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, int is_remount);
86 static void ocfs2_put_super(struct super_block *sb);
87 static int ocfs2_mount_volume(struct super_block *sb);
88 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
89 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
90 static int ocfs2_initialize_mem_caches(void);
91 static void ocfs2_free_mem_caches(void);
92 static void ocfs2_delete_osb(struct ocfs2_super *osb);
94 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
96 static int ocfs2_sync_fs(struct super_block *sb, int wait);
98 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
99 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
100 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
101 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
102 static int ocfs2_check_volume(struct ocfs2_super *osb);
103 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
104 struct buffer_head *bh,
106 static int ocfs2_initialize_super(struct super_block *sb,
107 struct buffer_head *bh,
109 static int ocfs2_get_sector(struct super_block *sb,
110 struct buffer_head **bh,
113 static void ocfs2_write_super(struct super_block *sb);
114 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
115 static void ocfs2_destroy_inode(struct inode *inode);
117 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
119 static const struct super_operations ocfs2_sops = {
120 .statfs = ocfs2_statfs,
121 .alloc_inode = ocfs2_alloc_inode,
122 .destroy_inode = ocfs2_destroy_inode,
123 .drop_inode = ocfs2_drop_inode,
124 .clear_inode = ocfs2_clear_inode,
125 .delete_inode = ocfs2_delete_inode,
126 .sync_fs = ocfs2_sync_fs,
127 .write_super = ocfs2_write_super,
128 .put_super = ocfs2_put_super,
129 .remount_fs = ocfs2_remount,
146 static match_table_t tokens = {
147 {Opt_barrier, "barrier=%u"},
148 {Opt_err_panic, "errors=panic"},
149 {Opt_err_ro, "errors=remount-ro"},
151 {Opt_nointr, "nointr"},
152 {Opt_hb_none, OCFS2_HB_NONE},
153 {Opt_hb_local, OCFS2_HB_LOCAL},
154 {Opt_data_ordered, "data=ordered"},
155 {Opt_data_writeback, "data=writeback"},
156 {Opt_atime_quantum, "atime_quantum=%u"},
161 * write_super and sync_fs ripped right out of ext3.
163 static void ocfs2_write_super(struct super_block *sb)
165 if (mutex_trylock(&sb->s_lock) != 0)
170 static int ocfs2_sync_fs(struct super_block *sb, int wait)
174 struct ocfs2_super *osb = OCFS2_SB(sb);
178 if (ocfs2_is_hard_readonly(osb))
182 status = ocfs2_flush_truncate_log(osb);
186 ocfs2_schedule_truncate_log_flush(osb, 0);
189 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
191 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
197 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
199 struct inode *new = NULL;
205 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
207 status = PTR_ERR(new);
211 osb->root_inode = new;
213 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
215 status = PTR_ERR(new);
219 osb->sys_root_inode = new;
221 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
222 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
223 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
225 ocfs2_release_system_inodes(osb);
228 /* FIXME: Should ERROR_RO_FS */
229 mlog(ML_ERROR, "Unable to load system inode %d, "
230 "possibly corrupt fs?", i);
233 // the array now has one ref, so drop this one
242 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
244 struct inode *new = NULL;
250 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
251 i < NUM_SYSTEM_INODES;
253 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
255 ocfs2_release_system_inodes(osb);
257 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
258 status, i, osb->slot_num);
261 /* the array now has one ref, so drop this one */
270 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
277 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
278 inode = osb->system_inodes[i];
281 osb->system_inodes[i] = NULL;
285 inode = osb->sys_root_inode;
288 osb->sys_root_inode = NULL;
291 inode = osb->root_inode;
294 osb->root_inode = NULL;
301 /* We're allocating fs objects, use GFP_NOFS */
302 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
304 struct ocfs2_inode_info *oi;
306 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
310 return &oi->vfs_inode;
313 static void ocfs2_destroy_inode(struct inode *inode)
315 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
318 /* From xfs_super.c:xfs_max_file_offset
319 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
321 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
323 unsigned int pagefactor = 1;
324 unsigned int bitshift = BITS_PER_LONG - 1;
326 /* Figure out maximum filesize, on Linux this can depend on
327 * the filesystem blocksize (on 32 bit platforms).
