git.oblomov.eu Git - linux-2.6/blob - fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  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.
  14  *
  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.
  19  *
  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.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/utsname.h>
  32 #include <linux/init.h>
  33 #include <linux/random.h>
  34 #include <linux/statfs.h>
  35 #include <linux/moduleparam.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/socket.h>
  38 #include <linux/inet.h>
  39 #include <linux/parser.h>
  40 #include <linux/crc32.h>
  41 #include <linux/debugfs.h>
  42 #include <linux/mount.h>
  43 #include <linux/seq_file.h>
  44
  45 #define MLOG_MASK_PREFIX ML_SUPER
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 /* this should be the only file to include a version 1 header */
  51 #include "ocfs1_fs_compat.h"
  52
  53 #include "alloc.h"
  54 #include "dlmglue.h"
  55 #include "export.h"
  56 #include "extent_map.h"
  57 #include "heartbeat.h"
  58 #include "inode.h"
  59 #include "journal.h"
  60 #include "localalloc.h"
  61 #include "namei.h"
  62 #include "slot_map.h"
  63 #include "super.h"
  64 #include "sysfile.h"
  65 #include "uptodate.h"
  66 #include "ver.h"
  67 #include "xattr.h"
  68 #include "quota.h"
  69
  70 #include "buffer_head_io.h"
  71
  72 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  73 struct kmem_cache *ocfs2_dquot_cachep;
  74 struct kmem_cache *ocfs2_qf_chunk_cachep;
  75
  76 /* OCFS2 needs to schedule several differnt types of work which
  77  * require cluster locking, disk I/O, recovery waits, etc. Since these
  78  * types of work tend to be heavy we avoid using the kernel events
  79  * workqueue and schedule on our own. */
  80 struct workqueue_struct *ocfs2_wq = NULL;
  81
  82 static struct dentry *ocfs2_debugfs_root = NULL;
  83
  84 MODULE_AUTHOR("Oracle");
  85 MODULE_LICENSE("GPL");
  86
  87 struct mount_options
  88 {
  89         unsigned long   commit_interval;
  90         unsigned long   mount_opt;
  91         unsigned int    atime_quantum;
  92         signed short    slot;
  93         unsigned int    localalloc_opt;
  94         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
  95 };
  96
  97 static int ocfs2_parse_options(struct super_block *sb, char *options,
  98                                struct mount_options *mopt,
  99                                int is_remount);
 100 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
 101 static void ocfs2_put_super(struct super_block *sb);
 102 static int ocfs2_mount_volume(struct super_block *sb);
 103 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 104 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 105 static int ocfs2_initialize_mem_caches(void);
 106 static void ocfs2_free_mem_caches(void);
 107 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 108
 109 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 110
 111 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 112
 113 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 114 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 115 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 116 static int ocfs2_check_volume(struct ocfs2_super *osb);
 117 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 118                                struct buffer_head *bh,
 119                                u32 sectsize);
 120 static int ocfs2_initialize_super(struct super_block *sb,
 121                                   struct buffer_head *bh,
 122                                   int sector_size);
 123 static int ocfs2_get_sector(struct super_block *sb,
 124                             struct buffer_head **bh,
 125                             int block,
 126                             int sect_size);
 127 static void ocfs2_write_super(struct super_block *sb);
 128 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 129 static void ocfs2_destroy_inode(struct inode *inode);
 130
 131 static const struct super_operations ocfs2_sops = {
 132         .statfs         = ocfs2_statfs,
 133         .alloc_inode    = ocfs2_alloc_inode,
 134         .destroy_inode  = ocfs2_destroy_inode,
 135         .drop_inode     = ocfs2_drop_inode,
 136         .clear_inode    = ocfs2_clear_inode,
 137         .delete_inode   = ocfs2_delete_inode,
 138         .sync_fs        = ocfs2_sync_fs,
 139         .write_super    = ocfs2_write_super,
 140         .put_super      = ocfs2_put_super,
 141         .remount_fs     = ocfs2_remount,
 142         .show_options   = ocfs2_show_options,
 143         .quota_read     = ocfs2_quota_read,
 144         .quota_write    = ocfs2_quota_write,
 145 };
 146
 147 enum {
 148         Opt_barrier,
 149         Opt_err_panic,
 150         Opt_err_ro,
 151         Opt_intr,
 152         Opt_nointr,
 153         Opt_hb_none,
 154         Opt_hb_local,
 155         Opt_data_ordered,
 156         Opt_data_writeback,
 157         Opt_atime_quantum,
 158         Opt_slot,
 159         Opt_commit,
 160         Opt_localalloc,
 161         Opt_localflocks,
 162         Opt_stack,
 163         Opt_user_xattr,
 164         Opt_nouser_xattr,
 165         Opt_inode64,
 166         Opt_acl,
 167         Opt_noacl,
 168         Opt_err,
 169 };
 170
 171 static const match_table_t tokens = {
 172         {Opt_barrier, "barrier=%u"},
 173         {Opt_err_panic, "errors=panic"},
 174         {Opt_err_ro, "errors=remount-ro"},
 175         {Opt_intr, "intr"},
 176         {Opt_nointr, "nointr"},
 177         {Opt_hb_none, OCFS2_HB_NONE},
 178         {Opt_hb_local, OCFS2_HB_LOCAL},
 179         {Opt_data_ordered, "data=ordered"},
 180         {Opt_data_writeback, "data=writeback"},
 181         {Opt_atime_quantum, "atime_quantum=%u"},
 182         {Opt_slot, "preferred_slot=%u"},
 183         {Opt_commit, "commit=%u"},
 184         {Opt_localalloc, "localalloc=%d"},
 185         {Opt_localflocks, "localflocks"},
 186         {Opt_stack, "cluster_stack=%s"},
 187         {Opt_user_xattr, "user_xattr"},
 188         {Opt_nouser_xattr, "nouser_xattr"},
 189         {Opt_inode64, "inode64"},
 190         {Opt_acl, "acl"},
 191         {Opt_noacl, "noacl"},
 192         {Opt_err, NULL}
 193 };
 194
 195 /*
 196  * write_super and sync_fs ripped right out of ext3.
