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 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,
143 Opt_tag, Opt_notag, Opt_tagid,
147 static match_table_t tokens = {
148 {Opt_barrier, "barrier=%u"},
149 {Opt_err_panic, "errors=panic"},
150 {Opt_err_ro, "errors=remount-ro"},
152 {Opt_nointr, "nointr"},
153 {Opt_hb_none, OCFS2_HB_NONE},
154 {Opt_hb_local, OCFS2_HB_LOCAL},
155 {Opt_data_ordered, "data=ordered"},
156 {Opt_data_writeback, "data=writeback"},
157 {Opt_atime_quantum, "atime_quantum=%u"},
160 {Opt_notag, "notag"},
161 {Opt_tagid, "tagid=%u"},
166 * write_super and sync_fs ripped right out of ext3.
168 static void ocfs2_write_super(struct super_block *sb)
170 if (mutex_trylock(&sb->s_lock) != 0)
175 static int ocfs2_sync_fs(struct super_block *sb, int wait)
179 struct ocfs2_super *osb = OCFS2_SB(sb);
183 if (ocfs2_is_hard_readonly(osb))
187 status = ocfs2_flush_truncate_log(osb);
191 ocfs2_schedule_truncate_log_flush(osb, 0);
194 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
196 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
202 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
204 struct inode *new = NULL;
210 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
212 status = PTR_ERR(new);
216 osb->root_inode = new;
218 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
220 status = PTR_ERR(new);
224 osb->sys_root_inode = new;
226 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
227 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
228 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
230 ocfs2_release_system_inodes(osb);
233 /* FIXME: Should ERROR_RO_FS */
234 mlog(ML_ERROR, "Unable to load system inode %d, "
235 "possibly corrupt fs?", i);
238 // the array now has one ref, so drop this one
247 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
249 struct inode *new = NULL;
255 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
256 i < NUM_SYSTEM_INODES;
258 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
260 ocfs2_release_system_inodes(osb);
262 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
263 status, i, osb->slot_num);
266 /* the array now has one ref, so drop this one */
275 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
282 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
283 inode = osb->system_inodes[i];
286 osb->system_inodes[i] = NULL;
290 inode = osb->sys_root_inode;
293 osb->sys_root_inode = NULL;
296 inode = osb->root_inode;
299 osb->root_inode = NULL;
306 /* We're allocating fs objects, use GFP_NOFS */
307 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
309 struct ocfs2_inode_info *oi;
311 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
315 return &oi->vfs_inode;
318 static void ocfs2_destroy_inode(struct inode *inode)
320 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
323 /* From xfs_super.c:xfs_max_file_offset
324 * Copyright (c) 2000-2004 Silicon Graphics, Inc.
326 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
328 unsigned int pagefactor = 1;
329 unsigned int bitshift = BITS_PER_LONG - 1;
331 /* Figure out maximum filesize, on Linux this can depend on
332 * the filesystem blocksize (on 32 bit platforms).
333 * __block_prepare_write does this in an [unsigned] long...
334 * page->index << (PAGE_CACHE_SHIFT - bbits)
335 * So, for page sized blocks (4K on 32 bit platforms),
336 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
337 * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
338 * but for smaller blocksizes it is less (bbits = log2 bsize).
339 * Note1: get_block_t takes a long (implicit cast from above)
340 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
341 * can optionally convert the [unsigned] long from above into
342 * an [unsigned] long long.
