Local heartbeat specified on readonly "
619 goto read_super_error;
622 status = ocfs2_check_journals_nolocks(osb);
624 if (status == -EROFS)
625 mlog(ML_ERROR, "Recovery required on readonly "
626 "file system, but write access is "
630 goto read_super_error;
633 ocfs2_set_ro_flag(osb, 1);
635 printk(KERN_NOTICE "Readonly device detected. No cluster "
636 "services will be utilized for this mount. Recovery "
637 "will be skipped.\n");
640 if (!ocfs2_is_hard_readonly(osb)) {
641 if (sb->s_flags & MS_RDONLY)
642 ocfs2_set_ro_flag(osb, 0);
645 status = ocfs2_verify_heartbeat(osb);
648 goto read_super_error;
651 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
653 if (!osb->osb_debug_root) {
655 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
656 goto read_super_error;
659 status = ocfs2_mount_volume(sb);
661 inode = igrab(osb->root_inode);
664 goto read_super_error;
669 goto read_super_error;
672 root = d_alloc_root(inode);
676 goto read_super_error;
681 ocfs2_complete_mount_recovery(osb);
683 if (ocfs2_mount_local(osb))
684 snprintf(nodestr, sizeof(nodestr), "local");
686 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
688 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
689 "with %s data mode.\n",
690 osb->dev_str, nodestr, osb->slot_num,
691 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
694 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
695 wake_up(&osb->osb_mount_event);
708 atomic_set(&osb->vol_state, VOLUME_DISABLED);
709 wake_up(&osb->osb_mount_event);
710 ocfs2_dismount_volume(sb, 1);
717 static int ocfs2_get_sb(struct file_system_type *fs_type,
719 const char *dev_name,
721 struct vfsmount *mnt)
723 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
727 static struct file_system_type ocfs2_fs_type = {
728 .owner = THIS_MODULE,
730 .get_sb = ocfs2_get_sb, /* is this called when we mount
732 .kill_sb = kill_block_super, /* set to the generic one
733 * right now, but do we
734 * need to change that? */
735 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
739 static int ocfs2_parse_options(struct super_block *sb,
741 struct mount_options *mopt,
747 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
748 options ? options : "(none)");
750 mopt->commit_interval = 0;
752 mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
753 mopt->slot = OCFS2_INVALID_SLOT;
754 mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
761 while ((p = strsep(&options, ",")) != NULL) {
763 substring_t args[MAX_OPT_ARGS];
768 token = match_token(p, tokens, args);
771 mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
774 mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
777 if (match_int(&args[0], &option)) {
782 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
784 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
787 mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
790 mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
793 mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
796 mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
798 case Opt_data_ordered:
799 mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
801 case Opt_data_writeback:
802 mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
804 case Opt_atime_quantum:
805 if (match_int(&args[0], &option)) {
810 mopt->atime_quantum = option;
814 if (match_int(&args[0], &option)) {
819 mopt->slot = (s16)option;
823 if (match_int(&args[0], &option)) {
830 option = JBD_DEFAULT_MAX_COMMIT_AGE;
831 mopt->commit_interval = HZ * option;
835 if (match_int(&args[0], &option)) {
839 if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
840 mopt->localalloc_opt = option;
842 case Opt_localflocks:
844 * Changing this during remount could race
845 * flock() requests, or "unbalance" existing
846 * ones (e.g., a lock is taken in one mode but
847 * dropped in the other). If users care enough
848 * to flip locking modes during remount, we
849 * could add a "local" flag to individual
850 * flock structures for proper tracking of
854 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
858 "Unrecognized mount option \"%s\" "
859 "or missing value\n", p);
872 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
874 struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
875 unsigned long opts = osb->s_mount_opt;
877 if (opts & OCFS2_MOUNT_HB_LOCAL)
878 seq_printf(s, ",_netdev,heartbeat=local");
880 seq_printf(s, ",heartbeat=none");
882 if (opts & OCFS2_MOUNT_NOINTR)
883 seq_printf(s, ",nointr");
885 if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
886 seq_printf(s, ",data=writeback");
888 seq_printf(s, ",data=ordered");
890 if (opts & OCFS2_MOUNT_BARRIER)
891 seq_printf(s, ",barrier=1");
893 if (opts & OCFS2_MOUNT_ERRORS_PANIC)
