1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * load/unload driver, mount/dismount volumes
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
43 #include <cluster/nodemanager.h>
45 #define MLOG_MASK_PREFIX ML_SUPER
46 #include <cluster/masklog.h>
50 /* this should be the only file to include a version 1 header */
51 #include "ocfs1_fs_compat.h"
56 #include "extent_map.h"
57 #include "heartbeat.h"
60 #include "localalloc.h"
69 #include "buffer_head_io.h"
71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
73 /* OCFS2 needs to schedule several differnt types of work which
74 * require cluster locking, disk I/O, recovery waits, etc. Since these
75 * types of work tend to be heavy we avoid using the kernel events
76 * workqueue and schedule on our own. */
77 struct workqueue_struct *ocfs2_wq = NULL;
79 static struct dentry *ocfs2_debugfs_root = NULL;
81 MODULE_AUTHOR("Oracle");
82 MODULE_LICENSE("GPL");
84 static int ocfs2_parse_options(struct super_block *sb, char *options,
85 unsigned long *mount_opt, s16 *slot,
87 static void ocfs2_put_super(struct super_block *sb);
88 static int ocfs2_mount_volume(struct super_block *sb);
89 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
90 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
91 static int ocfs2_initialize_mem_caches(void);
92 static void ocfs2_free_mem_caches(void);
93 static void ocfs2_delete_osb(struct ocfs2_super *osb);
95 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
97 static int ocfs2_sync_fs(struct super_block *sb, int wait);
99 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
100 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
101 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
102 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
103 static int ocfs2_check_volume(struct ocfs2_super *osb);
104 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
105 struct buffer_head *bh,
107 static int ocfs2_initialize_super(struct super_block *sb,
108 struct buffer_head *bh,
110 static int ocfs2_get_sector(struct super_block *sb,
111 struct buffer_head **bh,
114 static void ocfs2_write_super(struct super_block *sb);
115 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
116 static void ocfs2_destroy_inode(struct inode *inode);
118 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
120 static const struct super_operations ocfs2_sops = {
121 .statfs = ocfs2_statfs,
122 .alloc_inode = ocfs2_alloc_inode,
123 .destroy_inode = ocfs2_destroy_inode,
124 .drop_inode = ocfs2_drop_inode,
125 .clear_inode = ocfs2_clear_inode,
126 .delete_inode = ocfs2_delete_inode,
127 .sync_fs = ocfs2_sync_fs,
128 .write_super = ocfs2_write_super,
129 .put_super = ocfs2_put_super,
130 .remount_fs = ocfs2_remount,
148 static match_table_t tokens = {
149 {Opt_barrier, "barrier=%u"},
150 {Opt_err_panic, "errors=panic"},
151 {Opt_err_ro, "errors=remount-ro"},
153 {Opt_nointr, "nointr"},
154 {Opt_hb_none, OCFS2_HB_NONE},
155 {Opt_hb_local, OCFS2_HB_LOCAL},
156 {Opt_data_ordered, "data=ordered"},
157 {Opt_data_writeback, "data=writeback"},
158 {Opt_atime_quantum, "atime_quantum=%u"},
159 {Opt_slot, "preferred_slot=%u"},
164 * write_super and sync_fs ripped right out of ext3.
166 static void ocfs2_write_super(struct super_block *sb)
168 if (mutex_trylock(&sb->s_lock) != 0)
173 static int ocfs2_sync_fs(struct super_block *sb, int wait)
177 struct ocfs2_super *osb = OCFS2_SB(sb);
181 if (ocfs2_is_hard_readonly(osb))
185 status = ocfs2_flush_truncate_log(osb);
189 ocfs2_schedule_truncate_log_flush(osb, 0);
192 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
194 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
200 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
202 struct inode *new = NULL;
208 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
210 status = PTR_ERR(new);
214 osb->root_inode = new;
216 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
218 status = PTR_ERR(new);
222 osb->sys_root_inode = new;
224 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
225 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
226 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
228 ocfs2_release_system_inodes(osb);
231 /* FIXME: Should ERROR_RO_FS */
232 mlog(ML_ERROR, "Unable to load system inode %d, "
233 "possibly corrupt fs?", i);
236 // the array now has one ref, so drop this one
245 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
247 struct inode *new = NULL;
253 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
254 i < NUM_SYSTEM_INODES;
256 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
258 ocfs2_release_system_inodes(osb);
260 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
261 status, i, osb->slot_num);
264 /* the array now has one ref, so drop this one */
273 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
280 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
281 inode = osb->system_inodes[i];
284 osb->system_inodes[i] = NULL;
288 inode = osb->sys_root_inode;
291 osb->sys_root_inode = NULL;
294 inode = osb->root_inode;
297 osb->root_inode = NULL;
304 /* We're allocating fs objects, use GFP_NOFS */
305 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
307 struct ocfs2_inode_info *oi;
309 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
313 return &oi->vfs_inode;
316 static void ocfs2_destroy_inode(struct inode *inode)
318 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
321 /* From xfs_super.c:xfs_max_file_offset
322 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
324 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
326 unsigned int pagefactor = 1;
327 unsigned int bitshift = BITS_PER_LONG - 1;
329 /* Figure out maximum filesize, on Linux this can depend on
330 * the filesystem blocksize (on 32 bit platforms).
