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/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45 #include <linux/signal.h>
  46
  47 #define CREATE_TRACE_POINTS
  48 #include "ocfs2_trace.h"
  49
  50 #include <cluster/masklog.h>
  51
  52 #include "ocfs2.h"
  53
  54 /* this should be the only file to include a version 1 header */
  55 #include "ocfs1_fs_compat.h"
  56
  57 #include "alloc.h"
  58 #include "aops.h"
  59 #include "blockcheck.h"
  60 #include "dlmglue.h"
  61 #include "export.h"
  62 #include "extent_map.h"
  63 #include "heartbeat.h"
  64 #include "inode.h"
  65 #include "journal.h"
  66 #include "localalloc.h"
  67 #include "namei.h"
  68 #include "slot_map.h"
  69 #include "super.h"
  70 #include "sysfile.h"
  71 #include "uptodate.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78 #include "filecheck.h"
  79
  80 static struct kmem_cache *ocfs2_inode_cachep;
  81 struct kmem_cache *ocfs2_dquot_cachep;
  82 struct kmem_cache *ocfs2_qf_chunk_cachep;
  83
  84 static struct dentry *ocfs2_debugfs_root;
  85
  86 MODULE_AUTHOR("Oracle");
  87 MODULE_LICENSE("GPL");
  88 MODULE_DESCRIPTION("OCFS2 cluster file system");
  89
  90 struct mount_options
  91 {
  92         unsigned long   commit_interval;
  93         unsigned long   mount_opt;
  94         unsigned int    atime_quantum;
  95         signed short    slot;
  96         int             localalloc_opt;
  97         unsigned int    resv_level;
  98         int             dir_resv_level;
  99         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 100 };
 101
 102 static int ocfs2_parse_options(struct super_block *sb, char *options,
 103                                struct mount_options *mopt,
 104                                int is_remount);
 105 static int ocfs2_check_set_options(struct super_block *sb,
 106                                    struct mount_options *options);
 107 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 108 static void ocfs2_put_super(struct super_block *sb);
 109 static int ocfs2_mount_volume(struct super_block *sb);
 110 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 111 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 112 static int ocfs2_initialize_mem_caches(void);
 113 static void ocfs2_free_mem_caches(void);
 114 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 115
 116 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 117
 118 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 119
 120 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 121 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 122 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 123 static int ocfs2_check_volume(struct ocfs2_super *osb);
 124 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 125                                struct buffer_head *bh,
 126                                u32 sectsize,
 127                                struct ocfs2_blockcheck_stats *stats);
 128 static int ocfs2_initialize_super(struct super_block *sb,
 129                                   struct buffer_head *bh,
 130                                   int sector_size,
 131                                   struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_get_sector(struct super_block *sb,
 133                             struct buffer_head **bh,
 134                             int block,
 135                             int sect_size);
 136 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 137 static void ocfs2_destroy_inode(struct inode *inode);
 138 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 139 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 140 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 141
 142 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 143 {
 144         return OCFS2_I(inode)->i_dquot;
 145 }
 146
 147 static const struct super_operations ocfs2_sops = {
 148         .statfs         = ocfs2_statfs,
 149         .alloc_inode    = ocfs2_alloc_inode,
 150         .destroy_inode  = ocfs2_destroy_inode,
 151         .drop_inode     = ocfs2_drop_inode,
 152         .evict_inode    = ocfs2_evict_inode,
 153         .sync_fs        = ocfs2_sync_fs,
 154         .put_super      = ocfs2_put_super,
 155         .remount_fs     = ocfs2_remount,
 156         .show_options   = ocfs2_show_options,
 157         .quota_read     = ocfs2_quota_read,
 158         .quota_write    = ocfs2_quota_write,
 159         .get_dquots     = ocfs2_get_dquots,
 160 };
 161
 162 enum {
 163         Opt_barrier,
 164         Opt_err_panic,
 165         Opt_err_ro,
 166         Opt_intr,
 167         Opt_nointr,
 168         Opt_hb_none,
 169         Opt_hb_local,
 170         Opt_hb_global,
 171         Opt_data_ordered,
 172         Opt_data_writeback,
 173         Opt_atime_quantum,
 174         Opt_slot,
 175         Opt_commit,
 176         Opt_localalloc,
 177         Opt_localflocks,
 178         Opt_stack,
 179         Opt_user_xattr,
 180         Opt_nouser_xattr,
 181         Opt_inode64,
 182         Opt_acl,
 183         Opt_noacl,
 184         Opt_usrquota,
 185         Opt_grpquota,
 186         Opt_coherency_buffered,
 187         Opt_coherency_full,
 188         Opt_resv_level,
 189         Opt_dir_resv_level,
 190         Opt_journal_async_commit,
 191         Opt_err_cont,
 192         Opt_err,
 193 };
 194
 195 static const match_table_t tokens = {
 196         {Opt_barrier, "barrier=%u"},
 197         {Opt_err_panic, "errors=panic"},
 198         {Opt_err_ro, "errors=remount-ro"},
 199         {Opt_intr, "intr"},
 200         {Opt_nointr, "nointr"},
 201         {Opt_hb_none, OCFS2_HB_NONE},
 202         {Opt_hb_local, OCFS2_HB_LOCAL},
 203         {Opt_hb_global, OCFS2_HB_GLOBAL},
 204         {Opt_data_ordered, "data=ordered"},
 205         {Opt_data_writeback, "data=writeback"},
 206         {Opt_atime_quantum, "atime_quantum=%u"},
 207         {Opt_slot, "preferred_slot=%u"},
 208         {Opt_commit, "commit=%u"},
 209         {Opt_localalloc, "localalloc=%d"},
 210         {Opt_localflocks, "localflocks"},
 211         {Opt_stack, "cluster_stack=%s"},
 212         {Opt_user_xattr, "user_xattr"},
 213         {Opt_nouser_xattr, "nouser_xattr"},
 214         {Opt_inode64, "inode64"},
 215         {Opt_acl, "acl"},
 216         {Opt_noacl, "noacl"},
 217         {Opt_usrquota, "usrquota"},
 218         {Opt_grpquota, "grpquota"},
 219         {Opt_coherency_buffered, "coherency=buffered"},
 220         {Opt_coherency_full, "coherency=full"},
 221         {Opt_resv_level, "resv_level=%u"},
 222         {Opt_dir_resv_level, "dir_resv_level=%u"},
 223         {Opt_journal_async_commit, "journal_async_commit"},
 224         {Opt_err_cont, "errors=continue"},
 225         {Opt_err, NULL}
 226 };
 227
 228 #ifdef CONFIG_DEBUG_FS
 229 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 230 {
 231         struct ocfs2_cluster_connection *cconn = osb->cconn;
 232         struct ocfs2_recovery_map *rm = osb->recovery_map;
 233         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 234         int i, out = 0;
 235         unsigned long flags;
 236
 237         out += snprintf(buf + out, len - out,
 238                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 239                         "Device", osb->dev_str, osb->uuid_str,
 240                         osb->fs_generation, osb->vol_label);
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 244                         atomic_read(&osb->vol_state), osb->osb_flags);
 245
 246         out += snprintf(buf + out, len - out,
 247                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 248                         osb->sb->s_blocksize, osb->s_clustersize);
 249
 250         out += snprintf(buf + out, len - out,
 251                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 252                         "ROcompat: 0x%X\n",
 253                         "Features", osb->s_feature_compat,
 254                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 255
 256         out += snprintf(buf + out, len - out,
 257                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 258                         osb->s_mount_opt, osb->s_atime_quantum);
 259
 260         if (cconn) {
 261                 out += snprintf(buf + out, len - out,
 262                                 "%10s => Stack: %s  Name: %*s  "
 263                                 "Version: %d.%d\n", "Cluster",
 264                                 (*osb->osb_cluster_stack == '\0' ?
 265                                  "o2cb" : osb->osb_cluster_stack),
 266                                 cconn->cc_namelen, cconn->cc_name,
 267                                 cconn->cc_version.pv_major,
 268                                 cconn->cc_version.pv_minor);
 269         }
 270
 271         spin_lock_irqsave(&osb->dc_task_lock, flags);
 272         out += snprintf(buf + out, len - out,
 273                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 274                         "WorkSeq: %lu\n", "DownCnvt",
 275                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 276                         osb->blocked_lock_count, osb->dc_wake_sequence,
 277                         osb->dc_work_sequence);
 278         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 279
 280         spin_lock(&osb->osb_lock);
 281         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 282                         "Recovery",
 283                         (osb->recovery_thread_task ?