345 #if BITS_PER_LONG == 32
346 # if defined(CONFIG_LBD)
347 BUILD_BUG_ON(sizeof(sector_t) != 8);
348 pagefactor = PAGE_CACHE_SIZE;
349 bitshift = BITS_PER_LONG;
351 pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
355 return (((unsigned long long)pagefactor) << bitshift) - 1;
358 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
360 int incompat_features;
362 unsigned long parsed_options;
363 struct ocfs2_super *osb = OCFS2_SB(sb);
365 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
370 printk("ocfs2_remount: %lx,%lx\n", osb->s_mount_opt, sb->s_flags);
371 if ((parsed_options & OCFS2_MOUNT_TAGGED) &&
372 !(sb->s_flags & MS_TAGGED)) {
374 mlog(ML_ERROR, "Cannot change tagging on remount\n");
378 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
379 (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
381 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
385 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
386 (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
388 mlog(ML_ERROR, "Cannot change data mode on remount\n");
392 /* We're going to/from readonly mode. */
393 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
394 /* Lock here so the check of HARD_RO and the potential
395 * setting of SOFT_RO is atomic. */
396 spin_lock(&osb->osb_lock);
397 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
398 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
403 if (*flags & MS_RDONLY) {
404 mlog(0, "Going to ro mode.\n");
405 sb->s_flags |= MS_RDONLY;
406 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
408 mlog(0, "Making ro filesystem writeable.\n");
410 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
411 mlog(ML_ERROR, "Cannot remount RDWR "
412 "filesystem due to previous errors.\n");
416 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
417 if (incompat_features) {
418 mlog(ML_ERROR, "Cannot remount RDWR because "
419 "of unsupported optional features "
420 "(%x).\n", incompat_features);
424 sb->s_flags &= ~MS_RDONLY;
425 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
428 spin_unlock(&osb->osb_lock);
432 if (!ocfs2_is_hard_readonly(osb))
433 ocfs2_set_journal_params(osb);
435 /* Only save off the new mount options in case of a successful
437 osb->s_mount_opt = parsed_options;
443 static int ocfs2_sb_probe(struct super_block *sb,
444 struct buffer_head **bh,
447 int status = 0, tmpstat;
448 struct ocfs1_vol_disk_hdr *hdr;
449 struct ocfs2_dinode *di;
455 *sector_size = bdev_hardsect_size(sb->s_bdev);
456 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
457 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
458 *sector_size, OCFS2_MAX_BLOCKSIZE);
463 /* Can this really happen? */
464 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
465 *sector_size = OCFS2_MIN_BLOCKSIZE;
467 /* check block zero for old format */
468 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
473 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
474 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
475 mlog(ML_ERROR, "incompatible version: %u.%u\n",
476 hdr->major_version, hdr->minor_version);
479 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
480 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
481 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
488 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
489 "upgraded before mounting with ocfs v2\n");
494 * Now check at magic offset for 512, 1024, 2048, 4096
495 * blocksizes. 4096 is the maximum blocksize because it is
496 * the minimum clustersize.
499 for (blksize = *sector_size;
500 blksize <= OCFS2_MAX_BLOCKSIZE;
502 tmpstat = ocfs2_get_sector(sb, bh,
503 OCFS2_SUPER_BLOCK_BLKNO,
510 di = (struct ocfs2_dinode *) (*bh)->b_data;
511 status = ocfs2_verify_volume(di, *bh, blksize);
516 if (status != -EAGAIN)
524 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
526 if (ocfs2_mount_local(osb)) {
527 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