328 * __block_prepare_write does this in an [unsigned] long...
329 * page->index << (PAGE_CACHE_SHIFT - bbits)
330 * So, for page sized blocks (4K on 32 bit platforms),
331 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
332 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
333 * but for smaller blocksizes it is less (bbits = log2 bsize).
334 * Note1: get_block_t takes a long (implicit cast from above)
335 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
336 * can optionally convert the [unsigned] long from above into
337 * an [unsigned] long long.
340 #if BITS_PER_LONG == 32
341 # if defined(CONFIG_LBD)
342 BUILD_BUG_ON(sizeof(sector_t) != 8);
343 pagefactor = PAGE_CACHE_SIZE;
344 bitshift = BITS_PER_LONG;
346 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
350 return (((unsigned long long)pagefactor) << bitshift) - 1;
353 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
355 int incompat_features;
357 unsigned long parsed_options;
358 struct ocfs2_super *osb = OCFS2_SB(sb);
360 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
365 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
366 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
368 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
372 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
373 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
375 mlog(ML_ERROR, "Cannot change data mode on remount\n");
379 /* We're going to/from readonly mode. */
380 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
381 /* Lock here so the check of HARD_RO and the potential
382 * setting of SOFT_RO is atomic. */
383 spin_lock(&osb->osb_lock);
384 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
385 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
390 if (*flags & MS_RDONLY) {
391 mlog(0, "Going to ro mode.\n");
392 sb->s_flags |= MS_RDONLY;
393 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
395 mlog(0, "Making ro filesystem writeable.\n");
397 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
398 mlog(ML_ERROR, "Cannot remount RDWR "
399 "filesystem due to previous errors.\n");
403 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
404 if (incompat_features) {
405 mlog(ML_ERROR, "Cannot remount RDWR because "
406 "of unsupported optional features "
407 "(%x).\n", incompat_features);
411 sb->s_flags &= ~MS_RDONLY;
412 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
415 spin_unlock(&osb->osb_lock);
419 if (!ocfs2_is_hard_readonly(osb))
420 ocfs2_set_journal_params(osb);
422 /* Only save off the new mount options in case of a successful
424 osb->s_mount_opt = parsed_options;
430 static int ocfs2_sb_probe(struct super_block *sb,
431 struct buffer_head **bh,
434 int status = 0, tmpstat;
435 struct ocfs1_vol_disk_hdr *hdr;
436 struct ocfs2_dinode *di;
442 *sector_size = bdev_hardsect_size(sb->s_bdev);
443 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
444 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
445 *sector_size, OCFS2_MAX_BLOCKSIZE);
450 /* Can this really happen? */
451 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
452 *sector_size = OCFS2_MIN_BLOCKSIZE;
454 /* check block zero for old format */
455 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
460 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
461 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
462 mlog(ML_ERROR, "incompatible version: %u.%u\n",
463 hdr->major_version, hdr->minor_version);
466 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
467 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
468 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
475 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
476 "upgraded before mounting with ocfs v2\n");
481 * Now check at magic offset for 512, 1024, 2048, 4096
482 * blocksizes. 4096 is the maximum blocksize because it is
483 * the minimum clustersize.