 284                          task_pid_nr(osb->recovery_thread_task) : -1));
 285         if (rm->rm_used == 0)
 286                 out += snprintf(buf + out, len - out, " None\n");
 287         else {
 288                 for (i = 0; i < rm->rm_used; i++)
 289                         out += snprintf(buf + out, len - out, " %d",
 290                                         rm->rm_entries[i]);
 291                 out += snprintf(buf + out, len - out, "\n");
 292         }
 293         spin_unlock(&osb->osb_lock);
 294
 295         out += snprintf(buf + out, len - out,
 296                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 297                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 298                         osb->osb_commit_interval);
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 302                         "Journal", osb->journal->j_state,
 303                         osb->journal->j_trans_id,
 304                         atomic_read(&osb->journal->j_num_trans));
 305
 306         out += snprintf(buf + out, len - out,
 307                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 308                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 309                         "Stats",
 310                         atomic_read(&osb->alloc_stats.bitmap_data),
 311                         atomic_read(&osb->alloc_stats.local_data),
 312                         atomic_read(&osb->alloc_stats.bg_allocs),
 313                         atomic_read(&osb->alloc_stats.moves),
 314                         atomic_read(&osb->alloc_stats.bg_extends));
 315
 316         out += snprintf(buf + out, len - out,
 317                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 318                         "Default: %u bits\n",
 319                         "LocalAlloc", osb->local_alloc_state,
 320                         (unsigned long long)osb->la_last_gd,
 321                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 322
 323         spin_lock(&osb->osb_lock);
 324         out += snprintf(buf + out, len - out,
 325                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 326                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 327                         osb->s_inode_steal_slot,
 328                         atomic_read(&osb->s_num_inodes_stolen),
 329                         osb->s_meta_steal_slot,
 330                         atomic_read(&osb->s_num_meta_stolen));
 331         spin_unlock(&osb->osb_lock);
 332
 333         out += snprintf(buf + out, len - out, "OrphanScan => ");
 334         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 335                         os->os_count, os->os_seqno);
 336         out += snprintf(buf + out, len - out, " Last Scan: ");
 337         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 338                 out += snprintf(buf + out, len - out, "Disabled\n");
 339         else
 340                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 341                                 (unsigned long)(ktime_get_seconds() - os->os_scantime));
 342
 343         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 344                         "Slots", "Num", "RecoGen");
 345         for (i = 0; i < osb->max_slots; ++i) {
 346                 out += snprintf(buf + out, len - out,
 347                                 "%10s  %c %3d  %10d\n",
 348                                 " ",
 349                                 (i == osb->slot_num ? '*' : ' '),
 350                                 i, osb->slot_recovery_generations[i]);
 351         }
 352
 353         return out;
 354 }
 355
 356 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 357 {
 358         struct ocfs2_super *osb = inode->i_private;
 359         char *buf = NULL;
 360
 361         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 362         if (!buf)
 363                 goto bail;
 364
 365         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 366
 367         file->private_data = buf;
 368
 369         return 0;
 370 bail:
 371         return -ENOMEM;
 372 }
 373
 374 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 375 {
 376         kfree(file->private_data);
 377         return 0;
 378 }
 379
 380 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 381                                 size_t nbytes, loff_t *ppos)
 382 {
 383         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 384                                        i_size_read(file->f_mapping->host));
 385 }
 386 #else
 387 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 388 {
 389         return 0;
 390 }
 391 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 392 {
 393         return 0;
 394 }
 395 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 396                                 size_t nbytes, loff_t *ppos)
 397 {
 398         return 0;
 399 }
 400 #endif  /* CONFIG_DEBUG_FS */
 401
 402 static const struct file_operations ocfs2_osb_debug_fops = {
 403         .open =         ocfs2_osb_debug_open,
 404         .release =      ocfs2_debug_release,
 405         .read =         ocfs2_debug_read,
 406         .llseek =       generic_file_llseek,
 407 };
 408
 409 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 410 {
 411         int status;
 412         tid_t target;
 413         struct ocfs2_super *osb = OCFS2_SB(sb);
 414
 415         if (ocfs2_is_hard_readonly(osb))
 416                 return -EROFS;
 417
 418         if (wait) {
 419                 status = ocfs2_flush_truncate_log(osb);
 420                 if (status < 0)
 421                         mlog_errno(status);
 422         } else {
 423                 ocfs2_schedule_truncate_log_flush(osb, 0);
 424         }
 425
 426         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 427                                       &target)) {
 428                 if (wait)
 429                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 430                                              target);
 431         }
 432         return 0;
 433 }
 434
 435 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 436 {
 437         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 438             && (ino == USER_QUOTA_SYSTEM_INODE
 439                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 440                 return 0;
 441         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 442             && (ino == GROUP_QUOTA_SYSTEM_INODE
 443                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 444                 return 0;
 445         return 1;
 446 }
 447
 448 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 449 {
 450         struct inode *new = NULL;
 451         int status = 0;
 452         int i;
 453
 454         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 455         if (IS_ERR(new)) {
 456                 status = PTR_ERR(new);
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460         osb->root_inode = new;
 461
 462         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 463         if (IS_ERR(new)) {
 464                 status = PTR_ERR(new);
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468         osb->sys_root_inode = new;
 469
 470         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 471              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 472                 if (!ocfs2_need_system_inode(osb, i))
 473                         continue;
 474                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 475                 if (!new) {
 476                         ocfs2_release_system_inodes(osb);
 477                         status = -EINVAL;
 478                         mlog_errno(status);
 479                         /* FIXME: Should ERROR_RO_FS */
 480                         mlog(ML_ERROR, "Unable to load system inode %d, "
 481                              "possibly corrupt fs?", i);
 482                         goto bail;
 483                 }
 484                 // the array now has one ref, so drop this one
 485                 iput(new);
 486         }
 487
 488 bail:
 489         if (status)
 490                 mlog_errno(status);
 491         return status;
 492 }
 493
 494 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 495 {
 496         struct inode *new = NULL;
 497         int status = 0;
 498         int i;
 499
 500         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 501              i < NUM_SYSTEM_INODES;
 502              i++) {
 503                 if (!ocfs2_need_system_inode(osb, i))
 504                         continue;
 505                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 506                 if (!new) {
 507                         ocfs2_release_system_inodes(osb);
 508                         status = -EINVAL;
 509                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 510                              status, i, osb->slot_num);
 511                         goto bail;
 512                 }
 513                 /* the array now has one ref, so drop this one */
 514                 iput(new);
 515         }
 516
 517 bail:
 518         if (status)
 519                 mlog_errno(status);
 520         return status;
 521 }
 522
 523 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 524 {
 525         int i;
 526         struct inode *inode;
 527
 528         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 529                 inode = osb->global_system_inodes[i];
 530                 if (inode) {
 531                         iput(inode);
 532                         osb->global_system_inodes[i] = NULL;
 533                 }
 534         }
 535
 536         inode = osb->sys_root_inode;
 537         if (inode) {
 538                 iput(inode);
 539                 osb->sys_root_inode = NULL;
 540         }
 541
 542         inode = osb->root_inode;
 543         if (inode) {
 544                 iput(inode);
 545                 osb->root_inode = NULL;
 546         }
 547
 548         if (!osb->local_system_inodes)
 549                 return;
 550
 551         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 552                 if (osb->local_system_inodes[i]) {
 553                         iput(osb->local_system_inodes[i]);
 554                         osb->local_system_inodes[i] = NULL;
 555                 }
 556         }
 557
 558         kfree(osb->local_system_inodes);
 559         osb->local_system_inodes = NULL;
 560 }
 561
 562 /* We're allocating fs objects, use GFP_NOFS */
 563 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 564 {
 565         struct ocfs2_inode_info *oi;
 566
 567         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 568         if (!oi)
 569                 return NULL;
 570
 571         oi->i_sync_tid = 0;
 572         oi->i_datasync_tid = 0;
 573         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 574
 575         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 576         return &oi->vfs_inode;
 577 }
 578
 579 static void ocfs2_i_callback(struct rcu_head *head)
 580 {
 581         struct inode *inode = container_of(head, struct inode, i_rcu);
 582         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 583 }
 584
 585 static void ocfs2_destroy_inode(struct inode *inode)
 586 {
 587         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 588 }
 589
 590 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 591                                                 unsigned int cbits)
 592 {
 593         unsigned int bytes = 1 << cbits;
 594         unsigned int trim = bytes;
 595         unsigned int bitshift = 32;
 596
 597         /*
 598          * i_size and all block offsets in ocfs2 are always 64 bits
 599          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 600          * 64 bit platforms, cluster size will be the limiting factor.
 601          */
 602
 603 #if BITS_PER_LONG == 32
 604 # if defined(CONFIG_LBDAF)
 605         BUILD_BUG_ON(sizeof(sector_t) != 8);
 606         /*
 607          * We might be limited by page cache size.
 608          */
 609         if (bytes > PAGE_SIZE) {
 610                 bytes = PAGE_SIZE;
 611                 trim = 1;
 612                 /*
 613                  * Shift by 31 here so that we don't get larger than
 614                  * MAX_LFS_FILESIZE
 615                  */
 616                 bitshift = 31;
 617         }
 618 # else
 619         /*
 620          * We are limited by the size of sector_t. Use block size, as
 621          * that's what we expose to the VFS.
 622          */
 623         bytes = 1 << bbits;
 624         trim = 1;
 625         bitshift = 31;
 626 # endif
 627 #endif
 628
 629         /*
 630          * Trim by a whole cluster when we can actually approach the
 631          * on-disk limits. Otherwise we can overflow i_clusters when
 632          * an extent start is at the max offset.
