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
  68 #include "buffer_head_io.h"
  69
  70 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  71
  72 /* OCFS2 needs to schedule several differnt types of work which
  73  * require cluster locking, disk I/O, recovery waits, etc. Since these
  74  * types of work tend to be heavy we avoid using the kernel events
  75  * workqueue and schedule on our own. */
  76 struct workqueue_struct *ocfs2_wq = NULL;
  77
  78 static struct dentry *ocfs2_debugfs_root = NULL;
  79
  80 MODULE_AUTHOR("Oracle");
  81 MODULE_LICENSE("GPL");
  82
  83 struct mount_options
  84 {
  85         unsigned long   commit_interval;
  86         unsigned long   mount_opt;
  87         unsigned int    atime_quantum;
  88         signed short    slot;
  89         unsigned int    localalloc_opt;
  90         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
  91 };
  92
  93 static int ocfs2_parse_options(struct super_block *sb, char *options,
  94                                struct mount_options *mopt,
  95                                int is_remount);
  96 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
  97 static void ocfs2_put_super(struct super_block *sb);
  98 static int ocfs2_mount_volume(struct super_block *sb);
  99 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 100 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 101 static int ocfs2_initialize_mem_caches(void);
 102 static void ocfs2_free_mem_caches(void);
 103 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 104
 105 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 106
 107 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 108
 109 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 110 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 111 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 112 static int ocfs2_check_volume(struct ocfs2_super *osb);
 113 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 114                                struct buffer_head *bh,
 115                                u32 sectsize);
 116 static int ocfs2_initialize_super(struct super_block *sb,
 117                                   struct buffer_head *bh,
 118                                   int sector_size);
 119 static int ocfs2_get_sector(struct super_block *sb,
 120                             struct buffer_head **bh,
 121                             int block,
 122                             int sect_size);
 123 static void ocfs2_write_super(struct super_block *sb);
 124 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 125 static void ocfs2_destroy_inode(struct inode *inode);
 126
 127 static const struct super_operations ocfs2_sops = {
 128         .statfs         = ocfs2_statfs,
 129         .alloc_inode    = ocfs2_alloc_inode,
 130         .destroy_inode  = ocfs2_destroy_inode,
 131         .drop_inode     = ocfs2_drop_inode,
 132         .clear_inode    = ocfs2_clear_inode,
 133         .delete_inode   = ocfs2_delete_inode,
 134         .sync_fs        = ocfs2_sync_fs,
 135         .write_super    = ocfs2_write_super,
 136         .put_super      = ocfs2_put_super,
 137         .remount_fs     = ocfs2_remount,
 138         .show_options   = ocfs2_show_options,
 139 };
 140
 141 enum {
 142         Opt_barrier,
 143         Opt_err_panic,
 144         Opt_err_ro,
 145         Opt_intr,
 146         Opt_nointr,
 147         Opt_hb_none,
 148         Opt_hb_local,
 149         Opt_data_ordered,
 150         Opt_data_writeback,
 151         Opt_atime_quantum,
 152         Opt_slot,
 153         Opt_commit,
 154         Opt_localalloc,
 155         Opt_localflocks,
 156         Opt_stack,
 157         Opt_err,
 158 };
 159
 160 static const match_table_t tokens = {
 161         {Opt_barrier, "barrier=%u"},
 162         {Opt_err_panic, "errors=panic"},
 163         {Opt_err_ro, "errors=remount-ro"},
 164         {Opt_intr, "intr"},
 165         {Opt_nointr, "nointr"},
 166         {Opt_hb_none, OCFS2_HB_NONE},
 167         {Opt_hb_local, OCFS2_HB_LOCAL},
 168         {Opt_data_ordered, "data=ordered"},
 169         {Opt_data_writeback, "data=writeback"},
 170         {Opt_atime_quantum, "atime_quantum=%u"},
 171         {Opt_slot, "preferred_slot=%u"},
 172         {Opt_commit, "commit=%u"},
 173         {Opt_localalloc, "localalloc=%d"},
 174         {Opt_localflocks, "localflocks"},
 175         {Opt_stack, "cluster_stack=%s"},
 176         {Opt_err, NULL}
 177 };
 178
 179 /*
 180  * write_super and sync_fs ripped right out of ext3.
 181  */
 182 static void ocfs2_write_super(struct super_block *sb)
 183 {
 184         if (mutex_trylock(&sb->s_lock) != 0)
 185                 BUG();
 186         sb->s_dirt = 0;
 187 }
 188
 189 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 190 {
 191         int status;
 192         tid_t target;
 193         struct ocfs2_super *osb = OCFS2_SB(sb);
 194
 195         sb->s_dirt = 0;
 196
 197         if (ocfs2_is_hard_readonly(osb))
 198                 return -EROFS;
 199
 200         if (wait) {
 201                 status = ocfs2_flush_truncate_log(osb);
 202                 if (status < 0)
 203                         mlog_errno(status);
 204         } else {
 205                 ocfs2_schedule_truncate_log_flush(osb, 0);
 206         }
 207
 208         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
 209                 if (wait)
 210                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 211                                         target);
 212         }
 213         return 0;
 214 }
 215
 216 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 217 {
 218         struct inode *new = NULL;
 219         int status = 0;
 220         int i;
 221
 222         mlog_entry_void();
 223
 224         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 225         if (IS_ERR(new)) {
 226                 status = PTR_ERR(new);
 227                 mlog_errno(status);
 228                 goto bail;
 229         }
 230         osb->root_inode = new;
 231
 232         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 233         if (IS_ERR(new)) {
 234                 status = PTR_ERR(new);
 235                 mlog_errno(status);
 236                 goto bail;
 237         }
 238         osb->sys_root_inode = new;
 239
 240         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 241              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 242                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 243                 if (!new) {
 244                         ocfs2_release_system_inodes(osb);
 245                         status = -EINVAL;
 246                         mlog_errno(status);
 247                         /* FIXME: Should ERROR_RO_FS */
 248                         mlog(ML_ERROR, "Unable to load system inode %d, "
 249                              "possibly corrupt fs?", i);
 250                         goto bail;
 251                 }
 252                 // the array now has one ref, so drop this one
 253                 iput(new);
 254         }
 255
 256 bail:
 257         mlog_exit(status);
 258         return status;
 259 }
 260
 261 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 262 {
 263         struct inode *new = NULL;
 264         int status = 0;
 265         int i;
 266
 267         mlog_entry_void();
 268
 269         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 270              i < NUM_SYSTEM_INODES;
 271              i++) {
 272                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 273                 if (!new) {
 274                         ocfs2_release_system_inodes(osb);
 275                         status = -EINVAL;
 276                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 277                              status, i, osb->slot_num);
 278                         goto bail;
 279                 }
 280                 /* the array now has one ref, so drop this one */
 281                 iput(new);
 282         }
 283
 284 bail:
 285         mlog_exit(status);
 286         return status;
 287 }
 288
 289 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 290 {
 291         int i;
 292         struct inode *inode;
 293
 294         mlog_entry_void();
 295
 296         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
 297                 inode = osb->system_inodes[i];
 298                 if (inode) {
 299                         iput(inode);
 300                         osb->system_inodes[i] = NULL;
 301                 }
 302         }
 303
 304         inode = osb->sys_root_inode;
 305         if (inode) {
 306                 iput(inode);
 307                 osb->sys_root_inode = NULL;
 308         }
 309
 310         inode = osb->root_inode;
 311         if (inode) {
 312                 iput(inode);
 313                 osb->root_inode = NULL;
 314         }
 315
 316         mlog_exit(0);
 317 }
 318
 319 /* We're allocating fs objects, use GFP_NOFS */
 320 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 321 {
 322         struct ocfs2_inode_info *oi;
 323
 324         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 325         if (!oi)
 326                 return NULL;
 327
 328         return &oi->vfs_inode;
 329 }
 330
 331 static void ocfs2_destroy_inode(struct inode *inode)
 332 {
 333         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 334 }
 335
 336 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 337                                                 unsigned int cbits)
 338 {
 339         unsigned int bytes = 1 << cbits;
 340         unsigned int trim = bytes;
 341         unsigned int bitshift = 32;
 342
 343         /*
 344          * i_size and all block offsets in ocfs2 are always 64 bits
 345          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 346          * 64 bit platforms, cluster size will be the limiting factor.