486 for (blksize = *sector_size;
487 blksize <= OCFS2_MAX_BLOCKSIZE;
489 tmpstat = ocfs2_get_sector(sb, bh,
490 OCFS2_SUPER_BLOCK_BLKNO,
497 di = (struct ocfs2_dinode *) (*bh)->b_data;
498 status = ocfs2_verify_volume(di, *bh, blksize);
503 if (status != -EAGAIN)
511 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
513 if (ocfs2_mount_local(osb)) {
514 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
515 mlog(ML_ERROR, "Cannot heartbeat on a locally "
516 "mounted device.\n");
521 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
522 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
523 mlog(ML_ERROR, "Heartbeat has to be started to mount "
524 "a read-write clustered device.\n");
532 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
535 int status, sector_size;
536 unsigned long parsed_opt;
537 struct inode *inode = NULL;
538 struct ocfs2_super *osb = NULL;
539 struct buffer_head *bh = NULL;
542 mlog_entry("%p, %p, %i", sb, data, silent);
544 if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
546 goto read_super_error;
549 /* for now we only have one cluster/node, make sure we see it
550 * in the heartbeat universe */
551 if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
552 if (!o2hb_check_local_node_heartbeating()) {
554 goto read_super_error;
558 /* probe for superblock */
559 status = ocfs2_sb_probe(sb, &bh, §or_size);
561 mlog(ML_ERROR, "superblock probe failed!\n");
562 goto read_super_error;
565 status = ocfs2_initialize_super(sb, bh, sector_size);
569 goto read_super_error;
573 osb->s_mount_opt = parsed_opt;
575 sb->s_magic = OCFS2_SUPER_MAGIC;
577 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
579 if (bdev_read_only(sb->s_bdev)) {
580 if (!(sb->s_flags & MS_RDONLY)) {
582 mlog(ML_ERROR, "Readonly device detected but readonly "
583 "mount was not specified.\n");
584 goto read_super_error;
587 /* You should not be able to start a local heartbeat
588 * on a readonly device. */
589 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
591 mlog(ML_ERROR, "Local heartbeat specified on readonly "
593 goto read_super_error;
596 status = ocfs2_check_journals_nolocks(osb);
598 if (status == -EROFS)
599 mlog(ML_ERROR, "Recovery required on readonly "
600 "file system, but write access is "
604 goto read_super_error;
607 ocfs2_set_ro_flag(osb, 1);
609 printk(KERN_NOTICE "Readonly device detected. No cluster "
610 "services will be utilized for this mount. Recovery "
611 "will be skipped.\n");
614 if (!ocfs2_is_hard_readonly(osb)) {
615 if (sb->s_flags & MS_RDONLY)
616 ocfs2_set_ro_flag(osb, 0);
619 status = ocfs2_verify_heartbeat(osb);
622 goto read_super_error;
625 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
627 if (!osb->osb_debug_root) {
629 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
630 goto read_super_error;
633 status = ocfs2_mount_volume(sb);
635 inode = igrab(osb->root_inode);
638 goto read_super_error;
643 goto read_super_error;
646 root = d_alloc_root(inode);
650 goto read_super_error;
655 ocfs2_complete_mount_recovery(osb);
657 if (ocfs2_mount_local(osb))
658 snprintf(nodestr, sizeof(nodestr), "local");
660 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
662 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
663 "with %s data mode.\n",
664 osb->dev_str, nodestr, osb->slot_num,
665 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
668 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
669 wake_up(&osb->osb_mount_event);
682 atomic_set(&osb->vol_state, VOLUME_DISABLED);
683 wake_up(&osb->osb_mount_event);
684 ocfs2_dismount_volume(sb, 1);
691 static int ocfs2_get_sb(struct file_system_type *fs_type,
693 const char *dev_name,
695 struct vfsmount *mnt)
697 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
701 static struct file_system_type ocfs2_fs_type = {
702 .owner = THIS_MODULE,
704 .get_sb = ocfs2_get_sb, /* is this called when we mount
706 .kill_sb = kill_block_super, /* set to the generic one
707 * right now, but do we
708 * need to change that? */
709 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
