1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * load/unload driver, mount/dismount volumes
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
26 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/utsname.h>
32 #include <linux/init.h>
33 #include <linux/random.h>
34 #include <linux/statfs.h>
35 #include <linux/moduleparam.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/parser.h>
40 #include <linux/crc32.h>
41 #include <linux/debugfs.h>
42 #include <linux/mount.h>
44 #include <cluster/nodemanager.h>
46 #define MLOG_MASK_PREFIX ML_SUPER
47 #include <cluster/masklog.h>
51 /* this should be the only file to include a version 1 header */
52 #include "ocfs1_fs_compat.h"
57 #include "extent_map.h"
58 #include "heartbeat.h"
61 #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");
86 unsigned long mount_opt;
87 unsigned int atime_quantum;
91 static int ocfs2_parse_options(struct super_block *sb, char *options,
92 struct mount_options *mopt,
94 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
95 static void ocfs2_put_super(struct super_block *sb);
96 static int ocfs2_mount_volume(struct super_block *sb);
97 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
98 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
99 static int ocfs2_initialize_mem_caches(void);
100 static void ocfs2_free_mem_caches(void);
101 static void ocfs2_delete_osb(struct ocfs2_super *osb);
103 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
105 static int ocfs2_sync_fs(struct super_block *sb, int wait);
107 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
108 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
109 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
110 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
111 static int ocfs2_check_volume(struct ocfs2_super *osb);
112 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
113 struct buffer_head *bh,
115 static int ocfs2_initialize_super(struct super_block *sb,
116 struct buffer_head *bh,
118 static int ocfs2_get_sector(struct super_block *sb,
119 struct buffer_head **bh,
122 static void ocfs2_write_super(struct super_block *sb);
123 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
124 static void ocfs2_destroy_inode(struct inode *inode);
126 static const struct super_operations ocfs2_sops = {
127 .statfs = ocfs2_statfs,
128 .alloc_inode = ocfs2_alloc_inode,
129 .destroy_inode = ocfs2_destroy_inode,
130 .drop_inode = ocfs2_drop_inode,
131 .clear_inode = ocfs2_clear_inode,
132 .delete_inode = ocfs2_delete_inode,
133 .sync_fs = ocfs2_sync_fs,
134 .write_super = ocfs2_write_super,
135 .put_super = ocfs2_put_super,
136 .remount_fs = ocfs2_remount,
137 .show_options = ocfs2_show_options,
155 static match_table_t tokens = {
156 {Opt_barrier, "barrier=%u"},
157 {Opt_err_panic, "errors=panic"},
158 {Opt_err_ro, "errors=remount-ro"},
160 {Opt_nointr, "nointr"},
161 {Opt_hb_none, OCFS2_HB_NONE},
162 {Opt_hb_local, OCFS2_HB_LOCAL},
163 {Opt_data_ordered, "data=ordered"},
164 {Opt_data_writeback, "data=writeback"},
165 {Opt_atime_quantum, "atime_quantum=%u"},
166 {Opt_slot, "preferred_slot=%u"},
171 * write_super and sync_fs ripped right out of ext3.
