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
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "blockcheck.h"
  58 #include "dlmglue.h"
  59 #include "export.h"
  60 #include "extent_map.h"
  61 #include "heartbeat.h"
  62 #include "inode.h"
  63 #include "journal.h"
  64 #include "localalloc.h"
  65 #include "namei.h"
  66 #include "slot_map.h"
  67 #include "super.h"
  68 #include "sysfile.h"
  69 #include "uptodate.h"
  70 #include "ver.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root = NULL;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92
  93 struct mount_options
  94 {
  95         unsigned long   commit_interval;
  96         unsigned long   mount_opt;
  97         unsigned int    atime_quantum;
  98         signed short    slot;
  99         int             localalloc_opt;
 100         unsigned int    resv_level;
 101         int             dir_resv_level;
 102         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 103 };
 104
 105 static int ocfs2_parse_options(struct super_block *sb, char *options,
 106                                struct mount_options *mopt,
 107                                int is_remount);
 108 static int ocfs2_check_set_options(struct super_block *sb,
 109                                    struct mount_options *options);
 110 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
 111 static void ocfs2_put_super(struct super_block *sb);
 112 static int ocfs2_mount_volume(struct super_block *sb);
 113 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 114 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 115 static int ocfs2_initialize_mem_caches(void);
 116 static void ocfs2_free_mem_caches(void);
 117 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 118
 119 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 120
 121 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 122
 123 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 124 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 125 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 126 static int ocfs2_check_volume(struct ocfs2_super *osb);
 127 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 128                                struct buffer_head *bh,
 129                                u32 sectsize,
 130                                struct ocfs2_blockcheck_stats *stats);
 131 static int ocfs2_initialize_super(struct super_block *sb,
 132                                   struct buffer_head *bh,
 133                                   int sector_size,
 134                                   struct ocfs2_blockcheck_stats *stats);
 135 static int ocfs2_get_sector(struct super_block *sb,
 136                             struct buffer_head **bh,
 137                             int block,
 138                             int sect_size);
 139 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 140 static void ocfs2_destroy_inode(struct inode *inode);
 141 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 142 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 143 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 144
 145 static const struct super_operations ocfs2_sops = {
 146         .statfs         = ocfs2_statfs,
 147         .alloc_inode    = ocfs2_alloc_inode,
 148         .destroy_inode  = ocfs2_destroy_inode,
 149         .drop_inode     = ocfs2_drop_inode,
 150         .evict_inode    = ocfs2_evict_inode,
 151         .sync_fs        = ocfs2_sync_fs,
 152         .put_super      = ocfs2_put_super,
 153         .remount_fs     = ocfs2_remount,
 154         .show_options   = ocfs2_show_options,
 155         .quota_read     = ocfs2_quota_read,
 156         .quota_write    = ocfs2_quota_write,
 157 };
 158
 159 enum {
 160         Opt_barrier,
 161         Opt_err_panic,
 162         Opt_err_ro,
 163         Opt_intr,
 164         Opt_nointr,
 165         Opt_hb_none,
 166         Opt_hb_local,
 167         Opt_hb_global,
 168         Opt_data_ordered,
 169         Opt_data_writeback,
 170         Opt_atime_quantum,
 171         Opt_slot,
 172         Opt_commit,
 173         Opt_localalloc,
 174         Opt_localflocks,
 175         Opt_stack,
 176         Opt_user_xattr,
 177         Opt_nouser_xattr,
 178         Opt_inode64,
 179         Opt_acl,
 180         Opt_noacl,
 181         Opt_usrquota,
 182         Opt_grpquota,
 183         Opt_coherency_buffered,
 184         Opt_coherency_full,
 185         Opt_resv_level,
 186         Opt_dir_resv_level,
 187         Opt_err,
 188 };
 189
 190 static const match_table_t tokens = {
 191         {Opt_barrier, "barrier=%u"},
 192         {Opt_err_panic, "errors=panic"},
 193         {Opt_err_ro, "errors=remount-ro"},
 194         {Opt_intr, "intr"},
 195         {Opt_nointr, "nointr"},
 196         {Opt_hb_none, OCFS2_HB_NONE},
 197         {Opt_hb_local, OCFS2_HB_LOCAL},
 198         {Opt_hb_global, OCFS2_HB_GLOBAL},
 199         {Opt_data_ordered, "data=ordered"},
 200         {Opt_data_writeback, "data=writeback"},
 201         {Opt_atime_quantum, "atime_quantum=%u"},
 202         {Opt_slot, "preferred_slot=%u"},
 203         {Opt_commit, "commit=%u"},
 204         {Opt_localalloc, "localalloc=%d"},
 205         {Opt_localflocks, "localflocks"},
 206         {Opt_stack, "cluster_stack=%s"},
 207         {Opt_user_xattr, "user_xattr"},
 208         {Opt_nouser_xattr, "nouser_xattr"},
 209         {Opt_inode64, "inode64"},
 210         {Opt_acl, "acl"},
 211         {Opt_noacl, "noacl"},
 212         {Opt_usrquota, "usrquota"},
 213         {Opt_grpquota, "grpquota"},
 214         {Opt_coherency_buffered, "coherency=buffered"},
 215         {Opt_coherency_full, "coherency=full"},
 216         {Opt_resv_level, "resv_level=%u"},
 217         {Opt_dir_resv_level, "dir_resv_level=%u"},
 218         {Opt_err, NULL}
 219 };
 220
 221 #ifdef CONFIG_DEBUG_FS
 222 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 223 {
 224         struct ocfs2_cluster_connection *cconn = osb->cconn;
 225         struct ocfs2_recovery_map *rm = osb->recovery_map;
 226         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 227         int i, out = 0;
 228
 229         out += snprintf(buf + out, len - out,
 230                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 231                         "Device", osb->dev_str, osb->uuid_str,
 232                         osb->fs_generation, osb->vol_label);
 233
 234         out += snprintf(buf + out, len - out,
 235                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 236                         atomic_read(&osb->vol_state), osb->osb_flags);
 237
 238         out += snprintf(buf + out, len - out,
 239                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 240                         osb->sb->s_blocksize, osb->s_clustersize);
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 244                         "ROcompat: 0x%X\n",
 245                         "Features", osb->s_feature_compat,
 246                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 247
 248         out += snprintf(buf + out, len - out,
 249                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 250                         osb->s_mount_opt, osb->s_atime_quantum);
 251
 252         if (cconn) {
 253                 out += snprintf(buf + out, len - out,
 254                                 "%10s => Stack: %s  Name: %*s  "
 255                                 "Version: %d.%d\n", "Cluster",
 256                                 (*osb->osb_cluster_stack == '\0' ?
 257                                  "o2cb" : osb->osb_cluster_stack),
 258                                 cconn->cc_namelen, cconn->cc_name,
 259                                 cconn->cc_version.pv_major,
 260                                 cconn->cc_version.pv_minor);
 261         }
 262
 263         spin_lock(&osb->dc_task_lock);
 264         out += snprintf(buf + out, len - out,
 265                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 266                         "WorkSeq: %lu\n", "DownCnvt",
 267                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 268                         osb->blocked_lock_count, osb->dc_wake_sequence,
 269                         osb->dc_work_sequence);
 270         spin_unlock(&osb->dc_task_lock);
 271
 272         spin_lock(&osb->osb_lock);
 273         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 274                         "Recovery",
 275                         (osb->recovery_thread_task ?