331 * __block_prepare_write does this in an [unsigned] long...
332 * page->index << (PAGE_CACHE_SHIFT - bbits)
333 * So, for page sized blocks (4K on 32 bit platforms),
334 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
335 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
336 * but for smaller blocksizes it is less (bbits = log2 bsize).
337 * Note1: get_block_t takes a long (implicit cast from above)
338 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
339 * can optionally convert the [unsigned] long from above into
340 * an [unsigned] long long.
343 #if BITS_PER_LONG == 32
344 # if defined(CONFIG_LBD)
345 BUILD_BUG_ON(sizeof(sector_t) != 8);
346 pagefactor = PAGE_CACHE_SIZE;
347 bitshift = BITS_PER_LONG;
349 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
353 return (((unsigned long long)pagefactor) << bitshift) - 1;
356 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
358 int incompat_features;
360 unsigned long parsed_options;
362 struct ocfs2_super *osb = OCFS2_SB(sb);
364 if (!ocfs2_parse_options(sb, data, &parsed_options, &slot, 1)) {
369 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
370 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
372 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
376 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
377 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
379 mlog(ML_ERROR, "Cannot change data mode on remount\n");
383 /* We're going to/from readonly mode. */
384 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
385 /* Lock here so the check of HARD_RO and the potential
386 * setting of SOFT_RO is atomic. */
387 spin_lock(&osb->osb_lock);
388 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
389 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
394 if (*flags & MS_RDONLY) {
395 mlog(0, "Going to ro mode.\n");
396 sb->s_flags |= MS_RDONLY;
397 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
399 mlog(0, "Making ro filesystem writeable.\n");
401 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
402 mlog(ML_ERROR, "Cannot remount RDWR "
403 "filesystem due to previous errors.\n");
407 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
408 if (incompat_features) {
409 mlog(ML_ERROR, "Cannot remount RDWR because "
410 "of unsupported optional features "
411 "(%x).\n", incompat_features);
415 sb->s_flags &= ~MS_RDONLY;
416 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
419 spin_unlock(&osb->osb_lock);
423 if (!ocfs2_is_hard_readonly(osb))
424 ocfs2_set_journal_params(osb);
426 /* Only save off the new mount options in case of a successful
428 osb->s_mount_opt = parsed_options;
434 static int ocfs2_sb_probe(struct super_block *sb,
435 struct buffer_head **bh,
438 int status = 0, tmpstat;
439 struct ocfs1_vol_disk_hdr *hdr;
440 struct ocfs2_dinode *di;
446 *sector_size = bdev_hardsect_size(sb->s_bdev);
447 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
448 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
449 *sector_size, OCFS2_MAX_BLOCKSIZE);
454 /* Can this really happen? */
455 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
456 *sector_size = OCFS2_MIN_BLOCKSIZE;
458 /* check block zero for old format */
459 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
464 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
465 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
466 mlog(ML_ERROR, "incompatible version: %u.%u\n",
467 hdr->major_version, hdr->minor_version);
470 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
471 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
472 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
479 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
480 "upgraded before mounting with ocfs v2\n");
485 * Now check at magic offset for 512, 1024, 2048, 4096
486 * blocksizes. 4096 is the maximum blocksize because it is
487 * the minimum clustersize.