 197  */
 198 static void ocfs2_write_super(struct super_block *sb)
 199 {
 200         if (mutex_trylock(&sb->s_lock) != 0)
 201                 BUG();
 202         sb->s_dirt = 0;
 203 }
 204
 205 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 206 {
 207         int status;
 208         tid_t target;
 209         struct ocfs2_super *osb = OCFS2_SB(sb);
 210
 211         sb->s_dirt = 0;
 212
 213         if (ocfs2_is_hard_readonly(osb))
 214                 return -EROFS;
 215
 216         if (wait) {
 217                 status = ocfs2_flush_truncate_log(osb);
 218                 if (status < 0)
 219                         mlog_errno(status);
 220         } else {
 221                 ocfs2_schedule_truncate_log_flush(osb, 0);
 222         }
 223
 224         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 225                                       &target)) {
 226                 if (wait)
 227                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 228                                              target);
 229         }
 230         return 0;
 231 }
 232
 233 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 234 {
 235         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 236             && (ino == USER_QUOTA_SYSTEM_INODE
 237                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 238                 return 0;
 239         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 240             && (ino == GROUP_QUOTA_SYSTEM_INODE
 241                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 242                 return 0;
 243         return 1;
 244 }
 245
 246 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 247 {
 248         struct inode *new = NULL;
 249         int status = 0;
 250         int i;
 251
 252         mlog_entry_void();
 253
 254         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 255         if (IS_ERR(new)) {
 256                 status = PTR_ERR(new);
 257                 mlog_errno(status);
 258                 goto bail;
 259         }
 260         osb->root_inode = new;
 261
 262         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 263         if (IS_ERR(new)) {
 264                 status = PTR_ERR(new);
 265                 mlog_errno(status);
 266                 goto bail;
 267         }
 268         osb->sys_root_inode = new;
 269
 270         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 271              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 272                 if (!ocfs2_need_system_inode(osb, i))
 273                         continue;
 274                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 275                 if (!new) {
 276                         ocfs2_release_system_inodes(osb);
 277                         status = -EINVAL;
 278                         mlog_errno(status);
 279                         /* FIXME: Should ERROR_RO_FS */
 280                         mlog(ML_ERROR, "Unable to load system inode %d, "
 281                              "possibly corrupt fs?", i);
 282                         goto bail;
 283                 }
 284                 // the array now has one ref, so drop this one
 285                 iput(new);
 286         }
 287
 288 bail:
 289         mlog_exit(status);
 290         return status;
 291 }
 292
 293 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 294 {
 295         struct inode *new = NULL;
 296         int status = 0;
 297         int i;
 298
 299         mlog_entry_void();
 300
 301         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 302              i < NUM_SYSTEM_INODES;
 303              i++) {
 304                 if (!ocfs2_need_system_inode(osb, i))
 305                         continue;
 306                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 307                 if (!new) {
 308                         ocfs2_release_system_inodes(osb);
 309                         status = -EINVAL;
 310                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 311                              status, i, osb->slot_num);
 312                         goto bail;
 313                 }
 314                 /* the array now has one ref, so drop this one */
 315                 iput(new);
 316         }
 317
 318 bail:
 319         mlog_exit(status);
 320         return status;
 321 }
 322
 323 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 324 {
 325         int i;
 326         struct inode *inode;
 327
 328         mlog_entry_void();
 329
 330         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
 331                 inode = osb->system_inodes[i];
 332                 if (inode) {
 333                         iput(inode);
 334                         osb->system_inodes[i] = NULL;
 335                 }
 336         }
 337
 338         inode = osb->sys_root_inode;
 339         if (inode) {
 340                 iput(inode);
 341                 osb->sys_root_inode = NULL;
 342         }
 343
 344         inode = osb->root_inode;
 345         if (inode) {
 346                 iput(inode);
 347                 osb->root_inode = NULL;
 348         }
 349
 350         mlog_exit(0);
 351 }
 352
 353 /* We're allocating fs objects, use GFP_NOFS */
 354 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 355 {
 356         struct ocfs2_inode_info *oi;
 357
 358         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 359         if (!oi)
 360                 return NULL;
 361
 362         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 363         return &oi->vfs_inode;
 364 }
 365
 366 static void ocfs2_destroy_inode(struct inode *inode)
 367 {
 368         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 369 }
 370
 371 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 372                                                 unsigned int cbits)
 373 {
 374         unsigned int bytes = 1 << cbits;
 375         unsigned int trim = bytes;
 376         unsigned int bitshift = 32;
 377
 378         /*
 379          * i_size and all block offsets in ocfs2 are always 64 bits
 380          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 381          * 64 bit platforms, cluster size will be the limiting factor.
 382          */
 383
 384 #if BITS_PER_LONG == 32
 385 # if defined(CONFIG_LBD)
 386         BUILD_BUG_ON(sizeof(sector_t) != 8);
 387         /*
 388          * We might be limited by page cache size.
 389          */
 390         if (bytes > PAGE_CACHE_SIZE) {
 391                 bytes = PAGE_CACHE_SIZE;
 392                 trim = 1;
 393                 /*
 394                  * Shift by 31 here so that we don't get larger than
 395                  * MAX_LFS_FILESIZE
 396                  */
 397                 bitshift = 31;
 398         }
 399 # else
 400         /*
 401          * We are limited by the size of sector_t. Use block size, as
 402          * that's what we expose to the VFS.
 403          */
 404         bytes = 1 << bbits;
 405         trim = 1;
 406         bitshift = 31;
 407 # endif
 408 #endif
 409
 410         /*
 411          * Trim by a whole cluster when we can actually approach the
 412          * on-disk limits. Otherwise we can overflow i_clusters when
 413          * an extent start is at the max offset.
 414          */
 415         return (((unsigned long long)bytes) << bitshift) - trim;
 416 }
 417
 418 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 419 {
 420         int incompat_features;
 421         int ret = 0;
 422         struct mount_options parsed_options;
 423         struct ocfs2_super *osb = OCFS2_SB(sb);
 424
 425         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 426                 ret = -EINVAL;
 427                 goto out;
 428         }
 429
 430         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
 431             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 432                 ret = -EINVAL;
 433                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 434                 goto out;
 435         }
 436
 437         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 438             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 439                 ret = -EINVAL;
 440                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 441                 goto out;
 442         }
 443
 444         /* Probably don't want this on remount; it might
 445          * mess with other nodes */
 446         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 447             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 448                 ret = -EINVAL;
 449                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 450                 goto out;
 451         }
 452
 453         /* We're going to/from readonly mode. */
 454         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 455                 /* Lock here so the check of HARD_RO and the potential
 456                  * setting of SOFT_RO is atomic. */
 457                 spin_lock(&osb->osb_lock);
 458                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 459                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 460                         ret = -EROFS;
 461                         goto unlock_osb;
 462                 }
 463
 464                 if (*flags & MS_RDONLY) {
 465                         mlog(0, "Going to ro mode.\n");
 466                         sb->s_flags |= MS_RDONLY;
 467                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 468                 } else {
 469                         mlog(0, "Making ro filesystem writeable.\n");
 470
 471                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 472                                 mlog(ML_ERROR, "Cannot remount RDWR "
 473                                      "filesystem due to previous errors.\n");
 474                                 ret = -EROFS;
 475                                 goto unlock_osb;
 476                         }
 477                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 478                         if (incompat_features) {
 479                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 480                                      "of unsupported optional features "
 481                                      "(%x).\n", incompat_features);
 482                                 ret = -EINVAL;
 483                                 goto unlock_osb;
 484                         }
 485                         sb->s_flags &= ~MS_RDONLY;
 486                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 487                 }
 488 unlock_osb:
 489                 spin_unlock(&osb->osb_lock);
 490         }
 491
 492         if (!ret) {
 493                 /* Only save off the new mount options in case of a successful
 494                  * remount. */
 495                 if (!(osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_XATTR))
 496                         parsed_options.mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
 497                 osb->s_mount_opt = parsed_options.mount_opt;
 498                 osb->s_atime_quantum = parsed_options.atime_quantum;
 499                 osb->preferred_slot = parsed_options.slot;
 500                 if (parsed_options.commit_interval)
 501                         osb->osb_commit_interval = parsed_options.commit_interval;
 502
 503                 if (!ocfs2_is_hard_readonly(osb))
 504                         ocfs2_set_journal_params(osb);
 505         }
 506 out:
 507         return ret;
 508 }
 509
 510 static int ocfs2_sb_probe(struct super_block *sb,
 511                           struct buffer_head **bh,
 512                           int *sector_size)
 513 {
 514         int status, tmpstat;
 515         struct ocfs1_vol_disk_hdr *hdr;
 516         struct ocfs2_dinode *di;
 517         int blksize;
 518
 519         *bh = NULL;
 520
 521         /* may be > 512 */
 522         *sector_size = bdev_hardsect_size(sb->s_bdev);
 523         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 524                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 525                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 526                 status = -EINVAL;
 527                 goto bail;
 528         }
 529
 530         /* Can this really happen? */
 531         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 532                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 533
 534         /* check block zero for old format */
 535         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 536         if (status < 0) {
 537                 mlog_errno(status);
 538                 goto bail;
 539         }
 540         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 541         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 542                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 543                      hdr->major_version, hdr->minor_version);
 544                 status = -EINVAL;
 545         }
 546         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 547                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 548                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 549                      hdr->signature);
 550                 status = -EINVAL;
 551         }
 552         brelse(*bh);
 553         *bh = NULL;
 554         if (status < 0) {
 555                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 556                      "upgraded before mounting with ocfs v2\n");
 557                 goto bail;
 558         }
 559
 560         /*
 561          * Now check at magic offset for 512, 1024, 2048, 4096
 562          * blocksizes.  4096 is the maximum blocksize because it is
 563          * the minimum clustersize.