 347          */
 348
 349 #if BITS_PER_LONG == 32
 350 # if defined(CONFIG_LBD)
 351         BUILD_BUG_ON(sizeof(sector_t) != 8);
 352         /*
 353          * We might be limited by page cache size.
 354          */
 355         if (bytes > PAGE_CACHE_SIZE) {
 356                 bytes = PAGE_CACHE_SIZE;
 357                 trim = 1;
 358                 /*
 359                  * Shift by 31 here so that we don't get larger than
 360                  * MAX_LFS_FILESIZE
 361                  */
 362                 bitshift = 31;
 363         }
 364 # else
 365         /*
 366          * We are limited by the size of sector_t. Use block size, as
 367          * that's what we expose to the VFS.
 368          */
 369         bytes = 1 << bbits;
 370         trim = 1;
 371         bitshift = 31;
 372 # endif
 373 #endif
 374
 375         /*
 376          * Trim by a whole cluster when we can actually approach the
 377          * on-disk limits. Otherwise we can overflow i_clusters when
 378          * an extent start is at the max offset.
 379          */
 380         return (((unsigned long long)bytes) << bitshift) - trim;
 381 }
 382
 383 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 384 {
 385         int incompat_features;
 386         int ret = 0;
 387         struct mount_options parsed_options;
 388         struct ocfs2_super *osb = OCFS2_SB(sb);
 389
 390         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 391                 ret = -EINVAL;
 392                 goto out;
 393         }
 394
 395         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
 396             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 397                 ret = -EINVAL;
 398                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 399                 goto out;
 400         }
 401
 402         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 403             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 404                 ret = -EINVAL;
 405                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 406                 goto out;
 407         }
 408
 409         /* We're going to/from readonly mode. */
 410         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 411                 /* Lock here so the check of HARD_RO and the potential
 412                  * setting of SOFT_RO is atomic. */
 413                 spin_lock(&osb->osb_lock);
 414                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 415                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 416                         ret = -EROFS;
 417                         goto unlock_osb;
 418                 }
 419
 420                 if (*flags & MS_RDONLY) {
 421                         mlog(0, "Going to ro mode.\n");
 422                         sb->s_flags |= MS_RDONLY;
 423                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 424                 } else {
 425                         mlog(0, "Making ro filesystem writeable.\n");
 426
 427                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 428                                 mlog(ML_ERROR, "Cannot remount RDWR "
 429                                      "filesystem due to previous errors.\n");
 430                                 ret = -EROFS;
 431                                 goto unlock_osb;
 432                         }
 433                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 434                         if (incompat_features) {
 435                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 436                                      "of unsupported optional features "
 437                                      "(%x).\n", incompat_features);
 438                                 ret = -EINVAL;
 439                                 goto unlock_osb;
 440                         }
 441                         sb->s_flags &= ~MS_RDONLY;
 442                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 443                 }
 444 unlock_osb:
 445                 spin_unlock(&osb->osb_lock);
 446         }
 447
 448         if (!ret) {
 449                 /* Only save off the new mount options in case of a successful
 450                  * remount. */
 451                 osb->s_mount_opt = parsed_options.mount_opt;
 452                 osb->s_atime_quantum = parsed_options.atime_quantum;
 453                 osb->preferred_slot = parsed_options.slot;
 454                 if (parsed_options.commit_interval)
 455                         osb->osb_commit_interval = parsed_options.commit_interval;
 456
 457                 if (!ocfs2_is_hard_readonly(osb))
 458                         ocfs2_set_journal_params(osb);
 459         }
 460 out:
 461         return ret;
 462 }
 463
 464 static int ocfs2_sb_probe(struct super_block *sb,
 465                           struct buffer_head **bh,
 466                           int *sector_size)
 467 {
 468         int status, tmpstat;
 469         struct ocfs1_vol_disk_hdr *hdr;
 470         struct ocfs2_dinode *di;
 471         int blksize;
 472
 473         *bh = NULL;
 474
 475         /* may be > 512 */
 476         *sector_size = bdev_hardsect_size(sb->s_bdev);
 477         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 478                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 479                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 480                 status = -EINVAL;
 481                 goto bail;
 482         }
 483
 484         /* Can this really happen? */
 485         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 486                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 487
 488         /* check block zero for old format */
 489         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 490         if (status < 0) {
 491                 mlog_errno(status);
 492                 goto bail;
 493         }
 494         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 495         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 496                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 497                      hdr->major_version, hdr->minor_version);
 498                 status = -EINVAL;
 499         }
 500         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 501                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 502                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 503                      hdr->signature);
 504                 status = -EINVAL;
 505         }
 506         brelse(*bh);
 507         *bh = NULL;
 508         if (status < 0) {
 509                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 510                      "upgraded before mounting with ocfs v2\n");
 511                 goto bail;
 512         }
 513
 514         /*
 515          * Now check at magic offset for 512, 1024, 2048, 4096
 516          * blocksizes.  4096 is the maximum blocksize because it is
 517          * the minimum clustersize.