490 for (blksize = *sector_size;
491 blksize <= OCFS2_MAX_BLOCKSIZE;
493 tmpstat = ocfs2_get_sector(sb, bh,
494 OCFS2_SUPER_BLOCK_BLKNO,
501 di = (struct ocfs2_dinode *) (*bh)->b_data;
502 status = ocfs2_verify_volume(di, *bh, blksize);
507 if (status != -EAGAIN)
515 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
517 if (ocfs2_mount_local(osb)) {
518 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
519 mlog(ML_ERROR, "Cannot heartbeat on a locally "
520 "mounted device.\n");
525 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
526 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
527 mlog(ML_ERROR, "Heartbeat has to be started to mount "
528 "a read-write clustered device.\n");
536 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
539 int status, sector_size;
540 unsigned long parsed_opt;
542 struct inode *inode = NULL;
543 struct ocfs2_super *osb = NULL;
544 struct buffer_head *bh = NULL;
547 mlog_entry("%p, %p, %i", sb, data, silent);
549 if (!ocfs2_parse_options(sb, data, &parsed_opt, &slot, 0)) {
551 goto read_super_error;
554 /* for now we only have one cluster/node, make sure we see it
555 * in the heartbeat universe */
556 if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
557 if (!o2hb_check_local_node_heartbeating()) {
559 goto read_super_error;
563 /* probe for superblock */
564 status = ocfs2_sb_probe(sb, &bh, §or_size);
566 mlog(ML_ERROR, "superblock probe failed!\n");
567 goto read_super_error;
570 status = ocfs2_initialize_super(sb, bh, sector_size);
574 goto read_super_error;
578 osb->s_mount_opt = parsed_opt;
579 osb->preferred_slot = slot;
581 sb->s_magic = OCFS2_SUPER_MAGIC;
583 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
585 if (bdev_read_only(sb->s_bdev)) {
586 if (!(sb->s_flags & MS_RDONLY)) {
588 mlog(ML_ERROR, "Readonly device detected but readonly "
589 "mount was not specified.\n");
590 goto read_super_error;
593 /* You should not be able to start a local heartbeat
594 * on a readonly device. */
595 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
597 mlog(ML_ERROR, "Local heartbeat specified on readonly "
599 goto read_super_error;
602 status = ocfs2_check_journals_nolocks(osb);
604 if (status == -EROFS)
605 mlog(ML_ERROR, "Recovery required on readonly "
606 "file system, but write access is "
610 goto read_super_error;
613 ocfs2_set_ro_flag(osb, 1);
615 printk(KERN_NOTICE "Readonly device detected. No cluster "
616 "services will be utilized for this mount. Recovery "
617 "will be skipped.\n");
620 if (!ocfs2_is_hard_readonly(osb)) {
621 if (sb->s_flags & MS_RDONLY)
622 ocfs2_set_ro_flag(osb, 0);
625 status = ocfs2_verify_heartbeat(osb);
628 goto read_super_error;
631 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
633 if (!osb->osb_debug_root) {
635 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
636 goto read_super_error;
639 status = ocfs2_mount_volume(sb);
641 inode = igrab(osb->root_inode);
644 goto read_super_error;
649 goto read_super_error;
652 root = d_alloc_root(inode);
656 goto read_super_error;
661 ocfs2_complete_mount_recovery(osb);
663 if (ocfs2_mount_local(osb))
664 snprintf(nodestr, sizeof(nodestr), "local");
666 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
668 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
669 "with %s data mode.\n",
670 osb->dev_str, nodestr, osb->slot_num,
671 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
674 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
675 wake_up(&osb->osb_mount_event);
688 atomic_set(&osb->vol_state, VOLUME_DISABLED);
689 wake_up(&osb->osb_mount_event);
690 ocfs2_dismount_volume(sb, 1);
697 static int ocfs2_get_sb(struct file_system_type *fs_type,
699 const char *dev_name,
701 struct vfsmount *mnt)
703 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
707 static struct file_system_type ocfs2_fs_type = {
708 .owner = THIS_MODULE,
710 .get_sb = ocfs2_get_sb, /* is this called when we mount
712 .kill_sb = kill_block_super, /* set to the generic one
713 * right now, but do we
714 * need to change that? */
715 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