 564          */
 565         status = -EINVAL;
 566         for (blksize = *sector_size;
 567              blksize <= OCFS2_MAX_BLOCKSIZE;
 568              blksize <<= 1) {
 569                 tmpstat = ocfs2_get_sector(sb, bh,
 570                                            OCFS2_SUPER_BLOCK_BLKNO,
 571                                            blksize);
 572                 if (tmpstat < 0) {
 573                         status = tmpstat;
 574                         mlog_errno(status);
 575                         goto bail;
 576                 }
 577                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 578                 status = ocfs2_verify_volume(di, *bh, blksize);
 579                 if (status >= 0)
 580                         goto bail;
 581                 brelse(*bh);
 582                 *bh = NULL;
 583                 if (status != -EAGAIN)
 584                         break;
 585         }
 586
 587 bail:
 588         return status;
 589 }
 590
 591 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 592 {
 593         if (ocfs2_mount_local(osb)) {
 594                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 595                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 596                              "mounted device.\n");
 597                         return -EINVAL;
 598                 }
 599         }
 600
 601         if (ocfs2_userspace_stack(osb)) {
 602                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 603                         mlog(ML_ERROR, "Userspace stack expected, but "
 604                              "o2cb heartbeat arguments passed to mount\n");
 605                         return -EINVAL;
 606                 }
 607         }
 608
 609         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 610                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 611                     !ocfs2_userspace_stack(osb)) {
 612                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 613                              "a read-write clustered device.\n");
 614                         return -EINVAL;
 615                 }
 616         }
 617
 618         return 0;
 619 }
 620
 621 /*
 622  * If we're using a userspace stack, mount should have passed
 623  * a name that matches the disk.  If not, mount should not
 624  * have passed a stack.
 625  */
 626 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 627                                         struct mount_options *mopt)
 628 {
 629         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 630                 mlog(ML_ERROR,
 631                      "cluster stack passed to mount, but this filesystem "
 632                      "does not support it\n");
 633                 return -EINVAL;
 634         }
 635
 636         if (ocfs2_userspace_stack(osb) &&
 637             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 638                     OCFS2_STACK_LABEL_LEN)) {
 639                 mlog(ML_ERROR,
 640                      "cluster stack passed to mount (\"%s\") does not "
 641                      "match the filesystem (\"%s\")\n",
 642                      mopt->cluster_stack,
 643                      osb->osb_cluster_stack);
 644                 return -EINVAL;
 645         }
 646
 647         return 0;
 648 }
 649
 650 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 651 {
 652         struct dentry *root;
 653         int status, sector_size;
 654         struct mount_options parsed_options;
 655         struct inode *inode = NULL;
 656         struct ocfs2_super *osb = NULL;
 657         struct buffer_head *bh = NULL;
 658         char nodestr[8];
 659
 660         mlog_entry("%p, %p, %i", sb, data, silent);
 661
 662         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 663                 status = -EINVAL;
 664                 goto read_super_error;
 665         }
 666
 667         /* probe for superblock */
 668         status = ocfs2_sb_probe(sb, &bh, &sector_size);
 669         if (status < 0) {
 670                 mlog(ML_ERROR, "superblock probe failed!\n");
 671                 goto read_super_error;
 672         }
 673
 674         status = ocfs2_initialize_super(sb, bh, sector_size);
 675         osb = OCFS2_SB(sb);
 676         if (status < 0) {
 677                 mlog_errno(status);
 678                 goto read_super_error;
 679         }
 680         brelse(bh);
 681         bh = NULL;
 682
 683         if (!(osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_XATTR))
 684                 parsed_options.mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
 685
 686         osb->s_mount_opt = parsed_options.mount_opt;
 687         osb->s_atime_quantum = parsed_options.atime_quantum;
 688         osb->preferred_slot = parsed_options.slot;
 689         osb->osb_commit_interval = parsed_options.commit_interval;
 690         osb->local_alloc_default_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
 691         osb->local_alloc_bits = osb->local_alloc_default_bits;
 692
 693         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
 694         if (status)
 695                 goto read_super_error;
 696
 697         sb->s_magic = OCFS2_SUPER_MAGIC;
 698
 699         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 700                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
 701
 702         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
 703          * heartbeat=none */
 704         if (bdev_read_only(sb->s_bdev)) {
 705                 if (!(sb->s_flags & MS_RDONLY)) {
 706                         status = -EACCES;
 707                         mlog(ML_ERROR, "Readonly device detected but readonly "
 708                              "mount was not specified.\n");
 709                         goto read_super_error;
 710                 }
 711
 712                 /* You should not be able to start a local heartbeat
 713                  * on a readonly device. */
 714                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 715                         status = -EROFS;
 716                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
 717                              "device.\n");
 718                         goto read_super_error;
 719                 }
 720
 721                 status = ocfs2_check_journals_nolocks(osb);
 722                 if (status < 0) {
 723                         if (status == -EROFS)
 724                                 mlog(ML_ERROR, "Recovery required on readonly "
 725                                      "file system, but write access is "
 726                                      "unavailable.\n");
 727                         else
 728                                 mlog_errno(status);
 729                         goto read_super_error;
 730                 }
 731
 732                 ocfs2_set_ro_flag(osb, 1);
 733
 734                 printk(KERN_NOTICE "Readonly device detected. No cluster "
 735                        "services will be utilized for this mount. Recovery "
 736                        "will be skipped.\n");
 737         }
 738
 739         if (!ocfs2_is_hard_readonly(osb)) {
 740                 if (sb->s_flags & MS_RDONLY)
 741                         ocfs2_set_ro_flag(osb, 0);
 742         }
 743
 744         status = ocfs2_verify_heartbeat(osb);
 745         if (status < 0) {
 746                 mlog_errno(status);
 747                 goto read_super_error;
 748         }
 749
 750         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
 751                                                  ocfs2_debugfs_root);
 752         if (!osb->osb_debug_root) {
 753                 status = -EINVAL;
 754                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
 755                 goto read_super_error;
 756         }
 757
 758         status = ocfs2_mount_volume(sb);
 759         if (osb->root_inode)
 760                 inode = igrab(osb->root_inode);
 761
 762         if (status < 0)
 763                 goto read_super_error;
 764
 765         if (!inode) {
 766                 status = -EIO;
 767                 mlog_errno(status);
 768                 goto read_super_error;
 769         }
 770
 771         root = d_alloc_root(inode);
 772         if (!root) {