173 static void ocfs2_write_super(struct super_block *sb)
175 if (mutex_trylock(&sb->s_lock) != 0)
180 static int ocfs2_sync_fs(struct super_block *sb, int wait)
184 struct ocfs2_super *osb = OCFS2_SB(sb);
188 if (ocfs2_is_hard_readonly(osb))
192 status = ocfs2_flush_truncate_log(osb);
196 ocfs2_schedule_truncate_log_flush(osb, 0);
199 if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
201 log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
207 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
209 struct inode *new = NULL;
215 new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
217 status = PTR_ERR(new);
221 osb->root_inode = new;
223 new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
225 status = PTR_ERR(new);
229 osb->sys_root_inode = new;
231 for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
232 i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
233 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
235 ocfs2_release_system_inodes(osb);
238 /* FIXME: Should ERROR_RO_FS */
239 mlog(ML_ERROR, "Unable to load system inode %d, "
240 "possibly corrupt fs?", i);
243 // the array now has one ref, so drop this one
252 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
254 struct inode *new = NULL;
260 for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
261 i < NUM_SYSTEM_INODES;
263 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
265 ocfs2_release_system_inodes(osb);
267 mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
268 status, i, osb->slot_num);
271 /* the array now has one ref, so drop this one */
280 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
287 for (i = 0; i < NUM_SYSTEM_INODES; i++) {
288 inode = osb->system_inodes[i];
291 osb->system_inodes[i] = NULL;
295 inode = osb->sys_root_inode;
298 osb->sys_root_inode = NULL;
301 inode = osb->root_inode;
304 osb->root_inode = NULL;
310 /* We're allocating fs objects, use GFP_NOFS */
311 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
313 struct ocfs2_inode_info *oi;
315 oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
319 return &oi->vfs_inode;
322 static void ocfs2_destroy_inode(struct inode *inode)
324 kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
327 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
330 unsigned int bytes = 1 << cbits;
331 unsigned int trim = bytes;
332 unsigned int bitshift = 32;
335 * i_size and all block offsets in ocfs2 are always 64 bits
336 * wide. i_clusters is 32 bits, in cluster-sized units. So on
337 * 64 bit platforms, cluster size will be the limiting factor.
340 #if BITS_PER_LONG == 32
341 # if defined(CONFIG_LBD)
342 BUILD_BUG_ON(sizeof(sector_t) != 8);
344 * We might be limited by page cache size.
346 if (bytes > PAGE_CACHE_SIZE) {
347 bytes = PAGE_CACHE_SIZE;
350 * Shift by 31 here so that we don't get larger than
357 * We are limited by the size of sector_t. Use block size, as
358 * that's what we expose to the VFS.
367 * Trim by a whole cluster when we can actually approach the
368 * on-disk limits. Otherwise we can overflow i_clusters when
369 * an extent start is at the max offset.
371 return (((unsigned long long)bytes) << bitshift) - trim;
374 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
376 int incompat_features;
378 struct mount_options parsed_options;
379 struct ocfs2_super *osb = OCFS2_SB(sb);
381 if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
386 if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
387 (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
389 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
393 if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
394 (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
396 mlog(ML_ERROR, "Cannot change data mode on remount\n");
400 /* We're going to/from readonly mode. */
401 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
402 /* Lock here so the check of HARD_RO and the potential
403 * setting of SOFT_RO is atomic. */
404 spin_lock(&osb->osb_lock);
405 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
406 mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
411 if (*flags & MS_RDONLY) {
412 mlog(0, "Going to ro mode.\n");
413 sb->s_flags |= MS_RDONLY;
414 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
416 mlog(0, "Making ro filesystem writeable.\n");
418 if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
419 mlog(ML_ERROR, "Cannot remount RDWR "
420 "filesystem due to previous errors.\n");
424 incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
425 if (incompat_features) {
426 mlog(ML_ERROR, "Cannot remount RDWR because "
427 "of unsupported optional features "
428 "(%x).\n", incompat_features);
432 sb->s_flags &= ~MS_RDONLY;
433 osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
436 spin_unlock(&osb->osb_lock);
440 /* Only save off the new mount options in case of a successful
442 osb->s_mount_opt = parsed_options.mount_opt;
443 osb->s_atime_quantum = parsed_options.atime_quantum;
444 osb->preferred_slot = parsed_options.slot;
446 if (!ocfs2_is_hard_readonly(osb))
447 ocfs2_set_journal_params(osb);
453 static int ocfs2_sb_probe(struct super_block *sb,
454 struct buffer_head **bh,
458 struct ocfs1_vol_disk_hdr *hdr;
459 struct ocfs2_dinode *di;
465 *sector_size = bdev_hardsect_size(sb->s_bdev);
466 if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
467 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
468 *sector_size, OCFS2_MAX_BLOCKSIZE);
473 /* Can this really happen? */
474 if (*sector_size < OCFS2_MIN_BLOCKSIZE)
475 *sector_size = OCFS2_MIN_BLOCKSIZE;
477 /* check block zero for old format */
478 status = ocfs2_get_sector(sb, bh, 0, *sector_size);
483 hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
484 if (hdr->major_version == OCFS1_MAJOR_VERSION) {
485 mlog(ML_ERROR, "incompatible version: %u.%u\n",
486 hdr->major_version, hdr->minor_version);
489 if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
490 strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
491 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
498 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
499 "upgraded before mounting with ocfs v2\n");
504 * Now check at magic offset for 512, 1024, 2048, 4096
505 * blocksizes. 4096 is the maximum blocksize because it is
506 * the minimum clustersize.