 518          */
 519         status = -EINVAL;
 520         for (blksize = *sector_size;
 521              blksize <= OCFS2_MAX_BLOCKSIZE;
 522              blksize <<= 1) {
 523                 tmpstat = ocfs2_get_sector(sb, bh,
 524                                            OCFS2_SUPER_BLOCK_BLKNO,
 525                                            blksize);
 526                 if (tmpstat < 0) {
 527                         status = tmpstat;
 528                         mlog_errno(status);
 529                         goto bail;
 530                 }
 531                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 532                 status = ocfs2_verify_volume(di, *bh, blksize);
 533                 if (status >= 0)
 534                         goto bail;
 535                 brelse(*bh);
 536                 *bh = NULL;
 537                 if (status != -EAGAIN)
 538                         break;
 539         }
 540
 541 bail:
 542         return status;
 543 }
 544
 545 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 546 {
 547         if (ocfs2_mount_local(osb)) {
 548                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 549                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 550                              "mounted device.\n");
 551                         return -EINVAL;
 552                 }
 553         }
 554
 555         if (ocfs2_userspace_stack(osb)) {
 556                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 557                         mlog(ML_ERROR, "Userspace stack expected, but "
 558                              "o2cb heartbeat arguments passed to mount\n");
 559                         return -EINVAL;
 560                 }
 561         }
 562
 563         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 564                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 565                     !ocfs2_userspace_stack(osb)) {
 566                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 567                              "a read-write clustered device.\n");
 568                         return -EINVAL;
 569                 }
 570         }
 571
 572         return 0;
 573 }
 574
 575 /*
 576  * If we're using a userspace stack, mount should have passed
 577  * a name that matches the disk.  If not, mount should not
 578  * have passed a stack.
 579  */
 580 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 581                                         struct mount_options *mopt)
 582 {
 583         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 584                 mlog(ML_ERROR,
 585                      "cluster stack passed to mount, but this filesystem "
 586                      "does not support it\n");
 587                 return -EINVAL;
 588         }
 589
 590         if (ocfs2_userspace_stack(osb) &&
 591             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 592                     OCFS2_STACK_LABEL_LEN)) {
 593                 mlog(ML_ERROR,
 594                      "cluster stack passed to mount (\"%s\") does not "
 595                      "match the filesystem (\"%s\")\n",
 596                      mopt->cluster_stack,
 597                      osb->osb_cluster_stack);
 598                 return -EINVAL;
 599         }
 600
 601         return 0;
 602 }
 603
 604 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 605 {
 606         struct dentry *root;
 607         int status, sector_size;
 608         struct mount_options parsed_options;
 609         struct inode *inode = NULL;
 610         struct ocfs2_super *osb = NULL;
 611         struct buffer_head *bh = NULL;
 612         char nodestr[8];
 613
 614         mlog_entry("%p, %p, %i", sb, data, silent);
 615
 616         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 617                 status = -EINVAL;
 618                 goto read_super_error;
 619         }
 620
 621         /* probe for superblock */
 622         status = ocfs2_sb_probe(sb, &bh, &sector_size);
 623         if (status < 0) {
 624                 mlog(ML_ERROR, "superblock probe failed!\n");
 625                 goto read_super_error;
 626         }
 627
 628         status = ocfs2_initialize_super(sb, bh, sector_size);
 629         osb = OCFS2_SB(sb);
 630         if (status < 0) {
 631                 mlog_errno(status);
 632                 goto read_super_error;
 633         }
 634         brelse(bh);
 635         bh = NULL;
 636         osb->s_mount_opt = parsed_options.mount_opt;
 637         osb->s_atime_quantum = parsed_options.atime_quantum;
 638         osb->preferred_slot = parsed_options.slot;
 639         osb->osb_commit_interval = parsed_options.commit_interval;
 640         osb->local_alloc_default_bits = ocfs2_megabytes_to_clusters(sb, parsed_options.localalloc_opt);
 641         osb->local_alloc_bits = osb->local_alloc_default_bits;
 642
 643         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
 644         if (status)
 645                 goto read_super_error;
 646
 647         sb->s_magic = OCFS2_SUPER_MAGIC;
 648
 649         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
 650          * heartbeat=none */
 651         if (bdev_read_only(sb->s_bdev)) {
 652                 if (!(sb->s_flags & MS_RDONLY)) {
 653                         status = -EACCES;
 654                         mlog(ML_ERROR, "Readonly device detected but readonly "
 655                              "mount was not specified.\n");
 656                         goto read_super_error;
 657                 }
 658
 659                 /* You should not be able to start a local heartbeat
 660                  * on a readonly device. */
 661                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 662                         status = -EROFS;
 663                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
 664                              "device.\n");
 665                         goto read_super_error;
 666                 }
 667
 668                 status = ocfs2_check_journals_nolocks(osb);
 669                 if (status < 0) {
 670                         if (status == -EROFS)
 671                                 mlog(ML_ERROR, "Recovery required on readonly "
 672                                      "file system, but write access is "
 673                                      "unavailable.\n");
 674                         else
 675                                 mlog_errno(status);
 676                         goto read_super_error;
 677                 }
 678
 679                 ocfs2_set_ro_flag(osb, 1);
 680
 681                 printk(KERN_NOTICE "Readonly device detected. No cluster "
 682                        "services will be utilized for this mount. Recovery "
 683                        "will be skipped.\n");
 684         }
 685
 686         if (!ocfs2_is_hard_readonly(osb)) {
 687                 if (sb->s_flags & MS_RDONLY)
 688                         ocfs2_set_ro_flag(osb, 0);
 689         }
 690
 691         status = ocfs2_verify_heartbeat(osb);
 692         if (status < 0) {
 693                 mlog_errno(status);
 694                 goto read_super_error;
 695         }
 696
 697         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
 698                                                  ocfs2_debugfs_root);
 699         if (!osb->osb_debug_root) {
 700                 status = -EINVAL;
 701                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
 702                 goto read_super_error;
 703         }
 704
 705         status = ocfs2_mount_volume(sb);
 706         if (osb->root_inode)
 707                 inode = igrab(osb->root_inode);
 708
 709         if (status < 0)
 710                 goto read_super_error;
 711
 712         if (!inode) {
 713                 status = -EIO;
 714                 mlog_errno(status);
 715                 goto read_super_error;
 716         }
 717
 718         root = d_alloc_root(inode);
 719         if (!root) {
 720                 status = -ENOMEM;
 721                 mlog_errno(status);
 722                 goto read_super_error;
 723         }
 724
 725         sb->s_root = root;
 726
 727         ocfs2_complete_mount_recovery(osb);
 728
 729         if (ocfs2_mount_local(osb))
 730                 snprintf(nodestr, sizeof(nodestr), "local");
 731         else
 732                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
 733
 734         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
 735                "with %s data mode.\n",
 736                osb->dev_str, nodestr, osb->slot_num,
 737                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
 738                "ordered");
 739