719 static int ocfs2_parse_options(struct super_block *sb,
721 unsigned long *mount_opt,
728 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
729 options ? options : "(none)");
732 *slot = OCFS2_INVALID_SLOT;
739 while ((p = strsep(&options, ",")) != NULL) {
741 substring_t args[MAX_OPT_ARGS];
742 struct ocfs2_super * osb = OCFS2_SB(sb);
747 token = match_token(p, tokens, args);
750 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
753 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
756 if (match_int(&args[0], &option)) {
761 *mount_opt |= OCFS2_MOUNT_BARRIER;
763 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
766 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
769 *mount_opt |= OCFS2_MOUNT_NOINTR;
772 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
775 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
777 case Opt_data_ordered:
778 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
780 case Opt_data_writeback:
781 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
783 case Opt_atime_quantum:
784 if (match_int(&args[0], &option)) {
789 osb->s_atime_quantum = option;
791 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
795 if (match_int(&args[0], &option)) {
804 "Unrecognized mount option \"%s\" "
805 "or missing value\n", p);
818 static int __init ocfs2_init(void)
824 ocfs2_print_version();
826 status = init_ocfs2_uptodate_cache();
832 status = ocfs2_initialize_mem_caches();
838 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
844 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
845 if (!ocfs2_debugfs_root) {
847 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
852 ocfs2_free_mem_caches();
853 exit_ocfs2_uptodate_cache();
859 return register_filesystem(&ocfs2_fs_type);
864 static void __exit ocfs2_exit(void)
869 flush_workqueue(ocfs2_wq);
870 destroy_workqueue(ocfs2_wq);
873 debugfs_remove(ocfs2_debugfs_root);
875 ocfs2_free_mem_caches();
877 unregister_filesystem(&ocfs2_fs_type);
879 exit_ocfs2_uptodate_cache();
884 static void ocfs2_put_super(struct super_block *sb)
886 mlog_entry("(0x%p)\n", sb);
888 ocfs2_sync_blockdev(sb);
889 ocfs2_dismount_volume(sb, 0);
894 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
896 struct ocfs2_super *osb;
897 u32 numbits, freebits;
899 struct ocfs2_dinode *bm_lock;
900 struct buffer_head *bh = NULL;
901 struct inode *inode = NULL;
903 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
905 osb = OCFS2_SB(dentry->d_sb);
907 inode = ocfs2_get_system_file_inode(osb,
908 GLOBAL_BITMAP_SYSTEM_INODE,
911 mlog(ML_ERROR, "failed to get bitmap inode\n");
916 status = ocfs2_meta_lock(inode, &bh, 0);
922 bm_lock = (struct ocfs2_dinode *) bh->b_data;
924 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
925 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
927 buf->f_type = OCFS2_SUPER_MAGIC;
928 buf->f_bsize = dentry->d_sb->s_blocksize;
929 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
930 buf->f_blocks = ((sector_t) numbits) *
931 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
932 buf->f_bfree = ((sector_t) freebits) *
933 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
934 buf->f_bavail = buf->f_bfree;
935 buf->f_files = numbits;
936 buf->f_ffree = freebits;
940 ocfs2_meta_unlock(inode, 0);
951 static void ocfs2_inode_init_once(void *data,
952 struct kmem_cache *cachep,
955 struct ocfs2_inode_info *oi = data;
958 oi->ip_open_count = 0;
959 spin_lock_init(&oi->ip_lock);
960 ocfs2_extent_map_init(&oi->vfs_inode);
961 INIT_LIST_HEAD(&oi->ip_io_markers);
962 oi->ip_created_trans = 0;
963 oi->ip_last_trans = 0;
964 oi->ip_dir_start_lookup = 0;
966 init_rwsem(&oi->ip_alloc_sem);
967 mutex_init(&oi->ip_io_mutex);
972 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
973 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
974 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
975 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
977 ocfs2_metadata_cache_init(&oi->vfs_inode);
979 inode_init_once(&oi->vfs_inode);
982 static int ocfs2_initialize_mem_caches(void)