894 seq_printf(s, ",errors=panic");
896 seq_printf(s, ",errors=remount-ro");
898 if (osb->preferred_slot != OCFS2_INVALID_SLOT)
899 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
901 if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
902 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
904 if (osb->osb_commit_interval)
905 seq_printf(s, ",commit=%u",
906 (unsigned) (osb->osb_commit_interval / HZ));
908 if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
909 seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
911 if (opts & OCFS2_MOUNT_LOCALFLOCKS)
912 seq_printf(s, ",localflocks,");
917 static int __init ocfs2_init(void)
923 ocfs2_print_version();
925 dlmglue_init_stack();
927 status = init_ocfs2_uptodate_cache();
933 status = ocfs2_initialize_mem_caches();
939 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
945 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
946 if (!ocfs2_debugfs_root) {
948 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
953 ocfs2_free_mem_caches();
954 exit_ocfs2_uptodate_cache();
960 return register_filesystem(&ocfs2_fs_type);
965 static void __exit ocfs2_exit(void)
970 flush_workqueue(ocfs2_wq);
971 destroy_workqueue(ocfs2_wq);
974 debugfs_remove(ocfs2_debugfs_root);
976 ocfs2_free_mem_caches();
978 unregister_filesystem(&ocfs2_fs_type);
980 exit_ocfs2_uptodate_cache();
982 dlmglue_exit_stack();
987 static void ocfs2_put_super(struct super_block *sb)
989 mlog_entry("(0x%p)\n", sb);
991 ocfs2_sync_blockdev(sb);
992 ocfs2_dismount_volume(sb, 0);
997 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
999 struct ocfs2_super *osb;
1000 u32 numbits, freebits;
1002 struct ocfs2_dinode *bm_lock;
1003 struct buffer_head *bh = NULL;
1004 struct inode *inode = NULL;
1006 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1008 osb = OCFS2_SB(dentry->d_sb);
1010 inode = ocfs2_get_system_file_inode(osb,
1011 GLOBAL_BITMAP_SYSTEM_INODE,
1012 OCFS2_INVALID_SLOT);
1014 mlog(ML_ERROR, "failed to get bitmap inode\n");
1019 status = ocfs2_inode_lock(inode, &bh, 0);
1025 bm_lock = (struct ocfs2_dinode *) bh->b_data;
1027 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1028 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1030 buf->f_type = OCFS2_SUPER_MAGIC;
1031 buf->f_bsize = dentry->d_sb->s_blocksize;
1032 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1033 buf->f_blocks = ((sector_t) numbits) *
1034 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1035 buf->f_bfree = ((sector_t) freebits) *
1036 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1037 buf->f_bavail = buf->f_bfree;
1038 buf->f_files = numbits;
1039 buf->f_ffree = freebits;
1043 ocfs2_inode_unlock(inode, 0);
1054 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1056 struct ocfs2_inode_info *oi = data;
1059 oi->ip_open_count = 0;
1060 spin_lock_init(&oi->ip_lock);
1061 ocfs2_extent_map_init(&oi->vfs_inode);
1062 INIT_LIST_HEAD(&oi->ip_io_markers);
1063 oi->ip_created_trans = 0;
1064 oi->ip_last_trans = 0;
1065 oi->ip_dir_start_lookup = 0;
1067 init_rwsem(&oi->ip_alloc_sem);
1068 mutex_init(&oi->ip_io_mutex);
1070 oi->ip_blkno = 0ULL;
1071 oi->ip_clusters = 0;
1073 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1074 ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1075 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1077 ocfs2_metadata_cache_init(&oi->vfs_inode);
1079 inode_init_once(&oi->vfs_inode);
1082 static int ocfs2_initialize_mem_caches(void)
1084 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1085 sizeof(struct ocfs2_inode_info),
1087 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1089 ocfs2_inode_init_once);
1090 if (!ocfs2_inode_cachep)
1096 static void ocfs2_free_mem_caches(void)
1098 if (ocfs2_inode_cachep)
1099 kmem_cache_destroy(ocfs2_inode_cachep);
1101 ocfs2_inode_cachep = NULL;
1104 static int ocfs2_get_sector(struct super_block *sb,
1105 struct buffer_head **bh,
1109 if (!sb_set_blocksize(sb, sect_size)) {
1110 mlog(ML_ERROR, "unable to set blocksize\n");
1114 *bh = sb_getblk(sb, block);
1120 if (!buffer_dirty(*bh))
1121 clear_buffer_uptodate(*bh);
1123 ll_rw_block(READ, 1, bh);
1124 wait_on_buffer(*bh);
1128 static int ocfs2_mount_volume(struct super_block *sb)
1131 int unlock_super = 0;
1132 struct ocfs2_super *osb = OCFS2_SB(sb);
1136 if (ocfs2_is_hard_readonly(osb))
1139 status = ocfs2_dlm_init(osb);
1145 status = ocfs2_super_lock(osb, 1);
1152 /* This will load up the node map and add ourselves to it. */
1153 status = ocfs2_find_slot(osb);
1159 /* load all node-local system inodes */
1160 status = ocfs2_init_local_system_inodes(osb);
1166 status = ocfs2_check_volume(osb);
1172 status = ocfs2_truncate_log_init(osb);
1178 if (ocfs2_mount_local(osb))