 773                 status = -ENOMEM;
 774                 mlog_errno(status);
 775                 goto read_super_error;
 776         }
 777
 778         sb->s_root = root;
 779
 780         ocfs2_complete_mount_recovery(osb);
 781
 782         if (ocfs2_mount_local(osb))
 783                 snprintf(nodestr, sizeof(nodestr), "local");
 784         else
 785                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
 786
 787         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
 788                "with %s data mode.\n",
 789                osb->dev_str, nodestr, osb->slot_num,
 790                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
 791                "ordered");
 792
 793         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
 794         wake_up(&osb->osb_mount_event);
 795
 796         mlog_exit(status);
 797         return status;
 798
 799 read_super_error:
 800         brelse(bh);
 801
 802         if (inode)
 803                 iput(inode);
 804
 805         if (osb) {
 806                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
 807                 wake_up(&osb->osb_mount_event);
 808                 ocfs2_dismount_volume(sb, 1);
 809         }
 810
 811         mlog_exit(status);
 812         return status;
 813 }
 814
 815 static int ocfs2_get_sb(struct file_system_type *fs_type,
 816                         int flags,
 817                         const char *dev_name,
 818                         void *data,
 819                         struct vfsmount *mnt)
 820 {
 821         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
 822                            mnt);
 823 }
 824
 825 static struct file_system_type ocfs2_fs_type = {
 826         .owner          = THIS_MODULE,
 827         .name           = "ocfs2",
 828         .get_sb         = ocfs2_get_sb, /* is this called when we mount
 829                                         * the fs? */
 830         .kill_sb        = kill_block_super, /* set to the generic one
 831                                              * right now, but do we
 832                                              * need to change that? */
 833         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 834         .next           = NULL
 835 };
 836
 837 static int ocfs2_parse_options(struct super_block *sb,
 838                                char *options,
 839                                struct mount_options *mopt,
 840                                int is_remount)
 841 {
 842         int status;
 843         char *p;
 844
 845         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 846                    options ? options : "(none)");
 847
 848         mopt->commit_interval = 0;
 849         mopt->mount_opt = 0;
 850         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 851         mopt->slot = OCFS2_INVALID_SLOT;
 852         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
 853         mopt->cluster_stack[0] = '\0';
 854
 855         if (!options) {
 856                 status = 1;
 857                 goto bail;
 858         }
 859
 860         while ((p = strsep(&options, ",")) != NULL) {
 861                 int token, option;
 862                 substring_t args[MAX_OPT_ARGS];
 863
 864                 if (!*p)
 865                         continue;
 866
 867                 token = match_token(p, tokens, args);
 868                 switch (token) {
 869                 case Opt_hb_local:
 870                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 871                         break;
 872                 case Opt_hb_none:
 873                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 874                         break;
 875                 case Opt_barrier:
 876                         if (match_int(&args[0], &option)) {
 877                                 status = 0;
 878                                 goto bail;
 879                         }
 880                         if (option)
 881                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
 882                         else
 883                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
 884                         break;
 885                 case Opt_intr:
 886                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
 887                         break;
 888                 case Opt_nointr:
 889                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
 890                         break;
 891                 case Opt_err_panic:
 892                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 893                         break;
 894                 case Opt_err_ro:
 895                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 896                         break;
 897                 case Opt_data_ordered:
 898                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 899                         break;
 900                 case Opt_data_writeback:
 901                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 902                         break;
 903                 case Opt_user_xattr:
 904                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
 905                         break;
 906                 case Opt_nouser_xattr:
 907                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
 908                         break;
 909                 case Opt_atime_quantum:
 910                         if (match_int(&args[0], &option)) {
 911                                 status = 0;
 912                                 goto bail;
 913                         }
 914                         if (option >= 0)
 915                                 mopt->atime_quantum = option;
 916                         break;
 917                 case Opt_slot:
 918                         option = 0;
 919                         if (match_int(&args[0], &option)) {
 920                                 status = 0;
 921                                 goto bail;
 922                         }
 923                         if (option)
 924                                 mopt->slot = (s16)option;
 925                         break;
 926                 case Opt_commit:
 927                         option = 0;
 928                         if (match_int(&args[0], &option)) {
 929                                 status = 0;
 930                                 goto bail;
 931                         }
 932                         if (option < 0)
 933                                 return 0;
 934                         if (option == 0)
 935                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
 936                         mopt->commit_interval = HZ * option;
 937                         break;
 938                 case Opt_localalloc:
 939                         option = 0;
 940                         if (match_int(&args[0], &option)) {
 941                                 status = 0;
 942                                 goto bail;
 943                         }
 944                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
 945                                 mopt->localalloc_opt = option;
 946                         break;
 947                 case Opt_localflocks:
 948                         /*
 949                          * Changing this during remount could race
 950                          * flock() requests, or "unbalance" existing
 951                          * ones (e.g., a lock is taken in one mode but
 952                          * dropped in the other). If users care enough
 953                          * to flip locking modes during remount, we
 954                          * could add a "local" flag to individual
 955                          * flock structures for proper tracking of
 956                          * state.
 957                          */
 958                         if (!is_remount)
 959                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
 960                         break;
 961                 case Opt_stack:
 962                         /* Check both that the option we were passed
 963                          * is of the right length and that it is a proper
 964                          * string of the right length.