 740         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
 741         wake_up(&osb->osb_mount_event);
 742
 743         mlog_exit(status);
 744         return status;
 745
 746 read_super_error:
 747         if (bh != NULL)
 748                 brelse(bh);
 749
 750         if (inode)
 751                 iput(inode);
 752
 753         if (osb) {
 754                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
 755                 wake_up(&osb->osb_mount_event);
 756                 ocfs2_dismount_volume(sb, 1);
 757         }
 758
 759         mlog_exit(status);
 760         return status;
 761 }
 762
 763 static int ocfs2_get_sb(struct file_system_type *fs_type,
 764                         int flags,
 765                         const char *dev_name,
 766                         void *data,
 767                         struct vfsmount *mnt)
 768 {
 769         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
 770                            mnt);
 771 }
 772
 773 static struct file_system_type ocfs2_fs_type = {
 774         .owner          = THIS_MODULE,
 775         .name           = "ocfs2",
 776         .get_sb         = ocfs2_get_sb, /* is this called when we mount
 777                                         * the fs? */
 778         .kill_sb        = kill_block_super, /* set to the generic one
 779                                              * right now, but do we
 780                                              * need to change that? */
 781         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 782         .next           = NULL
 783 };
 784
 785 static int ocfs2_parse_options(struct super_block *sb,
 786                                char *options,
 787                                struct mount_options *mopt,
 788                                int is_remount)
 789 {
 790         int status;
 791         char *p;
 792
 793         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 794                    options ? options : "(none)");
 795
 796         mopt->commit_interval = 0;
 797         mopt->mount_opt = 0;
 798         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 799         mopt->slot = OCFS2_INVALID_SLOT;
 800         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
 801         mopt->cluster_stack[0] = '\0';
 802
 803         if (!options) {
 804                 status = 1;
 805                 goto bail;
 806         }
 807
 808         while ((p = strsep(&options, ",")) != NULL) {
 809                 int token, option;
 810                 substring_t args[MAX_OPT_ARGS];
 811
 812                 if (!*p)
 813                         continue;
 814
 815                 token = match_token(p, tokens, args);
 816                 switch (token) {
 817                 case Opt_hb_local:
 818                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 819                         break;
 820                 case Opt_hb_none:
 821                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 822                         break;
 823                 case Opt_barrier:
 824                         if (match_int(&args[0], &option)) {
 825                                 status = 0;
 826                                 goto bail;
 827                         }
 828                         if (option)
 829                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
 830                         else
 831                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
 832                         break;
 833                 case Opt_intr:
 834                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
 835                         break;
 836                 case Opt_nointr:
 837                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
 838                         break;
 839                 case Opt_err_panic:
 840                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 841                         break;
 842                 case Opt_err_ro:
 843                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 844                         break;
 845                 case Opt_data_ordered:
 846                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 847                         break;
 848                 case Opt_data_writeback:
 849                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 850                         break;
 851                 case Opt_atime_quantum:
 852                         if (match_int(&args[0], &option)) {
 853                                 status = 0;
 854                                 goto bail;
 855                         }
 856                         if (option >= 0)
 857                                 mopt->atime_quantum = option;
 858                         break;
 859                 case Opt_slot:
 860                         option = 0;
 861                         if (match_int(&args[0], &option)) {
 862                                 status = 0;
 863                                 goto bail;
 864                         }
 865                         if (option)
 866                                 mopt->slot = (s16)option;
 867                         break;
 868                 case Opt_commit:
 869                         option = 0;
 870                         if (match_int(&args[0], &option)) {
 871                                 status = 0;
 872                                 goto bail;
 873                         }
 874                         if (option < 0)
 875                                 return 0;
 876                         if (option == 0)
 877                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 878                         mopt->commit_interval = HZ * option;
 879                         break;
 880                 case Opt_localalloc:
 881                         option = 0;
 882                         if (match_int(&args[0], &option)) {
 883                                 status = 0;
 884                                 goto bail;
 885                         }
 886                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
 887                                 mopt->localalloc_opt = option;
 888                         break;
 889                 case Opt_localflocks:
 890                         /*
 891                          * Changing this during remount could race
 892                          * flock() requests, or "unbalance" existing
 893                          * ones (e.g., a lock is taken in one mode but
 894                          * dropped in the other). If users care enough
 895                          * to flip locking modes during remount, we
 896                          * could add a "local" flag to individual
 897                          * flock structures for proper tracking of
 898                          * state.
 899                          */
 900                         if (!is_remount)
 901                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
 902                         break;
 903                 case Opt_stack:
 904                         /* Check both that the option we were passed
 905                          * is of the right length and that it is a proper
 906                          * string of the right length.
 907                          */
 908                         if (((args[0].to - args[0].from) !=
 909                              OCFS2_STACK_LABEL_LEN) ||
 910                             (strnlen(args[0].from,
 911                                      OCFS2_STACK_LABEL_LEN) !=
 912                              OCFS2_STACK_LABEL_LEN)) {
 913                                 mlog(ML_ERROR,
 914                                      "Invalid cluster_stack option\n");
 915                                 status = 0;
 916                                 goto bail;
 917                         }
 918                         memcpy(mopt->cluster_stack, args[0].from,
 919                                OCFS2_STACK_LABEL_LEN);
 920                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
 921                         break;
 922                 default:
 923                         mlog(ML_ERROR,
 924                              "Unrecognized mount option \"%s\" "
 925                              "or missing value\n", p);
 926                         status = 0;
 927                         goto bail;
 928                 }
 929         }
 930
 931         status = 1;
 932
 933 bail:
 934         mlog_exit(status);
 935         return status;
 936 }
 937
 938 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 939 {
 940         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
