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 "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.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;
  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;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 147 {
 148         return OCFS2_I(inode)->i_dquot;
 149 }
 150
 151 static const struct super_operations ocfs2_sops = {
 152         .statfs         = ocfs2_statfs,
 153         .alloc_inode    = ocfs2_alloc_inode,
 154         .destroy_inode  = ocfs2_destroy_inode,
 155         .drop_inode     = ocfs2_drop_inode,
 156         .evict_inode    = ocfs2_evict_inode,
 157         .sync_fs        = ocfs2_sync_fs,
 158         .put_super      = ocfs2_put_super,
 159         .remount_fs     = ocfs2_remount,
 160         .show_options   = ocfs2_show_options,
 161         .quota_read     = ocfs2_quota_read,
 162         .quota_write    = ocfs2_quota_write,
 163         .get_dquots     = ocfs2_get_dquots,
 164 };
 165
 166 enum {
 167         Opt_barrier,
 168         Opt_err_panic,
 169         Opt_err_ro,
 170         Opt_intr,
 171         Opt_nointr,
 172         Opt_hb_none,
 173         Opt_hb_local,
 174         Opt_hb_global,
 175         Opt_data_ordered,
 176         Opt_data_writeback,
 177         Opt_atime_quantum,
 178         Opt_slot,
 179         Opt_commit,
 180         Opt_localalloc,
 181         Opt_localflocks,
 182         Opt_stack,
 183         Opt_user_xattr,
 184         Opt_nouser_xattr,
 185         Opt_inode64,
 186         Opt_acl,
 187         Opt_noacl,
 188         Opt_usrquota,
 189         Opt_grpquota,
 190         Opt_coherency_buffered,
 191         Opt_coherency_full,
 192         Opt_resv_level,
 193         Opt_dir_resv_level,
 194         Opt_err,
 195 };
 196
 197 static const match_table_t tokens = {
 198         {Opt_barrier, "barrier=%u"},
 199         {Opt_err_panic, "errors=panic"},
 200         {Opt_err_ro, "errors=remount-ro"},
 201         {Opt_intr, "intr"},
 202         {Opt_nointr, "nointr"},
 203         {Opt_hb_none, OCFS2_HB_NONE},
 204         {Opt_hb_local, OCFS2_HB_LOCAL},
 205         {Opt_hb_global, OCFS2_HB_GLOBAL},
 206         {Opt_data_ordered, "data=ordered"},
 207         {Opt_data_writeback, "data=writeback"},
 208         {Opt_atime_quantum, "atime_quantum=%u"},
 209         {Opt_slot, "preferred_slot=%u"},
 210         {Opt_commit, "commit=%u"},
 211         {Opt_localalloc, "localalloc=%d"},
 212         {Opt_localflocks, "localflocks"},
 213         {Opt_stack, "cluster_stack=%s"},
 214         {Opt_user_xattr, "user_xattr"},
 215         {Opt_nouser_xattr, "nouser_xattr"},
 216         {Opt_inode64, "inode64"},
 217         {Opt_acl, "acl"},
 218         {Opt_noacl, "noacl"},
 219         {Opt_usrquota, "usrquota"},
 220         {Opt_grpquota, "grpquota"},
 221         {Opt_coherency_buffered, "coherency=buffered"},
 222         {Opt_coherency_full, "coherency=full"},
 223         {Opt_resv_level, "resv_level=%u"},
 224         {Opt_dir_resv_level, "dir_resv_level=%u"},
 225         {Opt_err, NULL}
 226 };
 227
 228 #ifdef CONFIG_DEBUG_FS
 229 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 230 {
 231         struct ocfs2_cluster_connection *cconn = osb->cconn;
 232         struct ocfs2_recovery_map *rm = osb->recovery_map;
 233         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 234         int i, out = 0;
 235
 236         out += snprintf(buf + out, len - out,
 237                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 238                         "Device", osb->dev_str, osb->uuid_str,
 239                         osb->fs_generation, osb->vol_label);
 240
 241         out += snprintf(buf + out, len - out,
 242                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 243                         atomic_read(&osb->vol_state), osb->osb_flags);
 244
 245         out += snprintf(buf + out, len - out,
 246                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 247                         osb->sb->s_blocksize, osb->s_clustersize);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 251                         "ROcompat: 0x%X\n",
 252                         "Features", osb->s_feature_compat,
 253                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 254
 255         out += snprintf(buf + out, len - out,
 256                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 257                         osb->s_mount_opt, osb->s_atime_quantum);
 258
 259         if (cconn) {
 260                 out += snprintf(buf + out, len - out,
 261                                 "%10s => Stack: %s  Name: %*s  "
 262                                 "Version: %d.%d\n", "Cluster",
 263                                 (*osb->osb_cluster_stack == '\0' ?