713 static int ocfs2_parse_options(struct super_block *sb,
715 unsigned long *mount_opt,
721 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
722 options ? options : "(none)");
731 while ((p = strsep(&options, ",")) != NULL) {
733 substring_t args[MAX_OPT_ARGS];
734 struct ocfs2_super * osb = OCFS2_SB(sb);
739 token = match_token(p, tokens, args);
742 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
745 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
748 if (match_int(&args[0], &option)) {
753 *mount_opt |= OCFS2_MOUNT_BARRIER;
755 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
758 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
761 *mount_opt |= OCFS2_MOUNT_NOINTR;
764 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
767 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
769 case Opt_data_ordered:
770 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
772 case Opt_data_writeback:
773 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
775 case Opt_atime_quantum:
776 if (match_int(&args[0], &option)) {
781 osb->s_atime_quantum = option;
783 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
787 "Unrecognized mount option \"%s\" "
788 "or missing value\n", p);
801 static int __init ocfs2_init(void)
807 ocfs2_print_version();
809 status = init_ocfs2_uptodate_cache();
815 status = ocfs2_initialize_mem_caches();
821 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
827 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
828 if (!ocfs2_debugfs_root) {
830 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
835 ocfs2_free_mem_caches();
836 exit_ocfs2_uptodate_cache();
842 return register_filesystem(&ocfs2_fs_type);
847 static void __exit ocfs2_exit(void)
852 flush_workqueue(ocfs2_wq);
853 destroy_workqueue(ocfs2_wq);
856 debugfs_remove(ocfs2_debugfs_root);
858 ocfs2_free_mem_caches();
860 unregister_filesystem(&ocfs2_fs_type);
862 exit_ocfs2_uptodate_cache();
867 static void ocfs2_put_super(struct super_block *sb)
869 mlog_entry("(0x%p)\n", sb);
871 ocfs2_sync_blockdev(sb);
872 ocfs2_dismount_volume(sb, 0);
877 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
879 struct ocfs2_super *osb;
880 u32 numbits, freebits;
882 struct ocfs2_dinode *bm_lock;
883 struct buffer_head *bh = NULL;
884 struct inode *inode = NULL;
886 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
888 osb = OCFS2_SB(dentry->d_sb);
890 inode = ocfs2_get_system_file_inode(osb,
891 GLOBAL_BITMAP_SYSTEM_INODE,
894 mlog(ML_ERROR, "failed to get bitmap inode\n");
899 status = ocfs2_meta_lock(inode, &bh, 0);
905 bm_lock = (struct ocfs2_dinode *) bh->b_data;
907 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
908 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
910 buf->f_type = OCFS2_SUPER_MAGIC;
911 buf->f_bsize = dentry->d_sb->s_blocksize;
912 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
913 buf->f_blocks = ((sector_t) numbits) *
914 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
915 buf->f_bfree = ((sector_t) freebits) *
916 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
917 buf->f_bavail = buf->f_bfree;
918 buf->f_files = numbits;
919 buf->f_ffree = freebits;
923 ocfs2_meta_unlock(inode, 0);
934 static void ocfs2_inode_init_once(void *data,
935 struct kmem_cache *cachep,
938 struct ocfs2_inode_info *oi = data;
941 oi->ip_open_count = 0;
942 spin_lock_init(&oi->ip_lock);
943 ocfs2_extent_map_init(&oi->vfs_inode);
944 INIT_LIST_HEAD(&oi->ip_io_markers);
945 oi->ip_created_trans = 0;
946 oi->ip_last_trans = 0;
947 oi->ip_dir_start_lookup = 0;
949 init_rwsem(&oi->ip_alloc_sem);
950 mutex_init(&oi->ip_io_mutex);
955 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
956 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
957 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
958 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
960 ocfs2_metadata_cache_init(&oi->vfs_inode);
962 inode_init_once(&oi->vfs_inode);
965 static int ocfs2_initialize_mem_caches(void)
967 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