 276                          task_pid_nr(osb->recovery_thread_task) : -1));
 277         if (rm->rm_used == 0)
 278                 out += snprintf(buf + out, len - out, " None\n");
 279         else {
 280                 for (i = 0; i < rm->rm_used; i++)
 281                         out += snprintf(buf + out, len - out, " %d",
 282                                         rm->rm_entries[i]);
 283                 out += snprintf(buf + out, len - out, "\n");
 284         }
 285         spin_unlock(&osb->osb_lock);
 286
 287         out += snprintf(buf + out, len - out,
 288                         "%10s => Pid: %d  Interval: %lu  Needs: %d\n", "Commit",
 289                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 290                         osb->osb_commit_interval,
 291                         atomic_read(&osb->needs_checkpoint));
 292
 293         out += snprintf(buf + out, len - out,
 294                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 295                         "Journal", osb->journal->j_state,
 296                         osb->journal->j_trans_id,
 297                         atomic_read(&osb->journal->j_num_trans));
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 301                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 302                         "Stats",
 303                         atomic_read(&osb->alloc_stats.bitmap_data),
 304                         atomic_read(&osb->alloc_stats.local_data),
 305                         atomic_read(&osb->alloc_stats.bg_allocs),
 306                         atomic_read(&osb->alloc_stats.moves),
 307                         atomic_read(&osb->alloc_stats.bg_extends));
 308
 309         out += snprintf(buf + out, len - out,
 310                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 311                         "Default: %u bits\n",
 312                         "LocalAlloc", osb->local_alloc_state,
 313                         (unsigned long long)osb->la_last_gd,
 314                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 315
 316         spin_lock(&osb->osb_lock);
 317         out += snprintf(buf + out, len - out,
 318                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 319                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 320                         osb->s_inode_steal_slot,
 321                         atomic_read(&osb->s_num_inodes_stolen),
 322                         osb->s_meta_steal_slot,
 323                         atomic_read(&osb->s_num_meta_stolen));
 324         spin_unlock(&osb->osb_lock);
 325
 326         out += snprintf(buf + out, len - out, "OrphanScan => ");
 327         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 328                         os->os_count, os->os_seqno);
 329         out += snprintf(buf + out, len - out, " Last Scan: ");
 330         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 331                 out += snprintf(buf + out, len - out, "Disabled\n");
 332         else
 333                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 334                                 (get_seconds() - os->os_scantime.tv_sec));
 335
 336         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 337                         "Slots", "Num", "RecoGen");
 338         for (i = 0; i < osb->max_slots; ++i) {
 339                 out += snprintf(buf + out, len - out,
 340                                 "%10s  %c %3d  %10d\n",
 341                                 " ",
 342                                 (i == osb->slot_num ? '*' : ' '),
 343                                 i, osb->slot_recovery_generations[i]);
 344         }
 345
 346         return out;
 347 }
 348
 349 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 350 {
 351         struct ocfs2_super *osb = inode->i_private;
 352         char *buf = NULL;
 353
 354         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 355         if (!buf)
 356                 goto bail;
 357
 358         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 359
 360         file->private_data = buf;
 361
 362         return 0;
 363 bail:
 364         return -ENOMEM;
 365 }
 366
 367 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 368 {
 369         kfree(file->private_data);
 370         return 0;
 371 }
 372
 373 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 374                                 size_t nbytes, loff_t *ppos)
 375 {
 376         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 377                                        i_size_read(file->f_mapping->host));
 378 }
 379 #else
 380 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 381 {
 382         return 0;
 383 }
 384 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 385 {
 386         return 0;
 387 }
 388 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 389                                 size_t nbytes, loff_t *ppos)
 390 {
 391         return 0;
 392 }
 393 #endif  /* CONFIG_DEBUG_FS */
 394
 395 static const struct file_operations ocfs2_osb_debug_fops = {
 396         .open =         ocfs2_osb_debug_open,
 397         .release =      ocfs2_debug_release,
 398         .read =         ocfs2_debug_read,
 399         .llseek =       generic_file_llseek,
 400 };
 401
 402 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 403 {
 404         int status;
 405         tid_t target;
 406         struct ocfs2_super *osb = OCFS2_SB(sb);
 407
 408         if (ocfs2_is_hard_readonly(osb))
 409                 return -EROFS;
 410
 411         if (wait) {
 412                 status = ocfs2_flush_truncate_log(osb);
 413                 if (status < 0)
 414                         mlog_errno(status);
 415         } else {
 416                 ocfs2_schedule_truncate_log_flush(osb, 0);
 417         }
 418
 419         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 420                                       &target)) {
 421                 if (wait)
 422                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 423                                              target);
 424         }
 425         return 0;
 426 }
 427
 428 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 429 {
 430         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 431             && (ino == USER_QUOTA_SYSTEM_INODE
 432                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 433                 return 0;
 434         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 435             && (ino == GROUP_QUOTA_SYSTEM_INODE
 436                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 437                 return 0;
 438         return 1;
 439 }
 440
 441 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 442 {
 443         struct inode *new = NULL;
 444         int status = 0;
 445         int i;
 446
 447         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 448         if (IS_ERR(new)) {
 449                 status = PTR_ERR(new);
 450                 mlog_errno(status);
 451                 goto bail;
 452         }
 453         osb->root_inode = new;
 454
 455         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 456         if (IS_ERR(new)) {
 457                 status = PTR_ERR(new);
 458                 mlog_errno(status);
 459                 goto bail;
 460         }
 461         osb->sys_root_inode = new;
 462
 463         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 464              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 465                 if (!ocfs2_need_system_inode(osb, i))
 466                         continue;
 467                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 468                 if (!new) {
 469                         ocfs2_release_system_inodes(osb);
 470                         status = -EINVAL;
 471                         mlog_errno(status);
 472                         /* FIXME: Should ERROR_RO_FS */
 473                         mlog(ML_ERROR, "Unable to load system inode %d, "
 474                              "possibly corrupt fs?", i);
 475                         goto bail;
 476                 }
 477                 // the array now has one ref, so drop this one
 478                 iput(new);
 479         }
 480
 481 bail:
 482         if (status)
 483                 mlog_errno(status);
 484         return status;
 485 }
 486
 487 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 488 {
 489         struct inode *new = NULL;
 490         int status = 0;
 491         int i;
 492
 493         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 494              i < NUM_SYSTEM_INODES;
 495              i++) {
 496                 if (!ocfs2_need_system_inode(osb, i))
 497                         continue;
 498                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 499                 if (!new) {
 500                         ocfs2_release_system_inodes(osb);
 501                         status = -EINVAL;
 502                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 503                              status, i, osb->slot_num);
 504                         goto bail;
 505                 }
 506                 /* the array now has one ref, so drop this one */
 507                 iput(new);
 508         }
 509
 510 bail:
 511         if (status)
 512                 mlog_errno(status);
 513         return status;
 514 }
 515
 516 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 517 {
 518         int i;
 519         struct inode *inode;
 520
 521         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 522                 inode = osb->global_system_inodes[i];
 523                 if (inode) {
 524                         iput(inode);
 525                         osb->global_system_inodes[i] = NULL;
 526                 }
 527         }
 528
 529         inode = osb->sys_root_inode;
 530         if (inode) {
 531                 iput(inode);
 532                 osb->sys_root_inode = NULL;
 533         }
 534
 535         inode = osb->root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->root_inode = NULL;
 539         }
 540
 541         if (!osb->local_system_inodes)
 542                 return;
 543
 544         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 545                 if (osb->local_system_inodes[i]) {
 546                         iput(osb->local_system_inodes[i]);
 547                         osb->local_system_inodes[i] = NULL;
 548                 }
 549         }
 550
 551         kfree(osb->local_system_inodes);
 552         osb->local_system_inodes = NULL;
 553 }
 554
 555 /* We're allocating fs objects, use GFP_NOFS */
 556 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 557 {
 558         struct ocfs2_inode_info *oi;
 559
 560         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 561         if (!oi)
 562                 return NULL;
 563
 564         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 565         return &oi->vfs_inode;
 566 }
 567
 568 static void ocfs2_i_callback(struct rcu_head *head)
 569 {
 570         struct inode *inode = container_of(head, struct inode, i_rcu);
 571         INIT_LIST_HEAD(&inode->i_dentry);
 572         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 573 }
 574
 575 static void ocfs2_destroy_inode(struct inode *inode)
 576 {
 577         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 578 }
 579
 580 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 581                                                 unsigned int cbits)
 582 {
 583         unsigned int bytes = 1 << cbits;
 584         unsigned int trim = bytes;
 585         unsigned int bitshift = 32;
 586
 587         /*
 588          * i_size and all block offsets in ocfs2 are always 64 bits
 589          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 590          * 64 bit platforms, cluster size will be the limiting factor.
 591          */
 592
 593 #if BITS_PER_LONG == 32
 594 # if defined(CONFIG_LBDAF)
 595         BUILD_BUG_ON(sizeof(sector_t) != 8);
 596         /*
 597          * We might be limited by page cache size.
 598          */
 599         if (bytes > PAGE_CACHE_SIZE) {
 600                 bytes = PAGE_CACHE_SIZE;
 601                 trim = 1;
 602                 /*
 603                  * Shift by 31 here so that we don't get larger than
 604                  * MAX_LFS_FILESIZE
 605                  */
 606                 bitshift = 31;
 607         }
 608 # else
 609         /*
 610          * We are limited by the size of sector_t. Use block size, as
 611          * that's what we expose to the VFS.
 612          */
 613         bytes = 1 << bbits;
 614         trim = 1;
 615         bitshift = 31;
 616 # endif
 617 #endif
 618
 619         /*
 620          * Trim by a whole cluster when we can actually approach the
 621          * on-disk limits. Otherwise we can overflow i_clusters when
 622          * an extent start is at the max offset.