 264                                  "o2cb" : osb->osb_cluster_stack),
 265                                 cconn->cc_namelen, cconn->cc_name,
 266                                 cconn->cc_version.pv_major,
 267                                 cconn->cc_version.pv_minor);
 268         }
 269
 270         spin_lock(&osb->dc_task_lock);
 271         out += snprintf(buf + out, len - out,
 272                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 273                         "WorkSeq: %lu\n", "DownCnvt",
 274                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 275                         osb->blocked_lock_count, osb->dc_wake_sequence,
 276                         osb->dc_work_sequence);
 277         spin_unlock(&osb->dc_task_lock);
 278
 279         spin_lock(&osb->osb_lock);
 280         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 281                         "Recovery",
 282                         (osb->recovery_thread_task ?
 283                          task_pid_nr(osb->recovery_thread_task) : -1));
 284         if (rm->rm_used == 0)
 285                 out += snprintf(buf + out, len - out, " None\n");
 286         else {
 287                 for (i = 0; i < rm->rm_used; i++)
 288                         out += snprintf(buf + out, len - out, " %d",
 289                                         rm->rm_entries[i]);
 290                 out += snprintf(buf + out, len - out, "\n");
 291         }
 292         spin_unlock(&osb->osb_lock);
 293
 294         out += snprintf(buf + out, len - out,
 295                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 296                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 297                         osb->osb_commit_interval);
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 301                         "Journal", osb->journal->j_state,
 302                         osb->journal->j_trans_id,
 303                         atomic_read(&osb->journal->j_num_trans));
 304
 305         out += snprintf(buf + out, len - out,
 306                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 307                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 308                         "Stats",
 309                         atomic_read(&osb->alloc_stats.bitmap_data),
 310                         atomic_read(&osb->alloc_stats.local_data),
 311                         atomic_read(&osb->alloc_stats.bg_allocs),
 312                         atomic_read(&osb->alloc_stats.moves),
 313                         atomic_read(&osb->alloc_stats.bg_extends));
 314
 315         out += snprintf(buf + out, len - out,
 316                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 317                         "Default: %u bits\n",
 318                         "LocalAlloc", osb->local_alloc_state,
 319                         (unsigned long long)osb->la_last_gd,
 320                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 321
 322         spin_lock(&osb->osb_lock);
 323         out += snprintf(buf + out, len - out,
 324                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 325                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 326                         osb->s_inode_steal_slot,
 327                         atomic_read(&osb->s_num_inodes_stolen),
 328                         osb->s_meta_steal_slot,
 329                         atomic_read(&osb->s_num_meta_stolen));
 330         spin_unlock(&osb->osb_lock);
 331
 332         out += snprintf(buf + out, len - out, "OrphanScan => ");
 333         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 334                         os->os_count, os->os_seqno);
 335         out += snprintf(buf + out, len - out, " Last Scan: ");
 336         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 337                 out += snprintf(buf + out, len - out, "Disabled\n");
 338         else
 339                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 340                                 (get_seconds() - os->os_scantime.tv_sec));
 341
 342         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 343                         "Slots", "Num", "RecoGen");
 344         for (i = 0; i < osb->max_slots; ++i) {
 345                 out += snprintf(buf + out, len - out,
 346                                 "%10s  %c %3d  %10d\n",
 347                                 " ",
 348                                 (i == osb->slot_num ? '*' : ' '),
 349                                 i, osb->slot_recovery_generations[i]);
 350         }
 351
 352         return out;
 353 }
 354
 355 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 356 {
 357         struct ocfs2_super *osb = inode->i_private;
 358         char *buf = NULL;
 359
 360         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 361         if (!buf)
 362                 goto bail;
 363
 364         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 365
 366         file->private_data = buf;
 367
 368         return 0;
 369 bail:
 370         return -ENOMEM;
 371 }
 372
 373 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 374 {
 375         kfree(file->private_data);
 376         return 0;
 377 }
 378
 379 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 380                                 size_t nbytes, loff_t *ppos)
 381 {
 382         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 383                                        i_size_read(file->f_mapping->host));