 633          */
 634         return (((unsigned long long)bytes) << bitshift) - trim;
 635 }
 636
 637 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 638 {
 639         int incompat_features;
 640         int ret = 0;
 641         struct mount_options parsed_options;
 642         struct ocfs2_super *osb = OCFS2_SB(sb);
 643         u32 tmp;
 644
 645         sync_filesystem(sb);
 646
 647         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 648             !ocfs2_check_set_options(sb, &parsed_options)) {
 649                 ret = -EINVAL;
 650                 goto out;
 651         }
 652
 653         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 654                 OCFS2_MOUNT_HB_NONE;
 655         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 656                 ret = -EINVAL;
 657                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 658                 goto out;
 659         }
 660
 661         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 662             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 663                 ret = -EINVAL;
 664                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 665                 goto out;
 666         }
 667
 668         /* Probably don't want this on remount; it might
 669          * mess with other nodes */
 670         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 671             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 672                 ret = -EINVAL;
 673                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 674                 goto out;
 675         }
 676
 677         /* We're going to/from readonly mode. */
 678         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 679                 /* Disable quota accounting before remounting RO */
 680                 if (*flags & MS_RDONLY) {
 681                         ret = ocfs2_susp_quotas(osb, 0);
 682                         if (ret < 0)
 683                                 goto out;
 684                 }
 685                 /* Lock here so the check of HARD_RO and the potential
 686                  * setting of SOFT_RO is atomic. */
 687                 spin_lock(&osb->osb_lock);
 688                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 689                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 690                         ret = -EROFS;
 691                         goto unlock_osb;
 692                 }
 693
 694                 if (*flags & MS_RDONLY) {
 695                         sb->s_flags |= MS_RDONLY;
 696                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 697                 } else {
 698                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 699                                 mlog(ML_ERROR, "Cannot remount RDWR "
 700                                      "filesystem due to previous errors.\n");
 701                                 ret = -EROFS;
 702                                 goto unlock_osb;
 703                         }
 704                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 705                         if (incompat_features) {
 706                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 707                                      "of unsupported optional features "
 708                                      "(%x).\n", incompat_features);
 709                                 ret = -EINVAL;
 710                                 goto unlock_osb;
 711                         }
 712                         sb->s_flags &= ~MS_RDONLY;
 713                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 714                 }
 715                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 716 unlock_osb:
 717                 spin_unlock(&osb->osb_lock);
 718                 /* Enable quota accounting after remounting RW */
 719                 if (!ret && !(*flags & MS_RDONLY)) {
 720                         if (sb_any_quota_suspended(sb))
 721                                 ret = ocfs2_susp_quotas(osb, 1);
 722                         else
 723                                 ret = ocfs2_enable_quotas(osb);
 724                         if (ret < 0) {
 725                                 /* Return back changes... */
 726                                 spin_lock(&osb->osb_lock);
 727                                 sb->s_flags |= MS_RDONLY;
 728                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 729                                 spin_unlock(&osb->osb_lock);
 730                                 goto out;
 731                         }
 732                 }
 733         }
 734
 735         if (!ret) {
 736                 /* Only save off the new mount options in case of a successful
 737                  * remount. */
 738                 osb->s_mount_opt = parsed_options.mount_opt;
 739                 osb->s_atime_quantum = parsed_options.atime_quantum;
 740                 osb->preferred_slot = parsed_options.slot;
 741                 if (parsed_options.commit_interval)
 742                         osb->osb_commit_interval = parsed_options.commit_interval;
 743
 744                 if (!ocfs2_is_hard_readonly(osb))
 745                         ocfs2_set_journal_params(osb);
 746
 747                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 748                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 749                                                         MS_POSIXACL : 0);
 750         }
 751 out:
 752         return ret;
 753 }
 754
 755 static int ocfs2_sb_probe(struct super_block *sb,
 756                           struct buffer_head **bh,
 757                           int *sector_size,
 758                           struct ocfs2_blockcheck_stats *stats)
 759 {
 760         int status, tmpstat;
 761         struct ocfs1_vol_disk_hdr *hdr;
 762         struct ocfs2_dinode *di;
 763         int blksize;
 764
 765         *bh = NULL;
 766
 767         /* may be > 512 */
 768         *sector_size = bdev_logical_block_size(sb->s_bdev);
 769         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 770                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 771                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 772                 status = -EINVAL;
 773                 goto bail;
 774         }
 775
 776         /* Can this really happen? */
 777         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 778                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 779
 780         /* check block zero for old format */
 781         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 782         if (status < 0) {
 783                 mlog_errno(status);
 784                 goto bail;
 785         }
 786         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 787         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 788                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 789                      hdr->major_version, hdr->minor_version);
 790                 status = -EINVAL;
 791         }
 792         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 793                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 794                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 795                      hdr->signature);
 796                 status = -EINVAL;
 797         }
 798         brelse(*bh);
 799         *bh = NULL;
 800         if (status < 0) {
 801                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 802                      "upgraded before mounting with ocfs v2\n");
 803                 goto bail;
 804         }
 805
 806         /*
 807          * Now check at magic offset for 512, 1024, 2048, 4096
 808          * blocksizes.  4096 is the maximum blocksize because it is
 809          * the minimum clustersize.
 810          */
 811         status = -EINVAL;
 812         for (blksize = *sector_size;
 813              blksize <= OCFS2_MAX_BLOCKSIZE;
 814              blksize <<= 1) {
 815                 tmpstat = ocfs2_get_sector(sb, bh,
 816                                            OCFS2_SUPER_BLOCK_BLKNO,
 817                                            blksize);
 818                 if (tmpstat < 0) {
 819                         status = tmpstat;
 820                         mlog_errno(status);
 821                         break;
 822                 }
 823                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 824                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 825                 spin_lock_init(&stats->b_lock);
 826                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 827                 if (tmpstat < 0) {
 828                         brelse(*bh);
 829                         *bh = NULL;
 830                 }
 831                 if (tmpstat != -EAGAIN) {
 832                         status = tmpstat;
 833                         break;
 834                 }
 835         }
 836
 837 bail:
 838         return status;
 839 }
 840
 841 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 842 {
 843         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 844
 845         if (osb->s_mount_opt & hb_enabled) {
 846                 if (ocfs2_mount_local(osb)) {
 847                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 848                              "mounted device.\n");
 849                         return -EINVAL;
 850                 }
 851                 if (ocfs2_userspace_stack(osb)) {
 852                         mlog(ML_ERROR, "Userspace stack expected, but "
 853                              "o2cb heartbeat arguments passed to mount\n");
 854                         return -EINVAL;
 855                 }
 856                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 857                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 858                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 859                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 860                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 861                         return -EINVAL;
 862                 }
 863         }
 864
 865         if (!(osb->s_mount_opt & hb_enabled)) {
 866                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 867                     !ocfs2_userspace_stack(osb)) {
 868                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 869                              "a read-write clustered device.\n");
 870                         return -EINVAL;
 871                 }
 872         }
 873
 874         return 0;
 875 }
 876
 877 /*
 878  * If we're using a userspace stack, mount should have passed
 879  * a name that matches the disk.  If not, mount should not
 880  * have passed a stack.
 881  */
 882 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 883                                         struct mount_options *mopt)
 884 {
 885         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 886                 mlog(ML_ERROR,
 887                      "cluster stack passed to mount, but this filesystem "
 888                      "does not support it\n");
 889                 return -EINVAL;
 890         }
 891
 892         if (ocfs2_userspace_stack(osb) &&
 893             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 894                     OCFS2_STACK_LABEL_LEN)) {
 895                 mlog(ML_ERROR,
 896                      "cluster stack passed to mount (\"%s\") does not "
 897                      "match the filesystem (\"%s\")\n",
 898                      mopt->cluster_stack,
 899                      osb->osb_cluster_stack);
 900                 return -EINVAL;
 901         }
 902
 903         return 0;
 904 }
 905
 906 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 907 {
 908         int type;
 909         struct super_block *sb = osb->sb;
 910         unsigned int feature[OCFS2_MAXQUOTAS] = {
 911                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 912                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 913         int status = 0;
 914
 915         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 916                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 917                         continue;
 918                 if (unsuspend)
 919                         status = dquot_resume(sb, type);
 920                 else {
 921                         struct ocfs2_mem_dqinfo *oinfo;
 922
 923                         /* Cancel periodic syncing before suspending */
 924                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 925                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 926                         status = dquot_suspend(sb, type);
 927                 }
 928                 if (status < 0)
 929                         break;
 930         }
 931         if (status < 0)
 932                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 933                      "remount (error = %d).\n", status);
 934         return status;
 935 }
 936
 937 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 938 {
 939         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 940         struct super_block *sb = osb->sb;
 941         unsigned int feature[OCFS2_MAXQUOTAS] = {
 942                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 943                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 944         unsigned int ino[OCFS2_MAXQUOTAS] = {
 945                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 946                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 947         int status;
 948         int type;
 949
 950         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 951         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 952                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 953                         continue;
 954                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 955                                                         osb->slot_num);
 956                 if (!inode[type]) {
 957                         status = -ENOENT;
 958                         goto out_quota_off;
 959                 }
 960                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 961                                       DQUOT_USAGE_ENABLED);
 962                 if (status < 0)
 963                         goto out_quota_off;
 964         }
 965
 966         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 967                 iput(inode[type]);
 968         return 0;
 969 out_quota_off:
 970         ocfs2_disable_quotas(osb);
 971         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 972                 iput(inode[type]);
 973         mlog_errno(status);
 974         return status;
 975 }
 976
 977 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 978 {
 979         int type;
 980         struct inode *inode;
 981         struct super_block *sb = osb->sb;
 982         struct ocfs2_mem_dqinfo *oinfo;
 983
 984         /* We mostly ignore errors in this function because there's not much
 985          * we can do when we see them */
 986         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 987                 if (!sb_has_quota_loaded(sb, type))
 988                         continue;
 989                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 990                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 991                 inode = igrab(sb->s_dquot.files[type]);
 992                 /* Turn off quotas. This will remove all dquot structures from
 993                  * memory and so they will be automatically synced to global
 994                  * quota files */
 995                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 996                                         DQUOT_LIMITS_ENABLED);
 997                 if (!inode)
 998                         continue;
 999                 iput(inode);
1000         }
1001 }
1002
1003 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1004 {
1005         struct dentry *root;
1006         int status, sector_size;
1007         struct mount_options parsed_options;
1008         struct inode *inode = NULL;
1009         struct ocfs2_super *osb = NULL;
1010         struct buffer_head *bh = NULL;
1011         char nodestr[12];
1012         struct ocfs2_blockcheck_stats stats;
1013
1014         trace_ocfs2_fill_super(sb, data, silent);
1015
1016         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1017                 status = -EINVAL;
1018                 goto read_super_error;
1019         }
1020