984 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
985 sizeof(struct ocfs2_inode_info),
987 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
989 ocfs2_inode_init_once, NULL);
990 if (!ocfs2_inode_cachep)
996 static void ocfs2_free_mem_caches(void)
998 if (ocfs2_inode_cachep)
999 kmem_cache_destroy(ocfs2_inode_cachep);
1001 ocfs2_inode_cachep = NULL;
1004 static int ocfs2_get_sector(struct super_block *sb,
1005 struct buffer_head **bh,
1009 if (!sb_set_blocksize(sb, sect_size)) {
1010 mlog(ML_ERROR, "unable to set blocksize\n");
1014 *bh = sb_getblk(sb, block);
1020 if (!buffer_dirty(*bh))
1021 clear_buffer_uptodate(*bh);
1023 ll_rw_block(READ, 1, bh);
1024 wait_on_buffer(*bh);
1028 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1029 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1033 /* XXX hold a ref on the node while mounte? easy enough, if
1035 if (ocfs2_mount_local(osb))
1038 osb->node_num = o2nm_this_node();
1040 if (osb->node_num == O2NM_MAX_NODES) {
1041 mlog(ML_ERROR, "could not find this host's node number\n");
1046 mlog(0, "I am node %d\n", osb->node_num);
1053 static int ocfs2_mount_volume(struct super_block *sb)
1056 int unlock_super = 0;
1057 struct ocfs2_super *osb = OCFS2_SB(sb);
1061 if (ocfs2_is_hard_readonly(osb))
1064 status = ocfs2_fill_local_node_info(osb);
1070 status = ocfs2_register_hb_callbacks(osb);
1076 status = ocfs2_dlm_init(osb);
1082 /* requires vote_thread to be running. */
1083 status = ocfs2_register_net_handlers(osb);
1089 status = ocfs2_super_lock(osb, 1);
1096 /* This will load up the node map and add ourselves to it. */
1097 status = ocfs2_find_slot(osb);
1103 ocfs2_populate_mounted_map(osb);
1105 /* load all node-local system inodes */
1106 status = ocfs2_init_local_system_inodes(osb);
1112 status = ocfs2_check_volume(osb);
1118 status = ocfs2_truncate_log_init(osb);
1124 if (ocfs2_mount_local(osb))
1127 /* This should be sent *after* we recovered our journal as it
1128 * will cause other nodes to unmark us as needing
1129 * recovery. However, we need to send it *before* dropping the
1130 * super block lock as otherwise their recovery threads might
1131 * try to clean us up while we're live! */
1132 status = ocfs2_request_mount_vote(osb);
1138 ocfs2_super_unlock(osb, 1);
1144 /* we can't grab the goofy sem lock from inside wait_event, so we use
1145 * memory barriers to make sure that we'll see the null task before
1147 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1150 return osb->recovery_thread_task != NULL;
1153 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1156 struct ocfs2_super *osb = NULL;
1159 mlog_entry("(0x%p)\n", sb);
1165 ocfs2_shutdown_local_alloc(osb);
1167 ocfs2_truncate_log_shutdown(osb);
1169 /* disable any new recovery threads and wait for any currently
1170 * running ones to exit. Do this before setting the vol_state. */
1171 mutex_lock(&osb->recovery_lock);
1172 osb->disable_recovery = 1;
1173 mutex_unlock(&osb->recovery_lock);
1174 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1176 /* At this point, we know that no more recovery threads can be
1177 * launched, so wait for any recovery completion work to
1179 flush_workqueue(ocfs2_wq);
1181 ocfs2_journal_shutdown(osb);
1183 ocfs2_sync_blockdev(sb);
1185 /* No dlm means we've failed during mount, so skip all the
1186 * steps which depended on that to complete. */
1188 tmp = ocfs2_super_lock(osb, 1);
1194 tmp = ocfs2_request_umount_vote(osb);
1198 if (osb->slot_num != OCFS2_INVALID_SLOT)
1199 ocfs2_put_slot(osb);
1201 ocfs2_super_unlock(osb, 1);
1204 ocfs2_release_system_inodes(osb);
1207 ocfs2_unregister_net_handlers(osb);
1209 ocfs2_dlm_shutdown(osb);
1212 ocfs2_clear_hb_callbacks(osb);
1214 debugfs_remove(osb->osb_debug_root);
1217 ocfs2_stop_heartbeat(osb);
1219 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1221 if (ocfs2_mount_local(osb))
1222 snprintf(nodestr, sizeof(nodestr), "local");
1224 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1226 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1227 osb->dev_str, nodestr);