 384 }
 385 #else
 386 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 387 {
 388         return 0;
 389 }
 390 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 391 {
 392         return 0;
 393 }
 394 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 395                                 size_t nbytes, loff_t *ppos)
 396 {
 397         return 0;
 398 }
 399 #endif  /* CONFIG_DEBUG_FS */
 400
 401 static const struct file_operations ocfs2_osb_debug_fops = {
 402         .open =         ocfs2_osb_debug_open,
 403         .release =      ocfs2_debug_release,
 404         .read =         ocfs2_debug_read,
 405         .llseek =       generic_file_llseek,
 406 };
 407
 408 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 409 {
 410         int status;
 411         tid_t target;
 412         struct ocfs2_super *osb = OCFS2_SB(sb);
 413
 414         if (ocfs2_is_hard_readonly(osb))
 415                 return -EROFS;
 416
 417         if (wait) {
 418                 status = ocfs2_flush_truncate_log(osb);
 419                 if (status < 0)
 420                         mlog_errno(status);
 421         } else {
 422                 ocfs2_schedule_truncate_log_flush(osb, 0);
 423         }
 424
 425         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 426                                       &target)) {
 427                 if (wait)
 428                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 429                                              target);
 430         }
 431         return 0;
 432 }
 433
 434 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 435 {
 436         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 437             && (ino == USER_QUOTA_SYSTEM_INODE
 438                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 439                 return 0;
 440         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 441             && (ino == GROUP_QUOTA_SYSTEM_INODE
 442                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 443                 return 0;
 444         return 1;
 445 }
 446
 447 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 448 {
 449         struct inode *new = NULL;
 450         int status = 0;
 451         int i;
 452
 453         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 454         if (IS_ERR(new)) {
 455                 status = PTR_ERR(new);
 456                 mlog_errno(status);
 457                 goto bail;
 458         }
 459         osb->root_inode = new;
 460
 461         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 462         if (IS_ERR(new)) {
 463                 status = PTR_ERR(new);
 464                 mlog_errno(status);
 465                 goto bail;
 466         }
 467         osb->sys_root_inode = new;
 468
 469         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 470              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 471                 if (!ocfs2_need_system_inode(osb, i))
 472                         continue;
 473                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 474                 if (!new) {
 475                         ocfs2_release_system_inodes(osb);
 476                         status = -EINVAL;
 477                         mlog_errno(status);
 478                         /* FIXME: Should ERROR_RO_FS */
 479                         mlog(ML_ERROR, "Unable to load system inode %d, "
 480                              "possibly corrupt fs?", i);
 481                         goto bail;
 482                 }
 483                 // the array now has one ref, so drop this one
 484                 iput(new);
 485         }
 486
 487 bail:
 488         if (status)
 489                 mlog_errno(status);
 490         return status;
 491 }
 492
 493 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 494 {
 495         struct inode *new = NULL;
 496         int status = 0;
 497         int i;
 498
 499         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 500              i < NUM_SYSTEM_INODES;
 501              i++) {
 502                 if (!ocfs2_need_system_inode(osb, i))
 503                         continue;
 504                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 505                 if (!new) {
 506                         ocfs2_release_system_inodes(osb);
 507                         status = -EINVAL;
 508                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 509                              status, i, osb->slot_num);
 510                         goto bail;
 511                 }
 512                 /* the array now has one ref, so drop this one */
 513                 iput(new);
 514         }
 515
 516 bail:
 517         if (status)
 518                 mlog_errno(status);
 519         return status;
 520 }
 521
 522 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 523 {
 524         int i;
 525         struct inode *inode;
 526
 527         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 528                 inode = osb->global_system_inodes[i];
 529                 if (inode) {
 530                         iput(inode);
 531                         osb->global_system_inodes[i] = NULL;
 532                 }
 533         }
 534
 535         inode = osb->sys_root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->sys_root_inode = NULL;
 539         }
 540