528 mlog(ML_ERROR, "Cannot heartbeat on a locally "
529 "mounted device.\n");
534 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
535 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
536 mlog(ML_ERROR, "Heartbeat has to be started to mount "
537 "a read-write clustered device.\n");
545 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
548 int status, sector_size;
549 unsigned long parsed_opt;
550 struct inode *inode = NULL;
551 struct ocfs2_super *osb = NULL;
552 struct buffer_head *bh = NULL;
555 mlog_entry("%p, %p, %i", sb, data, silent);
557 if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
559 goto read_super_error;
562 /* for now we only have one cluster/node, make sure we see it
563 * in the heartbeat universe */
564 if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
565 if (!o2hb_check_local_node_heartbeating()) {
567 goto read_super_error;
571 /* probe for superblock */
572 status = ocfs2_sb_probe(sb, &bh, §or_size);
574 mlog(ML_ERROR, "superblock probe failed!\n");
575 goto read_super_error;
578 status = ocfs2_initialize_super(sb, bh, sector_size);
582 goto read_super_error;
586 osb->s_mount_opt = parsed_opt;
588 sb->s_magic = OCFS2_SUPER_MAGIC;
590 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
592 if (bdev_read_only(sb->s_bdev)) {
593 if (!(sb->s_flags & MS_RDONLY)) {
595 mlog(ML_ERROR, "Readonly device detected but readonly "
596 "mount was not specified.\n");
597 goto read_super_error;
600 /* You should not be able to start a local heartbeat
601 * on a readonly device. */
602 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
604 mlog(ML_ERROR, "Local heartbeat specified on readonly "
606 goto read_super_error;
609 status = ocfs2_check_journals_nolocks(osb);
611 if (status == -EROFS)
612 mlog(ML_ERROR, "Recovery required on readonly "
613 "file system, but write access is "
617 goto read_super_error;
620 ocfs2_set_ro_flag(osb, 1);
622 printk(KERN_NOTICE "Readonly device detected. No cluster "
623 "services will be utilized for this mount. Recovery "
624 "will be skipped.\n");
627 if (!ocfs2_is_hard_readonly(osb)) {
628 if (sb->s_flags & MS_RDONLY)
629 ocfs2_set_ro_flag(osb, 0);
632 status = ocfs2_verify_heartbeat(osb);
635 goto read_super_error;
638 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
640 if (!osb->osb_debug_root) {
642 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
643 goto read_super_error;
646 status = ocfs2_mount_volume(sb);
648 inode = igrab(osb->root_inode);
651 goto read_super_error;
656 goto read_super_error;
659 root = d_alloc_root(inode);
663 goto read_super_error;
668 ocfs2_complete_mount_recovery(osb);
670 if (osb->s_mount_opt & OCFS2_MOUNT_TAGGED)
671 sb->s_flags |= MS_TAGGED;
673 if (ocfs2_mount_local(osb))
674 snprintf(nodestr, sizeof(nodestr), "local");
676 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
678 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
679 "with %s data mode.\n",
680 osb->dev_str, nodestr, osb->slot_num,
681 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
684 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
685 wake_up(&osb->osb_mount_event);
698 atomic_set(&osb->vol_state, VOLUME_DISABLED);
699 wake_up(&osb->osb_mount_event);
700 ocfs2_dismount_volume(sb, 1);
707 static int ocfs2_get_sb(struct file_system_type *fs_type,
709 const char *dev_name,
711 struct vfsmount *mnt)
713 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
717 static struct file_system_type ocfs2_fs_type = {
718 .owner = THIS_MODULE,
720 .get_sb = ocfs2_get_sb, /* is this called when we mount
722 .kill_sb = kill_block_super, /* set to the generic one
723 * right now, but do we
724 * need to change that? */
725 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
729 static int ocfs2_parse_options(struct super_block *sb,
731 unsigned long *mount_opt,
737 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
738 options ? options : "(none)");
747 while ((p = strsep(&options, ",")) != NULL) {