1229 ocfs2_delete_osb(osb);
1232 sb->s_fs_info = NULL;
1235 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1236 unsigned uuid_bytes)
1241 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1243 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1244 if (osb->uuid_str == NULL)
1247 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1248 /* print with null */
1249 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1250 if (ret != 2) /* drop super cleans up */
1252 /* then only advance past the last char */
1259 static int ocfs2_initialize_super(struct super_block *sb,
1260 struct buffer_head *bh,
1265 struct ocfs2_dinode *di = NULL;
1266 struct inode *inode = NULL;
1267 struct buffer_head *bitmap_bh = NULL;
1268 struct ocfs2_journal *journal;
1269 __le32 uuid_net_key;
1270 struct ocfs2_super *osb;
1274 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1281 sb->s_fs_info = osb;
1282 sb->s_op = &ocfs2_sops;
1283 sb->s_export_op = &ocfs2_export_ops;
1284 sb->s_flags |= MS_NOATIME;
1285 /* this is needed to support O_LARGEFILE */
1286 sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1289 /* Save off for ocfs2_rw_direct */
1290 osb->s_sectsize_bits = blksize_bits(sector_size);
1291 BUG_ON(!osb->s_sectsize_bits);
1293 osb->net_response_ids = 0;
1294 spin_lock_init(&osb->net_response_lock);
1295 INIT_LIST_HEAD(&osb->net_response_list);
1297 INIT_LIST_HEAD(&osb->osb_net_handlers);
1298 init_waitqueue_head(&osb->recovery_event);
1299 spin_lock_init(&osb->vote_task_lock);
1300 init_waitqueue_head(&osb->vote_event);
1301 osb->vote_work_sequence = 0;
1302 osb->vote_wake_sequence = 0;
1303 INIT_LIST_HEAD(&osb->blocked_lock_list);
1304 osb->blocked_lock_count = 0;
1305 INIT_LIST_HEAD(&osb->vote_list);
1306 spin_lock_init(&osb->osb_lock);
1308 atomic_set(&osb->alloc_stats.moves, 0);
1309 atomic_set(&osb->alloc_stats.local_data, 0);
1310 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1311 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1312 atomic_set(&osb->alloc_stats.bg_extends, 0);
1314 ocfs2_init_node_maps(osb);
1316 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1317 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1319 mutex_init(&osb->recovery_lock);
1321 osb->disable_recovery = 0;
1322 osb->recovery_thread_task = NULL;
1324 init_waitqueue_head(&osb->checkpoint_event);
1325 atomic_set(&osb->needs_checkpoint, 0);
1327 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1329 osb->node_num = O2NM_INVALID_NODE_NUM;
1330 osb->slot_num = OCFS2_INVALID_SLOT;
1332 osb->local_alloc_state = OCFS2_LA_UNUSED;
1333 osb->local_alloc_bh = NULL;
1335 ocfs2_setup_hb_callbacks(osb);
1337 init_waitqueue_head(&osb->osb_mount_event);
1339 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1340 if (!osb->vol_label) {
1341 mlog(ML_ERROR, "unable to alloc vol label\n");
1346 di = (struct ocfs2_dinode *)bh->b_data;
1348 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1349 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1350 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1355 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1357 init_waitqueue_head(&osb->osb_wipe_event);
1358 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1359 sizeof(*osb->osb_orphan_wipes),
1361 if (!osb->osb_orphan_wipes) {
1367 osb->s_feature_compat =
1368 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1369 osb->s_feature_ro_compat =
1370 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1371 osb->s_feature_incompat =
1372 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1374 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1375 mlog(ML_ERROR, "couldn't mount because of unsupported "
1376 "optional features (%x).\n", i);
1380 if (!(osb->sb->s_flags & MS_RDONLY) &&
1381 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1382 mlog(ML_ERROR, "couldn't mount RDWR because of "
1383 "unsupported optional features (%x).\n", i);
1388 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1391 * This should be done in ocfs2_journal_init(), but unknown
1392 * ordering issues will cause the filesystem to crash.