509 for (blksize = *sector_size;
510 blksize <= OCFS2_MAX_BLOCKSIZE;
512 tmpstat = ocfs2_get_sector(sb, bh,
513 OCFS2_SUPER_BLOCK_BLKNO,
520 di = (struct ocfs2_dinode *) (*bh)->b_data;
521 status = ocfs2_verify_volume(di, *bh, blksize);
526 if (status != -EAGAIN)
534 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
536 if (ocfs2_mount_local(osb)) {
537 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
538 mlog(ML_ERROR, "Cannot heartbeat on a locally "
539 "mounted device.\n");
544 if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
545 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
546 mlog(ML_ERROR, "Heartbeat has to be started to mount "
547 "a read-write clustered device.\n");
555 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
558 int status, sector_size;
559 struct mount_options parsed_options;
560 struct inode *inode = NULL;
561 struct ocfs2_super *osb = NULL;
562 struct buffer_head *bh = NULL;
565 mlog_entry("%p, %p, %i", sb, data, silent);
567 if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
569 goto read_super_error;
572 /* for now we only have one cluster/node, make sure we see it
573 * in the heartbeat universe */
574 if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
575 if (!o2hb_check_local_node_heartbeating()) {
577 goto read_super_error;
581 /* probe for superblock */
582 status = ocfs2_sb_probe(sb, &bh, §or_size);
584 mlog(ML_ERROR, "superblock probe failed!\n");
585 goto read_super_error;
588 status = ocfs2_initialize_super(sb, bh, sector_size);
592 goto read_super_error;
596 osb->s_mount_opt = parsed_options.mount_opt;
597 osb->s_atime_quantum = parsed_options.atime_quantum;
598 osb->preferred_slot = parsed_options.slot;
600 sb->s_magic = OCFS2_SUPER_MAGIC;
602 /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
604 if (bdev_read_only(sb->s_bdev)) {
605 if (!(sb->s_flags & MS_RDONLY)) {
607 mlog(ML_ERROR, "Readonly device detected but readonly "
608 "mount was not specified.\n");
609 goto read_super_error;
612 /* You should not be able to start a local heartbeat
613 * on a readonly device. */
614 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
616 mlog(ML_ERROR, "Local heartbeat specified on readonly "
618 goto read_super_error;
621 status = ocfs2_check_journals_nolocks(osb);
623 if (status == -EROFS)
624 mlog(ML_ERROR, "Recovery required on readonly "
625 "file system, but write access is "
629 goto read_super_error;
632 ocfs2_set_ro_flag(osb, 1);
634 printk(KERN_NOTICE "Readonly device detected. No cluster "
635 "services will be utilized for this mount. Recovery "
636 "will be skipped.\n");
639 if (!ocfs2_is_hard_readonly(osb)) {
640 if (sb->s_flags & MS_RDONLY)
641 ocfs2_set_ro_flag(osb, 0);
644 status = ocfs2_verify_heartbeat(osb);
647 goto read_super_error;
650 osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
652 if (!osb->osb_debug_root) {
654 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
655 goto read_super_error;
658 status = ocfs2_mount_volume(sb);
660 inode = igrab(osb->root_inode);
663 goto read_super_error;
668 goto read_super_error;
671 root = d_alloc_root(inode);
675 goto read_super_error;
680 ocfs2_complete_mount_recovery(osb);
682 if (ocfs2_mount_local(osb))
683 snprintf(nodestr, sizeof(nodestr), "local");
685 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
687 printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
688 "with %s data mode.\n",
689 osb->dev_str, nodestr, osb->slot_num,
690 osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
693 atomic_set(&osb->vol_state, VOLUME_MOUNTED);
694 wake_up(&osb->osb_mount_event);
707 atomic_set(&osb->vol_state, VOLUME_DISABLED);
708 wake_up(&osb->osb_mount_event);
709 ocfs2_dismount_volume(sb, 1);
716 static int ocfs2_get_sb(struct file_system_type *fs_type,
718 const char *dev_name,
720 struct vfsmount *mnt)
722 return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
726 static struct file_system_type ocfs2_fs_type = {
727 .owner = THIS_MODULE,
729 .get_sb = ocfs2_get_sb, /* is this called when we mount
731 .kill_sb = kill_block_super, /* set to the generic one
732 * right now, but do we
733 * need to change that? */