1183 ocfs2_super_unlock(osb, 1);
1189 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1192 struct ocfs2_super *osb = NULL;
1195 mlog_entry("(0x%p)\n", sb);
1201 ocfs2_shutdown_local_alloc(osb);
1203 ocfs2_truncate_log_shutdown(osb);
1205 /* This will disable recovery and flush any recovery work. */
1206 ocfs2_recovery_exit(osb);
1208 ocfs2_journal_shutdown(osb);
1210 ocfs2_sync_blockdev(sb);
1212 /* No cluster connection means we've failed during mount, so skip
1213 * all the steps which depended on that to complete. */
1215 tmp = ocfs2_super_lock(osb, 1);
1222 if (osb->slot_num != OCFS2_INVALID_SLOT)
1223 ocfs2_put_slot(osb);
1226 ocfs2_super_unlock(osb, 1);
1228 ocfs2_release_system_inodes(osb);
1231 ocfs2_dlm_shutdown(osb);
1233 debugfs_remove(osb->osb_debug_root);
1236 * This is a small hack to move ocfs2_hb_ctl into stackglue.
1237 * If we're dismounting due to mount error, mount.ocfs2 will clean
1238 * up heartbeat. If we're a local mount, there is no heartbeat.
1239 * If we failed before we got a uuid_str yet, we can't stop
1240 * heartbeat. Otherwise, do it.
1242 if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1243 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1245 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1247 if (ocfs2_mount_local(osb))
1248 snprintf(nodestr, sizeof(nodestr), "local");
1250 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1252 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1253 osb->dev_str, nodestr);
1255 ocfs2_delete_osb(osb);
1258 sb->s_fs_info = NULL;
1261 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1262 unsigned uuid_bytes)
1267 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1269 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1270 if (osb->uuid_str == NULL)
1273 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1274 /* print with null */
1275 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1276 if (ret != 2) /* drop super cleans up */
1278 /* then only advance past the last char */
1285 static int ocfs2_initialize_super(struct super_block *sb,
1286 struct buffer_head *bh,
1290 int i, cbits, bbits;
1291 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1292 struct inode *inode = NULL;
1293 struct ocfs2_journal *journal;
1294 __le32 uuid_net_key;
1295 struct ocfs2_super *osb;
1299 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1306 sb->s_fs_info = osb;
1307 sb->s_op = &ocfs2_sops;
1308 sb->s_export_op = &ocfs2_export_ops;
1309 sb->s_time_gran = 1;
1310 sb->s_flags |= MS_NOATIME;
1311 /* this is needed to support O_LARGEFILE */
1312 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1313 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1314 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1317 /* Save off for ocfs2_rw_direct */
1318 osb->s_sectsize_bits = blksize_bits(sector_size);
1319 BUG_ON(!osb->s_sectsize_bits);
1321 spin_lock_init(&osb->dc_task_lock);
1322 init_waitqueue_head(&osb->dc_event);
1323 osb->dc_work_sequence = 0;
1324 osb->dc_wake_sequence = 0;
1325 INIT_LIST_HEAD(&osb->blocked_lock_list);
1326 osb->blocked_lock_count = 0;
1327 spin_lock_init(&osb->osb_lock);
1329 atomic_set(&osb->alloc_stats.moves, 0);
1330 atomic_set(&osb->alloc_stats.local_data, 0);
1331 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1332 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1333 atomic_set(&osb->alloc_stats.bg_extends, 0);
1335 ocfs2_init_node_maps(osb);
1337 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1338 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1340 status = ocfs2_recovery_init(osb);
1342 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1347 init_waitqueue_head(&osb->checkpoint_event);
1348 atomic_set(&osb->needs_checkpoint, 0);
1350 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1352 osb->slot_num = OCFS2_INVALID_SLOT;
1354 osb->local_alloc_state = OCFS2_LA_UNUSED;
1355 osb->local_alloc_bh = NULL;
1357 init_waitqueue_head(&osb->osb_mount_event);
1359 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1360 if (!osb->vol_label) {
1361 mlog(ML_ERROR, "unable to alloc vol label\n");
1366 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1367 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1368 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1373 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1375 init_waitqueue_head(&osb->osb_wipe_event);
1376 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1377 sizeof(*osb->osb_orphan_wipes),
1379 if (!osb->osb_orphan_wipes) {
1385 osb->s_feature_compat =
1386 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1387 osb->s_feature_ro_compat =
1388 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1389 osb->s_feature_incompat =