 965                          */
 966                         if (((args[0].to - args[0].from) !=
 967                              OCFS2_STACK_LABEL_LEN) ||
 968                             (strnlen(args[0].from,
 969                                      OCFS2_STACK_LABEL_LEN) !=
 970                              OCFS2_STACK_LABEL_LEN)) {
 971                                 mlog(ML_ERROR,
 972                                      "Invalid cluster_stack option\n");
 973                                 status = 0;
 974                                 goto bail;
 975                         }
 976                         memcpy(mopt->cluster_stack, args[0].from,
 977                                OCFS2_STACK_LABEL_LEN);
 978                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
 979                         break;
 980                 case Opt_inode64:
 981                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
 982                         break;
 983 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 984                 case Opt_acl:
 985                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
 986                         break;
 987                 case Opt_noacl:
 988                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
 989                         break;
 990 #else
 991                 case Opt_acl:
 992                 case Opt_noacl:
 993                         printk(KERN_INFO "ocfs2 (no)acl options not supported\n");
 994                         break;
 995 #endif
 996                 default:
 997                         mlog(ML_ERROR,
 998                              "Unrecognized mount option \"%s\" "
 999                              "or missing value\n", p);
1000                         status = 0;
1001                         goto bail;
1002                 }
1003         }
1004
1005         status = 1;
1006
1007 bail:
1008         mlog_exit(status);
1009         return status;
1010 }
1011
1012 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1013 {
1014         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
1015         unsigned long opts = osb->s_mount_opt;
1016         unsigned int local_alloc_megs;
1017
1018         if (opts & OCFS2_MOUNT_HB_LOCAL)
1019                 seq_printf(s, ",_netdev,heartbeat=local");
1020         else
1021                 seq_printf(s, ",heartbeat=none");
1022
1023         if (opts & OCFS2_MOUNT_NOINTR)
1024                 seq_printf(s, ",nointr");
1025
1026         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1027                 seq_printf(s, ",data=writeback");
1028         else
1029                 seq_printf(s, ",data=ordered");
1030
1031         if (opts & OCFS2_MOUNT_BARRIER)
1032                 seq_printf(s, ",barrier=1");
1033
1034         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1035                 seq_printf(s, ",errors=panic");
1036         else
1037                 seq_printf(s, ",errors=remount-ro");
1038
1039         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1040                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1041
1042         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
1043                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1044
1045         if (osb->osb_commit_interval)
1046                 seq_printf(s, ",commit=%u",
1047                            (unsigned) (osb->osb_commit_interval / HZ));
1048
1049         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1050         if (local_alloc_megs != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
1051                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1052
1053         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1054                 seq_printf(s, ",localflocks,");
1055
1056         if (osb->osb_cluster_stack[0])
1057                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1058                            osb->osb_cluster_stack);
1059
1060         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1061                 seq_printf(s, ",nouser_xattr");
1062         else
1063                 seq_printf(s, ",user_xattr");
1064
1065         if (opts & OCFS2_MOUNT_INODE64)
1066                 seq_printf(s, ",inode64");
1067
1068 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
1069         if (opts & OCFS2_MOUNT_POSIX_ACL)
1070                 seq_printf(s, ",acl");
1071         else
1072                 seq_printf(s, ",noacl");
1073 #endif
1074
1075         return 0;
1076 }
1077
1078 static int __init ocfs2_init(void)
1079 {
1080         int status;
1081
1082         mlog_entry_void();
1083
1084         ocfs2_print_version();
1085
1086         status = init_ocfs2_uptodate_cache();
1087         if (status < 0) {
1088                 mlog_errno(status);
1089                 goto leave;
1090         }
1091
1092         status = ocfs2_initialize_mem_caches();
1093         if (status < 0) {
1094                 mlog_errno(status);
1095                 goto leave;
1096         }
1097
1098         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1099         if (!ocfs2_wq) {
1100                 status = -ENOMEM;
1101                 goto leave;
1102         }
1103
1104         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1105         if (!ocfs2_debugfs_root) {
1106                 status = -EFAULT;
1107                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1108         }
1109
1110         ocfs2_set_locking_protocol();
1111
1112         status = register_quota_format(&ocfs2_quota_format);
1113 leave:
1114         if (status < 0) {
1115                 ocfs2_free_mem_caches();
1116                 exit_ocfs2_uptodate_cache();
1117         }
1118
1119         mlog_exit(status);
1120
1121         if (status >= 0) {
1122                 return register_filesystem(&ocfs2_fs_type);
1123         } else
1124                 return -1;
1125 }
1126
1127 static void __exit ocfs2_exit(void)
1128 {
1129         mlog_entry_void();
1130
1131         if (ocfs2_wq) {
1132                 flush_workqueue(ocfs2_wq);
1133                 destroy_workqueue(ocfs2_wq);
1134         }
1135
1136         unregister_quota_format(&ocfs2_quota_format);
1137
1138         debugfs_remove(ocfs2_debugfs_root);
1139
1140         ocfs2_free_mem_caches();
1141
1142         unregister_filesystem(&ocfs2_fs_type);
1143
1144         exit_ocfs2_uptodate_cache();
1145
1146         mlog_exit_void();
1147 }
1148
1149 static void ocfs2_put_super(struct super_block *sb)
1150 {
1151         mlog_entry("(0x%p)\n", sb);
1152
1153         ocfs2_sync_blockdev(sb);
1154         ocfs2_dismount_volume(sb, 0);
1155
1156         mlog_exit_void();
1157 }
1158
1159 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1160 {
1161         struct ocfs2_super *osb;
1162         u32 numbits, freebits;
1163         int status;
1164         struct ocfs2_dinode *bm_lock;
1165         struct buffer_head *bh = NULL;
1166         struct inode *inode = NULL;
1167
1168         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1169
1170         osb = OCFS2_SB(dentry->d_sb);
1171
1172         inode = ocfs2_get_system_file_inode(osb,
1173                                             GLOBAL_BITMAP_SYSTEM_INODE,
1174                                             OCFS2_INVALID_SLOT);
1175         if (!inode) {
1176                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1177                 status = -EIO;
1178                 goto bail;
1179         }
1180
1181         status = ocfs2_inode_lock(inode, &bh, 0);
1182         if (status < 0) {
1183                 mlog_errno(status);
1184                 goto bail;
1185         }
1186
1187         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1188
1189         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1190         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1191
1192         buf->f_type = OCFS2_SUPER_MAGIC;
1193         buf->f_bsize = dentry->d_sb->s_blocksize;
1194         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1195         buf->f_blocks = ((sector_t) numbits) *
1196                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1197         buf->f_bfree = ((sector_t) freebits) *
1198                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1199         buf->f_bavail = buf->f_bfree;
1200         buf->f_files = numbits;
1201         buf->f_ffree = freebits;
1202
1203         brelse(bh);
1204
1205         ocfs2_inode_unlock(inode, 0);
1206         status = 0;
1207 bail:
1208         if (inode)
1209                 iput(inode);
1210
1211         mlog_exit(status);
1212
1213         return status;
1214 }
1215
1216 static void ocfs2_inode_init_once(void *data)
1217 {
1218         struct ocfs2_inode_info *oi = data;
1219
1220         oi->ip_flags = 0;
1221         oi->ip_open_count = 0;
1222         spin_lock_init(&oi->ip_lock);
1223         ocfs2_extent_map_init(&oi->vfs_inode);
1224         INIT_LIST_HEAD(&oi->ip_io_markers);
1225         oi->ip_created_trans = 0;
1226         oi->ip_last_trans = 0;
1227         oi->ip_dir_start_lookup = 0;
1228
1229         init_rwsem(&oi->ip_alloc_sem);
1230         init_rwsem(&oi->ip_xattr_sem);
1231         mutex_init(&oi->ip_io_mutex);
1232
1233         oi->ip_blkno = 0ULL;
1234         oi->ip_clusters = 0;
1235
1236         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1237         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1238         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1239
1240         ocfs2_metadata_cache_init(&oi->vfs_inode);
1241
1242         inode_init_once(&oi->vfs_inode);
1243 }
1244
1245 static int ocfs2_initialize_mem_caches(void)
1246 {
1247         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1248                                        sizeof(struct ocfs2_inode_info),