1393 * If anyone wants to figure out what part of the code
1394 * refers to osb->journal before ocfs2_journal_init() is run,
1397 /* initialize our journal structure */
1399 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1401 mlog(ML_ERROR, "unable to alloc journal\n");
1405 osb->journal = journal;
1406 journal->j_osb = osb;
1408 atomic_set(&journal->j_num_trans, 0);
1409 init_rwsem(&journal->j_trans_barrier);
1410 init_waitqueue_head(&journal->j_checkpointed);
1411 spin_lock_init(&journal->j_lock);
1412 journal->j_trans_id = (unsigned long) 1;
1413 INIT_LIST_HEAD(&journal->j_la_cleanups);
1414 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1415 journal->j_state = OCFS2_JOURNAL_FREE;
1417 /* get some pseudo constants for clustersize bits */
1418 osb->s_clustersize_bits =
1419 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1420 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1421 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1423 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1424 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1425 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1426 osb->s_clustersize);
1431 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1433 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1434 "what jbd can address in 32 bits.\n");
1439 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1440 sizeof(di->id2.i_super.s_uuid))) {
1441 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1446 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1447 osb->net_key = le32_to_cpu(uuid_net_key);
1449 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1450 osb->vol_label[63] = '\0';
1451 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1452 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1453 osb->first_cluster_group_blkno =
1454 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1455 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1456 mlog(0, "vol_label: %s\n", osb->vol_label);
1457 mlog(0, "uuid: %s\n", osb->uuid_str);
1458 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1459 (unsigned long long)osb->root_blkno,
1460 (unsigned long long)osb->system_dir_blkno);
1462 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1463 if (!osb->osb_dlm_debug) {
1469 atomic_set(&osb->vol_state, VOLUME_INIT);
1471 /* load root, system_dir, and all global system inodes */
1472 status = ocfs2_init_global_system_inodes(osb);
1481 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1482 OCFS2_INVALID_SLOT);
1489 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1491 /* We don't have a cluster lock on the bitmap here because
1492 * we're only interested in static information and the extra
1493 * complexity at mount time isn't worht it. Don't pass the
1494 * inode in to the read function though as we don't want it to
1495 * be put in the cache. */
1496 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1504 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1505 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1507 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1508 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1510 status = ocfs2_init_slot_info(osb);
1522 * will return: -EAGAIN if it is ok to keep searching for superblocks
1523 * -EINVAL if there is a bad superblock
1526 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1527 struct buffer_head *bh,
1530 int status = -EAGAIN;
1534 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1535 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1537 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1538 mlog(ML_ERROR, "found superblock with incorrect block "
1539 "size: found %u, should be %u\n",
1540 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1542 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1543 OCFS2_MAJOR_REV_LEVEL ||
1544 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1545 OCFS2_MINOR_REV_LEVEL) {
1546 mlog(ML_ERROR, "found superblock with bad version: "
1547 "found %u.%u, should be %u.%u\n",
1548 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1549 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1550 OCFS2_MAJOR_REV_LEVEL,
1551 OCFS2_MINOR_REV_LEVEL);
1552 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1553 mlog(ML_ERROR, "bad block number on superblock: "
1554 "found %llu, should be %llu\n",
1555 (unsigned long long)le64_to_cpu(di->i_blkno),
1556 (unsigned long long)bh->b_blocknr);
1557 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1558 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1559 mlog(ML_ERROR, "bad cluster size found: %u\n",
1560 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1561 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1562 mlog(ML_ERROR, "bad root_blkno: 0\n");
1563 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1564 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1565 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1567 "Superblock slots found greater than file system "
1568 "maximum: found %u, max %u\n",
1569 le16_to_cpu(di->id2.i_super.s_max_slots),
1581 static int ocfs2_check_volume(struct ocfs2_super *osb)
1586 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1592 /* Init our journal object. */
1593 status = ocfs2_journal_init(osb->journal, &dirty);
1595 mlog(ML_ERROR, "Could not initialize journal!\n");
1599 /* If the journal was unmounted cleanly then we don't want to
1600 * recover anything. Otherwise, journal_load will do that
1601 * dirty work for us :) */
1603 status = ocfs2_journal_wipe(osb->journal, 0);
1609 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1610 "recovering volume.\n");
1613 local = ocfs2_mount_local(osb);
1615 /* will play back anything left in the journal. */
1616 ocfs2_journal_load(osb->journal, local);
1619 /* recover my local alloc if we didn't unmount cleanly. */
1620 status = ocfs2_begin_local_alloc_recovery(osb,
1627 /* we complete the recovery process after we've marked
1628 * ourselves as mounted. */
1631 mlog(0, "Journal loaded.\n");
1633 status = ocfs2_load_local_alloc(osb);
1640 /* Recovery will be completed after we've mounted the
1641 * rest of the volume. */
1643 osb->local_alloc_copy = local_alloc;
1647 /* go through each journal, trylock it and if you get the
1648 * lock, and it's marked as dirty, set the bit in the recover
1649 * map and launch a recovery thread for it. */
1650 status = ocfs2_mark_dead_nodes(osb);
1663 * The routine gets called from dismount or close whenever a dismount on
1664 * volume is requested and the osb open count becomes 1.
1665 * It will remove the osb from the global list and also free up all the
1666 * initialized resources and fileobject.
1668 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1672 /* This function assumes that the caller has the main osb resource */
1675 ocfs2_free_slot_info(osb->slot_info);
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);