 541         inode = osb->root_inode;
 542         if (inode) {
 543                 iput(inode);
 544                 osb->root_inode = NULL;
 545         }
 546
 547         if (!osb->local_system_inodes)
 548                 return;
 549
 550         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 551                 if (osb->local_system_inodes[i]) {
 552                         iput(osb->local_system_inodes[i]);
 553                         osb->local_system_inodes[i] = NULL;
 554                 }
 555         }
 556
 557         kfree(osb->local_system_inodes);
 558         osb->local_system_inodes = NULL;
 559 }
 560
 561 /* We're allocating fs objects, use GFP_NOFS */
 562 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 563 {
 564         struct ocfs2_inode_info *oi;
 565
 566         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 567         if (!oi)
 568                 return NULL;
 569
 570         oi->i_sync_tid = 0;
 571         oi->i_datasync_tid = 0;
 572         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 573
 574         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 575         return &oi->vfs_inode;
 576 }
 577
 578 static void ocfs2_i_callback(struct rcu_head *head)
 579 {
 580         struct inode *inode = container_of(head, struct inode, i_rcu);
 581         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 582 }
 583
 584 static void ocfs2_destroy_inode(struct inode *inode)
 585 {
 586         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 587 }
 588
 589 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 590                                                 unsigned int cbits)
 591 {
 592         unsigned int bytes = 1 << cbits;
 593         unsigned int trim = bytes;
 594         unsigned int bitshift = 32;
 595
 596         /*
 597          * i_size and all block offsets in ocfs2 are always 64 bits
 598          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 599          * 64 bit platforms, cluster size will be the limiting factor.
 600          */
 601
 602 #if BITS_PER_LONG == 32
 603 # if defined(CONFIG_LBDAF)
 604         BUILD_BUG_ON(sizeof(sector_t) != 8);
 605         /*
 606          * We might be limited by page cache size.
 607          */
 608         if (bytes > PAGE_CACHE_SIZE) {
 609                 bytes = PAGE_CACHE_SIZE;
 610                 trim = 1;
 611                 /*
 612                  * Shift by 31 here so that we don't get larger than
 613                  * MAX_LFS_FILESIZE
 614                  */
 615                 bitshift = 31;
 616         }
 617 # else
 618         /*
 619          * We are limited by the size of sector_t. Use block size, as
 620          * that's what we expose to the VFS.
 621          */
 622         bytes = 1 << bbits;
 623         trim = 1;
 624         bitshift = 31;
 625 # endif
 626 #endif
 627
 628         /*
 629          * Trim by a whole cluster when we can actually approach the
 630          * on-disk limits. Otherwise we can overflow i_clusters when
 631          * an extent start is at the max offset.
 632          */
 633         return (((unsigned long long)bytes) << bitshift) - trim;
 634 }
 635
 636 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 637 {
 638         int incompat_features;
 639         int ret = 0;
 640         struct mount_options parsed_options;
 641         struct ocfs2_super *osb = OCFS2_SB(sb);
 642         u32 tmp;
 643
 644         sync_filesystem(sb);
 645
 646         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 647             !ocfs2_check_set_options(sb, &parsed_options)) {
 648                 ret = -EINVAL;
 649                 goto out;
 650         }
 651
 652         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 653                 OCFS2_MOUNT_HB_NONE;
 654         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 657                 goto out;
 658         }
 659
 660         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 661             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 662                 ret = -EINVAL;
 663                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 664                 goto out;
 665         }
 666
 667         /* Probably don't want this on remount; it might
 668          * mess with other nodes */
 669         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 670             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 671                 ret = -EINVAL;
 672                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 673                 goto out;
 674         }
 675
 676         /* We're going to/from readonly mode. */
 677         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 678                 /* Disable quota accounting before remounting RO */
 679                 if (*flags & MS_RDONLY) {
 680                         ret = ocfs2_susp_quotas(osb, 0);
 681                         if (ret < 0)
 682                                 goto out;
 683                 }
 684                 /* Lock here so the check of HARD_RO and the potential
 685                  * setting of SOFT_RO is atomic. */
 686                 spin_lock(&osb->osb_lock);
 687                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 688                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 689                         ret = -EROFS;
 690                         goto unlock_osb;
 691                 }