1390 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1392 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1393 mlog(ML_ERROR, "couldn't mount because of unsupported "
1394 "optional features (%x).\n", i);
1398 if (!(osb->sb->s_flags & MS_RDONLY) &&
1399 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1400 mlog(ML_ERROR, "couldn't mount RDWR because of "
1401 "unsupported optional features (%x).\n", i);
1406 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1409 * This should be done in ocfs2_journal_init(), but unknown
1410 * ordering issues will cause the filesystem to crash.
1411 * If anyone wants to figure out what part of the code
1412 * refers to osb->journal before ocfs2_journal_init() is run,
1415 /* initialize our journal structure */
1417 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1419 mlog(ML_ERROR, "unable to alloc journal\n");
1423 osb->journal = journal;
1424 journal->j_osb = osb;
1426 atomic_set(&journal->j_num_trans, 0);
1427 init_rwsem(&journal->j_trans_barrier);
1428 init_waitqueue_head(&journal->j_checkpointed);
1429 spin_lock_init(&journal->j_lock);
1430 journal->j_trans_id = (unsigned long) 1;
1431 INIT_LIST_HEAD(&journal->j_la_cleanups);
1432 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1433 journal->j_state = OCFS2_JOURNAL_FREE;
1435 /* get some pseudo constants for clustersize bits */
1436 osb->s_clustersize_bits =
1437 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1438 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1439 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1441 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1442 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1443 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1444 osb->s_clustersize);
1449 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1451 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1452 "what jbd can address in 32 bits.\n");
1457 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1458 sizeof(di->id2.i_super.s_uuid))) {
1459 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1464 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1466 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1467 osb->vol_label[63] = '\0';
1468 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1469 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1470 osb->first_cluster_group_blkno =
1471 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1472 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1473 mlog(0, "vol_label: %s\n", osb->vol_label);
1474 mlog(0, "uuid: %s\n", osb->uuid_str);
1475 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1476 (unsigned long long)osb->root_blkno,
1477 (unsigned long long)osb->system_dir_blkno);
1479 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1480 if (!osb->osb_dlm_debug) {
1486 atomic_set(&osb->vol_state, VOLUME_INIT);
1488 /* load root, system_dir, and all global system inodes */
1489 status = ocfs2_init_global_system_inodes(osb);
1498 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1499 OCFS2_INVALID_SLOT);
1506 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1509 osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1511 status = ocfs2_init_slot_info(osb);
1523 * will return: -EAGAIN if it is ok to keep searching for superblocks
1524 * -EINVAL if there is a bad superblock
1527 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1528 struct buffer_head *bh,
1531 int status = -EAGAIN;
1535 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1536 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1538 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1539 mlog(ML_ERROR, "found superblock with incorrect block "
1540 "size: found %u, should be %u\n",
1541 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1543 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1544 OCFS2_MAJOR_REV_LEVEL ||
1545 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1546 OCFS2_MINOR_REV_LEVEL) {
1547 mlog(ML_ERROR, "found superblock with bad version: "
1548 "found %u.%u, should be %u.%u\n",
1549 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1550 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1551 OCFS2_MAJOR_REV_LEVEL,
1552 OCFS2_MINOR_REV_LEVEL);
1553 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1554 mlog(ML_ERROR, "bad block number on superblock: "
1555 "found %llu, should be %llu\n",
1556 (unsigned long long)le64_to_cpu(di->i_blkno),
1557 (unsigned long long)bh->b_blocknr);
1558 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1559 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1560 mlog(ML_ERROR, "bad cluster size found: %u\n",