734 .fs_flags = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
738 static int ocfs2_parse_options(struct super_block *sb,
740 struct mount_options *mopt,
746 mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
747 options ? options : "(none)");
750 mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
751 mopt->slot = OCFS2_INVALID_SLOT;
758 while ((p = strsep(&options, ",")) != NULL) {
760 substring_t args[MAX_OPT_ARGS];
765 token = match_token(p, tokens, args);
768 mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
771 mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
774 if (match_int(&args[0], &option)) {
779 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
781 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
784 mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
787 mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
790 mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
793 mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
795 case Opt_data_ordered:
796 mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
798 case Opt_data_writeback:
799 mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
801 case Opt_atime_quantum:
802 if (match_int(&args[0], &option)) {
807 mopt->atime_quantum = option;
811 if (match_int(&args[0], &option)) {
816 mopt->slot = (s16)option;
820 "Unrecognized mount option \"%s\" "
821 "or missing value\n", p);
834 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
836 struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
837 unsigned long opts = osb->s_mount_opt;
839 if (opts & OCFS2_MOUNT_HB_LOCAL)
840 seq_printf(s, ",_netdev,heartbeat=local");
842 seq_printf(s, ",heartbeat=none");
844 if (opts & OCFS2_MOUNT_NOINTR)
845 seq_printf(s, ",nointr");
847 if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
848 seq_printf(s, ",data=writeback");
850 seq_printf(s, ",data=ordered");
852 if (opts & OCFS2_MOUNT_BARRIER)
853 seq_printf(s, ",barrier=1");
855 if (opts & OCFS2_MOUNT_ERRORS_PANIC)
856 seq_printf(s, ",errors=panic");
858 seq_printf(s, ",errors=remount-ro");
860 if (osb->preferred_slot != OCFS2_INVALID_SLOT)
861 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
863 if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
864 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
869 static int __init ocfs2_init(void)
875 ocfs2_print_version();
877 status = init_ocfs2_uptodate_cache();
883 status = ocfs2_initialize_mem_caches();
889 ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
895 ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
896 if (!ocfs2_debugfs_root) {
898 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
903 ocfs2_free_mem_caches();
904 exit_ocfs2_uptodate_cache();
910 return register_filesystem(&ocfs2_fs_type);
915 static void __exit ocfs2_exit(void)
920 flush_workqueue(ocfs2_wq);
921 destroy_workqueue(ocfs2_wq);
924 debugfs_remove(ocfs2_debugfs_root);
926 ocfs2_free_mem_caches();
928 unregister_filesystem(&ocfs2_fs_type);
930 exit_ocfs2_uptodate_cache();
935 static void ocfs2_put_super(struct super_block *sb)
937 mlog_entry("(0x%p)\n", sb);
939 ocfs2_sync_blockdev(sb);
940 ocfs2_dismount_volume(sb, 0);
945 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
947 struct ocfs2_super *osb;
948 u32 numbits, freebits;
950 struct ocfs2_dinode *bm_lock;
951 struct buffer_head *bh = NULL;
952 struct inode *inode = NULL;
954 mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
956 osb = OCFS2_SB(dentry->d_sb);
958 inode = ocfs2_get_system_file_inode(osb,
959 GLOBAL_BITMAP_SYSTEM_INODE,
962 mlog(ML_ERROR, "failed to get bitmap inode\n");
967 status = ocfs2_inode_lock(inode, &bh, 0);
973 bm_lock = (struct ocfs2_dinode *) bh->b_data;
975 numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
976 freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
978 buf->f_type = OCFS2_SUPER_MAGIC;
979 buf->f_bsize = dentry->d_sb->s_blocksize;
980 buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
981 buf->f_blocks = ((sector_t) numbits) *
982 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
983 buf->f_bfree = ((sector_t) freebits) *
984 (osb->s_clustersize >> osb->sb->s_blocksize_bits);