 692
 693                 if (*flags & MS_RDONLY) {
 694                         sb->s_flags |= MS_RDONLY;
 695                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 696                 } else {
 697                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 698                                 mlog(ML_ERROR, "Cannot remount RDWR "
 699                                      "filesystem due to previous errors.\n");
 700                                 ret = -EROFS;
 701                                 goto unlock_osb;
 702                         }
 703                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 704                         if (incompat_features) {
 705                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 706                                      "of unsupported optional features "
 707                                      "(%x).\n", incompat_features);
 708                                 ret = -EINVAL;
 709                                 goto unlock_osb;
 710                         }
 711                         sb->s_flags &= ~MS_RDONLY;
 712                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 713                 }
 714                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 715 unlock_osb:
 716                 spin_unlock(&osb->osb_lock);
 717                 /* Enable quota accounting after remounting RW */
 718                 if (!ret && !(*flags & MS_RDONLY)) {
 719                         if (sb_any_quota_suspended(sb))
 720                                 ret = ocfs2_susp_quotas(osb, 1);
 721                         else
 722                                 ret = ocfs2_enable_quotas(osb);
 723                         if (ret < 0) {
 724                                 /* Return back changes... */
 725                                 spin_lock(&osb->osb_lock);
 726                                 sb->s_flags |= MS_RDONLY;
 727                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 728                                 spin_unlock(&osb->osb_lock);
 729                                 goto out;
 730                         }
 731                 }
 732         }
 733
 734         if (!ret) {
 735                 /* Only save off the new mount options in case of a successful
 736                  * remount. */
 737                 osb->s_mount_opt = parsed_options.mount_opt;
 738                 osb->s_atime_quantum = parsed_options.atime_quantum;
 739                 osb->preferred_slot = parsed_options.slot;
 740                 if (parsed_options.commit_interval)
 741                         osb->osb_commit_interval = parsed_options.commit_interval;
 742
 743                 if (!ocfs2_is_hard_readonly(osb))
 744                         ocfs2_set_journal_params(osb);
 745
 746                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 747                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 748                                                         MS_POSIXACL : 0);
 749         }
 750 out:
 751         return ret;
 752 }
 753
 754 static int ocfs2_sb_probe(struct super_block *sb,
 755                           struct buffer_head **bh,
 756                           int *sector_size,
 757                           struct ocfs2_blockcheck_stats *stats)
 758 {
 759         int status, tmpstat;
 760         struct ocfs1_vol_disk_hdr *hdr;
 761         struct ocfs2_dinode *di;
 762         int blksize;
 763
 764         *bh = NULL;
 765
 766         /* may be > 512 */
 767         *sector_size = bdev_logical_block_size(sb->s_bdev);
 768         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 769                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 770                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 771                 status = -EINVAL;
 772                 goto bail;
 773         }
 774
 775         /* Can this really happen? */
 776         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 777                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 778
 779         /* check block zero for old format */
 780         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 781         if (status < 0) {
 782                 mlog_errno(status);
 783                 goto bail;
 784         }
 785         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 786         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 787                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 788                      hdr->major_version, hdr->minor_version);
 789                 status = -EINVAL;
 790         }
 791         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 792                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 793                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 794                      hdr->signature);
 795                 status = -EINVAL;
 796         }
 797         brelse(*bh);
 798         *bh = NULL;
 799         if (status < 0) {
 800                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 801                      "upgraded before mounting with ocfs v2\n");
 802                 goto bail;
 803         }
 804
 805         /*
 806          * Now check at magic offset for 512, 1024, 2048, 4096
 807          * blocksizes.  4096 is the maximum blocksize because it is
 808          * the minimum clustersize.