 941         unsigned long opts = osb->s_mount_opt;
 942         unsigned int local_alloc_megs;
 943
 944         if (opts & OCFS2_MOUNT_HB_LOCAL)
 945                 seq_printf(s, ",_netdev,heartbeat=local");
 946         else
 947                 seq_printf(s, ",heartbeat=none");
 948
 949         if (opts & OCFS2_MOUNT_NOINTR)
 950                 seq_printf(s, ",nointr");
 951
 952         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
 953                 seq_printf(s, ",data=writeback");
 954         else
 955                 seq_printf(s, ",data=ordered");
 956
 957         if (opts & OCFS2_MOUNT_BARRIER)
 958                 seq_printf(s, ",barrier=1");
 959
 960         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
 961                 seq_printf(s, ",errors=panic");
 962         else
 963                 seq_printf(s, ",errors=remount-ro");
 964
 965         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
 966                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
 967
 968         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
 969                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
 970
 971         if (osb->osb_commit_interval)
 972                 seq_printf(s, ",commit=%u",
 973                            (unsigned) (osb->osb_commit_interval / HZ));
 974
 975         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
 976         if (local_alloc_megs != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
 977                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
 978
 979         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
 980                 seq_printf(s, ",localflocks,");
 981
 982         if (osb->osb_cluster_stack[0])
 983                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
 984                            osb->osb_cluster_stack);
 985
 986         return 0;
 987 }
 988
 989 static int __init ocfs2_init(void)
 990 {
 991         int status;
 992
 993         mlog_entry_void();
 994
 995         ocfs2_print_version();
 996
 997         status = init_ocfs2_uptodate_cache();
 998         if (status < 0) {
 999                 mlog_errno(status);
1000                 goto leave;
1001         }
1002
1003         status = ocfs2_initialize_mem_caches();
1004         if (status < 0) {
1005                 mlog_errno(status);
1006                 goto leave;
1007         }
1008
1009         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1010         if (!ocfs2_wq) {
1011                 status = -ENOMEM;
1012                 goto leave;
1013         }
1014
1015         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1016         if (!ocfs2_debugfs_root) {
1017                 status = -EFAULT;
1018                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1019         }
1020
1021         ocfs2_set_locking_protocol();
1022
1023 leave:
1024         if (status < 0) {
1025                 ocfs2_free_mem_caches();
1026                 exit_ocfs2_uptodate_cache();
1027         }
1028
1029         mlog_exit(status);
1030
1031         if (status >= 0) {
1032                 return register_filesystem(&ocfs2_fs_type);
1033         } else
1034                 return -1;
1035 }
1036
1037 static void __exit ocfs2_exit(void)
1038 {
1039         mlog_entry_void();
1040
1041         if (ocfs2_wq) {
1042                 flush_workqueue(ocfs2_wq);
1043                 destroy_workqueue(ocfs2_wq);
1044         }
1045
1046         debugfs_remove(ocfs2_debugfs_root);
1047
1048         ocfs2_free_mem_caches();
1049
1050         unregister_filesystem(&ocfs2_fs_type);
1051
1052         exit_ocfs2_uptodate_cache();
1053
1054         mlog_exit_void();
1055 }
1056
1057 static void ocfs2_put_super(struct super_block *sb)
1058 {
1059         mlog_entry("(0x%p)\n", sb);
1060
1061         ocfs2_sync_blockdev(sb);
1062         ocfs2_dismount_volume(sb, 0);
1063
1064         mlog_exit_void();
1065 }
1066
1067 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1068 {
1069         struct ocfs2_super *osb;
1070         u32 numbits, freebits;
1071         int status;
1072         struct ocfs2_dinode *bm_lock;
1073         struct buffer_head *bh = NULL;
1074         struct inode *inode = NULL;
1075
1076         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1077
1078         osb = OCFS2_SB(dentry->d_sb);
1079
1080         inode = ocfs2_get_system_file_inode(osb,
1081                                             GLOBAL_BITMAP_SYSTEM_INODE,
1082                                             OCFS2_INVALID_SLOT);
1083         if (!inode) {
1084                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1085                 status = -EIO;
1086                 goto bail;
1087         }
1088
1089         status = ocfs2_inode_lock(inode, &bh, 0);
1090         if (status < 0) {
1091                 mlog_errno(status);
1092                 goto bail;
1093         }
1094
1095         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1096
1097         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1098         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1099
1100         buf->f_type = OCFS2_SUPER_MAGIC;
1101         buf->f_bsize = dentry->d_sb->s_blocksize;
1102         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1103         buf->f_blocks = ((sector_t) numbits) *
1104                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1105         buf->f_bfree = ((sector_t) freebits) *
1106                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1107         buf->f_bavail = buf->f_bfree;
1108         buf->f_files = numbits;
1109         buf->f_ffree = freebits;
1110
1111         brelse(bh);
1112
1113         ocfs2_inode_unlock(inode, 0);
1114         status = 0;
1115 bail:
1116         if (inode)
1117                 iput(inode);
1118
1119         mlog_exit(status);
1120
1121         return status;
1122 }
1123
1124 static void ocfs2_inode_init_once(void *data)
1125 {
1126         struct ocfs2_inode_info *oi = data;
1127
1128         oi->ip_flags = 0;
1129         oi->ip_open_count = 0;
1130         spin_lock_init(&oi->ip_lock);
1131         ocfs2_extent_map_init(&oi->vfs_inode);
1132         INIT_LIST_HEAD(&oi->ip_io_markers);
1133         oi->ip_created_trans = 0;
1134         oi->ip_last_trans = 0;
1135         oi->ip_dir_start_lookup = 0;
1136
1137         init_rwsem(&oi->ip_alloc_sem);
1138         mutex_init(&oi->ip_io_mutex);
1139
1140         oi->ip_blkno = 0ULL;
1141         oi->ip_clusters = 0;
1142
1143         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1144         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1145         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1146
1147         ocfs2_metadata_cache_init(&oi->vfs_inode);
1148
1149         inode_init_once(&oi->vfs_inode);
1150 }
1151
1152 static int ocfs2_initialize_mem_caches(void)
1153 {
1154         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1155                                        sizeof(struct ocfs2_inode_info),
1156                                        0,
1157                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1158                                                 SLAB_MEM_SPREAD),
1159                                        ocfs2_inode_init_once);
1160         if (!ocfs2_inode_cachep)
1161                 return -ENOMEM;
1162
1163         return 0;
1164 }
1165
1166 static void ocfs2_free_mem_caches(void)
1167 {
1168         if (ocfs2_inode_cachep)
1169                 kmem_cache_destroy(ocfs2_inode_cachep);
1170
1171         ocfs2_inode_cachep = NULL;
1172 }
1173
1174 static int ocfs2_get_sector(struct super_block *sb,
1175                             struct buffer_head **bh,
1176                             int block,
1177                             int sect_size)
1178 {
1179         if (!sb_set_blocksize(sb, sect_size)) {
1180                 mlog(ML_ERROR, "unable to set blocksize\n");
1181                 return -EIO;
1182         }
1183
1184         *bh = sb_getblk(sb, block);
1185         if (!*bh) {
1186                 mlog_errno(-EIO);
1187                 return -EIO;
1188         }
1189         lock_buffer(*bh);
1190         if (!buffer_dirty(*bh))
1191                 clear_buffer_uptodate(*bh);
1192         unlock_buffer(*bh);
1193         ll_rw_block(READ, 1, bh);
1194         wait_on_buffer(*bh);
1195         return 0;
1196 }
1197
1198 static int ocfs2_mount_volume(struct super_block *sb)
1199 {
1200         int status = 0;
1201         int unlock_super = 0;
1202         struct ocfs2_super *osb = OCFS2_SB(sb);
1203
1204         mlog_entry_void();
1205
1206         if (ocfs2_is_hard_readonly(osb))
1207                 goto leave;
1208
1209         status = ocfs2_dlm_init(osb);