1021         /* probe for superblock */
1022         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1023         if (status < 0) {
1024                 mlog(ML_ERROR, "superblock probe failed!\n");
1025                 goto read_super_error;
1026         }
1027
1028         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1029         osb = OCFS2_SB(sb);
1030         if (status < 0) {
1031                 mlog_errno(status);
1032                 goto read_super_error;
1033         }
1034         brelse(bh);
1035         bh = NULL;
1036
1037         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1038                 status = -EINVAL;
1039                 goto read_super_error;
1040         }
1041         osb->s_mount_opt = parsed_options.mount_opt;
1042         osb->s_atime_quantum = parsed_options.atime_quantum;
1043         osb->preferred_slot = parsed_options.slot;
1044         osb->osb_commit_interval = parsed_options.commit_interval;
1045
1046         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1047         osb->osb_resv_level = parsed_options.resv_level;
1048         osb->osb_dir_resv_level = parsed_options.resv_level;
1049         if (parsed_options.dir_resv_level == -1)
1050                 osb->osb_dir_resv_level = parsed_options.resv_level;
1051         else
1052                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1053
1054         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1055         if (status)
1056                 goto read_super_error;
1057
1058         sb->s_magic = OCFS2_SUPER_MAGIC;
1059
1060         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1061                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1062
1063         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1064          * heartbeat=none */
1065         if (bdev_read_only(sb->s_bdev)) {
1066                 if (!(sb->s_flags & MS_RDONLY)) {
1067                         status = -EACCES;
1068                         mlog(ML_ERROR, "Readonly device detected but readonly "
1069                              "mount was not specified.\n");
1070                         goto read_super_error;
1071                 }
1072
1073                 /* You should not be able to start a local heartbeat
1074                  * on a readonly device. */
1075                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1076                         status = -EROFS;
1077                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1078                              "device.\n");
1079                         goto read_super_error;
1080                 }
1081
1082                 status = ocfs2_check_journals_nolocks(osb);
1083                 if (status < 0) {
1084                         if (status == -EROFS)
1085                                 mlog(ML_ERROR, "Recovery required on readonly "
1086                                      "file system, but write access is "
1087                                      "unavailable.\n");
1088                         else
1089                                 mlog_errno(status);
1090                         goto read_super_error;
1091                 }
1092
1093                 ocfs2_set_ro_flag(osb, 1);
1094
1095                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1096                        "Cluster services will not be used for this mount. "
1097                        "Recovery will be skipped.\n", osb->dev_str);
1098         }
1099
1100         if (!ocfs2_is_hard_readonly(osb)) {
1101                 if (sb->s_flags & MS_RDONLY)
1102                         ocfs2_set_ro_flag(osb, 0);
1103         }
1104
1105         status = ocfs2_verify_heartbeat(osb);
1106         if (status < 0) {
1107                 mlog_errno(status);
1108                 goto read_super_error;
1109         }
1110
1111         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1112                                                  ocfs2_debugfs_root);
1113         if (!osb->osb_debug_root) {
1114                 status = -EINVAL;
1115                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1116                 goto read_super_error;
1117         }
1118
1119         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1120                                             osb->osb_debug_root,
1121                                             osb,
1122                                             &ocfs2_osb_debug_fops);
1123         if (!osb->osb_ctxt) {
1124                 status = -EINVAL;
1125                 mlog_errno(status);
1126                 goto read_super_error;
1127         }
1128
1129         if (ocfs2_meta_ecc(osb)) {
1130                 status = ocfs2_blockcheck_stats_debugfs_install(
1131                                                 &osb->osb_ecc_stats,
1132                                                 osb->osb_debug_root);
1133                 if (status) {
1134                         mlog(ML_ERROR,
1135                              "Unable to create blockcheck statistics "
1136                              "files\n");
1137                         goto read_super_error;
1138                 }
1139         }
1140
1141         status = ocfs2_mount_volume(sb);
1142         if (status < 0)
1143                 goto read_super_error;
1144
1145         if (osb->root_inode)
1146                 inode = igrab(osb->root_inode);
1147
1148         if (!inode) {
1149                 status = -EIO;
1150                 mlog_errno(status);
1151                 goto read_super_error;
1152         }
1153
1154         root = d_make_root(inode);
1155         if (!root) {
1156                 status = -ENOMEM;
1157                 mlog_errno(status);
1158                 goto read_super_error;
1159         }
1160
1161         sb->s_root = root;
1162
1163         ocfs2_complete_mount_recovery(osb);
1164
1165         if (ocfs2_mount_local(osb))
1166                 snprintf(nodestr, sizeof(nodestr), "local");
1167         else
1168                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1169
1170         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1171                "with %s data mode.\n",
1172                osb->dev_str, nodestr, osb->slot_num,
1173                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1174                "ordered");
1175
1176         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1177         wake_up(&osb->osb_mount_event);
1178
1179         /* Now we can initialize quotas because we can afford to wait
1180          * for cluster locks recovery now. That also means that truncation
1181          * log recovery can happen but that waits for proper quota setup */
1182         if (!(sb->s_flags & MS_RDONLY)) {
1183                 status = ocfs2_enable_quotas(osb);
1184                 if (status < 0) {
1185                         /* We have to err-out specially here because
1186                          * s_root is already set */
1187                         mlog_errno(status);
1188                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1189                         wake_up(&osb->osb_mount_event);
1190                         return status;
1191                 }
1192         }
1193
1194         ocfs2_complete_quota_recovery(osb);
1195
1196         /* Now we wake up again for processes waiting for quotas */
1197         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1198         wake_up(&osb->osb_mount_event);
1199
1200         /* Start this when the mount is almost sure of being successful */
1201         ocfs2_orphan_scan_start(osb);
1202
1203         /* Create filecheck sysfile /sys/fs/ocfs2/<devname>/filecheck */
1204         ocfs2_filecheck_create_sysfs(sb);
1205
1206         return status;
1207
1208 read_super_error:
1209         brelse(bh);
1210
1211         if (osb) {
1212                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1213                 wake_up(&osb->osb_mount_event);
1214                 ocfs2_dismount_volume(sb, 1);
1215         }
1216
1217         if (status)
1218                 mlog_errno(status);
1219         return status;
1220 }
1221
1222 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1223                         int flags,
1224                         const char *dev_name,
1225                         void *data)
1226 {
1227         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1228 }
1229
1230 static struct file_system_type ocfs2_fs_type = {
1231         .owner          = THIS_MODULE,
1232         .name           = "ocfs2",
1233         .mount          = ocfs2_mount,
1234         .kill_sb        = kill_block_super,
1235         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1236         .next           = NULL
1237 };
1238 MODULE_ALIAS_FS("ocfs2");
1239
1240 static int ocfs2_check_set_options(struct super_block *sb,
1241                                    struct mount_options *options)
1242 {
1243         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1244             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1245                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1246                 mlog(ML_ERROR, "User quotas were requested, but this "
1247                      "filesystem does not have the feature enabled.\n");
1248                 return 0;
1249         }
1250         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1251             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1252                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1253                 mlog(ML_ERROR, "Group quotas were requested, but this "
1254                      "filesystem does not have the feature enabled.\n");
1255                 return 0;
1256         }
1257         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1258             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1259                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1260                      "feature is not enabled\n");
1261                 return 0;
1262         }
1263         /* No ACL setting specified? Use XATTR feature... */
1264         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1265                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1266                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1267                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1268                 else
1269                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1270         }
1271         return 1;
1272 }
1273
1274 static int ocfs2_parse_options(struct super_block *sb,
1275                                char *options,
1276                                struct mount_options *mopt,
1277                                int is_remount)
1278 {
1279         int status, user_stack = 0;
1280         char *p;
1281         u32 tmp;
1282         int token, option;
1283         substring_t args[MAX_OPT_ARGS];
1284
1285         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1286
1287         mopt->commit_interval = 0;
1288         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1289         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1290         mopt->slot = OCFS2_INVALID_SLOT;
1291         mopt->localalloc_opt = -1;
1292         mopt->cluster_stack[0] = '\0';
1293         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1294         mopt->dir_resv_level = -1;
1295
1296         if (!options) {
1297                 status = 1;
1298                 goto bail;
1299         }
1300
1301         while ((p = strsep(&options, ",")) != NULL) {
1302                 if (!*p)
1303                         continue;
1304
1305                 token = match_token(p, tokens, args);
1306                 switch (token) {
1307                 case Opt_hb_local:
1308                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1309                         break;
1310                 case Opt_hb_none:
1311                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1312                         break;
1313                 case Opt_hb_global:
1314                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1315                         break;
1316                 case Opt_barrier:
1317                         if (match_int(&args[0], &option)) {
1318                                 status = 0;
1319                                 goto bail;
1320                         }
1321                         if (option)
1322                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1323                         else
1324                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1325                         break;
1326                 case Opt_intr:
1327                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1328                         break;
1329                 case Opt_nointr:
1330                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1331                         break;
1332                 case Opt_err_panic:
1333                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1334                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1335                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1336                         break;
1337                 case Opt_err_ro:
1338                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1339                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1340                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1341                         break;
1342                 case Opt_err_cont:
1343                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1344                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1345                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1346                         break;
1347                 case Opt_data_ordered:
1348                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1349                         break;
1350                 case Opt_data_writeback:
1351                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1352                         break;
1353                 case Opt_user_xattr:
1354                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1355                         break;
1356                 case Opt_nouser_xattr:
1357                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1358                         break;
1359                 case Opt_atime_quantum:
1360                         if (match_int(&args[0], &option)) {
1361                                 status = 0;
1362                                 goto bail;
1363                         }
1364                         if (option >= 0)
1365                                 mopt->atime_quantum = option;
1366                         break;
1367                 case Opt_slot:
1368                         if (match_int(&args[0], &option)) {
1369                                 status = 0;
1370                                 goto bail;
1371                         }
1372                         if (option)
1373                                 mopt->slot = (s16)option;
1374                         break;
1375                 case Opt_commit:
1376                         if (match_int(&args[0], &option)) {
1377                                 status = 0;
1378                                 goto bail;
1379                         }
1380                         if (option < 0)
1381                                 return 0;
1382                         if (option == 0)
1383                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1384                         mopt->commit_interval = HZ * option;
1385                         break;
1386                 case Opt_localalloc:
1387                         if (match_int(&args[0], &option)) {
1388                                 status = 0;
1389                                 goto bail;
1390                         }
1391                         if (option >= 0)
1392                                 mopt->localalloc_opt = option;
1393                         break;
1394                 case Opt_localflocks:
1395                         /*
1396                          * Changing this during remount could race
1397                          * flock() requests, or "unbalance" existing
1398                          * ones (e.g., a lock is taken in one mode but
1399                          * dropped in the other). If users care enough
1400                          * to flip locking modes during remount, we
1401                          * could add a "local" flag to individual
1402                          * flock structures for proper tracking of
1403                          * state.