 809          */
 810         status = -EINVAL;
 811         for (blksize = *sector_size;
 812              blksize <= OCFS2_MAX_BLOCKSIZE;
 813              blksize <<= 1) {
 814                 tmpstat = ocfs2_get_sector(sb, bh,
 815                                            OCFS2_SUPER_BLOCK_BLKNO,
 816                                            blksize);
 817                 if (tmpstat < 0) {
 818                         status = tmpstat;
 819                         mlog_errno(status);
 820                         break;
 821                 }
 822                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 823                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 824                 spin_lock_init(&stats->b_lock);
 825                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 826                 if (tmpstat < 0) {
 827                         brelse(*bh);
 828                         *bh = NULL;
 829                 }
 830                 if (tmpstat != -EAGAIN) {
 831                         status = tmpstat;
 832                         break;
 833                 }
 834         }
 835
 836 bail:
 837         return status;
 838 }
 839
 840 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 841 {
 842         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 843
 844         if (osb->s_mount_opt & hb_enabled) {
 845                 if (ocfs2_mount_local(osb)) {
 846                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 847                              "mounted device.\n");
 848                         return -EINVAL;
 849                 }
 850                 if (ocfs2_userspace_stack(osb)) {
 851                         mlog(ML_ERROR, "Userspace stack expected, but "
 852                              "o2cb heartbeat arguments passed to mount\n");
 853                         return -EINVAL;
 854                 }
 855                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 856                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 857                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 858                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 859                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 860                         return -EINVAL;
 861                 }
 862         }
 863
 864         if (!(osb->s_mount_opt & hb_enabled)) {
 865                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 866                     !ocfs2_userspace_stack(osb)) {
 867                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 868                              "a read-write clustered device.\n");
 869                         return -EINVAL;
 870                 }
 871         }
 872
 873         return 0;
 874 }
 875
 876 /*
 877  * If we're using a userspace stack, mount should have passed
 878  * a name that matches the disk.  If not, mount should not
 879  * have passed a stack.
 880  */
 881 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 882                                         struct mount_options *mopt)
 883 {
 884         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 885                 mlog(ML_ERROR,
 886                      "cluster stack passed to mount, but this filesystem "
 887                      "does not support it\n");
 888                 return -EINVAL;
 889         }
 890
 891         if (ocfs2_userspace_stack(osb) &&
 892             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 893                     OCFS2_STACK_LABEL_LEN)) {
 894                 mlog(ML_ERROR,
 895                      "cluster stack passed to mount (\"%s\") does not "
 896                      "match the filesystem (\"%s\")\n",
 897                      mopt->cluster_stack,
 898                      osb->osb_cluster_stack);
 899                 return -EINVAL;
 900         }
 901
 902         return 0;
 903 }
 904
 905 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 906 {
 907         int type;
 908         struct super_block *sb = osb->sb;
 909         unsigned int feature[OCFS2_MAXQUOTAS] = {
 910                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 911                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 912         int status = 0;
 913
 914         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 915                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 916                         continue;
 917                 if (unsuspend)
 918                         status = dquot_resume(sb, type);
 919                 else {
 920                         struct ocfs2_mem_dqinfo *oinfo;
 921
 922                         /* Cancel periodic syncing before suspending */
 923                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 924                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 925                         status = dquot_suspend(sb, type);
 926                 }
 927                 if (status < 0)
 928                         break;
 929         }
 930         if (status < 0)
 931                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 932                      "remount (error = %d).\n", status);
 933         return status;
 934 }
 935
 936 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 937 {
 938         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 939         struct super_block *sb = osb->sb;
 940         unsigned int feature[OCFS2_MAXQUOTAS] = {
 941                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 942                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 943         unsigned int ino[OCFS2_MAXQUOTAS] = {
 944                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 945                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 946         int status;
 947         int type;
 948
 949         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 950         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 951                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 952                         continue;
 953                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 954                                                         osb->slot_num);