749 substring_t args[MAX_OPT_ARGS];
750 struct ocfs2_super * osb = OCFS2_SB(sb);
755 token = match_token(p, tokens, args);
758 *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
761 *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
764 if (match_int(&args[0], &option)) {
769 *mount_opt |= OCFS2_MOUNT_BARRIER;
771 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
774 *mount_opt &= ~OCFS2_MOUNT_NOINTR;
777 *mount_opt |= OCFS2_MOUNT_NOINTR;
780 *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
783 *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
785 case Opt_data_ordered:
786 *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
788 case Opt_data_writeback:
789 *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
791 case Opt_atime_quantum:
792 if (match_int(&args[0], &option)) {
797 osb->s_atime_quantum = option;
799 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
801 #ifndef CONFIG_TAGGING_NONE
803 *mount_opt |= OCFS2_MOUNT_TAGGED;
806 *mount_opt &= ~OCFS2_MOUNT_TAGGED;
809 #ifdef CONFIG_PROPAGATE
812 *mount_opt |= OCFS2_MOUNT_TAGGED;
817 "Unrecognized mount option \"%s\" "
818 "or missing value\n", p);
831 static int __init ocfs2_init(void)
837 ocfs2_print_version();
839 if (init_ocfs2_extent_maps())
842 status = init_ocfs2_uptodate_cache();
848 status = ocfs2_initialize_mem_caches();
854 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
860 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
861 if (!ocfs2_debugfs_root) {
863 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
868 ocfs2_free_mem_caches();
869 exit_ocfs2_uptodate_cache();
870 exit_ocfs2_extent_maps();
876 return register_filesystem(&ocfs2_fs_type);
881 static void __exit ocfs2_exit(void)
886 flush_workqueue(ocfs2_wq);
887 destroy_workqueue(ocfs2_wq);
890 debugfs_remove(ocfs2_debugfs_root);
892 ocfs2_free_mem_caches();
894 unregister_filesystem(&ocfs2_fs_type);
896 exit_ocfs2_extent_maps();
898 exit_ocfs2_uptodate_cache();
903 static void ocfs2_put_super(struct super_block *sb)
905 mlog_entry("(0x%p)\n", sb);
907 ocfs2_sync_blockdev(sb);
908 ocfs2_dismount_volume(sb, 0);
913 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
915 struct ocfs2_super *osb;
916 u32 numbits, freebits;
918 struct ocfs2_dinode *bm_lock;
919 struct buffer_head *bh = NULL;
920 struct inode *inode = NULL;
922 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
924 osb = OCFS2_SB(dentry->d_sb);
926 inode = ocfs2_get_system_file_inode(osb,
927 GLOBAL_BITMAP_SYSTEM_INODE,
930 mlog(ML_ERROR, "failed to get bitmap inode\n");
935 status = ocfs2_meta_lock(inode, &bh, 0);
941 bm_lock = (struct ocfs2_dinode *) bh->b_data;
943 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
944 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
946 buf->f_type = OCFS2_SUPER_MAGIC;
947 buf->f_bsize = dentry->d_sb->s_blocksize;
948 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
949 buf->f_blocks = ((sector_t) numbits) *
950 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
951 buf->f_bfree = ((sector_t) freebits) *
952 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
953 buf->f_bavail = buf->f_bfree;
954 buf->f_files = numbits;
955 buf->f_ffree = freebits;
959 ocfs2_meta_unlock(inode, 0);
970 static void ocfs2_inode_init_once(void *data,
971 struct kmem_cache *cachep,
974 struct ocfs2_inode_info *oi = data;
976 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
977 SLAB_CTOR_CONSTRUCTOR) {
979 oi->ip_open_count = 0;
980 spin_lock_init(&oi->ip_lock);
981 ocfs2_extent_map_init(&oi->vfs_inode);
982 INIT_LIST_HEAD(&oi->ip_io_markers);
983 oi->ip_created_trans = 0;
984 oi->ip_last_trans = 0;
985 oi->ip_dir_start_lookup = 0;
987 init_rwsem(&oi->ip_alloc_sem);
988 mutex_init(&oi->ip_io_mutex);
993 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
994 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
995 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