1404                          */
1405                         if (!is_remount)
1406                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1407                         break;
1408                 case Opt_stack:
1409                         /* Check both that the option we were passed
1410                          * is of the right length and that it is a proper
1411                          * string of the right length.
1412                          */
1413                         if (((args[0].to - args[0].from) !=
1414                              OCFS2_STACK_LABEL_LEN) ||
1415                             (strnlen(args[0].from,
1416                                      OCFS2_STACK_LABEL_LEN) !=
1417                              OCFS2_STACK_LABEL_LEN)) {
1418                                 mlog(ML_ERROR,
1419                                      "Invalid cluster_stack option\n");
1420                                 status = 0;
1421                                 goto bail;
1422                         }
1423                         memcpy(mopt->cluster_stack, args[0].from,
1424                                OCFS2_STACK_LABEL_LEN);
1425                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1426                         /*
1427                          * Open code the memcmp here as we don't have
1428                          * an osb to pass to
1429                          * ocfs2_userspace_stack().
1430                          */
1431                         if (memcmp(mopt->cluster_stack,
1432                                    OCFS2_CLASSIC_CLUSTER_STACK,
1433                                    OCFS2_STACK_LABEL_LEN))
1434                                 user_stack = 1;
1435                         break;
1436                 case Opt_inode64:
1437                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1438                         break;
1439                 case Opt_usrquota:
1440                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1441                         break;
1442                 case Opt_grpquota:
1443                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1444                         break;
1445                 case Opt_coherency_buffered:
1446                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1447                         break;
1448                 case Opt_coherency_full:
1449                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1450                         break;
1451                 case Opt_acl:
1452                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1453                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1454                         break;
1455                 case Opt_noacl:
1456                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1457                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1458                         break;
1459                 case Opt_resv_level:
1460                         if (is_remount)
1461                                 break;
1462                         if (match_int(&args[0], &option)) {
1463                                 status = 0;
1464                                 goto bail;
1465                         }
1466                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1467                             option < OCFS2_MAX_RESV_LEVEL)
1468                                 mopt->resv_level = option;
1469                         break;
1470                 case Opt_dir_resv_level:
1471                         if (is_remount)
1472                                 break;
1473                         if (match_int(&args[0], &option)) {
1474                                 status = 0;
1475                                 goto bail;
1476                         }
1477                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1478                             option < OCFS2_MAX_RESV_LEVEL)
1479                                 mopt->dir_resv_level = option;
1480                         break;
1481                 case Opt_journal_async_commit:
1482                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1483                         break;
1484                 default:
1485                         mlog(ML_ERROR,
1486                              "Unrecognized mount option \"%s\" "
1487                              "or missing value\n", p);
1488                         status = 0;
1489                         goto bail;
1490                 }
1491         }
1492
1493         if (user_stack == 0) {
1494                 /* Ensure only one heartbeat mode */
1495                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1496                                          OCFS2_MOUNT_HB_GLOBAL |
1497                                          OCFS2_MOUNT_HB_NONE);
1498                 if (hweight32(tmp) != 1) {
1499                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1500                         status = 0;
1501                         goto bail;
1502                 }
1503         }
1504
1505         status = 1;
1506
1507 bail:
1508         return status;
1509 }
1510
1511 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1512 {
1513         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1514         unsigned long opts = osb->s_mount_opt;
1515         unsigned int local_alloc_megs;
1516
1517         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1518                 seq_printf(s, ",_netdev");
1519                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1520                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1521                 else
1522                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1523         } else
1524                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1525
1526         if (opts & OCFS2_MOUNT_NOINTR)
1527                 seq_printf(s, ",nointr");
1528
1529         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1530                 seq_printf(s, ",data=writeback");
1531         else
1532                 seq_printf(s, ",data=ordered");
1533
1534         if (opts & OCFS2_MOUNT_BARRIER)
1535                 seq_printf(s, ",barrier=1");
1536
1537         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1538                 seq_printf(s, ",errors=panic");
1539         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1540                 seq_printf(s, ",errors=continue");
1541         else
1542                 seq_printf(s, ",errors=remount-ro");
1543
1544         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1545                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1546
1547         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1548
1549         if (osb->osb_commit_interval)
1550                 seq_printf(s, ",commit=%u",
1551                            (unsigned) (osb->osb_commit_interval / HZ));
1552
1553         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1554         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1555                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1556
1557         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1558                 seq_printf(s, ",localflocks,");
1559
1560         if (osb->osb_cluster_stack[0])
1561                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1562                                   OCFS2_STACK_LABEL_LEN);
1563         if (opts & OCFS2_MOUNT_USRQUOTA)
1564                 seq_printf(s, ",usrquota");
1565         if (opts & OCFS2_MOUNT_GRPQUOTA)
1566                 seq_printf(s, ",grpquota");
1567
1568         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1569                 seq_printf(s, ",coherency=buffered");
1570         else
1571                 seq_printf(s, ",coherency=full");
1572
1573         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1574                 seq_printf(s, ",nouser_xattr");
1575         else
1576                 seq_printf(s, ",user_xattr");
1577
1578         if (opts & OCFS2_MOUNT_INODE64)
1579                 seq_printf(s, ",inode64");
1580
1581         if (opts & OCFS2_MOUNT_POSIX_ACL)
1582                 seq_printf(s, ",acl");
1583         else
1584                 seq_printf(s, ",noacl");
1585
1586         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1587                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1588
1589         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1590                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1591
1592         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1593                 seq_printf(s, ",journal_async_commit");
1594
1595         return 0;
1596 }
1597
1598 static int __init ocfs2_init(void)
1599 {
1600         int status;
1601
1602         status = init_ocfs2_uptodate_cache();
1603         if (status < 0)
1604                 goto out1;
1605
1606         status = ocfs2_initialize_mem_caches();
1607         if (status < 0)
1608                 goto out2;
1609
1610         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1611         if (!ocfs2_debugfs_root) {
1612                 status = -ENOMEM;
1613                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1614                 goto out3;
1615         }
1616
1617         ocfs2_set_locking_protocol();
1618
1619         status = register_quota_format(&ocfs2_quota_format);
1620         if (status < 0)
1621                 goto out3;
1622         status = register_filesystem(&ocfs2_fs_type);
1623         if (!status)
1624                 return 0;
1625
1626         unregister_quota_format(&ocfs2_quota_format);
1627 out3:
1628         debugfs_remove(ocfs2_debugfs_root);
1629         ocfs2_free_mem_caches();
1630 out2:
1631         exit_ocfs2_uptodate_cache();
1632 out1:
1633         mlog_errno(status);
1634         return status;
1635 }
1636
1637 static void __exit ocfs2_exit(void)
1638 {
1639         unregister_quota_format(&ocfs2_quota_format);
1640
1641         debugfs_remove(ocfs2_debugfs_root);
1642
1643         ocfs2_free_mem_caches();
1644
1645         unregister_filesystem(&ocfs2_fs_type);
1646
1647         exit_ocfs2_uptodate_cache();
1648 }
1649
1650 static void ocfs2_put_super(struct super_block *sb)
1651 {
1652         trace_ocfs2_put_super(sb);
1653
1654         ocfs2_sync_blockdev(sb);
1655         ocfs2_dismount_volume(sb, 0);
1656         ocfs2_filecheck_remove_sysfs(sb);
1657 }
1658
1659 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1660 {
1661         struct ocfs2_super *osb;
1662         u32 numbits, freebits;
1663         int status;
1664         struct ocfs2_dinode *bm_lock;
1665         struct buffer_head *bh = NULL;
1666         struct inode *inode = NULL;
1667
1668         trace_ocfs2_statfs(dentry->d_sb, buf);
1669
1670         osb = OCFS2_SB(dentry->d_sb);
1671
1672         inode = ocfs2_get_system_file_inode(osb,
1673                                             GLOBAL_BITMAP_SYSTEM_INODE,
1674                                             OCFS2_INVALID_SLOT);
1675         if (!inode) {
1676                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1677                 status = -EIO;
1678                 goto bail;
1679         }
1680
1681         status = ocfs2_inode_lock(inode, &bh, 0);
1682         if (status < 0) {
1683                 mlog_errno(status);
1684                 goto bail;
1685         }
1686
1687         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1688
1689         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1690         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1691
1692         buf->f_type = OCFS2_SUPER_MAGIC;
1693         buf->f_bsize = dentry->d_sb->s_blocksize;
1694         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1695         buf->f_blocks = ((sector_t) numbits) *
1696                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1697         buf->f_bfree = ((sector_t) freebits) *
1698                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1699         buf->f_bavail = buf->f_bfree;
1700         buf->f_files = numbits;
1701         buf->f_ffree = freebits;
1702         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1703                                 & 0xFFFFFFFFUL;
1704         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1705                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1706
1707         brelse(bh);
1708
1709         ocfs2_inode_unlock(inode, 0);
1710         status = 0;
1711 bail:
1712         iput(inode);
1713
1714         if (status)
1715                 mlog_errno(status);
1716
1717         return status;
1718 }
1719
1720 static void ocfs2_inode_init_once(void *data)
1721 {
1722         struct ocfs2_inode_info *oi = data;
1723
1724         oi->ip_flags = 0;
1725         oi->ip_open_count = 0;
1726         spin_lock_init(&oi->ip_lock);
1727         ocfs2_extent_map_init(&oi->vfs_inode);
1728         INIT_LIST_HEAD(&oi->ip_io_markers);
1729         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1730         oi->ip_dir_start_lookup = 0;
1731         init_rwsem(&oi->ip_alloc_sem);
1732         init_rwsem(&oi->ip_xattr_sem);
1733         mutex_init(&oi->ip_io_mutex);
1734
1735         oi->ip_blkno = 0ULL;
1736         oi->ip_clusters = 0;
1737
1738         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1739
1740         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1741         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1742         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1743
1744         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1745                                   &ocfs2_inode_caching_ops);
1746
1747         inode_init_once(&oi->vfs_inode);
1748 }
1749
1750 static int ocfs2_initialize_mem_caches(void)
1751 {
1752         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1753                                        sizeof(struct ocfs2_inode_info),
1754                                        0,
1755                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1756                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1757                                        ocfs2_inode_init_once);
1758         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1759                                         sizeof(struct ocfs2_dquot),
1760                                         0,
1761                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1762                                                 SLAB_MEM_SPREAD),
1763                                         NULL);
1764         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1765                                         sizeof(struct ocfs2_quota_chunk),
1766                                         0,
1767                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1768                                         NULL);
1769         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1770             !ocfs2_qf_chunk_cachep) {
1771                 if (ocfs2_inode_cachep)
1772                         kmem_cache_destroy(ocfs2_inode_cachep);
1773                 if (ocfs2_dquot_cachep)
1774                         kmem_cache_destroy(ocfs2_dquot_cachep);
1775                 if (ocfs2_qf_chunk_cachep)
1776                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1777                 return -ENOMEM;
1778         }
1779
1780         return 0;
1781 }
1782
1783 static void ocfs2_free_mem_caches(void)
1784 {
1785         /*
1786          * Make sure all delayed rcu free inodes are flushed before we
1787          * destroy cache.