1210         if (status < 0) {
1211                 mlog_errno(status);
1212                 goto leave;
1213         }
1214
1215         status = ocfs2_super_lock(osb, 1);
1216         if (status < 0) {
1217                 mlog_errno(status);
1218                 goto leave;
1219         }
1220         unlock_super = 1;
1221
1222         /* This will load up the node map and add ourselves to it. */
1223         status = ocfs2_find_slot(osb);
1224         if (status < 0) {
1225                 mlog_errno(status);
1226                 goto leave;
1227         }
1228
1229         /* load all node-local system inodes */
1230         status = ocfs2_init_local_system_inodes(osb);
1231         if (status < 0) {
1232                 mlog_errno(status);
1233                 goto leave;
1234         }
1235
1236         status = ocfs2_check_volume(osb);
1237         if (status < 0) {
1238                 mlog_errno(status);
1239                 goto leave;
1240         }
1241
1242         status = ocfs2_truncate_log_init(osb);
1243         if (status < 0) {
1244                 mlog_errno(status);
1245                 goto leave;
1246         }
1247
1248         if (ocfs2_mount_local(osb))
1249                 goto leave;
1250
1251 leave:
1252         if (unlock_super)
1253                 ocfs2_super_unlock(osb, 1);
1254
1255         mlog_exit(status);
1256         return status;
1257 }
1258
1259 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1260 {
1261         int tmp, hangup_needed = 0;
1262         struct ocfs2_super *osb = NULL;
1263         char nodestr[8];
1264
1265         mlog_entry("(0x%p)\n", sb);
1266
1267         BUG_ON(!sb);
1268         osb = OCFS2_SB(sb);
1269         BUG_ON(!osb);
1270
1271         ocfs2_shutdown_local_alloc(osb);
1272
1273         ocfs2_truncate_log_shutdown(osb);
1274
1275         /* This will disable recovery and flush any recovery work. */
1276         ocfs2_recovery_exit(osb);
1277
1278         ocfs2_journal_shutdown(osb);
1279
1280         ocfs2_sync_blockdev(sb);
1281
1282         /* No cluster connection means we've failed during mount, so skip
1283          * all the steps which depended on that to complete. */
1284         if (osb->cconn) {
1285                 tmp = ocfs2_super_lock(osb, 1);
1286                 if (tmp < 0) {
1287                         mlog_errno(tmp);
1288                         return;
1289                 }
1290         }
1291
1292         if (osb->slot_num != OCFS2_INVALID_SLOT)
1293                 ocfs2_put_slot(osb);
1294
1295         if (osb->cconn)
1296                 ocfs2_super_unlock(osb, 1);
1297
1298         ocfs2_release_system_inodes(osb);
1299
1300         /*
1301          * If we're dismounting due to mount error, mount.ocfs2 will clean
1302          * up heartbeat.  If we're a local mount, there is no heartbeat.
1303          * If we failed before we got a uuid_str yet, we can't stop
1304          * heartbeat.  Otherwise, do it.
1305          */
1306         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1307                 hangup_needed = 1;
1308
1309         if (osb->cconn)
1310                 ocfs2_dlm_shutdown(osb, hangup_needed);
1311
1312         debugfs_remove(osb->osb_debug_root);
1313
1314         if (hangup_needed)
1315                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1316
1317         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1318
1319         if (ocfs2_mount_local(osb))
1320                 snprintf(nodestr, sizeof(nodestr), "local");
1321         else
1322                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1323
1324         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1325                osb->dev_str, nodestr);
1326
1327         ocfs2_delete_osb(osb);
1328         kfree(osb);
1329         sb->s_dev = 0;
1330         sb->s_fs_info = NULL;
1331 }
1332
1333 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1334                                 unsigned uuid_bytes)
1335 {
1336         int i, ret;
1337         char *ptr;
1338
1339         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1340
1341         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1342         if (osb->uuid_str == NULL)
1343                 return -ENOMEM;
1344
1345         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1346                 /* print with null */
1347                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1348                 if (ret != 2) /* drop super cleans up */
1349                         return -EINVAL;
1350                 /* then only advance past the last char */
1351                 ptr += 2;
1352         }
1353
1354         return 0;
1355 }
1356
1357 static int ocfs2_initialize_super(struct super_block *sb,
1358                                   struct buffer_head *bh,
1359                                   int sector_size)
1360 {
1361         int status;
1362         int i, cbits, bbits;
1363         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1364         struct inode *inode = NULL;
1365         struct ocfs2_journal *journal;
1366         __le32 uuid_net_key;
1367         struct ocfs2_super *osb;
1368
1369         mlog_entry_void();
1370
1371         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1372         if (!osb) {
1373                 status = -ENOMEM;
1374                 mlog_errno(status);
1375                 goto bail;
1376         }
1377
1378         sb->s_fs_info = osb;
1379         sb->s_op = &ocfs2_sops;
1380         sb->s_export_op = &ocfs2_export_ops;
1381         sb->s_time_gran = 1;
1382         sb->s_flags |= MS_NOATIME;
1383         /* this is needed to support O_LARGEFILE */
1384         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1385         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1386         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1387
1388         osb->sb = sb;
1389         /* Save off for ocfs2_rw_direct */
1390         osb->s_sectsize_bits = blksize_bits(sector_size);
1391         BUG_ON(!osb->s_sectsize_bits);
1392
1393         spin_lock_init(&osb->dc_task_lock);
1394         init_waitqueue_head(&osb->dc_event);
1395         osb->dc_work_sequence = 0;
1396         osb->dc_wake_sequence = 0;
1397         INIT_LIST_HEAD(&osb->blocked_lock_list);
1398         osb->blocked_lock_count = 0;
1399         spin_lock_init(&osb->osb_lock);
1400         ocfs2_init_inode_steal_slot(osb);
1401
1402         atomic_set(&osb->alloc_stats.moves, 0);
1403         atomic_set(&osb->alloc_stats.local_data, 0);
1404         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1405         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1406         atomic_set(&osb->alloc_stats.bg_extends, 0);
1407
1408         ocfs2_init_node_maps(osb);
1409
1410         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1411                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1412
1413         status = ocfs2_recovery_init(osb);
1414         if (status) {
1415                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1416                 mlog_errno(status);
1417                 goto bail;
1418         }
1419
1420         init_waitqueue_head(&osb->checkpoint_event);
1421         atomic_set(&osb->needs_checkpoint, 0);
1422
1423         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1424
1425         osb->slot_num = OCFS2_INVALID_SLOT;
1426
1427         osb->s_xattr_inline_size = OCFS2_MIN_XATTR_INLINE_SIZE;
1428
1429         osb->local_alloc_state = OCFS2_LA_UNUSED;
1430         osb->local_alloc_bh = NULL;
1431         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
1432
1433         init_waitqueue_head(&osb->osb_mount_event);
1434
1435         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1436         if (!osb->vol_label) {
1437                 mlog(ML_ERROR, "unable to alloc vol label\n");
1438                 status = -ENOMEM;
1439                 goto bail;
1440         }
1441
1442         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1443         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1444                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1445                      osb->max_slots);
1446                 status = -EINVAL;
1447                 goto bail;
1448         }
1449         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1450
1451         osb->slot_recovery_generations =
1452                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
1453                         GFP_KERNEL);
1454         if (!osb->slot_recovery_generations) {
1455                 status = -ENOMEM;
1456                 mlog_errno(status);
1457                 goto bail;
1458         }
1459