997 ocfs2_metadata_cache_init(&oi->vfs_inode);
999 inode_init_once(&oi->vfs_inode);
1003 static int ocfs2_initialize_mem_caches(void)
1005 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1006 sizeof(struct ocfs2_inode_info),
1008 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1010 ocfs2_inode_init_once, NULL);
1011 if (!ocfs2_inode_cachep)
1017 static void ocfs2_free_mem_caches(void)
1019 if (ocfs2_inode_cachep)
1020 kmem_cache_destroy(ocfs2_inode_cachep);
1022 ocfs2_inode_cachep = NULL;
1025 static int ocfs2_get_sector(struct super_block *sb,
1026 struct buffer_head **bh,
1030 if (!sb_set_blocksize(sb, sect_size)) {
1031 mlog(ML_ERROR, "unable to set blocksize\n");
1035 *bh = sb_getblk(sb, block);
1041 if (!buffer_dirty(*bh))
1042 clear_buffer_uptodate(*bh);
1044 ll_rw_block(READ, 1, bh);
1045 wait_on_buffer(*bh);
1049 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1050 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1054 /* XXX hold a ref on the node while mounte? easy enough, if
1056 if (ocfs2_mount_local(osb))
1059 osb->node_num = o2nm_this_node();
1061 if (osb->node_num == O2NM_MAX_NODES) {
1062 mlog(ML_ERROR, "could not find this host's node number\n");
1067 mlog(0, "I am node %d\n", osb->node_num);
1074 static int ocfs2_mount_volume(struct super_block *sb)
1077 int unlock_super = 0;
1078 struct ocfs2_super *osb = OCFS2_SB(sb);
1082 if (ocfs2_is_hard_readonly(osb))
1085 status = ocfs2_fill_local_node_info(osb);
1091 status = ocfs2_register_hb_callbacks(osb);
1097 status = ocfs2_dlm_init(osb);
1103 /* requires vote_thread to be running. */
1104 status = ocfs2_register_net_handlers(osb);
1110 status = ocfs2_super_lock(osb, 1);
1117 /* This will load up the node map and add ourselves to it. */
1118 status = ocfs2_find_slot(osb);
1124 ocfs2_populate_mounted_map(osb);
1126 /* load all node-local system inodes */
1127 status = ocfs2_init_local_system_inodes(osb);
1133 status = ocfs2_check_volume(osb);
1139 status = ocfs2_truncate_log_init(osb);
1145 if (ocfs2_mount_local(osb))
1148 /* This should be sent *after* we recovered our journal as it
1149 * will cause other nodes to unmark us as needing
1150 * recovery. However, we need to send it *before* dropping the
1151 * super block lock as otherwise their recovery threads might
1152 * try to clean us up while we're live! */
1153 status = ocfs2_request_mount_vote(osb);
1159 ocfs2_super_unlock(osb, 1);
1165 /* we can't grab the goofy sem lock from inside wait_event, so we use
1166 * memory barriers to make sure that we'll see the null task before
1168 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1171 return osb->recovery_thread_task != NULL;
1174 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1177 struct ocfs2_super *osb = NULL;
1180 mlog_entry("(0x%p)\n", sb);
1186 ocfs2_shutdown_local_alloc(osb);
1188 ocfs2_truncate_log_shutdown(osb);
1190 /* disable any new recovery threads and wait for any currently
1191 * running ones to exit. Do this before setting the vol_state. */
1192 mutex_lock(&osb->recovery_lock);
1193 osb->disable_recovery = 1;
1194 mutex_unlock(&osb->recovery_lock);
1195 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1197 /* At this point, we know that no more recovery threads can be
1198 * launched, so wait for any recovery completion work to
1200 flush_workqueue(ocfs2_wq);
1202 ocfs2_journal_shutdown(osb);
1204 ocfs2_sync_blockdev(sb);
1206 /* No dlm means we've failed during mount, so skip all the
1207 * steps which depended on that to complete. */
1209 tmp = ocfs2_super_lock(osb, 1);
1215 tmp = ocfs2_request_umount_vote(osb);
1219 if (osb->slot_num != OCFS2_INVALID_SLOT)
1220 ocfs2_put_slot(osb);
1222 ocfs2_super_unlock(osb, 1);
1225 ocfs2_release_system_inodes(osb);
1228 ocfs2_unregister_net_handlers(osb);
1230 ocfs2_dlm_shutdown(osb);
1233 ocfs2_clear_hb_callbacks(osb);
1235 debugfs_remove(osb->osb_debug_root);
1238 ocfs2_stop_heartbeat(osb);