1788          */
1789         rcu_barrier();
1790         if (ocfs2_inode_cachep)
1791                 kmem_cache_destroy(ocfs2_inode_cachep);
1792         ocfs2_inode_cachep = NULL;
1793
1794         if (ocfs2_dquot_cachep)
1795                 kmem_cache_destroy(ocfs2_dquot_cachep);
1796         ocfs2_dquot_cachep = NULL;
1797
1798         if (ocfs2_qf_chunk_cachep)
1799                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1800         ocfs2_qf_chunk_cachep = NULL;
1801 }
1802
1803 static int ocfs2_get_sector(struct super_block *sb,
1804                             struct buffer_head **bh,
1805                             int block,
1806                             int sect_size)
1807 {
1808         if (!sb_set_blocksize(sb, sect_size)) {
1809                 mlog(ML_ERROR, "unable to set blocksize\n");
1810                 return -EIO;
1811         }
1812
1813         *bh = sb_getblk(sb, block);
1814         if (!*bh) {
1815                 mlog_errno(-ENOMEM);
1816                 return -ENOMEM;
1817         }
1818         lock_buffer(*bh);
1819         if (!buffer_dirty(*bh))
1820                 clear_buffer_uptodate(*bh);
1821         unlock_buffer(*bh);
1822         ll_rw_block(REQ_OP_READ, 0, 1, bh);
1823         wait_on_buffer(*bh);
1824         if (!buffer_uptodate(*bh)) {
1825                 mlog_errno(-EIO);
1826                 brelse(*bh);
1827                 *bh = NULL;
1828                 return -EIO;
1829         }
1830
1831         return 0;
1832 }
1833
1834 static int ocfs2_mount_volume(struct super_block *sb)
1835 {
1836         int status = 0;
1837         int unlock_super = 0;
1838         struct ocfs2_super *osb = OCFS2_SB(sb);
1839
1840         if (ocfs2_is_hard_readonly(osb))
1841                 goto leave;
1842
1843         status = ocfs2_dlm_init(osb);
1844         if (status < 0) {
1845                 mlog_errno(status);
1846                 goto leave;
1847         }
1848
1849         status = ocfs2_super_lock(osb, 1);
1850         if (status < 0) {
1851                 mlog_errno(status);
1852                 goto leave;
1853         }
1854         unlock_super = 1;
1855
1856         /* This will load up the node map and add ourselves to it. */
1857         status = ocfs2_find_slot(osb);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto leave;
1861         }
1862
1863         /* load all node-local system inodes */
1864         status = ocfs2_init_local_system_inodes(osb);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 goto leave;
1868         }
1869
1870         status = ocfs2_check_volume(osb);
1871         if (status < 0) {
1872                 mlog_errno(status);
1873                 goto leave;
1874         }
1875
1876         status = ocfs2_truncate_log_init(osb);
1877         if (status < 0)
1878                 mlog_errno(status);
1879
1880 leave:
1881         if (unlock_super)
1882                 ocfs2_super_unlock(osb, 1);
1883
1884         return status;
1885 }
1886
1887 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1888 {
1889         int tmp, hangup_needed = 0;
1890         struct ocfs2_super *osb = NULL;
1891         char nodestr[12];
1892
1893         trace_ocfs2_dismount_volume(sb);
1894
1895         BUG_ON(!sb);
1896         osb = OCFS2_SB(sb);
1897         BUG_ON(!osb);
1898
1899         debugfs_remove(osb->osb_ctxt);
1900
1901         /* Orphan scan should be stopped as early as possible */
1902         ocfs2_orphan_scan_stop(osb);
1903
1904         ocfs2_disable_quotas(osb);
1905
1906         /* All dquots should be freed by now */
1907         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1908         /* Wait for worker to be done with the work structure in osb */
1909         cancel_work_sync(&osb->dquot_drop_work);
1910
1911         ocfs2_shutdown_local_alloc(osb);
1912
1913         ocfs2_truncate_log_shutdown(osb);
1914
1915         /* This will disable recovery and flush any recovery work. */
1916         ocfs2_recovery_exit(osb);
1917
1918         ocfs2_journal_shutdown(osb);
1919
1920         ocfs2_sync_blockdev(sb);
1921
1922         ocfs2_purge_refcount_trees(osb);
1923
1924         /* No cluster connection means we've failed during mount, so skip
1925          * all the steps which depended on that to complete. */
1926         if (osb->cconn) {
1927                 tmp = ocfs2_super_lock(osb, 1);
1928                 if (tmp < 0) {
1929                         mlog_errno(tmp);
1930                         return;
1931                 }
1932         }
1933
1934         if (osb->slot_num != OCFS2_INVALID_SLOT)
1935                 ocfs2_put_slot(osb);
1936
1937         if (osb->cconn)
1938                 ocfs2_super_unlock(osb, 1);
1939
1940         ocfs2_release_system_inodes(osb);
1941
1942         /*
1943          * If we're dismounting due to mount error, mount.ocfs2 will clean
1944          * up heartbeat.  If we're a local mount, there is no heartbeat.
1945          * If we failed before we got a uuid_str yet, we can't stop
1946          * heartbeat.  Otherwise, do it.