985 buf->f_bavail = buf->f_bfree;
986 buf->f_files = numbits;
987 buf->f_ffree = freebits;
991 ocfs2_inode_unlock(inode, 0);
1002 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1004 struct ocfs2_inode_info *oi = data;
1007 oi->ip_open_count = 0;
1008 spin_lock_init(&oi->ip_lock);
1009 ocfs2_extent_map_init(&oi->vfs_inode);
1010 INIT_LIST_HEAD(&oi->ip_io_markers);
1011 oi->ip_created_trans = 0;
1012 oi->ip_last_trans = 0;
1013 oi->ip_dir_start_lookup = 0;
1015 init_rwsem(&oi->ip_alloc_sem);
1016 mutex_init(&oi->ip_io_mutex);
1018 oi->ip_blkno = 0ULL;
1019 oi->ip_clusters = 0;
1021 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1022 ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1023 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1025 ocfs2_metadata_cache_init(&oi->vfs_inode);
1027 inode_init_once(&oi->vfs_inode);
1030 static int ocfs2_initialize_mem_caches(void)
1032 ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1033 sizeof(struct ocfs2_inode_info),
1035 (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1037 ocfs2_inode_init_once);
1038 if (!ocfs2_inode_cachep)
1044 static void ocfs2_free_mem_caches(void)
1046 if (ocfs2_inode_cachep)
1047 kmem_cache_destroy(ocfs2_inode_cachep);
1049 ocfs2_inode_cachep = NULL;
1052 static int ocfs2_get_sector(struct super_block *sb,
1053 struct buffer_head **bh,
1057 if (!sb_set_blocksize(sb, sect_size)) {
1058 mlog(ML_ERROR, "unable to set blocksize\n");
1062 *bh = sb_getblk(sb, block);
1068 if (!buffer_dirty(*bh))
1069 clear_buffer_uptodate(*bh);
1071 ll_rw_block(READ, 1, bh);
1072 wait_on_buffer(*bh);
1076 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1077 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1081 /* XXX hold a ref on the node while mounte? easy enough, if
1083 if (ocfs2_mount_local(osb))
1086 osb->node_num = o2nm_this_node();
1088 if (osb->node_num == O2NM_MAX_NODES) {
1089 mlog(ML_ERROR, "could not find this host's node number\n");
1094 mlog(0, "I am node %d\n", osb->node_num);
1101 static int ocfs2_mount_volume(struct super_block *sb)
1104 int unlock_super = 0;
1105 struct ocfs2_super *osb = OCFS2_SB(sb);
1109 if (ocfs2_is_hard_readonly(osb))
1112 status = ocfs2_fill_local_node_info(osb);
1118 status = ocfs2_dlm_init(osb);
1124 status = ocfs2_super_lock(osb, 1);
1131 /* This will load up the node map and add ourselves to it. */
1132 status = ocfs2_find_slot(osb);
1138 /* load all node-local system inodes */
1139 status = ocfs2_init_local_system_inodes(osb);
1145 status = ocfs2_check_volume(osb);
1151 status = ocfs2_truncate_log_init(osb);
1157 if (ocfs2_mount_local(osb))
1162 ocfs2_super_unlock(osb, 1);
1168 /* we can't grab the goofy sem lock from inside wait_event, so we use
1169 * memory barriers to make sure that we'll see the null task before
1171 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1174 return osb->recovery_thread_task != NULL;
1177 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1180 struct ocfs2_super *osb = NULL;
1183 mlog_entry("(0x%p)\n", sb);
1189 ocfs2_shutdown_local_alloc(osb);
1191 ocfs2_truncate_log_shutdown(osb);
1193 /* disable any new recovery threads and wait for any currently
1194 * running ones to exit. Do this before setting the vol_state. */
1195 mutex_lock(&osb->recovery_lock);
1196 osb->disable_recovery = 1;
1197 mutex_unlock(&osb->recovery_lock);
1198 wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1200 /* At this point, we know that no more recovery threads can be
1201 * launched, so wait for any recovery completion work to
1203 flush_workqueue(ocfs2_wq);
1205 ocfs2_journal_shutdown(osb);
1207 ocfs2_sync_blockdev(sb);
1209 /* No dlm means we've failed during mount, so skip all the
1210 * steps which depended on that to complete. */
1212 tmp = ocfs2_super_lock(osb, 1);
1219 if (osb->slot_num != OCFS2_INVALID_SLOT)
1220 ocfs2_put_slot(osb);
1223 ocfs2_super_unlock(osb, 1);
1225 ocfs2_release_system_inodes(osb);
1228 ocfs2_dlm_shutdown(osb);
1230 debugfs_remove(osb->osb_debug_root);
1233 ocfs2_stop_heartbeat(osb);
1235 atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1237 if (ocfs2_mount_local(osb))
1238 snprintf(nodestr, sizeof(nodestr), "local");