 623          */
 624         return (((unsigned long long)bytes) << bitshift) - trim;
 625 }
 626
 627 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 628 {
 629         int incompat_features;
 630         int ret = 0;
 631         struct mount_options parsed_options;
 632         struct ocfs2_super *osb = OCFS2_SB(sb);
 633         u32 tmp;
 634
 635         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 636             !ocfs2_check_set_options(sb, &parsed_options)) {
 637                 ret = -EINVAL;
 638                 goto out;
 639         }
 640
 641         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 642                 OCFS2_MOUNT_HB_NONE;
 643         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 644                 ret = -EINVAL;
 645                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 646                 goto out;
 647         }
 648
 649         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 650             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 651                 ret = -EINVAL;
 652                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 653                 goto out;
 654         }
 655
 656         /* Probably don't want this on remount; it might
 657          * mess with other nodes */
 658         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 659             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 660                 ret = -EINVAL;
 661                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 662                 goto out;
 663         }
 664
 665         /* We're going to/from readonly mode. */
 666         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 667                 /* Disable quota accounting before remounting RO */
 668                 if (*flags & MS_RDONLY) {
 669                         ret = ocfs2_susp_quotas(osb, 0);
 670                         if (ret < 0)
 671                                 goto out;
 672                 }
 673                 /* Lock here so the check of HARD_RO and the potential
 674                  * setting of SOFT_RO is atomic. */
 675                 spin_lock(&osb->osb_lock);
 676                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 677                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 678                         ret = -EROFS;
 679                         goto unlock_osb;
 680                 }
 681
 682                 if (*flags & MS_RDONLY) {
 683                         sb->s_flags |= MS_RDONLY;
 684                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 685                 } else {
 686                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 687                                 mlog(ML_ERROR, "Cannot remount RDWR "
 688                                      "filesystem due to previous errors.\n");
 689                                 ret = -EROFS;
 690                                 goto unlock_osb;
 691                         }
 692                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 693                         if (incompat_features) {
 694                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 695                                      "of unsupported optional features "
 696                                      "(%x).\n", incompat_features);
 697                                 ret = -EINVAL;
 698                                 goto unlock_osb;
 699                         }
 700                         sb->s_flags &= ~MS_RDONLY;
 701                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 702                 }
 703                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 704 unlock_osb:
 705                 spin_unlock(&osb->osb_lock);
 706                 /* Enable quota accounting after remounting RW */
 707                 if (!ret && !(*flags & MS_RDONLY)) {
 708                         if (sb_any_quota_suspended(sb))
 709                                 ret = ocfs2_susp_quotas(osb, 1);
 710                         else
 711                                 ret = ocfs2_enable_quotas(osb);
 712                         if (ret < 0) {
 713                                 /* Return back changes... */
 714                                 spin_lock(&osb->osb_lock);
 715                                 sb->s_flags |= MS_RDONLY;
 716                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 717                                 spin_unlock(&osb->osb_lock);
 718                                 goto out;
 719                         }
 720                 }
 721         }
 722
 723         if (!ret) {
 724                 /* Only save off the new mount options in case of a successful
 725                  * remount. */
 726                 osb->s_mount_opt = parsed_options.mount_opt;
 727                 osb->s_atime_quantum = parsed_options.atime_quantum;
 728                 osb->preferred_slot = parsed_options.slot;
 729                 if (parsed_options.commit_interval)
 730                         osb->osb_commit_interval = parsed_options.commit_interval;
 731
 732                 if (!ocfs2_is_hard_readonly(osb))
 733                         ocfs2_set_journal_params(osb);
 734
 735                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 736                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 737                                                         MS_POSIXACL : 0);
 738         }
 739 out:
 740         return ret;
 741 }
 742
 743 static int ocfs2_sb_probe(struct super_block *sb,
 744                           struct buffer_head **bh,
 745                           int *sector_size,
 746                           struct ocfs2_blockcheck_stats *stats)
 747 {
 748         int status, tmpstat;
 749         struct ocfs1_vol_disk_hdr *hdr;
 750         struct ocfs2_dinode *di;
 751         int blksize;
 752
 753         *bh = NULL;
 754
 755         /* may be > 512 */
 756         *sector_size = bdev_logical_block_size(sb->s_bdev);
 757         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 758                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 759                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 760                 status = -EINVAL;
 761                 goto bail;
 762         }
 763
 764         /* Can this really happen? */
 765         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 766                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 767
 768         /* check block zero for old format */
 769         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 770         if (status < 0) {
 771                 mlog_errno(status);
 772                 goto bail;
 773         }
 774         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 775         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 776                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 777                      hdr->major_version, hdr->minor_version);
 778                 status = -EINVAL;
 779         }
 780         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 781                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 782                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 783                      hdr->signature);
 784                 status = -EINVAL;
 785         }
 786         brelse(*bh);
 787         *bh = NULL;
 788         if (status < 0) {
 789                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 790                      "upgraded before mounting with ocfs v2\n");
 791                 goto bail;
 792         }
 793
 794         /*
 795          * Now check at magic offset for 512, 1024, 2048, 4096
 796          * blocksizes.  4096 is the maximum blocksize because it is
 797          * the minimum clustersize.
 798          */
 799         status = -EINVAL;
 800         for (blksize = *sector_size;
 801              blksize <= OCFS2_MAX_BLOCKSIZE;
 802              blksize <<= 1) {
 803                 tmpstat = ocfs2_get_sector(sb, bh,
 804                                            OCFS2_SUPER_BLOCK_BLKNO,
 805                                            blksize);
 806                 if (tmpstat < 0) {
 807                         status = tmpstat;
 808                         mlog_errno(status);
 809                         break;
 810                 }
 811                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 812                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 813                 spin_lock_init(&stats->b_lock);
 814                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 815                 if (tmpstat < 0) {
 816                         brelse(*bh);
 817                         *bh = NULL;
 818                 }
 819                 if (tmpstat != -EAGAIN) {
 820                         status = tmpstat;
 821                         break;
 822                 }
 823         }
 824
 825 bail:
 826         return status;
 827 }
 828
 829 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 830 {
 831         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 832
 833         if (osb->s_mount_opt & hb_enabled) {
 834                 if (ocfs2_mount_local(osb)) {
 835                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 836                              "mounted device.\n");
 837                         return -EINVAL;
 838                 }
 839                 if (ocfs2_userspace_stack(osb)) {
 840                         mlog(ML_ERROR, "Userspace stack expected, but "
 841                              "o2cb heartbeat arguments passed to mount\n");
 842                         return -EINVAL;
 843                 }
 844                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 845                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 846                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 847                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 848                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 849                         return -EINVAL;
 850                 }
 851         }
 852
 853         if (!(osb->s_mount_opt & hb_enabled)) {
 854                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 855                     !ocfs2_userspace_stack(osb)) {
 856                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 857                              "a read-write clustered device.\n");
 858                         return -EINVAL;
 859                 }
 860         }
 861
 862         return 0;
 863 }
 864
 865 /*
 866  * If we're using a userspace stack, mount should have passed
 867  * a name that matches the disk.  If not, mount should not
 868  * have passed a stack.
 869  */
 870 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 871                                         struct mount_options *mopt)
 872 {
 873         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 874                 mlog(ML_ERROR,
 875                      "cluster stack passed to mount, but this filesystem "
 876                      "does not support it\n");
 877                 return -EINVAL;
 878         }
 879
 880         if (ocfs2_userspace_stack(osb) &&
 881             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 882                     OCFS2_STACK_LABEL_LEN)) {
 883                 mlog(ML_ERROR,
 884                      "cluster stack passed to mount (\"%s\") does not "
 885                      "match the filesystem (\"%s\")\n",
 886                      mopt->cluster_stack,
 887                      osb->osb_cluster_stack);
 888                 return -EINVAL;
 889         }
 890
 891         return 0;
 892 }
 893
 894 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 895 {
 896         int type;
 897         struct super_block *sb = osb->sb;
 898         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 899                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 900         int status = 0;
 901
 902         for (type = 0; type < MAXQUOTAS; type++) {
 903                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 904                         continue;
 905                 if (unsuspend)
 906                         status = dquot_resume(sb, type);
 907                 else {
 908                         struct ocfs2_mem_dqinfo *oinfo;
 909
 910                         /* Cancel periodic syncing before suspending */
 911                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 912                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 913                         status = dquot_suspend(sb, type);
 914                 }
 915                 if (status < 0)
 916                         break;
 917         }
 918         if (status < 0)
 919                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 920                      "remount (error = %d).\n", status);
 921         return status;
 922 }
 923
 924 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 925 {
 926         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 927         struct super_block *sb = osb->sb;
 928         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 929                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 930         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 931                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 932         int status;
 933         int type;
 934
 935         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 936         for (type = 0; type < MAXQUOTAS; type++) {
 937                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 938                         continue;
 939                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 940                                                         osb->slot_num);
 941                 if (!inode[type]) {
 942                         status = -ENOENT;
 943                         goto out_quota_off;
 944                 }
 945                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 946                                       DQUOT_USAGE_ENABLED);
 947                 if (status < 0)
 948                         goto out_quota_off;
 949         }
 950
 951         for (type = 0; type < MAXQUOTAS; type++)
 952                 iput(inode[type]);
 953         return 0;
 954 out_quota_off:
 955         ocfs2_disable_quotas(osb);
 956         for (type = 0; type < MAXQUOTAS; type++)
 957                 iput(inode[type]);
 958         mlog_errno(status);
 959         return status;
 960 }
 961
 962 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 963 {
 964         int type;
 965         struct inode *inode;
 966         struct super_block *sb = osb->sb;
 967         struct ocfs2_mem_dqinfo *oinfo;
 968
 969         /* We mostly ignore errors in this function because there's not much
 970          * we can do when we see them */
 971         for (type = 0; type < MAXQUOTAS; type++) {
 972                 if (!sb_has_quota_loaded(sb, type))
 973                         continue;
 974                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 975                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 976                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 977                 inode = igrab(sb->s_dquot.files[type]);
 978                 /* Turn off quotas. This will remove all dquot structures from
 979                  * memory and so they will be automatically synced to global
 980                  * quota files */
 981                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 982                                         DQUOT_LIMITS_ENABLED);
 983                 if (!inode)
 984                         continue;
 985                 iput(inode);
 986         }
 987 }
 988
 989 /* Handle quota on quotactl */
 990 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 991 {
 992         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 993                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 994
 995         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 996                 return -EINVAL;
 997
 998         return dquot_enable(sb_dqopt(sb)->files[type], type,
 999                             format_id, DQUOT_LIMITS_ENABLED);
1000 }
1001
1002 /* Handle quota off quotactl */
1003 static int ocfs2_quota_off(struct super_block *sb, int type)
1004 {
1005         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1006 }
1007
1008 static const struct quotactl_ops ocfs2_quotactl_ops = {
1009         .quota_on_meta  = ocfs2_quota_on,
1010         .quota_off      = ocfs2_quota_off,
1011         .quota_sync     = dquot_quota_sync,
1012         .get_info       = dquot_get_dqinfo,
1013         .set_info       = dquot_set_dqinfo,
1014         .get_dqblk      = dquot_get_dqblk,
1015         .set_dqblk      = dquot_set_dqblk,
1016 };
1017
1018 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1019 {
1020         struct dentry *root;
1021         int status, sector_size;
1022         struct mount_options parsed_options;
1023         struct inode *inode = NULL;