1947          */
1948         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1949             !ocfs2_is_hard_readonly(osb))
1950                 hangup_needed = 1;
1951
1952         if (osb->cconn)
1953                 ocfs2_dlm_shutdown(osb, hangup_needed);
1954
1955         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1956         debugfs_remove(osb->osb_debug_root);
1957
1958         if (hangup_needed)
1959                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1960
1961         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1962
1963         if (ocfs2_mount_local(osb))
1964                 snprintf(nodestr, sizeof(nodestr), "local");
1965         else
1966                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1967
1968         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1969                osb->dev_str, nodestr);
1970
1971         ocfs2_delete_osb(osb);
1972         kfree(osb);
1973         sb->s_dev = 0;
1974         sb->s_fs_info = NULL;
1975 }
1976
1977 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1978                                 unsigned uuid_bytes)
1979 {
1980         int i, ret;
1981         char *ptr;
1982
1983         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1984
1985         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1986         if (osb->uuid_str == NULL)
1987                 return -ENOMEM;
1988
1989         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1990                 /* print with null */
1991                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1992                 if (ret != 2) /* drop super cleans up */
1993                         return -EINVAL;
1994                 /* then only advance past the last char */
1995                 ptr += 2;
1996         }
1997
1998         return 0;
1999 }
2000
2001 /* Make sure entire volume is addressable by our journal.  Requires
2002    osb_clusters_at_boot to be valid and for the journal to have been
2003    initialized by ocfs2_journal_init(). */
2004 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2005 {
2006         int status = 0;
2007         u64 max_block =
2008                 ocfs2_clusters_to_blocks(osb->sb,
2009                                          osb->osb_clusters_at_boot) - 1;
2010
2011         /* 32-bit block number is always OK. */
2012         if (max_block <= (u32)~0ULL)
2013                 goto out;
2014
2015         /* Volume is "huge", so see if our journal is new enough to
2016            support it. */
2017         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2018                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2019               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2020                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2021                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2022                      "Enable the 'block64' journal option with tunefs.ocfs2");
2023                 status = -EFBIG;
2024                 goto out;
2025         }
2026
2027  out:
2028         return status;
2029 }
2030
2031 static int ocfs2_initialize_super(struct super_block *sb,
2032                                   struct buffer_head *bh,
2033                                   int sector_size,
2034                                   struct ocfs2_blockcheck_stats *stats)
2035 {
2036         int status;
2037         int i, cbits, bbits;
2038         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2039         struct inode *inode = NULL;
2040         struct ocfs2_journal *journal;
2041         struct ocfs2_super *osb;
2042         u64 total_blocks;
2043
2044         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2045         if (!osb) {
2046                 status = -ENOMEM;
2047                 mlog_errno(status);
2048                 goto bail;
2049         }
2050
2051         sb->s_fs_info = osb;
2052         sb->s_op = &ocfs2_sops;
2053         sb->s_d_op = &ocfs2_dentry_ops;
2054         sb->s_export_op = &ocfs2_export_ops;
2055         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2056         sb->dq_op = &ocfs2_quota_operations;
2057         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2058         sb->s_xattr = ocfs2_xattr_handlers;
2059         sb->s_time_gran = 1;
2060         sb->s_flags |= MS_NOATIME;
2061         /* this is needed to support O_LARGEFILE */
2062         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2063         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2064         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2065         memcpy(&sb->s_uuid, di->id2.i_super.s_uuid,
2066                sizeof(di->id2.i_super.s_uuid));
2067
2068         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2069
2070         for (i = 0; i < 3; i++)
2071                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2072         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2073
2074         osb->sb = sb;
2075         osb->s_sectsize_bits = blksize_bits(sector_size);
2076         BUG_ON(!osb->s_sectsize_bits);
2077
2078         spin_lock_init(&osb->dc_task_lock);
2079         init_waitqueue_head(&osb->dc_event);
2080         osb->dc_work_sequence = 0;
2081         osb->dc_wake_sequence = 0;
2082         INIT_LIST_HEAD(&osb->blocked_lock_list);
2083         osb->blocked_lock_count = 0;
2084         spin_lock_init(&osb->osb_lock);
2085         spin_lock_init(&osb->osb_xattr_lock);
2086         ocfs2_init_steal_slots(osb);
2087
2088         mutex_init(&osb->system_file_mutex);
2089
2090         atomic_set(&osb->alloc_stats.moves, 0);
2091         atomic_set(&osb->alloc_stats.local_data, 0);
2092         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2093         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2094         atomic_set(&osb->alloc_stats.bg_extends, 0);
2095
2096         /* Copy the blockcheck stats from the superblock probe */
2097         osb->osb_ecc_stats = *stats;
2098
2099         ocfs2_init_node_maps(osb);
2100
2101         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2102                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2103
2104         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2105         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2106                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2107                      osb->max_slots);
2108                 status = -EINVAL;
2109                 goto bail;
2110         }
2111
2112         ocfs2_orphan_scan_init(osb);
2113
2114         status = ocfs2_recovery_init(osb);
2115         if (status) {
2116                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2117                 mlog_errno(status);
2118                 goto bail;
2119         }
2120
2121         init_waitqueue_head(&osb->checkpoint_event);
2122
2123         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2124
2125         osb->slot_num = OCFS2_INVALID_SLOT;
2126
2127         osb->s_xattr_inline_size = le16_to_cpu(
2128                                         di->id2.i_super.s_xattr_inline_size);
2129
2130         osb->local_alloc_state = OCFS2_LA_UNUSED;
2131         osb->local_alloc_bh = NULL;
2132         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2133
2134         init_waitqueue_head(&osb->osb_mount_event);
2135
2136         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2137         if (status) {
2138                 mlog_errno(status);
2139                 goto bail;
2140         }
2141
2142         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2143         if (!osb->vol_label) {
2144                 mlog(ML_ERROR, "unable to alloc vol label\n");
2145                 status = -ENOMEM;
2146                 goto bail;
2147         }
2148
2149         osb->slot_recovery_generations =
2150                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2151                         GFP_KERNEL);
2152         if (!osb->slot_recovery_generations) {
2153                 status = -ENOMEM;
2154                 mlog_errno(status);
2155                 goto bail;
2156         }
2157
2158         init_waitqueue_head(&osb->osb_wipe_event);
2159         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2160                                         sizeof(*osb->osb_orphan_wipes),
2161                                         GFP_KERNEL);
2162         if (!osb->osb_orphan_wipes) {
2163                 status = -ENOMEM;
2164                 mlog_errno(status);
2165                 goto bail;
2166         }
2167
2168         osb->osb_rf_lock_tree = RB_ROOT;
2169
2170         osb->s_feature_compat =
2171                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2172         osb->s_feature_ro_compat =
2173                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2174         osb->s_feature_incompat =
2175                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2176
2177         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2178                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2179                      "optional features (%x).\n", i);
2180                 status = -EINVAL;
2181                 goto bail;
2182         }
2183         if (!(osb->sb->s_flags & MS_RDONLY) &&
2184             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2185                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2186                      "unsupported optional features (%x).\n", i);
2187                 status = -EINVAL;
2188                 goto bail;
2189         }
2190
2191         if (ocfs2_clusterinfo_valid(osb)) {
2192                 osb->osb_stackflags =
2193                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2194                 strlcpy(osb->osb_cluster_stack,
2195                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2196                        OCFS2_STACK_LABEL_LEN + 1);
2197                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2198                         mlog(ML_ERROR,
2199                              "couldn't mount because of an invalid "
2200                              "cluster stack label (%s) \n",
2201                              osb->osb_cluster_stack);
2202                         status = -EINVAL;
2203                         goto bail;
2204                 }
2205                 strlcpy(osb->osb_cluster_name,
2206                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2207                         OCFS2_CLUSTER_NAME_LEN + 1);
2208         } else {
2209                 /* The empty string is identical with classic tools that
2210                  * don't know about s_cluster_info. */
2211                 osb->osb_cluster_stack[0] = '\0';
2212         }
2213
2214         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2215
2216         /* FIXME
2217          * This should be done in ocfs2_journal_init(), but unknown
2218          * ordering issues will cause the filesystem to crash.
2219          * If anyone wants to figure out what part of the code
2220          * refers to osb->journal before ocfs2_journal_init() is run,
2221          * be my guest.
2222          */
2223         /* initialize our journal structure */
2224
2225         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2226         if (!journal) {
2227                 mlog(ML_ERROR, "unable to alloc journal\n");
2228                 status = -ENOMEM;
2229                 goto bail;
2230         }
2231         osb->journal = journal;
2232         journal->j_osb = osb;
2233
2234         atomic_set(&journal->j_num_trans, 0);
2235         init_rwsem(&journal->j_trans_barrier);
2236         init_waitqueue_head(&journal->j_checkpointed);
2237         spin_lock_init(&journal->j_lock);
2238         journal->j_trans_id = (unsigned long) 1;
2239         INIT_LIST_HEAD(&journal->j_la_cleanups);
2240         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2241         journal->j_state = OCFS2_JOURNAL_FREE;
2242
2243         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2244         init_llist_head(&osb->dquot_drop_list);
2245
2246         /* get some pseudo constants for clustersize bits */
2247         osb->s_clustersize_bits =
2248                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2249         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2250
2251         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2252             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2253                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2254                      osb->s_clustersize);
2255                 status = -EINVAL;
2256                 goto bail;
2257         }
2258
2259         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2260                                                 le32_to_cpu(di->i_clusters));
2261
2262         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2263                                            total_blocks);
2264         if (status) {
2265                 mlog(ML_ERROR, "Volume too large "
2266                      "to mount safely on this system");
2267                 status = -EFBIG;
2268                 goto bail;
2269         }
2270
2271         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2272                                  sizeof(di->id2.i_super.s_uuid))) {
2273                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2274                 status = -ENOMEM;
2275                 goto bail;
2276         }
2277
2278         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2279                 OCFS2_MAX_VOL_LABEL_LEN);
2280         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2281         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2282         osb->first_cluster_group_blkno =
2283                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2284         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2285         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2286         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2287                                      (unsigned long long)osb->root_blkno,
2288                                      (unsigned long long)osb->system_dir_blkno,
2289                                      osb->s_clustersize_bits);
2290
2291         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2292         if (!osb->osb_dlm_debug) {
2293                 status = -ENOMEM;
2294                 mlog_errno(status);
2295                 goto bail;
2296         }
2297
2298         atomic_set(&osb->vol_state, VOLUME_INIT);
2299