1249                                        0,
1250                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1251                                                 SLAB_MEM_SPREAD),
1252                                        ocfs2_inode_init_once);
1253         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1254                                         sizeof(struct ocfs2_dquot),
1255                                         0,
1256                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1257                                                 SLAB_MEM_SPREAD),
1258                                         NULL);
1259         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1260                                         sizeof(struct ocfs2_quota_chunk),
1261                                         0,
1262                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1263                                         NULL);
1264         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1265             !ocfs2_qf_chunk_cachep) {
1266                 if (ocfs2_inode_cachep)
1267                         kmem_cache_destroy(ocfs2_inode_cachep);
1268                 if (ocfs2_dquot_cachep)
1269                         kmem_cache_destroy(ocfs2_dquot_cachep);
1270                 if (ocfs2_qf_chunk_cachep)
1271                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1272                 return -ENOMEM;
1273         }
1274
1275         return 0;
1276 }
1277
1278 static void ocfs2_free_mem_caches(void)
1279 {
1280         if (ocfs2_inode_cachep)
1281                 kmem_cache_destroy(ocfs2_inode_cachep);
1282         ocfs2_inode_cachep = NULL;
1283
1284         if (ocfs2_dquot_cachep)
1285                 kmem_cache_destroy(ocfs2_dquot_cachep);
1286         ocfs2_dquot_cachep = NULL;
1287
1288         if (ocfs2_qf_chunk_cachep)
1289                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1290         ocfs2_qf_chunk_cachep = NULL;
1291 }
1292
1293 static int ocfs2_get_sector(struct super_block *sb,
1294                             struct buffer_head **bh,
1295                             int block,
1296                             int sect_size)
1297 {
1298         if (!sb_set_blocksize(sb, sect_size)) {
1299                 mlog(ML_ERROR, "unable to set blocksize\n");
1300                 return -EIO;
1301         }
1302
1303         *bh = sb_getblk(sb, block);
1304         if (!*bh) {
1305                 mlog_errno(-EIO);
1306                 return -EIO;
1307         }
1308         lock_buffer(*bh);
1309         if (!buffer_dirty(*bh))
1310                 clear_buffer_uptodate(*bh);
1311         unlock_buffer(*bh);
1312         ll_rw_block(READ, 1, bh);
1313         wait_on_buffer(*bh);
1314         return 0;
1315 }
1316
1317 static int ocfs2_mount_volume(struct super_block *sb)
1318 {
1319         int status = 0;
1320         int unlock_super = 0;
1321         struct ocfs2_super *osb = OCFS2_SB(sb);
1322
1323         mlog_entry_void();
1324
1325         if (ocfs2_is_hard_readonly(osb))
1326                 goto leave;
1327
1328         status = ocfs2_dlm_init(osb);
1329         if (status < 0) {
1330                 mlog_errno(status);
1331                 goto leave;
1332         }
1333
1334         status = ocfs2_super_lock(osb, 1);
1335         if (status < 0) {
1336                 mlog_errno(status);
1337                 goto leave;
1338         }
1339         unlock_super = 1;
1340
1341         /* This will load up the node map and add ourselves to it. */
1342         status = ocfs2_find_slot(osb);
1343         if (status < 0) {
1344                 mlog_errno(status);
1345                 goto leave;
1346         }
1347
1348         /* load all node-local system inodes */
1349         status = ocfs2_init_local_system_inodes(osb);
1350         if (status < 0) {
1351                 mlog_errno(status);
1352                 goto leave;
1353         }
1354
1355         status = ocfs2_check_volume(osb);
1356         if (status < 0) {
1357                 mlog_errno(status);
1358                 goto leave;
1359         }
1360
1361         status = ocfs2_truncate_log_init(osb);
1362         if (status < 0) {
1363                 mlog_errno(status);
1364                 goto leave;
1365         }
1366
1367         if (ocfs2_mount_local(osb))
1368                 goto leave;
1369
1370 leave:
1371         if (unlock_super)
1372                 ocfs2_super_unlock(osb, 1);
1373
1374         mlog_exit(status);
1375         return status;
1376 }
1377
1378 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1379 {
1380         int tmp, hangup_needed = 0;
1381         struct ocfs2_super *osb = NULL;
1382         char nodestr[8];
1383
1384         mlog_entry("(0x%p)\n", sb);
1385
1386         BUG_ON(!sb);
1387         osb = OCFS2_SB(sb);
1388         BUG_ON(!osb);
1389
1390         ocfs2_shutdown_local_alloc(osb);
1391
1392         ocfs2_truncate_log_shutdown(osb);
1393
1394         /* This will disable recovery and flush any recovery work. */
1395         ocfs2_recovery_exit(osb);
1396
1397         ocfs2_journal_shutdown(osb);
1398
1399         ocfs2_sync_blockdev(sb);
1400
1401         /* No cluster connection means we've failed during mount, so skip
1402          * all the steps which depended on that to complete. */
1403         if (osb->cconn) {
1404                 tmp = ocfs2_super_lock(osb, 1);
1405                 if (tmp < 0) {
1406                         mlog_errno(tmp);
1407                         return;
1408                 }
1409         }
1410
1411         if (osb->slot_num != OCFS2_INVALID_SLOT)
1412                 ocfs2_put_slot(osb);
1413
1414         if (osb->cconn)
1415                 ocfs2_super_unlock(osb, 1);
1416
1417         ocfs2_release_system_inodes(osb);
1418
1419         /*
1420          * If we're dismounting due to mount error, mount.ocfs2 will clean
1421          * up heartbeat.  If we're a local mount, there is no heartbeat.
1422          * If we failed before we got a uuid_str yet, we can't stop
1423          * heartbeat.  Otherwise, do it.
1424          */
1425         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1426                 hangup_needed = 1;
1427
1428         if (osb->cconn)
1429                 ocfs2_dlm_shutdown(osb, hangup_needed);
1430
1431         debugfs_remove(osb->osb_debug_root);
1432
1433         if (hangup_needed)
1434                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1435
1436         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1437
1438         if (ocfs2_mount_local(osb))
1439                 snprintf(nodestr, sizeof(nodestr), "local");
1440         else
1441                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1442
1443         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1444                osb->dev_str, nodestr);
1445
1446         ocfs2_delete_osb(osb);
1447         kfree(osb);
1448         sb->s_dev = 0;
1449         sb->s_fs_info = NULL;
1450 }
1451
1452 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1453                                 unsigned uuid_bytes)
1454 {
1455         int i, ret;
1456         char *ptr;
1457
1458         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1459
1460         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1461         if (osb->uuid_str == NULL)
1462                 return -ENOMEM;
1463
1464         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1465                 /* print with null */
1466                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1467                 if (ret != 2) /* drop super cleans up */
1468                         return -EINVAL;
1469                 /* then only advance past the last char */
1470                 ptr += 2;
1471         }
1472
1473         return 0;
1474 }
1475
1476 static int ocfs2_initialize_super(struct super_block *sb,
1477                                   struct buffer_head *bh,
1478                                   int sector_size)
1479 {
1480         int status;
1481         int i, cbits, bbits;
1482         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1483         struct inode *inode = NULL;
1484         struct ocfs2_journal *journal;
1485         __le32 uuid_net_key;
1486         struct ocfs2_super *osb;
1487
1488         mlog_entry_void();
1489
1490         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1491         if (!osb) {
1492                 status = -ENOMEM;
1493                 mlog_errno(status);
1494                 goto bail;
1495         }
1496
1497         sb->s_fs_info = osb;
1498         sb->s_op = &ocfs2_sops;
1499         sb->s_export_op = &ocfs2_export_ops;
1500         sb->s_xattr = ocfs2_xattr_handlers;
1501         sb->s_time_gran = 1;
1502         sb->s_flags |= MS_NOATIME;
1503         /* this is needed to support O_LARGEFILE */
1504         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1505         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1506         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1507
1508         osb->sb = sb;
1509         /* Save off for ocfs2_rw_direct */
1510         osb->s_sectsize_bits = blksize_bits(sector_size);
1511         BUG_ON(!osb->s_sectsize_bits);
1512
1513         spin_lock_init(&osb->dc_task_lock);
1514         init_waitqueue_head(&osb->dc_event);
1515         osb->dc_work_sequence = 0;
1516         osb->dc_wake_sequence = 0;
1517         INIT_LIST_HEAD(&osb->blocked_lock_list);
1518         osb->blocked_lock_count = 0;
1519         spin_lock_init(&osb->osb_lock);
1520         ocfs2_init_inode_steal_slot(osb);
1521
1522         atomic_set(&osb->alloc_stats.moves, 0);
1523         atomic_set(&osb->alloc_stats.local_data, 0);