1024         struct ocfs2_super *osb = NULL;
1025         struct buffer_head *bh = NULL;
1026         char nodestr[8];
1027         struct ocfs2_blockcheck_stats stats;
1028
1029         trace_ocfs2_fill_super(sb, data, silent);
1030
1031         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1032                 status = -EINVAL;
1033                 goto read_super_error;
1034         }
1035
1036         /* probe for superblock */
1037         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1038         if (status < 0) {
1039                 mlog(ML_ERROR, "superblock probe failed!\n");
1040                 goto read_super_error;
1041         }
1042
1043         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1044         osb = OCFS2_SB(sb);
1045         if (status < 0) {
1046                 mlog_errno(status);
1047                 goto read_super_error;
1048         }
1049         brelse(bh);
1050         bh = NULL;
1051
1052         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1053                 status = -EINVAL;
1054                 goto read_super_error;
1055         }
1056         osb->s_mount_opt = parsed_options.mount_opt;
1057         osb->s_atime_quantum = parsed_options.atime_quantum;
1058         osb->preferred_slot = parsed_options.slot;
1059         osb->osb_commit_interval = parsed_options.commit_interval;
1060
1061         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1062         osb->osb_resv_level = parsed_options.resv_level;
1063         osb->osb_dir_resv_level = parsed_options.resv_level;
1064         if (parsed_options.dir_resv_level == -1)
1065                 osb->osb_dir_resv_level = parsed_options.resv_level;
1066         else
1067                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1068
1069         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1070         if (status)
1071                 goto read_super_error;
1072
1073         sb->s_magic = OCFS2_SUPER_MAGIC;
1074
1075         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1076                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1077
1078         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1079          * heartbeat=none */
1080         if (bdev_read_only(sb->s_bdev)) {
1081                 if (!(sb->s_flags & MS_RDONLY)) {
1082                         status = -EACCES;
1083                         mlog(ML_ERROR, "Readonly device detected but readonly "
1084                              "mount was not specified.\n");
1085                         goto read_super_error;
1086                 }
1087
1088                 /* You should not be able to start a local heartbeat
1089                  * on a readonly device. */
1090                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1091                         status = -EROFS;
1092                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1093                              "device.\n");
1094                         goto read_super_error;
1095                 }
1096
1097                 status = ocfs2_check_journals_nolocks(osb);
1098                 if (status < 0) {
1099                         if (status == -EROFS)
1100                                 mlog(ML_ERROR, "Recovery required on readonly "
1101                                      "file system, but write access is "
1102                                      "unavailable.\n");
1103                         else
1104                                 mlog_errno(status);
1105                         goto read_super_error;
1106                 }
1107
1108                 ocfs2_set_ro_flag(osb, 1);
1109
1110                 printk(KERN_NOTICE "Readonly device detected. No cluster "
1111                        "services will be utilized for this mount. Recovery "
1112                        "will be skipped.\n");
1113         }
1114
1115         if (!ocfs2_is_hard_readonly(osb)) {
1116                 if (sb->s_flags & MS_RDONLY)
1117                         ocfs2_set_ro_flag(osb, 0);
1118         }
1119
1120         status = ocfs2_verify_heartbeat(osb);
1121         if (status < 0) {
1122                 mlog_errno(status);
1123                 goto read_super_error;
1124         }
1125
1126         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1127                                                  ocfs2_debugfs_root);
1128         if (!osb->osb_debug_root) {
1129                 status = -EINVAL;
1130                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1131                 goto read_super_error;
1132         }
1133
1134         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1135                                             osb->osb_debug_root,
1136                                             osb,
1137                                             &ocfs2_osb_debug_fops);
1138         if (!osb->osb_ctxt) {
1139                 status = -EINVAL;
1140                 mlog_errno(status);
1141                 goto read_super_error;
1142         }
1143
1144         if (ocfs2_meta_ecc(osb)) {
1145                 status = ocfs2_blockcheck_stats_debugfs_install(
1146                                                 &osb->osb_ecc_stats,
1147                                                 osb->osb_debug_root);
1148                 if (status) {
1149                         mlog(ML_ERROR,
1150                              "Unable to create blockcheck statistics "
1151                              "files\n");
1152                         goto read_super_error;
1153                 }
1154         }
1155
1156         status = ocfs2_mount_volume(sb);
1157         if (osb->root_inode)
1158                 inode = igrab(osb->root_inode);
1159
1160         if (status < 0)
1161                 goto read_super_error;
1162
1163         if (!inode) {
1164                 status = -EIO;
1165                 mlog_errno(status);
1166                 goto read_super_error;
1167         }
1168
1169         root = d_alloc_root(inode);
1170         if (!root) {
1171                 status = -ENOMEM;
1172                 mlog_errno(status);
1173                 goto read_super_error;
1174         }
1175
1176         sb->s_root = root;
1177
1178         ocfs2_complete_mount_recovery(osb);
1179
1180         if (ocfs2_mount_local(osb))
1181                 snprintf(nodestr, sizeof(nodestr), "local");
1182         else
1183                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1184
1185         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1186                "with %s data mode.\n",
1187                osb->dev_str, nodestr, osb->slot_num,
1188                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1189                "ordered");
1190
1191         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1192         wake_up(&osb->osb_mount_event);
1193
1194         /* Now we can initialize quotas because we can afford to wait
1195          * for cluster locks recovery now. That also means that truncation
1196          * log recovery can happen but that waits for proper quota setup */
1197         if (!(sb->s_flags & MS_RDONLY)) {
1198                 status = ocfs2_enable_quotas(osb);
1199                 if (status < 0) {
1200                         /* We have to err-out specially here because
1201                          * s_root is already set */
1202                         mlog_errno(status);
1203                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1204                         wake_up(&osb->osb_mount_event);
1205                         return status;
1206                 }
1207         }
1208
1209         ocfs2_complete_quota_recovery(osb);
1210
1211         /* Now we wake up again for processes waiting for quotas */
1212         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1213         wake_up(&osb->osb_mount_event);
1214
1215         /* Start this when the mount is almost sure of being successful */
1216         ocfs2_orphan_scan_start(osb);
1217
1218         return status;
1219
1220 read_super_error:
1221         brelse(bh);
1222
1223         if (inode)
1224                 iput(inode);
1225
1226         if (osb) {
1227                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1228                 wake_up(&osb->osb_mount_event);
1229                 ocfs2_dismount_volume(sb, 1);
1230         }
1231
1232         if (status)
1233                 mlog_errno(status);
1234         return status;
1235 }
1236
1237 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1238                         int flags,
1239                         const char *dev_name,
1240                         void *data)
1241 {
1242         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1243 }
1244
1245 static void ocfs2_kill_sb(struct super_block *sb)
1246 {
1247         struct ocfs2_super *osb = OCFS2_SB(sb);
1248
1249         /* Failed mount? */
1250         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1251                 goto out;
1252
1253         /* Prevent further queueing of inode drop events */
1254         spin_lock(&dentry_list_lock);
1255         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1256         spin_unlock(&dentry_list_lock);
1257         /* Wait for work to finish and/or remove it */
1258         cancel_work_sync(&osb->dentry_lock_work);
1259 out:
1260         kill_block_super(sb);
1261 }
1262
1263 static struct file_system_type ocfs2_fs_type = {
1264         .owner          = THIS_MODULE,
1265         .name           = "ocfs2",
1266         .mount          = ocfs2_mount,
1267         .kill_sb        = ocfs2_kill_sb,
1268
1269         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1270         .next           = NULL
1271 };
1272
1273 static int ocfs2_check_set_options(struct super_block *sb,
1274                                    struct mount_options *options)
1275 {
1276         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1277             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1278                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1279                 mlog(ML_ERROR, "User quotas were requested, but this "
1280                      "filesystem does not have the feature enabled.\n");
1281                 return 0;
1282         }
1283         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1284             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1285                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1286                 mlog(ML_ERROR, "Group quotas were requested, but this "
1287                      "filesystem does not have the feature enabled.\n");
1288                 return 0;
1289         }
1290         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1291             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1292                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1293                      "feature is not enabled\n");
1294                 return 0;
1295         }
1296         /* No ACL setting specified? Use XATTR feature... */
1297         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1298                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1299                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1300                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1301                 else
1302                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1303         }
1304         return 1;
1305 }
1306
1307 static int ocfs2_parse_options(struct super_block *sb,
1308                                char *options,
1309                                struct mount_options *mopt,
1310                                int is_remount)
1311 {
1312         int status, user_stack = 0;
1313         char *p;
1314         u32 tmp;
1315
1316         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1317
1318         mopt->commit_interval = 0;
1319         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1320         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1321         mopt->slot = OCFS2_INVALID_SLOT;
1322         mopt->localalloc_opt = -1;
1323         mopt->cluster_stack[0] = '\0';
1324         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1325         mopt->dir_resv_level = -1;
1326
1327         if (!options) {
1328                 status = 1;
1329                 goto bail;
1330         }
1331
1332         while ((p = strsep(&options, ",")) != NULL) {
1333                 int token, option;
1334                 substring_t args[MAX_OPT_ARGS];
1335
1336                 if (!*p)
1337                         continue;
1338
1339                 token = match_token(p, tokens, args);
1340                 switch (token) {
1341                 case Opt_hb_local:
1342                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1343                         break;
1344                 case Opt_hb_none:
1345                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1346                         break;
1347                 case Opt_hb_global:
1348                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1349                         break;
1350                 case Opt_barrier:
1351                         if (match_int(&args[0], &option)) {
1352                                 status = 0;
1353                                 goto bail;
1354                         }
1355                         if (option)
1356                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1357                         else
1358                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1359                         break;
1360                 case Opt_intr:
1361                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1362                         break;
1363                 case Opt_nointr:
1364                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1365                         break;
1366                 case Opt_err_panic:
1367                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1368                         break;
1369                 case Opt_err_ro:
1370                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1371                         break;
1372                 case Opt_data_ordered:
1373                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1374                         break;
1375                 case Opt_data_writeback:
1376                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1377                         break;
1378                 case Opt_user_xattr:
1379                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1380                         break;
1381                 case Opt_nouser_xattr:
1382                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1383                         break;
1384                 case Opt_atime_quantum:
1385                         if (match_int(&args[0], &option)) {
1386                                 status = 0;
1387                                 goto bail;
1388                         }
1389                         if (option >= 0)
1390                                 mopt->atime_quantum = option;
1391                         break;
1392                 case Opt_slot:
1393                         option = 0;
1394                         if (match_int(&args[0], &option)) {
1395                                 status = 0;
1396                                 goto bail;
1397                         }
1398                         if (option)
1399                                 mopt->slot = (s16)option;
1400                         break;
1401                 case Opt_commit:
1402                         option = 0;
1403                         if (match_int(&args[0], &option)) {
1404                                 status = 0;
1405                                 goto bail;
1406                         }
1407                         if (option < 0)
1408                                 return 0;
1409                         if (option == 0)
1410                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1411                         mopt->commit_interval = HZ * option;
1412                         break;
1413                 case Opt_localalloc:
1414                         option = 0;
1415                         if (match_int(&args[0], &option)) {
1416                                 status = 0;
1417                                 goto bail;
1418                         }
1419                         if (option >= 0)
1420                                 mopt->localalloc_opt = option;
1421                         break;
1422                 case Opt_localflocks:
1423                         /*
1424                          * Changing this during remount could race
1425                          * flock() requests, or "unbalance" existing
1426                          * ones (e.g., a lock is taken in one mode but
1427                          * dropped in the other). If users care enough
1428                          * to flip locking modes during remount, we
1429                          * could add a "local" flag to individual
1430                          * flock structures for proper tracking of
1431                          * state.