1240 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1242 printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1243 osb->dev_str, nodestr);
1245 ocfs2_delete_osb(osb);
1248 sb->s_fs_info = NULL;
1251 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1252 unsigned uuid_bytes)
1257 BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1259 osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1260 if (osb->uuid_str == NULL)
1263 for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1264 /* print with null */
1265 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1266 if (ret != 2) /* drop super cleans up */
1268 /* then only advance past the last char */
1275 static int ocfs2_initialize_super(struct super_block *sb,
1276 struct buffer_head *bh,
1280 int i, cbits, bbits;
1281 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1282 struct inode *inode = NULL;
1283 struct ocfs2_journal *journal;
1284 __le32 uuid_net_key;
1285 struct ocfs2_super *osb;
1289 osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1296 sb->s_fs_info = osb;
1297 sb->s_op = &ocfs2_sops;
1298 sb->s_export_op = &ocfs2_export_ops;
1299 sb->s_time_gran = 1;
1300 sb->s_flags |= MS_NOATIME;
1301 /* this is needed to support O_LARGEFILE */
1302 cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1303 bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1304 sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1307 /* Save off for ocfs2_rw_direct */
1308 osb->s_sectsize_bits = blksize_bits(sector_size);
1309 BUG_ON(!osb->s_sectsize_bits);
1311 init_waitqueue_head(&osb->recovery_event);
1312 spin_lock_init(&osb->dc_task_lock);
1313 init_waitqueue_head(&osb->dc_event);
1314 osb->dc_work_sequence = 0;
1315 osb->dc_wake_sequence = 0;
1316 INIT_LIST_HEAD(&osb->blocked_lock_list);
1317 osb->blocked_lock_count = 0;
1318 spin_lock_init(&osb->osb_lock);
1320 atomic_set(&osb->alloc_stats.moves, 0);
1321 atomic_set(&osb->alloc_stats.local_data, 0);
1322 atomic_set(&osb->alloc_stats.bitmap_data, 0);
1323 atomic_set(&osb->alloc_stats.bg_allocs, 0);
1324 atomic_set(&osb->alloc_stats.bg_extends, 0);
1326 ocfs2_init_node_maps(osb);
1328 snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1329 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1331 mutex_init(&osb->recovery_lock);
1333 osb->disable_recovery = 0;
1334 osb->recovery_thread_task = NULL;
1336 init_waitqueue_head(&osb->checkpoint_event);
1337 atomic_set(&osb->needs_checkpoint, 0);
1339 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1341 osb->node_num = O2NM_INVALID_NODE_NUM;
1342 osb->slot_num = OCFS2_INVALID_SLOT;
1344 osb->local_alloc_state = OCFS2_LA_UNUSED;
1345 osb->local_alloc_bh = NULL;
1347 ocfs2_setup_hb_callbacks(osb);
1349 init_waitqueue_head(&osb->osb_mount_event);
1351 osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1352 if (!osb->vol_label) {
1353 mlog(ML_ERROR, "unable to alloc vol label\n");
1358 osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1359 if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1360 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1365 mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1367 init_waitqueue_head(&osb->osb_wipe_event);
1368 osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1369 sizeof(*osb->osb_orphan_wipes),
1371 if (!osb->osb_orphan_wipes) {
1377 osb->s_feature_compat =
1378 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1379 osb->s_feature_ro_compat =
1380 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1381 osb->s_feature_incompat =
1382 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1384 if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1385 mlog(ML_ERROR, "couldn't mount because of unsupported "
1386 "optional features (%x).\n", i);
1390 if (!(osb->sb->s_flags & MS_RDONLY) &&
1391 (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1392 mlog(ML_ERROR, "couldn't mount RDWR because of "
1393 "unsupported optional features (%x).\n", i);
1398 get_random_bytes(&osb->s_next_generation, sizeof(u32));
1401 * This should be done in ocfs2_journal_init(), but unknown
1402 * ordering issues will cause the filesystem to crash.