1460         init_waitqueue_head(&osb->osb_wipe_event);
1461         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1462                                         sizeof(*osb->osb_orphan_wipes),
1463                                         GFP_KERNEL);
1464         if (!osb->osb_orphan_wipes) {
1465                 status = -ENOMEM;
1466                 mlog_errno(status);
1467                 goto bail;
1468         }
1469
1470         osb->s_feature_compat =
1471                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1472         osb->s_feature_ro_compat =
1473                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1474         osb->s_feature_incompat =
1475                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1476
1477         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1478                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1479                      "optional features (%x).\n", i);
1480                 status = -EINVAL;
1481                 goto bail;
1482         }
1483         if (!(osb->sb->s_flags & MS_RDONLY) &&
1484             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1485                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1486                      "unsupported optional features (%x).\n", i);
1487                 status = -EINVAL;
1488                 goto bail;
1489         }
1490
1491         if (ocfs2_userspace_stack(osb)) {
1492                 memcpy(osb->osb_cluster_stack,
1493                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
1494                        OCFS2_STACK_LABEL_LEN);
1495                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1496                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
1497                         mlog(ML_ERROR,
1498                              "couldn't mount because of an invalid "
1499                              "cluster stack label (%s) \n",
1500                              osb->osb_cluster_stack);
1501                         status = -EINVAL;
1502                         goto bail;
1503                 }
1504         } else {
1505                 /* The empty string is identical with classic tools that
1506                  * don't know about s_cluster_info. */
1507                 osb->osb_cluster_stack[0] = '\0';
1508         }
1509
1510         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1511
1512         /* FIXME
1513          * This should be done in ocfs2_journal_init(), but unknown
1514          * ordering issues will cause the filesystem to crash.
1515          * If anyone wants to figure out what part of the code
1516          * refers to osb->journal before ocfs2_journal_init() is run,
1517          * be my guest.
1518          */
1519         /* initialize our journal structure */
1520
1521         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1522         if (!journal) {
1523                 mlog(ML_ERROR, "unable to alloc journal\n");
1524                 status = -ENOMEM;
1525                 goto bail;
1526         }
1527         osb->journal = journal;
1528         journal->j_osb = osb;
1529
1530         atomic_set(&journal->j_num_trans, 0);
1531         init_rwsem(&journal->j_trans_barrier);
1532         init_waitqueue_head(&journal->j_checkpointed);
1533         spin_lock_init(&journal->j_lock);
1534         journal->j_trans_id = (unsigned long) 1;
1535         INIT_LIST_HEAD(&journal->j_la_cleanups);
1536         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1537         journal->j_state = OCFS2_JOURNAL_FREE;
1538
1539         /* get some pseudo constants for clustersize bits */
1540         osb->s_clustersize_bits =
1541                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1542         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1543         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1544
1545         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1546             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1547                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1548                      osb->s_clustersize);
1549                 status = -EINVAL;
1550                 goto bail;
1551         }
1552
1553         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1554             > (u32)~0UL) {
1555                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1556                      "what jbd can address in 32 bits.\n");
1557                 status = -EINVAL;
1558                 goto bail;
1559         }
1560
1561         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1562                                  sizeof(di->id2.i_super.s_uuid))) {
1563                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1564                 status = -ENOMEM;
1565                 goto bail;
1566         }
1567
1568         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1569
1570         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1571         osb->vol_label[63] = '\0';
1572         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1573         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1574         osb->first_cluster_group_blkno =
1575                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1576         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1577         mlog(0, "vol_label: %s\n", osb->vol_label);
1578         mlog(0, "uuid: %s\n", osb->uuid_str);
1579         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1580              (unsigned long long)osb->root_blkno,
1581              (unsigned long long)osb->system_dir_blkno);
1582
1583         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1584         if (!osb->osb_dlm_debug) {
1585                 status = -ENOMEM;
1586                 mlog_errno(status);
1587                 goto bail;
1588         }
1589
1590         atomic_set(&osb->vol_state, VOLUME_INIT);
1591
1592         /* load root, system_dir, and all global system inodes */
1593         status = ocfs2_init_global_system_inodes(osb);
1594         if (status < 0) {
1595                 mlog_errno(status);
1596                 goto bail;
1597         }
1598
1599         /*
1600          * global bitmap
1601          */
1602         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1603                                             OCFS2_INVALID_SLOT);
1604         if (!inode) {
1605                 status = -EINVAL;
1606                 mlog_errno(status);
1607                 goto bail;
1608         }
1609
1610         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1611         iput(inode);
1612
1613         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1614
1615         status = ocfs2_init_slot_info(osb);
1616         if (status < 0) {
1617                 mlog_errno(status);
1618                 goto bail;
1619         }
1620
1621 bail:
1622         mlog_exit(status);
1623         return status;
1624 }
1625
1626 /*
1627  * will return: -EAGAIN if it is ok to keep searching for superblocks
1628  *              -EINVAL if there is a bad superblock
1629  *              0 on success
1630  */
1631 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1632                                struct buffer_head *bh,
1633                                u32 blksz)
1634 {
1635         int status = -EAGAIN;
1636
1637         mlog_entry_void();
1638
1639         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1640                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1641                 status = -EINVAL;
1642                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1643                         mlog(ML_ERROR, "found superblock with incorrect block "
1644                              "size: found %u, should be %u\n",
1645                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1646                                blksz);
1647                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1648                            OCFS2_MAJOR_REV_LEVEL ||
1649                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1650                            OCFS2_MINOR_REV_LEVEL) {
1651                         mlog(ML_ERROR, "found superblock with bad version: "
1652                              "found %u.%u, should be %u.%u\n",
1653                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1654                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1655                              OCFS2_MAJOR_REV_LEVEL,
1656                              OCFS2_MINOR_REV_LEVEL);
1657                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1658                         mlog(ML_ERROR, "bad block number on superblock: "
