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