1403 * If anyone wants to figure out what part of the code
1404 * refers to osb->journal before ocfs2_journal_init() is run,
1407 /* initialize our journal structure */
1409 journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1411 mlog(ML_ERROR, "unable to alloc journal\n");
1415 osb->journal = journal;
1416 journal->j_osb = osb;
1418 atomic_set(&journal->j_num_trans, 0);
1419 init_rwsem(&journal->j_trans_barrier);
1420 init_waitqueue_head(&journal->j_checkpointed);
1421 spin_lock_init(&journal->j_lock);
1422 journal->j_trans_id = (unsigned long) 1;
1423 INIT_LIST_HEAD(&journal->j_la_cleanups);
1424 INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1425 journal->j_state = OCFS2_JOURNAL_FREE;
1427 /* get some pseudo constants for clustersize bits */
1428 osb->s_clustersize_bits =
1429 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1430 osb->s_clustersize = 1 << osb->s_clustersize_bits;
1431 mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1433 if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1434 osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1435 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1436 osb->s_clustersize);
1441 if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1443 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1444 "what jbd can address in 32 bits.\n");
1449 if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1450 sizeof(di->id2.i_super.s_uuid))) {
1451 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1456 memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1458 strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1459 osb->vol_label[63] = '\0';
1460 osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1461 osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1462 osb->first_cluster_group_blkno =
1463 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1464 osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1465 mlog(0, "vol_label: %s\n", osb->vol_label);
1466 mlog(0, "uuid: %s\n", osb->uuid_str);
1467 mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1468 (unsigned long long)osb->root_blkno,
1469 (unsigned long long)osb->system_dir_blkno);
1471 osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1472 if (!osb->osb_dlm_debug) {
1478 atomic_set(&osb->vol_state, VOLUME_INIT);
1480 /* load root, system_dir, and all global system inodes */
1481 status = ocfs2_init_global_system_inodes(osb);
1490 inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1491 OCFS2_INVALID_SLOT);
1498 osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1501 osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1503 status = ocfs2_init_slot_info(osb);
1515 * will return: -EAGAIN if it is ok to keep searching for superblocks
1516 * -EINVAL if there is a bad superblock
1519 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1520 struct buffer_head *bh,
1523 int status = -EAGAIN;
1527 if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1528 strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1530 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1531 mlog(ML_ERROR, "found superblock with incorrect block "
1532 "size: found %u, should be %u\n",
1533 1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1535 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1536 OCFS2_MAJOR_REV_LEVEL ||
1537 le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1538 OCFS2_MINOR_REV_LEVEL) {
1539 mlog(ML_ERROR, "found superblock with bad version: "
1540 "found %u.%u, should be %u.%u\n",
1541 le16_to_cpu(di->id2.i_super.s_major_rev_level),
1542 le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1543 OCFS2_MAJOR_REV_LEVEL,
1544 OCFS2_MINOR_REV_LEVEL);
1545 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1546 mlog(ML_ERROR, "bad block number on superblock: "
1547 "found %llu, should be %llu\n",
1548 (unsigned long long)le64_to_cpu(di->i_blkno),
1549 (unsigned long long)bh->b_blocknr);
1550 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1551 le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1552 mlog(ML_ERROR, "bad cluster size found: %u\n",
1553 1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1554 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1555 mlog(ML_ERROR, "bad root_blkno: 0\n");
1556 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1557 mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1558 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1560 "Superblock slots found greater than file system "