968 sizeof(struct ocfs2_inode_info),
970 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
972 ocfs2_inode_init_once, NULL);
973 if (!ocfs2_inode_cachep)
979 static void ocfs2_free_mem_caches(void)
981 if (ocfs2_inode_cachep)
982 kmem_cache_destroy(ocfs2_inode_cachep);
984 ocfs2_inode_cachep = NULL;
987 static int ocfs2_get_sector(struct super_block *sb,
988 struct buffer_head **bh,
992 if (!sb_set_blocksize(sb, sect_size)) {
993 mlog(ML_ERROR, "unable to set blocksize\n");
997 *bh = sb_getblk(sb, block);
1003 if (!buffer_dirty(*bh))
1004 clear_buffer_uptodate(*bh);
1006 ll_rw_block(READ, 1, bh);
1007 wait_on_buffer(*bh);
1011 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1012 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1016 /* XXX hold a ref on the node while mounte? easy enough, if
1018 if (ocfs2_mount_local(osb))
1021 osb->node_num = o2nm_this_node();
1023 if (osb->node_num == O2NM_MAX_NODES) {
1024 mlog(ML_ERROR, "could not find this host's node number\n");
1029 mlog(0, "I am node %d\n", osb->node_num);
1036 static int ocfs2_mount_volume(struct super_block *sb)
1039 int unlock_super = 0;
1040 struct ocfs2_super *osb = OCFS2_SB(sb);
1044 if (ocfs2_is_hard_readonly(osb))
1047 status = ocfs2_fill_local_node_info(osb);
1053 status = ocfs2_register_hb_callbacks(osb);
1059 status = ocfs2_dlm_init(osb);
1065 /* requires vote_thread to be running. */
1066 status = ocfs2_register_net_handlers(osb);
1072 status = ocfs2_super_lock(osb, 1);
1079 /* This will load up the node map and add ourselves to it. */
1080 status = ocfs2_find_slot(osb);
1086 ocfs2_populate_mounted_map(osb);
1088 /* load all node-local system inodes */
1089 status = ocfs2_init_local_system_inodes(osb);
1095 status = ocfs2_check_volume(osb);
1101 status = ocfs2_truncate_log_init(osb);
1107 if (ocfs2_mount_local(osb))
1110 /* This should be sent *after* we recovered our journal as it
1111 * will cause other nodes to unmark us as needing
1112 * recovery. However, we need to send it *before* dropping the
1113 * super block lock as otherwise their recovery threads might
1114 * try to clean us up while we're live! */
1115 status = ocfs2_request_mount_vote(osb);
1121 ocfs2_super_unlock(osb, 1);
1127 /* we can't grab the goofy sem lock from inside wait_event, so we use
1128 * memory barriers to make sure that we'll see the null task before
1130 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1133 return osb->recovery_thread_task != NULL;
1136 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1139 struct ocfs2_super *osb = NULL;
1142 mlog_entry("(0x%p)\n", sb);
1148 ocfs2_shutdown_local_alloc(osb);
1150 ocfs2_truncate_log_shutdown(osb);
1152 /* disable any new recovery threads and wait for any currently
1153 * running ones to exit. Do this before setting the vol_state. */
1154 mutex_lock(&osb->recovery_lock);
1155 osb->disable_recovery = 1;
1156 mutex_unlock(&osb->recovery_lock);
1157 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1159 /* At this point, we know that no more recovery threads can be
1160 * launched, so wait for any recovery completion work to
1162 flush_workqueue(ocfs2_wq);
1164 ocfs2_journal_shutdown(osb);
1166 ocfs2_sync_blockdev(sb);
1168 /* No dlm means we've failed during mount, so skip all the
1169 * steps which depended on that to complete. */
1171 tmp = ocfs2_super_lock(osb, 1);
1177 tmp = ocfs2_request_umount_vote(osb);
1181 if (osb->slot_num != OCFS2_INVALID_SLOT)
1182 ocfs2_put_slot(osb);
1184 ocfs2_super_unlock(osb, 1);
1187 ocfs2_release_system_inodes(osb);
1190 ocfs2_unregister_net_handlers(osb);
1192 ocfs2_dlm_shutdown(osb);
1195 ocfs2_clear_hb_callbacks(osb);
1197 debugfs_remove(osb->osb_debug_root);
1200 ocfs2_stop_heartbeat(osb);
1202 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1204 if (ocfs2_mount_local(osb))
1205 snprintf(nodestr, sizeof(nodestr), "local");
1207 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1209 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1210 osb->dev_str, nodestr);
1212 ocfs2_delete_osb(osb);