 955                 if (!inode[type]) {
 956                         status = -ENOENT;
 957                         goto out_quota_off;
 958                 }
 959                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 960                                       DQUOT_USAGE_ENABLED);
 961                 if (status < 0)
 962                         goto out_quota_off;
 963         }
 964
 965         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 966                 iput(inode[type]);
 967         return 0;
 968 out_quota_off:
 969         ocfs2_disable_quotas(osb);
 970         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 971                 iput(inode[type]);
 972         mlog_errno(status);
 973         return status;
 974 }
 975
 976 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 977 {
 978         int type;
 979         struct inode *inode;
 980         struct super_block *sb = osb->sb;
 981         struct ocfs2_mem_dqinfo *oinfo;
 982
 983         /* We mostly ignore errors in this function because there's not much
 984          * we can do when we see them */
 985         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 986                 if (!sb_has_quota_loaded(sb, type))
 987                         continue;
 988                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 989                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 990                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 991                 inode = igrab(sb->s_dquot.files[type]);
 992                 /* Turn off quotas. This will remove all dquot structures from
 993                  * memory and so they will be automatically synced to global
 994                  * quota files */
 995                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 996                                         DQUOT_LIMITS_ENABLED);
 997                 if (!inode)
 998                         continue;
 999                 iput(inode);
1000         }
1001 }
1002
1003 /* Handle quota on quotactl */
1004 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
1005 {
1006         unsigned int feature[OCFS2_MAXQUOTAS] = {
1007                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
1008                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
1009
1010         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
1011                 return -EINVAL;
1012
1013         return dquot_enable(sb_dqopt(sb)->files[type], type,
1014                             format_id, DQUOT_LIMITS_ENABLED);
1015 }
1016
1017 /* Handle quota off quotactl */
1018 static int ocfs2_quota_off(struct super_block *sb, int type)
1019 {
1020         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1021 }
1022
1023 static const struct quotactl_ops ocfs2_quotactl_ops = {
1024         .quota_on_meta  = ocfs2_quota_on,
1025         .quota_off      = ocfs2_quota_off,
1026         .quota_sync     = dquot_quota_sync,
1027         .get_info       = dquot_get_dqinfo,
1028         .set_info       = dquot_set_dqinfo,
1029         .get_dqblk      = dquot_get_dqblk,
1030         .set_dqblk      = dquot_set_dqblk,
1031 };
1032
1033 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1034 {
1035         struct dentry *root;
1036         int status, sector_size;
1037         struct mount_options parsed_options;
1038         struct inode *inode = NULL;
1039         struct ocfs2_super *osb = NULL;
1040         struct buffer_head *bh = NULL;
1041         char nodestr[12];
1042         struct ocfs2_blockcheck_stats stats;
1043
1044         trace_ocfs2_fill_super(sb, data, silent);
1045
1046         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1047                 status = -EINVAL;
1048                 goto read_super_error;
1049         }
1050
1051         /* probe for superblock */
1052         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1053         if (status < 0) {
1054                 mlog(ML_ERROR, "superblock probe failed!\n");
1055                 goto read_super_error;
1056         }
1057
1058         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1059         osb = OCFS2_SB(sb);
1060         if (status < 0) {
1061                 mlog_errno(status);
1062                 goto read_super_error;
1063         }
1064         brelse(bh);
1065         bh = NULL;
1066
1067         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1068                 status = -EINVAL;
1069                 goto read_super_error;
1070         }
1071         osb->s_mount_opt = parsed_options.mount_opt;
1072         osb->s_atime_quantum = parsed_options.atime_quantum;
1073         osb->preferred_slot = parsed_options.slot;
1074         osb->osb_commit_interval = parsed_options.commit_interval;
1075
1076         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1077         osb->osb_resv_level = parsed_options.resv_level;
1078         osb->osb_dir_resv_level = parsed_options.resv_level;
1079         if (parsed_options.dir_resv_level == -1)
1080                 osb->osb_dir_resv_level = parsed_options.resv_level;
1081         else
1082                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1083
1084         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1085         if (status)
1086                 goto read_super_error;
1087
1088         sb->s_magic = OCFS2_SUPER_MAGIC;
1089
1090         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1091                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1092
1093         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1094          * heartbeat=none */
1095         if (bdev_read_only(sb->s_bdev)) {
1096                 if (!(sb->s_flags & MS_RDONLY)) {
1097                         status = -EACCES;
1098                         mlog(ML_ERROR, "Readonly device detected but readonly "
1099                              "mount was not specified.\n");
1100                         goto read_super_error;
1101                 }
1102
1103                 /* You should not be able to start a local heartbeat
1104                  * on a readonly device. */
1105                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1106                         status = -EROFS;