1432                          */
1433                         if (!is_remount)
1434                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1435                         break;
1436                 case Opt_stack:
1437                         /* Check both that the option we were passed
1438                          * is of the right length and that it is a proper
1439                          * string of the right length.
1440                          */
1441                         if (((args[0].to - args[0].from) !=
1442                              OCFS2_STACK_LABEL_LEN) ||
1443                             (strnlen(args[0].from,
1444                                      OCFS2_STACK_LABEL_LEN) !=
1445                              OCFS2_STACK_LABEL_LEN)) {
1446                                 mlog(ML_ERROR,
1447                                      "Invalid cluster_stack option\n");
1448                                 status = 0;
1449                                 goto bail;
1450                         }
1451                         memcpy(mopt->cluster_stack, args[0].from,
1452                                OCFS2_STACK_LABEL_LEN);
1453                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1454                         /*
1455                          * Open code the memcmp here as we don't have
1456                          * an osb to pass to
1457                          * ocfs2_userspace_stack().
1458                          */
1459                         if (memcmp(mopt->cluster_stack,
1460                                    OCFS2_CLASSIC_CLUSTER_STACK,
1461                                    OCFS2_STACK_LABEL_LEN))
1462                                 user_stack = 1;
1463                         break;
1464                 case Opt_inode64:
1465                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1466                         break;
1467                 case Opt_usrquota:
1468                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1469                         break;
1470                 case Opt_grpquota:
1471                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1472                         break;
1473                 case Opt_coherency_buffered:
1474                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1475                         break;
1476                 case Opt_coherency_full:
1477                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1478                         break;
1479                 case Opt_acl:
1480                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1481                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1482                         break;
1483                 case Opt_noacl:
1484                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1485                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1486                         break;
1487                 case Opt_resv_level:
1488                         if (is_remount)
1489                                 break;
1490                         if (match_int(&args[0], &option)) {
1491                                 status = 0;
1492                                 goto bail;
1493                         }
1494                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1495                             option < OCFS2_MAX_RESV_LEVEL)
1496                                 mopt->resv_level = option;
1497                         break;
1498                 case Opt_dir_resv_level:
1499                         if (is_remount)
1500                                 break;
1501                         if (match_int(&args[0], &option)) {
1502                                 status = 0;
1503                                 goto bail;
1504                         }
1505                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1506                             option < OCFS2_MAX_RESV_LEVEL)
1507                                 mopt->dir_resv_level = option;
1508                         break;
1509                 default:
1510                         mlog(ML_ERROR,
1511                              "Unrecognized mount option \"%s\" "
1512                              "or missing value\n", p);
1513                         status = 0;
1514                         goto bail;
1515                 }
1516         }
1517
1518         if (user_stack == 0) {
1519                 /* Ensure only one heartbeat mode */
1520                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1521                                          OCFS2_MOUNT_HB_GLOBAL |
1522                                          OCFS2_MOUNT_HB_NONE);
1523                 if (hweight32(tmp) != 1) {
1524                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1525                         status = 0;
1526                         goto bail;
1527                 }
1528         }
1529
1530         status = 1;
1531
1532 bail:
1533         return status;
1534 }
1535
1536 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1537 {
1538         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
1539         unsigned long opts = osb->s_mount_opt;
1540         unsigned int local_alloc_megs;
1541
1542         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1543                 seq_printf(s, ",_netdev");
1544                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1545                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1546                 else
1547                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1548         } else
1549                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1550
1551         if (opts & OCFS2_MOUNT_NOINTR)
1552                 seq_printf(s, ",nointr");
1553
1554         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1555                 seq_printf(s, ",data=writeback");
1556         else
1557                 seq_printf(s, ",data=ordered");
1558
1559         if (opts & OCFS2_MOUNT_BARRIER)
1560                 seq_printf(s, ",barrier=1");
1561
1562         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1563                 seq_printf(s, ",errors=panic");
1564         else
1565                 seq_printf(s, ",errors=remount-ro");
1566
1567         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1568                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1569
1570         if (!(mnt->mnt_flags & MNT_NOATIME) && !(mnt->mnt_flags & MNT_RELATIME))
1571                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1572
1573         if (osb->osb_commit_interval)
1574                 seq_printf(s, ",commit=%u",
1575                            (unsigned) (osb->osb_commit_interval / HZ));
1576
1577         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1578         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1579                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1580
1581         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1582                 seq_printf(s, ",localflocks,");
1583
1584         if (osb->osb_cluster_stack[0])
1585                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1586                            osb->osb_cluster_stack);
1587         if (opts & OCFS2_MOUNT_USRQUOTA)
1588                 seq_printf(s, ",usrquota");
1589         if (opts & OCFS2_MOUNT_GRPQUOTA)
1590                 seq_printf(s, ",grpquota");
1591
1592         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1593                 seq_printf(s, ",coherency=buffered");
1594         else
1595                 seq_printf(s, ",coherency=full");
1596
1597         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1598                 seq_printf(s, ",nouser_xattr");
1599         else
1600                 seq_printf(s, ",user_xattr");
1601
1602         if (opts & OCFS2_MOUNT_INODE64)
1603                 seq_printf(s, ",inode64");
1604
1605         if (opts & OCFS2_MOUNT_POSIX_ACL)
1606                 seq_printf(s, ",acl");
1607         else
1608                 seq_printf(s, ",noacl");
1609
1610         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1611                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1612
1613         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1614                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1615
1616         return 0;
1617 }
1618
1619 static int __init ocfs2_init(void)
1620 {
1621         int status;
1622
1623         ocfs2_print_version();
1624
1625         status = init_ocfs2_uptodate_cache();
1626         if (status < 0) {
1627                 mlog_errno(status);
1628                 goto leave;
1629         }
1630
1631         status = ocfs2_initialize_mem_caches();
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto leave;
1635         }
1636
1637         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1638         if (!ocfs2_wq) {
1639                 status = -ENOMEM;
1640                 goto leave;
1641         }
1642
1643         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1644         if (!ocfs2_debugfs_root) {
1645                 status = -EFAULT;
1646                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1647         }
1648
1649         ocfs2_set_locking_protocol();
1650
1651         status = register_quota_format(&ocfs2_quota_format);
1652 leave:
1653         if (status < 0) {
1654                 ocfs2_free_mem_caches();
1655                 exit_ocfs2_uptodate_cache();
1656                 mlog_errno(status);
1657         }
1658
1659         if (status >= 0) {
1660                 return register_filesystem(&ocfs2_fs_type);
1661         } else
1662                 return -1;
1663 }
1664
1665 static void __exit ocfs2_exit(void)
1666 {
1667         if (ocfs2_wq) {
1668                 flush_workqueue(ocfs2_wq);
1669                 destroy_workqueue(ocfs2_wq);
1670         }
1671
1672         unregister_quota_format(&ocfs2_quota_format);
1673
1674         debugfs_remove(ocfs2_debugfs_root);
1675
1676         ocfs2_free_mem_caches();
1677
1678         unregister_filesystem(&ocfs2_fs_type);
1679
1680         exit_ocfs2_uptodate_cache();