1215 sb->s_fs_info = NULL;
1218 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1219 unsigned uuid_bytes)
1224 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1226 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1227 if (osb->uuid_str == NULL)
1230 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1231 /* print with null */
1232 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1233 if (ret != 2) /* drop super cleans up */
1235 /* then only advance past the last char */
1242 static int ocfs2_initialize_super(struct super_block *sb,
1243 struct buffer_head *bh,
1248 struct ocfs2_dinode *di = NULL;
1249 struct inode *inode = NULL;
1250 struct buffer_head *bitmap_bh = NULL;
1251 struct ocfs2_journal *journal;
1252 __le32 uuid_net_key;
1253 struct ocfs2_super *osb;
1257 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1264 sb->s_fs_info = osb;
1265 sb->s_op = &ocfs2_sops;
1266 sb->s_export_op = &ocfs2_export_ops;
1267 sb->s_flags |= MS_NOATIME;
1268 /* this is needed to support O_LARGEFILE */
1269 sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1272 /* Save off for ocfs2_rw_direct */
1273 osb->s_sectsize_bits = blksize_bits(sector_size);
1274 BUG_ON(!osb->s_sectsize_bits);
1276 osb->net_response_ids = 0;
1277 spin_lock_init(&osb->net_response_lock);
1278 INIT_LIST_HEAD(&osb->net_response_list);
1280 INIT_LIST_HEAD(&osb->osb_net_handlers);
1281 init_waitqueue_head(&osb->recovery_event);
1282 spin_lock_init(&osb->vote_task_lock);
1283 init_waitqueue_head(&osb->vote_event);
1284 osb->vote_work_sequence = 0;
1285 osb->vote_wake_sequence = 0;
1286 INIT_LIST_HEAD(&osb->blocked_lock_list);
1287 osb->blocked_lock_count = 0;
1288 INIT_LIST_HEAD(&osb->vote_list);
1289 spin_lock_init(&osb->osb_lock);
1291 atomic_set(&osb->alloc_stats.moves, 0);
1292 atomic_set(&osb->alloc_stats.local_data, 0);
1293 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1294 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1295 atomic_set(&osb->alloc_stats.bg_extends, 0);
1297 ocfs2_init_node_maps(osb);
1299 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1300 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1302 mutex_init(&osb->recovery_lock);
1304 osb->disable_recovery = 0;
1305 osb->recovery_thread_task = NULL;
1307 init_waitqueue_head(&osb->checkpoint_event);
1308 atomic_set(&osb->needs_checkpoint, 0);
1310 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1312 osb->node_num = O2NM_INVALID_NODE_NUM;
1313 osb->slot_num = OCFS2_INVALID_SLOT;
1315 osb->local_alloc_state = OCFS2_LA_UNUSED;
1316 osb->local_alloc_bh = NULL;
1318 ocfs2_setup_hb_callbacks(osb);
1320 init_waitqueue_head(&osb->osb_mount_event);
1322 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1323 if (!osb->vol_label) {
1324 mlog(ML_ERROR, "unable to alloc vol label\n");
1329 di = (struct ocfs2_dinode *)bh->b_data;
1331 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1332 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1333 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1338 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1340 init_waitqueue_head(&osb->osb_wipe_event);
1341 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1342 sizeof(*osb->osb_orphan_wipes),
1344 if (!osb->osb_orphan_wipes) {
1350 osb->s_feature_compat =
1351 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1352 osb->s_feature_ro_compat =
1353 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1354 osb->s_feature_incompat =
1355 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1357 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1358 mlog(ML_ERROR, "couldn't mount because of unsupported "
1359 "optional features (%x).\n", i);
1363 if (!(osb->sb->s_flags & MS_RDONLY) &&
1364 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1365 mlog(ML_ERROR, "couldn't mount RDWR because of "
1366 "unsupported optional features (%x).\n", i);
1371 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1374 * This should be done in ocfs2_journal_init(), but unknown
1375 * ordering issues will cause the filesystem to crash.