1524         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1525         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1526         atomic_set(&osb->alloc_stats.bg_extends, 0);
1527
1528         ocfs2_init_node_maps(osb);
1529
1530         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1531                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1532
1533         status = ocfs2_recovery_init(osb);
1534         if (status) {
1535                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1536                 mlog_errno(status);
1537                 goto bail;
1538         }
1539
1540         init_waitqueue_head(&osb->checkpoint_event);
1541         atomic_set(&osb->needs_checkpoint, 0);
1542
1543         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1544
1545         osb->slot_num = OCFS2_INVALID_SLOT;
1546
1547         osb->s_xattr_inline_size = le16_to_cpu(
1548                                         di->id2.i_super.s_xattr_inline_size);
1549
1550         osb->local_alloc_state = OCFS2_LA_UNUSED;
1551         osb->local_alloc_bh = NULL;
1552         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
1553
1554         init_waitqueue_head(&osb->osb_mount_event);
1555
1556         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1557         if (!osb->vol_label) {
1558                 mlog(ML_ERROR, "unable to alloc vol label\n");
1559                 status = -ENOMEM;
1560                 goto bail;
1561         }
1562
1563         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1564         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1565                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1566                      osb->max_slots);
1567                 status = -EINVAL;
1568                 goto bail;
1569         }
1570         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1571
1572         osb->slot_recovery_generations =
1573                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
1574                         GFP_KERNEL);
1575         if (!osb->slot_recovery_generations) {
1576                 status = -ENOMEM;
1577                 mlog_errno(status);
1578                 goto bail;
1579         }
1580
1581         init_waitqueue_head(&osb->osb_wipe_event);
1582         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1583                                         sizeof(*osb->osb_orphan_wipes),
1584                                         GFP_KERNEL);
1585         if (!osb->osb_orphan_wipes) {
1586                 status = -ENOMEM;
1587                 mlog_errno(status);
1588                 goto bail;
1589         }
1590
1591         osb->s_feature_compat =
1592                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1593         osb->s_feature_ro_compat =
1594                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1595         osb->s_feature_incompat =
1596                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1597
1598         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1599                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1600                      "optional features (%x).\n", i);
1601                 status = -EINVAL;
1602                 goto bail;
1603         }
1604         if (!(osb->sb->s_flags & MS_RDONLY) &&
1605             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1606                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1607                      "unsupported optional features (%x).\n", i);
1608                 status = -EINVAL;
1609                 goto bail;
1610         }
1611
1612         if (ocfs2_userspace_stack(osb)) {
1613                 memcpy(osb->osb_cluster_stack,
1614                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
1615                        OCFS2_STACK_LABEL_LEN);
1616                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1617                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
1618                         mlog(ML_ERROR,
1619                              "couldn't mount because of an invalid "
1620                              "cluster stack label (%s) \n",
1621                              osb->osb_cluster_stack);
1622                         status = -EINVAL;
1623                         goto bail;
1624                 }
1625         } else {
1626                 /* The empty string is identical with classic tools that
1627                  * don't know about s_cluster_info. */
1628                 osb->osb_cluster_stack[0] = '\0';
1629         }
1630
1631         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1632
1633         /* FIXME
1634          * This should be done in ocfs2_journal_init(), but unknown
1635          * ordering issues will cause the filesystem to crash.
1636          * If anyone wants to figure out what part of the code
1637          * refers to osb->journal before ocfs2_journal_init() is run,
1638          * be my guest.
1639          */
1640         /* initialize our journal structure */
1641
1642         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1643         if (!journal) {
1644                 mlog(ML_ERROR, "unable to alloc journal\n");
1645                 status = -ENOMEM;
1646                 goto bail;
1647         }
1648         osb->journal = journal;
1649         journal->j_osb = osb;
1650
1651         atomic_set(&journal->j_num_trans, 0);
1652         init_rwsem(&journal->j_trans_barrier);
1653         init_waitqueue_head(&journal->j_checkpointed);
1654         spin_lock_init(&journal->j_lock);
1655         journal->j_trans_id = (unsigned long) 1;
1656         INIT_LIST_HEAD(&journal->j_la_cleanups);
1657         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1658         journal->j_state = OCFS2_JOURNAL_FREE;
1659
1660         /* get some pseudo constants for clustersize bits */
1661         osb->s_clustersize_bits =
1662                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1663         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1664         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1665
1666         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1667             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1668                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1669                      osb->s_clustersize);
1670                 status = -EINVAL;
1671                 goto bail;
1672         }
1673
1674         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1675             > (u32)~0UL) {
1676                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1677                      "what jbd can address in 32 bits.\n");
1678                 status = -EINVAL;
1679                 goto bail;
1680         }
1681
1682         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1683                                  sizeof(di->id2.i_super.s_uuid))) {
1684                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1685                 status = -ENOMEM;
1686                 goto bail;
1687         }
1688
1689         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1690
1691         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1692         osb->vol_label[63] = '\0';
1693         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1694         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1695         osb->first_cluster_group_blkno =
1696                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1697         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1698         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
1699         mlog(0, "vol_label: %s\n", osb->vol_label);
1700         mlog(0, "uuid: %s\n", osb->uuid_str);
1701         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1702              (unsigned long long)osb->root_blkno,
1703              (unsigned long long)osb->system_dir_blkno);
1704
1705         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1706         if (!osb->osb_dlm_debug) {
1707                 status = -ENOMEM;
1708                 mlog_errno(status);
1709                 goto bail;
1710         }
1711
1712         atomic_set(&osb->vol_state, VOLUME_INIT);
1713
1714         /* load root, system_dir, and all global system inodes */
1715         status = ocfs2_init_global_system_inodes(osb);
1716         if (status < 0) {
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         /*
1722          * global bitmap
1723          */
1724         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1725                                             OCFS2_INVALID_SLOT);
1726         if (!inode) {
1727                 status = -EINVAL;
1728                 mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1733         iput(inode);
1734
1735         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1736
1737         status = ocfs2_init_slot_info(osb);
1738         if (status < 0) {
1739                 mlog_errno(status);
1740                 goto bail;
1741         }
1742
1743 bail:
1744         mlog_exit(status);
1745         return status;
1746 }
1747
1748 /*
1749  * will return: -EAGAIN if it is ok to keep searching for superblocks
1750  *              -EINVAL if there is a bad superblock
1751  *              0 on success
1752  */
1753 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1754                                struct buffer_head *bh,
1755                                u32 blksz)
1756 {
1757         int status = -EAGAIN;
1758
1759         mlog_entry_void();
1760
1761         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1762                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1763                 status = -EINVAL;