1240 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1242 if (ocfs2_mount_local(osb))
1243 snprintf(nodestr, sizeof(nodestr), "local");
1245 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1247 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1248 osb->dev_str, nodestr);
1250 ocfs2_delete_osb(osb);
1253 sb->s_fs_info = NULL;
1256 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1257 unsigned uuid_bytes)
1262 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1264 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1265 if (osb->uuid_str == NULL)
1268 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1269 /* print with null */
1270 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1271 if (ret != 2) /* drop super cleans up */
1273 /* then only advance past the last char */
1280 static int ocfs2_initialize_super(struct super_block *sb,
1281 struct buffer_head *bh,
1286 struct ocfs2_dinode *di = NULL;
1287 struct inode *inode = NULL;
1288 struct buffer_head *bitmap_bh = NULL;
1289 struct ocfs2_journal *journal;
1290 __le32 uuid_net_key;
1291 struct ocfs2_super *osb;
1295 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1302 sb->s_fs_info = osb;
1303 sb->s_op = &ocfs2_sops;
1304 sb->s_export_op = &ocfs2_export_ops;
1305 sb->s_flags |= MS_NOATIME;
1306 /* this is needed to support O_LARGEFILE */
1307 sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1310 /* Save off for ocfs2_rw_direct */
1311 osb->s_sectsize_bits = blksize_bits(sector_size);
1312 BUG_ON(!osb->s_sectsize_bits);
1314 osb->net_response_ids = 0;
1315 spin_lock_init(&osb->net_response_lock);
1316 INIT_LIST_HEAD(&osb->net_response_list);
1318 INIT_LIST_HEAD(&osb->osb_net_handlers);
1319 init_waitqueue_head(&osb->recovery_event);
1320 spin_lock_init(&osb->vote_task_lock);
1321 init_waitqueue_head(&osb->vote_event);
1322 osb->vote_work_sequence = 0;
1323 osb->vote_wake_sequence = 0;
1324 INIT_LIST_HEAD(&osb->blocked_lock_list);
1325 osb->blocked_lock_count = 0;
1326 INIT_LIST_HEAD(&osb->vote_list);
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 mutex_init(&osb->recovery_lock);
1342 osb->disable_recovery = 0;
1343 osb->recovery_thread_task = NULL;
1345 init_waitqueue_head(&osb->checkpoint_event);
1346 atomic_set(&osb->needs_checkpoint, 0);
1348 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1350 osb->node_num = O2NM_INVALID_NODE_NUM;
1351 osb->slot_num = OCFS2_INVALID_SLOT;
1353 osb->local_alloc_state = OCFS2_LA_UNUSED;
1354 osb->local_alloc_bh = NULL;
1356 ocfs2_setup_hb_callbacks(osb);
1358 init_waitqueue_head(&osb->osb_mount_event);
1360 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1361 if (!osb->vol_label) {
1362 mlog(ML_ERROR, "unable to alloc vol label\n");
1367 di = (struct ocfs2_dinode *)bh->b_data;
1369 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1370 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1371 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1376 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1378 init_waitqueue_head(&osb->osb_wipe_event);
1379 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1380 sizeof(*osb->osb_orphan_wipes),
1382 if (!osb->osb_orphan_wipes) {
1388 osb->s_feature_compat =
1389 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1390 osb->s_feature_ro_compat =
1391 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1392 osb->s_feature_incompat =
1393 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1395 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1396 mlog(ML_ERROR, "couldn't mount because of unsupported "
1397 "optional features (%x).\n", i);
1401 if (!(osb->sb->s_flags & MS_RDONLY) &&
1402 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1403 mlog(ML_ERROR, "couldn't mount RDWR because of "
1404 "unsupported optional features (%x).\n", i);
1409 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1412 * This should be done in ocfs2_journal_init(), but unknown
1413 * ordering issues will cause the filesystem to crash.