1681 }
1682
1683 static void ocfs2_put_super(struct super_block *sb)
1684 {
1685         trace_ocfs2_put_super(sb);
1686
1687         ocfs2_sync_blockdev(sb);
1688         ocfs2_dismount_volume(sb, 0);
1689 }
1690
1691 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1692 {
1693         struct ocfs2_super *osb;
1694         u32 numbits, freebits;
1695         int status;
1696         struct ocfs2_dinode *bm_lock;
1697         struct buffer_head *bh = NULL;
1698         struct inode *inode = NULL;
1699
1700         trace_ocfs2_statfs(dentry->d_sb, buf);
1701
1702         osb = OCFS2_SB(dentry->d_sb);
1703
1704         inode = ocfs2_get_system_file_inode(osb,
1705                                             GLOBAL_BITMAP_SYSTEM_INODE,
1706                                             OCFS2_INVALID_SLOT);
1707         if (!inode) {
1708                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1709                 status = -EIO;
1710                 goto bail;
1711         }
1712
1713         status = ocfs2_inode_lock(inode, &bh, 0);
1714         if (status < 0) {
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718
1719         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1720
1721         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1722         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1723
1724         buf->f_type = OCFS2_SUPER_MAGIC;
1725         buf->f_bsize = dentry->d_sb->s_blocksize;
1726         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1727         buf->f_blocks = ((sector_t) numbits) *
1728                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1729         buf->f_bfree = ((sector_t) freebits) *
1730                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1731         buf->f_bavail = buf->f_bfree;
1732         buf->f_files = numbits;
1733         buf->f_ffree = freebits;
1734         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1735                                 & 0xFFFFFFFFUL;
1736         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1737                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1738
1739         brelse(bh);
1740
1741         ocfs2_inode_unlock(inode, 0);
1742         status = 0;
1743 bail:
1744         if (inode)
1745                 iput(inode);
1746
1747         if (status)
1748                 mlog_errno(status);
1749
1750         return status;
1751 }
1752
1753 static void ocfs2_inode_init_once(void *data)
1754 {
1755         struct ocfs2_inode_info *oi = data;
1756
1757         oi->ip_flags = 0;
1758         oi->ip_open_count = 0;
1759         spin_lock_init(&oi->ip_lock);
1760         ocfs2_extent_map_init(&oi->vfs_inode);
1761         INIT_LIST_HEAD(&oi->ip_io_markers);
1762         oi->ip_dir_start_lookup = 0;
1763
1764         init_rwsem(&oi->ip_alloc_sem);
1765         init_rwsem(&oi->ip_xattr_sem);
1766         mutex_init(&oi->ip_io_mutex);
1767
1768         oi->ip_blkno = 0ULL;
1769         oi->ip_clusters = 0;
1770
1771         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1772
1773         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1774         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1775         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1776
1777         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1778                                   &ocfs2_inode_caching_ops);
1779
1780         inode_init_once(&oi->vfs_inode);
1781 }
1782
1783 static int ocfs2_initialize_mem_caches(void)
1784 {
1785         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1786                                        sizeof(struct ocfs2_inode_info),
1787                                        0,
1788                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1789                                                 SLAB_MEM_SPREAD),
1790                                        ocfs2_inode_init_once);
1791         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1792                                         sizeof(struct ocfs2_dquot),
1793                                         0,
1794                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1795                                                 SLAB_MEM_SPREAD),
1796                                         NULL);
1797         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1798                                         sizeof(struct ocfs2_quota_chunk),
1799                                         0,
1800                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1801                                         NULL);
1802         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1803             !ocfs2_qf_chunk_cachep) {
1804                 if (ocfs2_inode_cachep)
1805                         kmem_cache_destroy(ocfs2_inode_cachep);
1806                 if (ocfs2_dquot_cachep)
1807                         kmem_cache_destroy(ocfs2_dquot_cachep);
1808                 if (ocfs2_qf_chunk_cachep)
1809                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1810                 return -ENOMEM;
1811         }
1812
1813         return 0;
1814 }
1815
1816 static void ocfs2_free_mem_caches(void)
1817 {
1818         if (ocfs2_inode_cachep)
1819                 kmem_cache_destroy(ocfs2_inode_cachep);
1820         ocfs2_inode_cachep = NULL;
1821
1822         if (ocfs2_dquot_cachep)
1823                 kmem_cache_destroy(ocfs2_dquot_cachep);
1824         ocfs2_dquot_cachep = NULL;
1825
1826         if (ocfs2_qf_chunk_cachep)
1827                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1828         ocfs2_qf_chunk_cachep = NULL;
1829 }
1830
1831 static int ocfs2_get_sector(struct super_block *sb,
1832                             struct buffer_head **bh,
1833                             int block,
1834                             int sect_size)
1835 {
1836         if (!sb_set_blocksize(sb, sect_size)) {
1837                 mlog(ML_ERROR, "unable to set blocksize\n");
1838                 return -EIO;
1839         }
1840
1841         *bh = sb_getblk(sb, block);
1842         if (!*bh) {
1843                 mlog_errno(-EIO);
1844                 return -EIO;
1845         }
1846         lock_buffer(*bh);
1847         if (!buffer_dirty(*bh))
1848                 clear_buffer_uptodate(*bh);
1849         unlock_buffer(*bh);
1850         ll_rw_block(READ, 1, bh);
1851         wait_on_buffer(*bh);
1852         if (!buffer_uptodate(*bh)) {
1853                 mlog_errno(-EIO);
1854                 brelse(*bh);
1855                 *bh = NULL;
1856                 return -EIO;
1857         }
1858
1859         return 0;
1860 }
1861
1862 static int ocfs2_mount_volume(struct super_block *sb)
1863 {
1864         int status = 0;
1865         int unlock_super = 0;
1866         struct ocfs2_super *osb = OCFS2_SB(sb);
1867
1868         if (ocfs2_is_hard_readonly(osb))
1869                 goto leave;
1870
1871         status = ocfs2_dlm_init(osb);
1872         if (status < 0) {
1873                 mlog_errno(status);
1874                 goto leave;
1875         }
1876
1877         status = ocfs2_super_lock(osb, 1);
1878         if (status < 0) {
1879                 mlog_errno(status);
1880                 goto leave;
1881         }
1882         unlock_super = 1;
1883
1884         /* This will load up the node map and add ourselves to it. */
1885         status = ocfs2_find_slot(osb);
1886         if (status < 0) {
1887                 mlog_errno(status);
1888                 goto leave;
1889         }
1890
1891         /* load all node-local system inodes */
1892         status = ocfs2_init_local_system_inodes(osb);
1893         if (status < 0) {
1894                 mlog_errno(status);
1895                 goto leave;
1896         }
1897
1898         status = ocfs2_check_volume(osb);
1899         if (status < 0) {
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         status = ocfs2_truncate_log_init(osb);
1905         if (status < 0)
1906                 mlog_errno(status);
1907
1908 leave:
1909         if (unlock_super)
1910                 ocfs2_super_unlock(osb, 1);
1911
1912         return status;
1913 }
1914
1915 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1916 {
1917         int tmp, hangup_needed = 0;
1918         struct ocfs2_super *osb = NULL;
1919         char nodestr[8];
1920
1921         trace_ocfs2_dismount_volume(sb);
1922
1923         BUG_ON(!sb);
1924         osb = OCFS2_SB(sb);
1925         BUG_ON(!osb);
1926
1927         debugfs_remove(osb->osb_ctxt);
1928
1929         /*
1930          * Flush inode dropping work queue so that deletes are
1931          * performed while the filesystem is still working
1932          */
1933         ocfs2_drop_all_dl_inodes(osb);
1934
1935         /* Orphan scan should be stopped as early as possible */
1936         ocfs2_orphan_scan_stop(osb);
1937
1938         ocfs2_disable_quotas(osb);
1939
1940         ocfs2_shutdown_local_alloc(osb);
1941
1942         ocfs2_truncate_log_shutdown(osb);
1943
1944         /* This will disable recovery and flush any recovery work. */
1945         ocfs2_recovery_exit(osb);
1946
1947         ocfs2_journal_shutdown(osb);
1948
1949         ocfs2_sync_blockdev(sb);
1950
1951         ocfs2_purge_refcount_trees(osb);
1952
1953         /* No cluster connection means we've failed during mount, so skip
1954          * all the steps which depended on that to complete. */
1955         if (osb->cconn) {
1956                 tmp = ocfs2_super_lock(osb, 1);
1957                 if (tmp < 0) {
1958                         mlog_errno(tmp);
1959                         return;
1960                 }
1961         }
1962
1963         if (osb->slot_num != OCFS2_INVALID_SLOT)
1964                 ocfs2_put_slot(osb);
1965
1966         if (osb->cconn)
1967                 ocfs2_super_unlock(osb, 1);
1968
1969         ocfs2_release_system_inodes(osb);
1970
1971         /*
1972          * If we're dismounting due to mount error, mount.ocfs2 will clean
1973          * up heartbeat.  If we're a local mount, there is no heartbeat.
1974          * If we failed before we got a uuid_str yet, we can't stop
1975          * heartbeat.  Otherwise, do it.