1561 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1562 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1563 mlog(ML_ERROR, "bad root_blkno: 0\n");
1564 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1565 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1566 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1568 "Superblock slots found greater than file system "
1569 "maximum: found %u, max %u\n",
1570 le16_to_cpu(di->id2.i_super.s_max_slots),
1582 static int ocfs2_check_volume(struct ocfs2_super *osb)
1587 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1593 /* Init our journal object. */
1594 status = ocfs2_journal_init(osb->journal, &dirty);
1596 mlog(ML_ERROR, "Could not initialize journal!\n");
1600 /* If the journal was unmounted cleanly then we don't want to
1601 * recover anything. Otherwise, journal_load will do that
1602 * dirty work for us :) */
1604 status = ocfs2_journal_wipe(osb->journal, 0);
1610 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1611 "recovering volume.\n");
1614 local = ocfs2_mount_local(osb);
1616 /* will play back anything left in the journal. */
1617 ocfs2_journal_load(osb->journal, local);
1620 /* recover my local alloc if we didn't unmount cleanly. */
1621 status = ocfs2_begin_local_alloc_recovery(osb,
1628 /* we complete the recovery process after we've marked
1629 * ourselves as mounted. */
1632 mlog(0, "Journal loaded.\n");
1634 status = ocfs2_load_local_alloc(osb);
1641 /* Recovery will be completed after we've mounted the
1642 * rest of the volume. */
1644 osb->local_alloc_copy = local_alloc;
1648 /* go through each journal, trylock it and if you get the
1649 * lock, and it's marked as dirty, set the bit in the recover
1650 * map and launch a recovery thread for it. */
1651 status = ocfs2_mark_dead_nodes(osb);
1664 * The routine gets called from dismount or close whenever a dismount on
1665 * volume is requested and the osb open count becomes 1.
1666 * It will remove the osb from the global list and also free up all the
1667 * initialized resources and fileobject.
1669 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1673 /* This function assumes that the caller has the main osb resource */
1675 ocfs2_free_slot_info(osb);
1677 kfree(osb->osb_orphan_wipes);
1679 * This belongs in journal shutdown, but because we have to
1680 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1683 kfree(osb->journal);
1684 if (osb->local_alloc_copy)
1685 kfree(osb->local_alloc_copy);
1686 kfree(osb->uuid_str);
1687 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1688 memset(osb, 0, sizeof(struct ocfs2_super));
1693 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1694 * panic(). We do not support continue-on-error operation. */
1695 static void ocfs2_handle_error(struct super_block *sb)
1697 struct ocfs2_super *osb = OCFS2_SB(sb);
1699 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1700 panic("OCFS2: (device %s): panic forced after error\n",
1703 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1705 if (sb->s_flags & MS_RDONLY &&
1706 (ocfs2_is_soft_readonly(osb) ||
1707 ocfs2_is_hard_readonly(osb)))
1710 printk(KERN_CRIT "File system is now read-only due to the potential "
1711 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1712 "system is unmounted.\n");
1713 sb->s_flags |= MS_RDONLY;
1714 ocfs2_set_ro_flag(osb, 0);
1717 static char error_buf[1024];
1719 void __ocfs2_error(struct super_block *sb,
1720 const char *function,
1721 const char *fmt, ...)
1725 va_start(args, fmt);
1726 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1729 /* Not using mlog here because we want to show the actual
1730 * function the error came from. */
1731 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1732 sb->s_id, function, error_buf);
1734 ocfs2_handle_error(sb);
1737 /* Handle critical errors. This is intentionally more drastic than
1738 * ocfs2_handle_error, so we only use for things like journal errors,
1740 void __ocfs2_abort(struct super_block* sb,
1741 const char *function,
1742 const char *fmt, ...)
1746 va_start(args, fmt);
1747 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1750 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1751 sb->s_id, function, error_buf);
1753 /* We don't have the cluster support yet to go straight to
1754 * hard readonly in here. Until then, we want to keep
1755 * ocfs2_abort() so that we can at least mark critical
1758 * TODO: This should abort the journal and alert other nodes
1759 * that our slot needs recovery. */
1761 /* Force a panic(). This stinks, but it's better than letting
1762 * things continue without having a proper hard readonly
1764 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1765 ocfs2_handle_error(sb);
1768 module_init(ocfs2_init);
1769 module_exit(ocfs2_exit);