1414 * If anyone wants to figure out what part of the code
1415 * refers to osb->journal before ocfs2_journal_init() is run,
1418 /* initialize our journal structure */
1420 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1422 mlog(ML_ERROR, "unable to alloc journal\n");
1426 osb->journal = journal;
1427 journal->j_osb = osb;
1429 atomic_set(&journal->j_num_trans, 0);
1430 init_rwsem(&journal->j_trans_barrier);
1431 init_waitqueue_head(&journal->j_checkpointed);
1432 spin_lock_init(&journal->j_lock);
1433 journal->j_trans_id = (unsigned long) 1;
1434 INIT_LIST_HEAD(&journal->j_la_cleanups);
1435 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1436 journal->j_state = OCFS2_JOURNAL_FREE;
1438 /* get some pseudo constants for clustersize bits */
1439 osb->s_clustersize_bits =
1440 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1441 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1442 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1444 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1445 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1446 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1447 osb->s_clustersize);
1452 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1454 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1455 "what jbd can address in 32 bits.\n");
1460 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1461 sizeof(di->id2.i_super.s_uuid))) {
1462 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1467 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1468 osb->net_key = le32_to_cpu(uuid_net_key);
1470 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1471 osb->vol_label[63] = '\0';
1472 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1473 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1474 osb->first_cluster_group_blkno =
1475 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1476 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1477 mlog(0, "vol_label: %s\n", osb->vol_label);
1478 mlog(0, "uuid: %s\n", osb->uuid_str);
1479 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1480 (unsigned long long)osb->root_blkno,
1481 (unsigned long long)osb->system_dir_blkno);
1483 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1484 if (!osb->osb_dlm_debug) {
1490 atomic_set(&osb->vol_state, VOLUME_INIT);
1492 /* load root, system_dir, and all global system inodes */
1493 status = ocfs2_init_global_system_inodes(osb);
1502 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1503 OCFS2_INVALID_SLOT);
1510 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1512 /* We don't have a cluster lock on the bitmap here because
1513 * we're only interested in static information and the extra
1514 * complexity at mount time isn't worht it. Don't pass the
1515 * inode in to the read function though as we don't want it to
1516 * be put in the cache. */
1517 status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1525 di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1526 osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1528 mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1529 (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1531 status = ocfs2_init_slot_info(osb);
1543 * will return: -EAGAIN if it is ok to keep searching for superblocks
1544 * -EINVAL if there is a bad superblock
1547 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1548 struct buffer_head *bh,
1551 int status = -EAGAIN;
1555 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1556 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1558 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1559 mlog(ML_ERROR, "found superblock with incorrect block "
1560 "size: found %u, should be %u\n",
1561 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1563 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1564 OCFS2_MAJOR_REV_LEVEL ||
1565 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1566 OCFS2_MINOR_REV_LEVEL) {
1567 mlog(ML_ERROR, "found superblock with bad version: "
1568 "found %u.%u, should be %u.%u\n",
1569 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1570 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1571 OCFS2_MAJOR_REV_LEVEL,
1572 OCFS2_MINOR_REV_LEVEL);
1573 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1574 mlog(ML_ERROR, "bad block number on superblock: "
1575 "found %llu, should be %llu\n",
1576 (unsigned long long)di->i_blkno,
1577 (unsigned long long)bh->b_blocknr);
1578 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1579 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1580 mlog(ML_ERROR, "bad cluster size found: %u\n",