1976          */
1977         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1978                 hangup_needed = 1;
1979
1980         if (osb->cconn)
1981                 ocfs2_dlm_shutdown(osb, hangup_needed);
1982
1983         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1984         debugfs_remove(osb->osb_debug_root);
1985
1986         if (hangup_needed)
1987                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1988
1989         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1990
1991         if (ocfs2_mount_local(osb))
1992                 snprintf(nodestr, sizeof(nodestr), "local");
1993         else
1994                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1995
1996         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1997                osb->dev_str, nodestr);
1998
1999         ocfs2_delete_osb(osb);
2000         kfree(osb);
2001         sb->s_dev = 0;
2002         sb->s_fs_info = NULL;
2003 }
2004
2005 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2006                                 unsigned uuid_bytes)
2007 {
2008         int i, ret;
2009         char *ptr;
2010
2011         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2012
2013         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2014         if (osb->uuid_str == NULL)
2015                 return -ENOMEM;
2016
2017         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2018                 /* print with null */
2019                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2020                 if (ret != 2) /* drop super cleans up */
2021                         return -EINVAL;
2022                 /* then only advance past the last char */
2023                 ptr += 2;
2024         }
2025
2026         return 0;
2027 }
2028
2029 /* Make sure entire volume is addressable by our journal.  Requires
2030    osb_clusters_at_boot to be valid and for the journal to have been
2031    initialized by ocfs2_journal_init(). */
2032 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2033 {
2034         int status = 0;
2035         u64 max_block =
2036                 ocfs2_clusters_to_blocks(osb->sb,
2037                                          osb->osb_clusters_at_boot) - 1;
2038
2039         /* 32-bit block number is always OK. */
2040         if (max_block <= (u32)~0ULL)
2041                 goto out;
2042
2043         /* Volume is "huge", so see if our journal is new enough to
2044            support it. */
2045         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2046                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2047               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2048                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2049                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2050                      "Enable the 'block64' journal option with tunefs.ocfs2");
2051                 status = -EFBIG;
2052                 goto out;
2053         }
2054
2055  out:
2056         return status;
2057 }
2058
2059 static int ocfs2_initialize_super(struct super_block *sb,
2060                                   struct buffer_head *bh,
2061                                   int sector_size,
2062                                   struct ocfs2_blockcheck_stats *stats)
2063 {
2064         int status;
2065         int i, cbits, bbits;
2066         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2067         struct inode *inode = NULL;
2068         struct ocfs2_journal *journal;
2069         __le32 uuid_net_key;
2070         struct ocfs2_super *osb;
2071         u64 total_blocks;
2072
2073         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2074         if (!osb) {
2075                 status = -ENOMEM;
2076                 mlog_errno(status);
2077                 goto bail;
2078         }
2079
2080         sb->s_fs_info = osb;
2081         sb->s_op = &ocfs2_sops;
2082         sb->s_d_op = &ocfs2_dentry_ops;
2083         sb->s_export_op = &ocfs2_export_ops;
2084         sb->s_qcop = &ocfs2_quotactl_ops;
2085         sb->dq_op = &ocfs2_quota_operations;
2086         sb->s_xattr = ocfs2_xattr_handlers;
2087         sb->s_time_gran = 1;
2088         sb->s_flags |= MS_NOATIME;
2089         /* this is needed to support O_LARGEFILE */
2090         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2091         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2092         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2093
2094         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2095
2096         for (i = 0; i < 3; i++)
2097                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2098         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2099
2100         osb->sb = sb;
2101         /* Save off for ocfs2_rw_direct */
2102         osb->s_sectsize_bits = blksize_bits(sector_size);
2103         BUG_ON(!osb->s_sectsize_bits);
2104
2105         spin_lock_init(&osb->dc_task_lock);
2106         init_waitqueue_head(&osb->dc_event);
2107         osb->dc_work_sequence = 0;
2108         osb->dc_wake_sequence = 0;
2109         INIT_LIST_HEAD(&osb->blocked_lock_list);
2110         osb->blocked_lock_count = 0;
2111         spin_lock_init(&osb->osb_lock);
2112         spin_lock_init(&osb->osb_xattr_lock);
2113         ocfs2_init_steal_slots(osb);
2114
2115         atomic_set(&osb->alloc_stats.moves, 0);
2116         atomic_set(&osb->alloc_stats.local_data, 0);
2117         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2118         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2119         atomic_set(&osb->alloc_stats.bg_extends, 0);
2120
2121         /* Copy the blockcheck stats from the superblock probe */
2122         osb->osb_ecc_stats = *stats;
2123
2124         ocfs2_init_node_maps(osb);
2125
2126         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2127                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2128
2129         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2130         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2131                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2132                      osb->max_slots);
2133                 status = -EINVAL;
2134                 goto bail;
2135         }
2136
2137         ocfs2_orphan_scan_init(osb);
2138
2139         status = ocfs2_recovery_init(osb);
2140         if (status) {
2141                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2142                 mlog_errno(status);
2143                 goto bail;
2144         }
2145
2146         init_waitqueue_head(&osb->checkpoint_event);
2147         atomic_set(&osb->needs_checkpoint, 0);
2148
2149         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2150
2151         osb->slot_num = OCFS2_INVALID_SLOT;
2152
2153         osb->s_xattr_inline_size = le16_to_cpu(
2154                                         di->id2.i_super.s_xattr_inline_size);
2155
2156         osb->local_alloc_state = OCFS2_LA_UNUSED;
2157         osb->local_alloc_bh = NULL;
2158         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2159
2160         init_waitqueue_head(&osb->osb_mount_event);
2161
2162         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2163         if (status) {
2164                 mlog_errno(status);
2165                 goto bail;
2166         }
2167
2168         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2169         if (!osb->vol_label) {
2170                 mlog(ML_ERROR, "unable to alloc vol label\n");
2171                 status = -ENOMEM;
2172                 goto bail;
2173         }
2174
2175         osb->slot_recovery_generations =
2176                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2177                         GFP_KERNEL);
2178         if (!osb->slot_recovery_generations) {
2179                 status = -ENOMEM;
2180                 mlog_errno(status);
2181                 goto bail;
2182         }
2183
2184         init_waitqueue_head(&osb->osb_wipe_event);
2185         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2186                                         sizeof(*osb->osb_orphan_wipes),
2187                                         GFP_KERNEL);
2188         if (!osb->osb_orphan_wipes) {
2189                 status = -ENOMEM;
2190                 mlog_errno(status);
2191                 goto bail;
2192         }
2193
2194         osb->osb_rf_lock_tree = RB_ROOT;
2195
2196         osb->s_feature_compat =
2197                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2198         osb->s_feature_ro_compat =
2199                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2200         osb->s_feature_incompat =
2201                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2202
2203         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2204                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2205                      "optional features (%x).\n", i);
2206                 status = -EINVAL;
2207                 goto bail;
2208         }
2209         if (!(osb->sb->s_flags & MS_RDONLY) &&
2210             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2211                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2212                      "unsupported optional features (%x).\n", i);
2213                 status = -EINVAL;
2214                 goto bail;
2215         }
2216
2217         if (ocfs2_clusterinfo_valid(osb)) {
2218                 osb->osb_stackflags =
2219                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2220                 memcpy(osb->osb_cluster_stack,
2221                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2222                        OCFS2_STACK_LABEL_LEN);
2223                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
2224                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2225                         mlog(ML_ERROR,
2226                              "couldn't mount because of an invalid "
2227                              "cluster stack label (%s) \n",
2228                              osb->osb_cluster_stack);
2229                         status = -EINVAL;
2230                         goto bail;
2231                 }
2232         } else {
2233                 /* The empty string is identical with classic tools that
2234                  * don't know about s_cluster_info. */
2235                 osb->osb_cluster_stack[0] = '\0';
2236         }
2237
2238         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2239
2240         /* FIXME
2241          * This should be done in ocfs2_journal_init(), but unknown
2242          * ordering issues will cause the filesystem to crash.
2243          * If anyone wants to figure out what part of the code
2244          * refers to osb->journal before ocfs2_journal_init() is run,
2245          * be my guest.