2300         /* load root, system_dir, and all global system inodes */
2301         status = ocfs2_init_global_system_inodes(osb);
2302         if (status < 0) {
2303                 mlog_errno(status);
2304                 goto bail;
2305         }
2306
2307         /*
2308          * global bitmap
2309          */
2310         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2311                                             OCFS2_INVALID_SLOT);
2312         if (!inode) {
2313                 status = -EINVAL;
2314                 mlog_errno(status);
2315                 goto bail;
2316         }
2317
2318         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2319         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2320         iput(inode);
2321
2322         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2323                                  osb->s_feature_incompat) * 8;
2324
2325         status = ocfs2_init_slot_info(osb);
2326         if (status < 0) {
2327                 mlog_errno(status);
2328                 goto bail;
2329         }
2330         cleancache_init_shared_fs(sb);
2331
2332         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2333         if (!osb->ocfs2_wq) {
2334                 status = -ENOMEM;
2335                 mlog_errno(status);
2336         }
2337
2338 bail:
2339         return status;
2340 }
2341
2342 /*
2343  * will return: -EAGAIN if it is ok to keep searching for superblocks
2344  *              -EINVAL if there is a bad superblock
2345  *              0 on success
2346  */
2347 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2348                                struct buffer_head *bh,
2349                                u32 blksz,
2350                                struct ocfs2_blockcheck_stats *stats)
2351 {
2352         int status = -EAGAIN;
2353
2354         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2355                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2356                 /* We have to do a raw check of the feature here */
2357                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2358                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2359                         status = ocfs2_block_check_validate(bh->b_data,
2360                                                             bh->b_size,
2361                                                             &di->i_check,
2362                                                             stats);
2363                         if (status)
2364                                 goto out;
2365                 }
2366                 status = -EINVAL;
2367                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2368                         mlog(ML_ERROR, "found superblock with incorrect block "
2369                              "size: found %u, should be %u\n",
2370                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2371                                blksz);
2372                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2373                            OCFS2_MAJOR_REV_LEVEL ||
2374                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2375                            OCFS2_MINOR_REV_LEVEL) {
2376                         mlog(ML_ERROR, "found superblock with bad version: "
2377                              "found %u.%u, should be %u.%u\n",
2378                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2379                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2380                              OCFS2_MAJOR_REV_LEVEL,
2381                              OCFS2_MINOR_REV_LEVEL);
2382                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2383                         mlog(ML_ERROR, "bad block number on superblock: "
2384                              "found %llu, should be %llu\n",
2385                              (unsigned long long)le64_to_cpu(di->i_blkno),
2386                              (unsigned long long)bh->b_blocknr);
2387                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2388                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2389                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2390                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2391                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2392                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2393                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2394                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2395                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2396                         mlog(ML_ERROR,
2397                              "Superblock slots found greater than file system "
2398                              "maximum: found %u, max %u\n",
2399                              le16_to_cpu(di->id2.i_super.s_max_slots),
2400                              OCFS2_MAX_SLOTS);
2401                 } else {
2402                         /* found it! */
2403                         status = 0;
2404                 }
2405         }
2406
2407 out:
2408         if (status && status != -EAGAIN)
2409                 mlog_errno(status);
2410         return status;
2411 }
2412
2413 static int ocfs2_check_volume(struct ocfs2_super *osb)
2414 {
2415         int status;
2416         int dirty;
2417         int local;
2418         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2419                                                   * recover
2420                                                   * ourselves. */
2421
2422         /* Init our journal object. */
2423         status = ocfs2_journal_init(osb->journal, &dirty);
2424         if (status < 0) {
2425                 mlog(ML_ERROR, "Could not initialize journal!\n");
2426                 goto finally;
2427         }
2428
2429         /* Now that journal has been initialized, check to make sure
2430            entire volume is addressable. */
2431         status = ocfs2_journal_addressable(osb);
2432         if (status)
2433                 goto finally;
2434
2435         /* If the journal was unmounted cleanly then we don't want to
2436          * recover anything. Otherwise, journal_load will do that
2437          * dirty work for us :) */
2438         if (!dirty) {
2439                 status = ocfs2_journal_wipe(osb->journal, 0);
2440                 if (status < 0) {
2441                         mlog_errno(status);
2442                         goto finally;
2443                 }
2444         } else {
2445                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2446                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2447         }
2448
2449         local = ocfs2_mount_local(osb);
2450
2451         /* will play back anything left in the journal. */
2452         status = ocfs2_journal_load(osb->journal, local, dirty);
2453         if (status < 0) {
2454                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2455                 goto finally;
2456         }
2457
2458         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2459                 jbd2_journal_set_features(osb->journal->j_journal,
2460                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2461                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2462         else
2463                 jbd2_journal_clear_features(osb->journal->j_journal,
2464                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2465                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2466
2467         if (dirty) {
2468                 /* recover my local alloc if we didn't unmount cleanly. */
2469                 status = ocfs2_begin_local_alloc_recovery(osb,
2470                                                           osb->slot_num,
2471                                                           &local_alloc);
2472                 if (status < 0) {
2473                         mlog_errno(status);
2474                         goto finally;
2475                 }
2476                 /* we complete the recovery process after we've marked
2477                  * ourselves as mounted. */
2478         }
2479
2480         status = ocfs2_load_local_alloc(osb);
2481         if (status < 0) {
2482                 mlog_errno(status);
2483                 goto finally;
2484         }
2485
2486         if (dirty) {
2487                 /* Recovery will be completed after we've mounted the
2488                  * rest of the volume. */
2489                 osb->dirty = 1;
2490                 osb->local_alloc_copy = local_alloc;
2491                 local_alloc = NULL;
2492         }
2493
2494         /* go through each journal, trylock it and if you get the
2495          * lock, and it's marked as dirty, set the bit in the recover
2496          * map and launch a recovery thread for it. */
2497         status = ocfs2_mark_dead_nodes(osb);
2498         if (status < 0) {
2499                 mlog_errno(status);
2500                 goto finally;
2501         }
2502
2503         status = ocfs2_compute_replay_slots(osb);
2504         if (status < 0)
2505                 mlog_errno(status);
2506
2507 finally:
2508         kfree(local_alloc);
2509
2510         if (status)
2511                 mlog_errno(status);
2512         return status;
2513 }
2514
2515 /*
2516  * The routine gets called from dismount or close whenever a dismount on
2517  * volume is requested and the osb open count becomes 1.
2518  * It will remove the osb from the global list and also free up all the
2519  * initialized resources and fileobject.
2520  */
2521 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2522 {
2523         /* This function assumes that the caller has the main osb resource */
2524
2525         /* ocfs2_initializer_super have already created this workqueue */
2526         if (osb->ocfs2_wq) {
2527                 flush_workqueue(osb->ocfs2_wq);
2528                 destroy_workqueue(osb->ocfs2_wq);
2529         }
2530
2531         ocfs2_free_slot_info(osb);
2532
2533         kfree(osb->osb_orphan_wipes);
2534         kfree(osb->slot_recovery_generations);
2535         /* FIXME
2536          * This belongs in journal shutdown, but because we have to
2537          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2538          * we free it here.
2539          */
2540         kfree(osb->journal);
2541         kfree(osb->local_alloc_copy);
2542         kfree(osb->uuid_str);
2543         kfree(osb->vol_label);
2544         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2545         memset(osb, 0, sizeof(struct ocfs2_super));
2546 }
2547
2548 /* Depending on the mount option passed, perform one of the following:
2549  * Put OCFS2 into a readonly state (default)
2550  * Return EIO so that only the process errs
2551  * Fix the error as if fsck.ocfs2 -y
2552  * panic
2553  */
2554 static int ocfs2_handle_error(struct super_block *sb)
2555 {
2556         struct ocfs2_super *osb = OCFS2_SB(sb);
2557         int rv = 0;
2558
2559         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2560         pr_crit("On-disk corruption discovered. "
2561                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2562
2563         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2564                 panic("OCFS2: (device %s): panic forced after error\n",
2565                       sb->s_id);
2566         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2567                 pr_crit("OCFS2: Returning error to the calling process.\n");
2568                 rv = -EIO;
2569         } else { /* default option */
2570                 rv = -EROFS;
2571                 if (sb->s_flags & MS_RDONLY &&
2572                                 (ocfs2_is_soft_readonly(osb) ||
2573                                  ocfs2_is_hard_readonly(osb)))
2574                         return rv;
2575
2576                 pr_crit("OCFS2: File system is now read-only.\n");
2577                 sb->s_flags |= MS_RDONLY;
2578                 ocfs2_set_ro_flag(osb, 0);
2579         }
2580
2581         return rv;
2582 }
2583
2584 int __ocfs2_error(struct super_block *sb, const char *function,
2585                   const char *fmt, ...)
2586 {
2587         struct va_format vaf;
2588         va_list args;
2589
2590         va_start(args, fmt);
2591         vaf.fmt = fmt;
2592         vaf.va = &args;
2593
2594         /* Not using mlog here because we want to show the actual
2595          * function the error came from. */
2596         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2597                sb->s_id, function, &vaf);
2598
2599         va_end(args);
2600
2601         return ocfs2_handle_error(sb);
2602 }
2603
2604 /* Handle critical errors. This is intentionally more drastic than
2605  * ocfs2_handle_error, so we only use for things like journal errors,
2606  * etc. */
2607 void __ocfs2_abort(struct super_block *sb, const char *function,
2608                    const char *fmt, ...)
2609 {
2610         struct va_format vaf;
2611         va_list args;
2612
2613         va_start(args, fmt);
2614
2615         vaf.fmt = fmt;
2616         vaf.va = &args;
2617
2618         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2619                sb->s_id, function, &vaf);
2620
2621         va_end(args);
2622
2623         /* We don't have the cluster support yet to go straight to
2624          * hard readonly in here. Until then, we want to keep
2625          * ocfs2_abort() so that we can at least mark critical
2626          * errors.
2627          *
2628          * TODO: This should abort the journal and alert other nodes
2629          * that our slot needs recovery. */
2630
2631         /* Force a panic(). This stinks, but it's better than letting
2632          * things continue without having a proper hard readonly
2633          * here. */
2634         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2635                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2636         ocfs2_handle_error(sb);
2637 }
2638
2639 /*
2640  * Void signal blockers, because in-kernel sigprocmask() only fails
2641  * when SIG_* is wrong.
2642  */
2643 void ocfs2_block_signals(sigset_t *oldset)
2644 {
2645         int rc;
2646         sigset_t blocked;
2647
2648         sigfillset(&blocked);
2649         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2650         BUG_ON(rc);
2651 }
2652
2653 void ocfs2_unblock_signals(sigset_t *oldset)
2654 {
2655         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2656         BUG_ON(rc);
2657 }
2658
2659 module_init(ocfs2_init);
2660 module_exit(ocfs2_exit);