1764                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1765                         mlog(ML_ERROR, "found superblock with incorrect block "
1766                              "size: found %u, should be %u\n",
1767                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1768                                blksz);
1769                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1770                            OCFS2_MAJOR_REV_LEVEL ||
1771                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1772                            OCFS2_MINOR_REV_LEVEL) {
1773                         mlog(ML_ERROR, "found superblock with bad version: "
1774                              "found %u.%u, should be %u.%u\n",
1775                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1776                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1777                              OCFS2_MAJOR_REV_LEVEL,
1778                              OCFS2_MINOR_REV_LEVEL);
1779                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1780                         mlog(ML_ERROR, "bad block number on superblock: "
1781                              "found %llu, should be %llu\n",
1782                              (unsigned long long)le64_to_cpu(di->i_blkno),
1783                              (unsigned long long)bh->b_blocknr);
1784                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1785                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1786                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1787                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1788                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1789                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1790                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1791                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1792                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1793                         mlog(ML_ERROR,
1794                              "Superblock slots found greater than file system "
1795                              "maximum: found %u, max %u\n",
1796                              le16_to_cpu(di->id2.i_super.s_max_slots),
1797                              OCFS2_MAX_SLOTS);
1798                 } else {
1799                         /* found it! */
1800                         status = 0;
1801                 }
1802         }
1803
1804         mlog_exit(status);
1805         return status;
1806 }
1807
1808 static int ocfs2_check_volume(struct ocfs2_super *osb)
1809 {
1810         int status;
1811         int dirty;
1812         int local;
1813         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1814                                                   * recover
1815                                                   * ourselves. */
1816
1817         mlog_entry_void();
1818
1819         /* Init our journal object. */
1820         status = ocfs2_journal_init(osb->journal, &dirty);
1821         if (status < 0) {
1822                 mlog(ML_ERROR, "Could not initialize journal!\n");
1823                 goto finally;
1824         }
1825
1826         /* If the journal was unmounted cleanly then we don't want to
1827          * recover anything. Otherwise, journal_load will do that
1828          * dirty work for us :) */
1829         if (!dirty) {
1830                 status = ocfs2_journal_wipe(osb->journal, 0);
1831                 if (status < 0) {
1832                         mlog_errno(status);
1833                         goto finally;
1834                 }
1835         } else {
1836                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1837                      "recovering volume.\n");
1838         }
1839
1840         local = ocfs2_mount_local(osb);
1841
1842         /* will play back anything left in the journal. */
1843         status = ocfs2_journal_load(osb->journal, local, dirty);
1844         if (status < 0) {
1845                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
1846                 goto finally;
1847         }
1848
1849         if (dirty) {
1850                 /* recover my local alloc if we didn't unmount cleanly. */
1851                 status = ocfs2_begin_local_alloc_recovery(osb,
1852                                                           osb->slot_num,
1853                                                           &local_alloc);
1854                 if (status < 0) {
1855                         mlog_errno(status);
1856                         goto finally;
1857                 }
1858                 /* we complete the recovery process after we've marked
1859                  * ourselves as mounted. */
1860         }
1861
1862         mlog(0, "Journal loaded.\n");
1863
1864         status = ocfs2_load_local_alloc(osb);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 goto finally;
1868         }
1869
1870         if (dirty) {
1871                 /* Recovery will be completed after we've mounted the
1872                  * rest of the volume. */
1873                 osb->dirty = 1;
1874                 osb->local_alloc_copy = local_alloc;
1875                 local_alloc = NULL;
1876         }
1877
1878         /* go through each journal, trylock it and if you get the
1879          * lock, and it's marked as dirty, set the bit in the recover
1880          * map and launch a recovery thread for it. */
1881         status = ocfs2_mark_dead_nodes(osb);
1882         if (status < 0)
1883                 mlog_errno(status);
1884
1885 finally:
1886         if (local_alloc)
1887                 kfree(local_alloc);
1888
1889         mlog_exit(status);
1890         return status;
1891 }
1892
1893 /*
1894  * The routine gets called from dismount or close whenever a dismount on
1895  * volume is requested and the osb open count becomes 1.
1896  * It will remove the osb from the global list and also free up all the
1897  * initialized resources and fileobject.
1898  */
1899 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1900 {
1901         mlog_entry_void();
1902
1903         /* This function assumes that the caller has the main osb resource */
1904
1905         ocfs2_free_slot_info(osb);
1906
1907         kfree(osb->osb_orphan_wipes);
1908         kfree(osb->slot_recovery_generations);
1909         /* FIXME
1910          * This belongs in journal shutdown, but because we have to
1911          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1912          * we free it here.
1913          */
1914         kfree(osb->journal);
1915         if (osb->local_alloc_copy)
1916                 kfree(osb->local_alloc_copy);
1917         kfree(osb->uuid_str);
1918         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1919         memset(osb, 0, sizeof(struct ocfs2_super));
1920
1921         mlog_exit_void();
1922 }
1923
1924 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1925  * panic(). We do not support continue-on-error operation. */
1926 static void ocfs2_handle_error(struct super_block *sb)
1927 {
1928         struct ocfs2_super *osb = OCFS2_SB(sb);
1929
1930         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1931                 panic("OCFS2: (device %s): panic forced after error\n",
1932                       sb->s_id);
1933
1934         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1935
1936         if (sb->s_flags & MS_RDONLY &&
1937             (ocfs2_is_soft_readonly(osb) ||
1938              ocfs2_is_hard_readonly(osb)))
1939                 return;
1940
1941         printk(KERN_CRIT "File system is now read-only due to the potential "
1942                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1943                "system is unmounted.\n");
1944         sb->s_flags |= MS_RDONLY;
1945         ocfs2_set_ro_flag(osb, 0);
1946 }
1947
1948 static char error_buf[1024];
1949
1950 void __ocfs2_error(struct super_block *sb,
1951                    const char *function,
1952                    const char *fmt, ...)
1953 {
1954         va_list args;
1955
1956         va_start(args, fmt);
1957         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1958         va_end(args);
1959
1960         /* Not using mlog here because we want to show the actual
1961          * function the error came from. */
1962         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1963                sb->s_id, function, error_buf);
1964
1965         ocfs2_handle_error(sb);
1966 }
1967
1968 /* Handle critical errors. This is intentionally more drastic than
1969  * ocfs2_handle_error, so we only use for things like journal errors,
1970  * etc. */
1971 void __ocfs2_abort(struct super_block* sb,
1972                    const char *function,
1973                    const char *fmt, ...)
1974 {
1975         va_list args;
1976
1977         va_start(args, fmt);
1978         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1979         va_end(args);
1980
1981         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1982                sb->s_id, function, error_buf);
1983
1984         /* We don't have the cluster support yet to go straight to
1985          * hard readonly in here. Until then, we want to keep
1986          * ocfs2_abort() so that we can at least mark critical
1987          * errors.
1988          *
1989          * TODO: This should abort the journal and alert other nodes
1990          * that our slot needs recovery. */
1991
1992         /* Force a panic(). This stinks, but it's better than letting
1993          * things continue without having a proper hard readonly
1994          * here. */
1995         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1996         ocfs2_handle_error(sb);
1997 }
1998
1999 module_init(ocfs2_init);
2000 module_exit(ocfs2_exit);