2246          */
2247         /* initialize our journal structure */
2248
2249         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2250         if (!journal) {
2251                 mlog(ML_ERROR, "unable to alloc journal\n");
2252                 status = -ENOMEM;
2253                 goto bail;
2254         }
2255         osb->journal = journal;
2256         journal->j_osb = osb;
2257
2258         atomic_set(&journal->j_num_trans, 0);
2259         init_rwsem(&journal->j_trans_barrier);
2260         init_waitqueue_head(&journal->j_checkpointed);
2261         spin_lock_init(&journal->j_lock);
2262         journal->j_trans_id = (unsigned long) 1;
2263         INIT_LIST_HEAD(&journal->j_la_cleanups);
2264         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2265         journal->j_state = OCFS2_JOURNAL_FREE;
2266
2267         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2268         osb->dentry_lock_list = NULL;
2269
2270         /* get some pseudo constants for clustersize bits */
2271         osb->s_clustersize_bits =
2272                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2273         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2274
2275         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2276             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2277                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2278                      osb->s_clustersize);
2279                 status = -EINVAL;
2280                 goto bail;
2281         }
2282
2283         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2284                                                 le32_to_cpu(di->i_clusters));
2285
2286         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2287                                            total_blocks);
2288         if (status) {
2289                 mlog(ML_ERROR, "Volume too large "
2290                      "to mount safely on this system");
2291                 status = -EFBIG;
2292                 goto bail;
2293         }
2294
2295         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2296                                  sizeof(di->id2.i_super.s_uuid))) {
2297                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2298                 status = -ENOMEM;
2299                 goto bail;
2300         }
2301
2302         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2303
2304         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2305         osb->vol_label[63] = '\0';
2306         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2307         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2308         osb->first_cluster_group_blkno =
2309                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2310         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2311         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2312         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2313                                      (unsigned long long)osb->root_blkno,
2314                                      (unsigned long long)osb->system_dir_blkno,
2315                                      osb->s_clustersize_bits);
2316
2317         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2318         if (!osb->osb_dlm_debug) {
2319                 status = -ENOMEM;
2320                 mlog_errno(status);
2321                 goto bail;
2322         }
2323
2324         atomic_set(&osb->vol_state, VOLUME_INIT);
2325
2326         /* load root, system_dir, and all global system inodes */
2327         status = ocfs2_init_global_system_inodes(osb);
2328         if (status < 0) {
2329                 mlog_errno(status);
2330                 goto bail;
2331         }
2332
2333         /*
2334          * global bitmap
2335          */
2336         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2337                                             OCFS2_INVALID_SLOT);
2338         if (!inode) {
2339                 status = -EINVAL;
2340                 mlog_errno(status);
2341                 goto bail;
2342         }
2343
2344         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2345         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2346         iput(inode);
2347
2348         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2349                                  osb->s_feature_incompat) * 8;
2350
2351         status = ocfs2_init_slot_info(osb);
2352         if (status < 0) {
2353                 mlog_errno(status);
2354                 goto bail;
2355         }
2356         cleancache_init_shared_fs((char *)&uuid_net_key, sb);
2357
2358 bail:
2359         return status;
2360 }
2361
2362 /*
2363  * will return: -EAGAIN if it is ok to keep searching for superblocks
2364  *              -EINVAL if there is a bad superblock
2365  *              0 on success
2366  */
2367 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2368                                struct buffer_head *bh,
2369                                u32 blksz,
2370                                struct ocfs2_blockcheck_stats *stats)
2371 {
2372         int status = -EAGAIN;
2373
2374         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2375                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2376                 /* We have to do a raw check of the feature here */
2377                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2378                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2379                         status = ocfs2_block_check_validate(bh->b_data,
2380                                                             bh->b_size,
2381                                                             &di->i_check,
2382                                                             stats);
2383                         if (status)
2384                                 goto out;
2385                 }
2386                 status = -EINVAL;
2387                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2388                         mlog(ML_ERROR, "found superblock with incorrect block "
2389                              "size: found %u, should be %u\n",
2390                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2391                                blksz);
2392                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2393                            OCFS2_MAJOR_REV_LEVEL ||
2394                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2395                            OCFS2_MINOR_REV_LEVEL) {
2396                         mlog(ML_ERROR, "found superblock with bad version: "
2397                              "found %u.%u, should be %u.%u\n",
2398                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2399                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2400                              OCFS2_MAJOR_REV_LEVEL,
2401                              OCFS2_MINOR_REV_LEVEL);
2402                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2403                         mlog(ML_ERROR, "bad block number on superblock: "
2404                              "found %llu, should be %llu\n",
2405                              (unsigned long long)le64_to_cpu(di->i_blkno),
2406                              (unsigned long long)bh->b_blocknr);
2407                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2408                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2409                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2410                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2411                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2412                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2413                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2414                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2415                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2416                         mlog(ML_ERROR,
2417                              "Superblock slots found greater than file system "
2418                              "maximum: found %u, max %u\n",
2419                              le16_to_cpu(di->id2.i_super.s_max_slots),
2420                              OCFS2_MAX_SLOTS);
2421                 } else {
2422                         /* found it! */
2423                         status = 0;
2424                 }
2425         }
2426
2427 out:
2428         if (status && status != -EAGAIN)
2429                 mlog_errno(status);
2430         return status;
2431 }
2432
2433 static int ocfs2_check_volume(struct ocfs2_super *osb)
2434 {
2435         int status;
2436         int dirty;
2437         int local;
2438         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2439                                                   * recover
2440                                                   * ourselves. */
2441
2442         /* Init our journal object. */
2443         status = ocfs2_journal_init(osb->journal, &dirty);
2444         if (status < 0) {
2445                 mlog(ML_ERROR, "Could not initialize journal!\n");
2446                 goto finally;
2447         }
2448
2449         /* Now that journal has been initialized, check to make sure
2450            entire volume is addressable. */
2451         status = ocfs2_journal_addressable(osb);
2452         if (status)
2453                 goto finally;
2454
2455         /* If the journal was unmounted cleanly then we don't want to
2456          * recover anything. Otherwise, journal_load will do that
2457          * dirty work for us :) */
2458         if (!dirty) {
2459                 status = ocfs2_journal_wipe(osb->journal, 0);
2460                 if (status < 0) {
2461                         mlog_errno(status);
2462                         goto finally;
2463                 }
2464         } else {
2465                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
2466                      "recovering volume.\n");
2467         }
2468
2469         local = ocfs2_mount_local(osb);
2470
2471         /* will play back anything left in the journal. */
2472         status = ocfs2_journal_load(osb->journal, local, dirty);
2473         if (status < 0) {
2474                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2475                 goto finally;
2476         }
2477
2478         if (dirty) {
2479                 /* recover my local alloc if we didn't unmount cleanly. */
2480                 status = ocfs2_begin_local_alloc_recovery(osb,
2481                                                           osb->slot_num,
2482                                                           &local_alloc);
2483                 if (status < 0) {
2484                         mlog_errno(status);
2485                         goto finally;
2486                 }
2487                 /* we complete the recovery process after we've marked
2488                  * ourselves as mounted. */
2489         }
2490
2491         status = ocfs2_load_local_alloc(osb);
2492         if (status < 0) {
2493                 mlog_errno(status);
2494                 goto finally;
2495         }
2496
2497         if (dirty) {
2498                 /* Recovery will be completed after we've mounted the
2499                  * rest of the volume. */
2500                 osb->dirty = 1;
2501                 osb->local_alloc_copy = local_alloc;
2502                 local_alloc = NULL;
2503         }
2504
2505         /* go through each journal, trylock it and if you get the
2506          * lock, and it's marked as dirty, set the bit in the recover
2507          * map and launch a recovery thread for it. */
2508         status = ocfs2_mark_dead_nodes(osb);
2509         if (status < 0) {
2510                 mlog_errno(status);
2511                 goto finally;
2512         }
2513
2514         status = ocfs2_compute_replay_slots(osb);
2515         if (status < 0)
2516                 mlog_errno(status);
2517
2518 finally:
2519         if (local_alloc)
2520                 kfree(local_alloc);
2521
2522         if (status)
2523                 mlog_errno(status);
2524         return status;
2525 }
2526
2527 /*
2528  * The routine gets called from dismount or close whenever a dismount on
2529  * volume is requested and the osb open count becomes 1.
2530  * It will remove the osb from the global list and also free up all the
2531  * initialized resources and fileobject.
2532  */
2533 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2534 {
2535         /* This function assumes that the caller has the main osb resource */
2536
2537         ocfs2_free_slot_info(osb);
2538
2539         kfree(osb->osb_orphan_wipes);
2540         kfree(osb->slot_recovery_generations);
2541         /* FIXME
2542          * This belongs in journal shutdown, but because we have to
2543          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2544          * we free it here.
2545          */
2546         kfree(osb->journal);
2547         if (osb->local_alloc_copy)
2548                 kfree(osb->local_alloc_copy);
2549         kfree(osb->uuid_str);
2550         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2551         memset(osb, 0, sizeof(struct ocfs2_super));
2552 }
2553
2554 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2555  * panic(). We do not support continue-on-error operation. */
2556 static void ocfs2_handle_error(struct super_block *sb)
2557 {
2558         struct ocfs2_super *osb = OCFS2_SB(sb);
2559
2560         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2561                 panic("OCFS2: (device %s): panic forced after error\n",
2562                       sb->s_id);
2563
2564         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2565
2566         if (sb->s_flags & MS_RDONLY &&
2567             (ocfs2_is_soft_readonly(osb) ||
2568              ocfs2_is_hard_readonly(osb)))
2569                 return;
2570
2571         printk(KERN_CRIT "File system is now read-only due to the potential "
2572                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2573                "system is unmounted.\n");
2574         sb->s_flags |= MS_RDONLY;
2575         ocfs2_set_ro_flag(osb, 0);
2576 }
2577
2578 static char error_buf[1024];
2579
2580 void __ocfs2_error(struct super_block *sb,
2581                    const char *function,
2582                    const char *fmt, ...)
2583 {
2584         va_list args;
2585
2586         va_start(args, fmt);
2587         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2588         va_end(args);
2589
2590         /* Not using mlog here because we want to show the actual
2591          * function the error came from. */
2592         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2593                sb->s_id, function, error_buf);
2594
2595         ocfs2_handle_error(sb);
2596 }
2597
2598 /* Handle critical errors. This is intentionally more drastic than
2599  * ocfs2_handle_error, so we only use for things like journal errors,
2600  * etc. */
2601 void __ocfs2_abort(struct super_block* sb,
2602                    const char *function,
2603                    const char *fmt, ...)
2604 {
2605         va_list args;
2606
2607         va_start(args, fmt);
2608         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2609         va_end(args);
2610
2611         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2612                sb->s_id, function, error_buf);
2613
2614         /* We don't have the cluster support yet to go straight to
2615          * hard readonly in here. Until then, we want to keep
2616          * ocfs2_abort() so that we can at least mark critical
2617          * errors.
2618          *
2619          * TODO: This should abort the journal and alert other nodes
2620          * that our slot needs recovery. */
2621
2622         /* Force a panic(). This stinks, but it's better than letting
2623          * things continue without having a proper hard readonly
2624          * here. */
2625         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2626                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2627         ocfs2_handle_error(sb);
2628 }
2629
2630 /*
2631  * Void signal blockers, because in-kernel sigprocmask() only fails
2632  * when SIG_* is wrong.
2633  */
2634 void ocfs2_block_signals(sigset_t *oldset)
2635 {
2636         int rc;
2637         sigset_t blocked;
2638
2639         sigfillset(&blocked);
2640         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2641         BUG_ON(rc);
2642 }
2643
2644 void ocfs2_unblock_signals(sigset_t *oldset)
2645 {
2646         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2647         BUG_ON(rc);
2648 }
2649
2650 module_init(ocfs2_init);
2651 module_exit(ocfs2_exit);