1376 * If anyone wants to figure out what part of the code
1377 * refers to osb->journal before ocfs2_journal_init() is run,
1380 /* initialize our journal structure */
1382 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1384 mlog(ML_ERROR, "unable to alloc journal\n");
1388 osb->journal = journal;
1389 journal->j_osb = osb;
1391 atomic_set(&journal->j_num_trans, 0);
1392 init_rwsem(&journal->j_trans_barrier);
1393 init_waitqueue_head(&journal->j_checkpointed);
1394 spin_lock_init(&journal->j_lock);
1395 journal->j_trans_id = (unsigned long) 1;
1396 INIT_LIST_HEAD(&journal->j_la_cleanups);
1397 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1398 journal->j_state = OCFS2_JOURNAL_FREE;
1400 /* get some pseudo constants for clustersize bits */
1401 osb->s_clustersize_bits =
1402 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1403 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1404 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1406 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1407 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1408 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1409 osb->s_clustersize);
1414 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1416 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1417 "what jbd can address in 32 bits.\n");
1422 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1423 sizeof(di->id2.i_super.s_uuid))) {
1424 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1429 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1430 osb->net_key = le32_to_cpu(uuid_net_key);
1432 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1433 osb->vol_label[63] = '\0';
1434 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1435 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1436 osb->first_cluster_group_blkno =
1437 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1438 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1439 mlog(0, "vol_label: %s\n", osb->vol_label);
1440 mlog(0, "uuid: %s\n", osb->uuid_str);
1441 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1442 (unsigned long long)osb->root_blkno,
1443 (unsigned long long)osb->system_dir_blkno);
1445 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1446 if (!osb->osb_dlm_debug) {
1452 atomic_set(&osb->vol_state, VOLUME_INIT);
1454 /* load root, system_dir, and all global system inodes */
1455 status = ocfs2_init_global_system_inodes(osb);
1464 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1465 OCFS2_INVALID_SLOT);
1472 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1474 /* We don't have a cluster lock on the bitmap here because
1475 * we're only interested in static information and the extra
1476 * complexity at mount time isn't worht it. Don't pass the
1477 * inode in to the read function though as we don't want it to
1478 * be put in the cache. */
1479 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1487 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1488 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1490 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1491 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1493 status = ocfs2_init_slot_info(osb);
1505 * will return: -EAGAIN if it is ok to keep searching for superblocks
1506 * -EINVAL if there is a bad superblock
1509 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1510 struct buffer_head *bh,
1513 int status = -EAGAIN;
1517 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1518 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1520 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1521 mlog(ML_ERROR, "found superblock with incorrect block "
1522 "size: found %u, should be %u\n",
1523 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1525 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1526 OCFS2_MAJOR_REV_LEVEL ||
1527 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1528 OCFS2_MINOR_REV_LEVEL) {
1529 mlog(ML_ERROR, "found superblock with bad version: "
1530 "found %u.%u, should be %u.%u\n",
1531 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1532 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1533 OCFS2_MAJOR_REV_LEVEL,
1534 OCFS2_MINOR_REV_LEVEL);
1535 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1536 mlog(ML_ERROR, "bad block number on superblock: "
1537 "found %llu, should be %llu\n",
1538 (unsigned long long)le64_to_cpu(di->i_blkno),
1539 (unsigned long long)bh->b_blocknr);
1540 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1541 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1542 mlog(ML_ERROR, "bad cluster size found: %u\n",
1543 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1544 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1545 mlog(ML_ERROR, "bad root_blkno: 0\n");