1659                              "found %llu, should be %llu\n",
1660                              (unsigned long long)le64_to_cpu(di->i_blkno),
1661                              (unsigned long long)bh->b_blocknr);
1662                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1663                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1664                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1665                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1666                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1667                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1668                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1669                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1670                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1671                         mlog(ML_ERROR,
1672                              "Superblock slots found greater than file system "
1673                              "maximum: found %u, max %u\n",
1674                              le16_to_cpu(di->id2.i_super.s_max_slots),
1675                              OCFS2_MAX_SLOTS);
1676                 } else {
1677                         /* found it! */
1678                         status = 0;
1679                 }
1680         }
1681
1682         mlog_exit(status);
1683         return status;
1684 }
1685
1686 static int ocfs2_check_volume(struct ocfs2_super *osb)
1687 {
1688         int status;
1689         int dirty;
1690         int local;
1691         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1692                                                   * recover
1693                                                   * ourselves. */
1694
1695         mlog_entry_void();
1696
1697         /* Init our journal object. */
1698         status = ocfs2_journal_init(osb->journal, &dirty);
1699         if (status < 0) {
1700                 mlog(ML_ERROR, "Could not initialize journal!\n");
1701                 goto finally;
1702         }
1703
1704         /* If the journal was unmounted cleanly then we don't want to
1705          * recover anything. Otherwise, journal_load will do that
1706          * dirty work for us :) */
1707         if (!dirty) {
1708                 status = ocfs2_journal_wipe(osb->journal, 0);
1709                 if (status < 0) {
1710                         mlog_errno(status);
1711                         goto finally;
1712                 }
1713         } else {
1714                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1715                      "recovering volume.\n");
1716         }
1717
1718         local = ocfs2_mount_local(osb);
1719
1720         /* will play back anything left in the journal. */
1721         status = ocfs2_journal_load(osb->journal, local, dirty);
1722         if (status < 0) {
1723                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
1724                 goto finally;
1725         }
1726
1727         if (dirty) {
1728                 /* recover my local alloc if we didn't unmount cleanly. */
1729                 status = ocfs2_begin_local_alloc_recovery(osb,
1730                                                           osb->slot_num,
1731                                                           &local_alloc);
1732                 if (status < 0) {
1733                         mlog_errno(status);
1734                         goto finally;
1735                 }
1736                 /* we complete the recovery process after we've marked
1737                  * ourselves as mounted. */
1738         }
1739
1740         mlog(0, "Journal loaded.\n");
1741
1742         status = ocfs2_load_local_alloc(osb);
1743         if (status < 0) {
1744                 mlog_errno(status);
1745                 goto finally;
1746         }
1747
1748         if (dirty) {
1749                 /* Recovery will be completed after we've mounted the
1750                  * rest of the volume. */
1751                 osb->dirty = 1;
1752                 osb->local_alloc_copy = local_alloc;
1753                 local_alloc = NULL;
1754         }
1755
1756         /* go through each journal, trylock it and if you get the
1757          * lock, and it's marked as dirty, set the bit in the recover
1758          * map and launch a recovery thread for it. */
1759         status = ocfs2_mark_dead_nodes(osb);
1760         if (status < 0)
1761                 mlog_errno(status);
1762
1763 finally:
1764         if (local_alloc)
1765                 kfree(local_alloc);
1766
1767         mlog_exit(status);
1768         return status;
1769 }
1770
1771 /*
1772  * The routine gets called from dismount or close whenever a dismount on
1773  * volume is requested and the osb open count becomes 1.
1774  * It will remove the osb from the global list and also free up all the
1775  * initialized resources and fileobject.
1776  */
1777 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1778 {
1779         mlog_entry_void();
1780
1781         /* This function assumes that the caller has the main osb resource */
1782
1783         ocfs2_free_slot_info(osb);
1784
1785         kfree(osb->osb_orphan_wipes);
1786         kfree(osb->slot_recovery_generations);
1787         /* FIXME
1788          * This belongs in journal shutdown, but because we have to
1789          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1790          * we free it here.
1791          */
1792         kfree(osb->journal);
1793         if (osb->local_alloc_copy)
1794                 kfree(osb->local_alloc_copy);
1795         kfree(osb->uuid_str);
1796         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1797         memset(osb, 0, sizeof(struct ocfs2_super));
1798
1799         mlog_exit_void();
1800 }
1801
1802 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1803  * panic(). We do not support continue-on-error operation. */
1804 static void ocfs2_handle_error(struct super_block *sb)
1805 {
1806         struct ocfs2_super *osb = OCFS2_SB(sb);
1807
1808         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1809                 panic("OCFS2: (device %s): panic forced after error\n",
1810                       sb->s_id);
1811
1812         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1813
1814         if (sb->s_flags & MS_RDONLY &&
1815             (ocfs2_is_soft_readonly(osb) ||
1816              ocfs2_is_hard_readonly(osb)))
1817                 return;
1818
1819         printk(KERN_CRIT "File system is now read-only due to the potential "
1820                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1821                "system is unmounted.\n");
1822         sb->s_flags |= MS_RDONLY;
1823         ocfs2_set_ro_flag(osb, 0);
1824 }
1825
1826 static char error_buf[1024];
1827
1828 void __ocfs2_error(struct super_block *sb,
1829                    const char *function,
1830                    const char *fmt, ...)
1831 {
1832         va_list args;
1833
1834         va_start(args, fmt);
1835         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1836         va_end(args);
1837
1838         /* Not using mlog here because we want to show the actual
1839          * function the error came from. */
1840         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1841                sb->s_id, function, error_buf);
1842
1843         ocfs2_handle_error(sb);
1844 }
1845
1846 /* Handle critical errors. This is intentionally more drastic than
1847  * ocfs2_handle_error, so we only use for things like journal errors,
1848  * etc. */
1849 void __ocfs2_abort(struct super_block* sb,
1850                    const char *function,
1851                    const char *fmt, ...)
1852 {
1853         va_list args;
1854
1855         va_start(args, fmt);
1856         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1857         va_end(args);
1858
1859         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1860                sb->s_id, function, error_buf);
1861
1862         /* We don't have the cluster support yet to go straight to
1863          * hard readonly in here. Until then, we want to keep
1864          * ocfs2_abort() so that we can at least mark critical
1865          * errors.
1866          *
1867          * TODO: This should abort the journal and alert other nodes
1868          * that our slot needs recovery. */
1869
1870         /* Force a panic(). This stinks, but it's better than letting
1871          * things continue without having a proper hard readonly
1872          * here. */
1873         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1874         ocfs2_handle_error(sb);
1875 }
1876
1877 module_init(ocfs2_init);
1878 module_exit(ocfs2_exit);