1581 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1582 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1583 mlog(ML_ERROR, "bad root_blkno: 0\n");
1584 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1585 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1586 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1588 "Superblock slots found greater than file system "
1589 "maximum: found %u, max %u\n",
1590 le16_to_cpu(di->id2.i_super.s_max_slots),
1602 static int ocfs2_check_volume(struct ocfs2_super *osb)
1607 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1613 /* Init our journal object. */
1614 status = ocfs2_journal_init(osb->journal, &dirty);
1616 mlog(ML_ERROR, "Could not initialize journal!\n");
1620 /* If the journal was unmounted cleanly then we don't want to
1621 * recover anything. Otherwise, journal_load will do that
1622 * dirty work for us :) */
1624 status = ocfs2_journal_wipe(osb->journal, 0);
1630 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1631 "recovering volume.\n");
1634 local = ocfs2_mount_local(osb);
1636 /* will play back anything left in the journal. */
1637 ocfs2_journal_load(osb->journal, local);
1640 /* recover my local alloc if we didn't unmount cleanly. */
1641 status = ocfs2_begin_local_alloc_recovery(osb,
1648 /* we complete the recovery process after we've marked
1649 * ourselves as mounted. */
1652 mlog(0, "Journal loaded.\n");
1654 status = ocfs2_load_local_alloc(osb);
1661 /* Recovery will be completed after we've mounted the
1662 * rest of the volume. */
1664 osb->local_alloc_copy = local_alloc;
1668 /* go through each journal, trylock it and if you get the
1669 * lock, and it's marked as dirty, set the bit in the recover
1670 * map and launch a recovery thread for it. */
1671 status = ocfs2_mark_dead_nodes(osb);
1684 * The routine gets called from dismount or close whenever a dismount on
1685 * volume is requested and the osb open count becomes 1.
1686 * It will remove the osb from the global list and also free up all the
1687 * initialized resources and fileobject.
1689 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1693 /* This function assumes that the caller has the main osb resource */
1696 ocfs2_free_slot_info(osb->slot_info);
1698 kfree(osb->osb_orphan_wipes);
1700 * This belongs in journal shutdown, but because we have to
1701 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1704 kfree(osb->journal);
1705 if (osb->local_alloc_copy)
1706 kfree(osb->local_alloc_copy);
1707 kfree(osb->uuid_str);
1708 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1709 memset(osb, 0, sizeof(struct ocfs2_super));
1714 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1715 * panic(). We do not support continue-on-error operation. */
1716 static void ocfs2_handle_error(struct super_block *sb)
1718 struct ocfs2_super *osb = OCFS2_SB(sb);
1720 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1721 panic("OCFS2: (device %s): panic forced after error\n",
1724 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1726 if (sb->s_flags & MS_RDONLY &&
1727 (ocfs2_is_soft_readonly(osb) ||
1728 ocfs2_is_hard_readonly(osb)))
1731 printk(KERN_CRIT "File system is now read-only due to the potential "
1732 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1733 "system is unmounted.\n");
1734 sb->s_flags |= MS_RDONLY;
1735 ocfs2_set_ro_flag(osb, 0);
1738 static char error_buf[1024];
1740 void __ocfs2_error(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 /* Not using mlog here because we want to show the actual
1751 * function the error came from. */
1752 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1753 sb->s_id, function, error_buf);
1755 ocfs2_handle_error(sb);
1758 /* Handle critical errors. This is intentionally more drastic than
1759 * ocfs2_handle_error, so we only use for things like journal errors,
1761 void __ocfs2_abort(struct super_block* sb,
1762 const char *function,
1763 const char *fmt, ...)
1767 va_start(args, fmt);
1768 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1771 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1772 sb->s_id, function, error_buf);
1774 /* We don't have the cluster support yet to go straight to
1775 * hard readonly in here. Until then, we want to keep
1776 * ocfs2_abort() so that we can at least mark critical
1779 * TODO: This should abort the journal and alert other nodes
1780 * that our slot needs recovery. */
1782 /* Force a panic(). This stinks, but it's better than letting
1783 * things continue without having a proper hard readonly
1785 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1786 ocfs2_handle_error(sb);
1789 module_init(ocfs2_init);
1790 module_exit(ocfs2_exit);