1561 "maximum: found %u, max %u\n",
1562 le16_to_cpu(di->id2.i_super.s_max_slots),
1574 static int ocfs2_check_volume(struct ocfs2_super *osb)
1579 struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1585 /* Init our journal object. */
1586 status = ocfs2_journal_init(osb->journal, &dirty);
1588 mlog(ML_ERROR, "Could not initialize journal!\n");
1592 /* If the journal was unmounted cleanly then we don't want to
1593 * recover anything. Otherwise, journal_load will do that
1594 * dirty work for us :) */
1596 status = ocfs2_journal_wipe(osb->journal, 0);
1602 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1603 "recovering volume.\n");
1606 local = ocfs2_mount_local(osb);
1608 /* will play back anything left in the journal. */
1609 ocfs2_journal_load(osb->journal, local);
1612 /* recover my local alloc if we didn't unmount cleanly. */
1613 status = ocfs2_begin_local_alloc_recovery(osb,
1620 /* we complete the recovery process after we've marked
1621 * ourselves as mounted. */
1624 mlog(0, "Journal loaded.\n");
1626 status = ocfs2_load_local_alloc(osb);
1633 /* Recovery will be completed after we've mounted the
1634 * rest of the volume. */
1636 osb->local_alloc_copy = local_alloc;
1640 /* go through each journal, trylock it and if you get the
1641 * lock, and it's marked as dirty, set the bit in the recover
1642 * map and launch a recovery thread for it. */
1643 status = ocfs2_mark_dead_nodes(osb);
1656 * The routine gets called from dismount or close whenever a dismount on
1657 * volume is requested and the osb open count becomes 1.
1658 * It will remove the osb from the global list and also free up all the
1659 * initialized resources and fileobject.
1661 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1665 /* This function assumes that the caller has the main osb resource */
1668 ocfs2_free_slot_info(osb->slot_info);
1670 kfree(osb->osb_orphan_wipes);
1672 * This belongs in journal shutdown, but because we have to
1673 * allocate osb->journal at the start of ocfs2_initalize_osb(),
1676 kfree(osb->journal);
1677 if (osb->local_alloc_copy)
1678 kfree(osb->local_alloc_copy);
1679 kfree(osb->uuid_str);
1680 ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1681 memset(osb, 0, sizeof(struct ocfs2_super));
1686 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1687 * panic(). We do not support continue-on-error operation. */
1688 static void ocfs2_handle_error(struct super_block *sb)
1690 struct ocfs2_super *osb = OCFS2_SB(sb);
1692 if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1693 panic("OCFS2: (device %s): panic forced after error\n",
1696 ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1698 if (sb->s_flags & MS_RDONLY &&
1699 (ocfs2_is_soft_readonly(osb) ||
1700 ocfs2_is_hard_readonly(osb)))
1703 printk(KERN_CRIT "File system is now read-only due to the potential "
1704 "of on-disk corruption. Please run fsck.ocfs2 once the file "
1705 "system is unmounted.\n");
1706 sb->s_flags |= MS_RDONLY;
1707 ocfs2_set_ro_flag(osb, 0);
1710 static char error_buf[1024];
1712 void __ocfs2_error(struct super_block *sb,
1713 const char *function,
1714 const char *fmt, ...)
1718 va_start(args, fmt);
1719 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1722 /* Not using mlog here because we want to show the actual
1723 * function the error came from. */
1724 printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1725 sb->s_id, function, error_buf);
1727 ocfs2_handle_error(sb);
1730 /* Handle critical errors. This is intentionally more drastic than
1731 * ocfs2_handle_error, so we only use for things like journal errors,
1733 void __ocfs2_abort(struct super_block* sb,
1734 const char *function,
1735 const char *fmt, ...)
1739 va_start(args, fmt);
1740 vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1743 printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1744 sb->s_id, function, error_buf);
1746 /* We don't have the cluster support yet to go straight to
1747 * hard readonly in here. Until then, we want to keep
1748 * ocfs2_abort() so that we can at least mark critical
1751 * TODO: This should abort the journal and alert other nodes
1752 * that our slot needs recovery. */
1754 /* Force a panic(). This stinks, but it's better than letting
1755 * things continue without having a proper hard readonly
1757 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1758 ocfs2_handle_error(sb);
1761 module_init(ocfs2_init);
1762 module_exit(ocfs2_exit);