1546 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1547 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1548 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1550 "Superblock slots found greater than file system "
1551 "maximum: found %u, max %u\n",
1552 le16_to_cpu(di->id2.i_super.s_max_slots),
1564 static int ocfs2_check_volume(struct ocfs2_super *osb)
1569 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1575 /* Init our journal object. */
1576 status = ocfs2_journal_init(osb->journal, &dirty);
1578 mlog(ML_ERROR, "Could not initialize journal!\n");
1582 /* If the journal was unmounted cleanly then we don't want to
1583 * recover anything. Otherwise, journal_load will do that
1584 * dirty work for us :) */
1586 status = ocfs2_journal_wipe(osb->journal, 0);
1592 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1593 "recovering volume.\n");
1596 local = ocfs2_mount_local(osb);
1598 /* will play back anything left in the journal. */
1599 ocfs2_journal_load(osb->journal, local);
1602 /* recover my local alloc if we didn't unmount cleanly. */
1603 status = ocfs2_begin_local_alloc_recovery(osb,
1610 /* we complete the recovery process after we've marked
1611 * ourselves as mounted. */
1614 mlog(0, "Journal loaded.\n");
1616 status = ocfs2_load_local_alloc(osb);
1623 /* Recovery will be completed after we've mounted the
1624 * rest of the volume. */
1626 osb->local_alloc_copy = local_alloc;
1630 /* go through each journal, trylock it and if you get the
1631 * lock, and it's marked as dirty, set the bit in the recover
1632 * map and launch a recovery thread for it. */
1633 status = ocfs2_mark_dead_nodes(osb);
1646 * The routine gets called from dismount or close whenever a dismount on
1647 * volume is requested and the osb open count becomes 1.
1648 * It will remove the osb from the global list and also free up all the
1649 * initialized resources and fileobject.
1651 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1655 /* This function assumes that the caller has the main osb resource */
1658 ocfs2_free_slot_info(osb->slot_info);
1660 kfree(osb->osb_orphan_wipes);
1662 * This belongs in journal shutdown, but because we have to
1663 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1666 kfree(osb->journal);
1667 if (osb->local_alloc_copy)
1668 kfree(osb->local_alloc_copy);
1669 kfree(osb->uuid_str);
1670 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1671 memset(osb, 0, sizeof(struct ocfs2_super));
1676 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1677 * panic(). We do not support continue-on-error operation. */
1678 static void ocfs2_handle_error(struct super_block *sb)
1680 struct ocfs2_super *osb = OCFS2_SB(sb);
1682 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1683 panic("OCFS2: (device %s): panic forced after error\n",
1686 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1688 if (sb->s_flags & MS_RDONLY &&
1689 (ocfs2_is_soft_readonly(osb) ||
1690 ocfs2_is_hard_readonly(osb)))
1693 printk(KERN_CRIT "File system is now read-only due to the potential "
1694 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1695 "system is unmounted.\n");
1696 sb->s_flags |= MS_RDONLY;
1697 ocfs2_set_ro_flag(osb, 0);
1700 static char error_buf[1024];
1702 void __ocfs2_error(struct super_block *sb,
1703 const char *function,
1704 const char *fmt, ...)
1708 va_start(args, fmt);
1709 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1712 /* Not using mlog here because we want to show the actual
1713 * function the error came from. */
1714 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1715 sb->s_id, function, error_buf);
1717 ocfs2_handle_error(sb);
1720 /* Handle critical errors. This is intentionally more drastic than
1721 * ocfs2_handle_error, so we only use for things like journal errors,
1723 void __ocfs2_abort(struct super_block* sb,
1724 const char *function,
1725 const char *fmt, ...)
1729 va_start(args, fmt);
1730 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1733 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1734 sb->s_id, function, error_buf);
1736 /* We don't have the cluster support yet to go straight to
1737 * hard readonly in here. Until then, we want to keep
1738 * ocfs2_abort() so that we can at least mark critical
1741 * TODO: This should abort the journal and alert other nodes
1742 * that our slot needs recovery. */
1744 /* Force a panic(). This stinks, but it's better than letting
1745 * things continue without having a proper hard readonly
1747 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1748 ocfs2_handle_error(sb);
1751 module_init(ocfs2_init);
1752 module_exit(ocfs2_exit);