1107                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1108                              "device.\n");
1109                         goto read_super_error;
1110                 }
1111
1112                 status = ocfs2_check_journals_nolocks(osb);
1113                 if (status < 0) {
1114                         if (status == -EROFS)
1115                                 mlog(ML_ERROR, "Recovery required on readonly "
1116                                      "file system, but write access is "
1117                                      "unavailable.\n");
1118                         else
1119                                 mlog_errno(status);
1120                         goto read_super_error;
1121                 }
1122
1123                 ocfs2_set_ro_flag(osb, 1);
1124
1125                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1126                        "Cluster services will not be used for this mount. "
1127                        "Recovery will be skipped.\n", osb->dev_str);
1128         }
1129
1130         if (!ocfs2_is_hard_readonly(osb)) {
1131                 if (sb->s_flags & MS_RDONLY)
1132                         ocfs2_set_ro_flag(osb, 0);
1133         }
1134
1135         status = ocfs2_verify_heartbeat(osb);
1136         if (status < 0) {
1137                 mlog_errno(status);
1138                 goto read_super_error;
1139         }
1140
1141         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1142                                                  ocfs2_debugfs_root);
1143         if (!osb->osb_debug_root) {
1144                 status = -EINVAL;
1145                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1146                 goto read_super_error;
1147         }
1148
1149         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1150                                             osb->osb_debug_root,
1151                                             osb,
1152                                             &ocfs2_osb_debug_fops);
1153         if (!osb->osb_ctxt) {
1154                 status = -EINVAL;
1155                 mlog_errno(status);
1156                 goto read_super_error;
1157         }
1158
1159         if (ocfs2_meta_ecc(osb)) {
1160                 status = ocfs2_blockcheck_stats_debugfs_install(
1161                                                 &osb->osb_ecc_stats,
1162                                                 osb->osb_debug_root);
1163                 if (status) {
1164                         mlog(ML_ERROR,
1165                              "Unable to create blockcheck statistics "
1166                              "files\n");
1167                         goto read_super_error;
1168                 }
1169         }
1170
1171         status = ocfs2_mount_volume(sb);
1172         if (status < 0)
1173                 goto read_super_error;
1174
1175         if (osb->root_inode)
1176                 inode = igrab(osb->root_inode);
1177
1178         if (!inode) {
1179                 status = -EIO;
1180                 mlog_errno(status);
1181                 goto read_super_error;
1182         }
1183
1184         root = d_make_root(inode);
1185         if (!root) {
1186                 status = -ENOMEM;
1187                 mlog_errno(status);
1188                 goto read_super_error;
1189         }
1190
1191         sb->s_root = root;
1192
1193         ocfs2_complete_mount_recovery(osb);
1194
1195         if (ocfs2_mount_local(osb))
1196                 snprintf(nodestr, sizeof(nodestr), "local");
1197         else
1198                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1199
1200         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1201                "with %s data mode.\n",
1202                osb->dev_str, nodestr, osb->slot_num,
1203                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1204                "ordered");
1205
1206         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1207         wake_up(&osb->osb_mount_event);
1208
1209         /* Now we can initialize quotas because we can afford to wait
1210          * for cluster locks recovery now. That also means that truncation
1211          * log recovery can happen but that waits for proper quota setup */
1212         if (!(sb->s_flags & MS_RDONLY)) {
1213                 status = ocfs2_enable_quotas(osb);
1214                 if (status < 0) {
1215                         /* We have to err-out specially here because
1216                          * s_root is already set */
1217                         mlog_errno(status);
1218                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1219                         wake_up(&osb->osb_mount_event);
1220                         return status;
1221                 }
1222         }
1223
1224         ocfs2_complete_quota_recovery(osb);
1225
1226         /* Now we wake up again for processes waiting for quotas */
1227         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1228         wake_up(&osb->osb_mount_event);
1229
1230         /* Start this when the mount is almost sure of being successful */
1231         ocfs2_orphan_scan_start(osb);
1232
1233         return status;
1234
1235 read_super_error:
1236         brelse(bh);
1237
1238         if (osb) {
1239                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1240                 wake_up(&osb->osb_mount_event);
1241                 ocfs2_dismount_volume(sb, 1);
1242         }
1243
1244         if (status)
1245                 mlog_errno(status);
1246         return status;
1247 }
1248
1249 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1250                         int flags,
1251                         const char *dev_name,
1252                         void *data)
1253 {
1254         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1255 }
1256
1257 static struct file_system_type ocfs2_fs_type = {
1258         .owner          = THIS_MODULE,
1259         .name           = "ocfs2",
1260         .mount          = ocfs2_mount,
1261         .kill_sb        = kill_block_super,
1262         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1263         .next           = NULL
1264 };
1265 MODULE_ALIAS_FS("ocfs2");
1266
1267 static int ocfs2_check_set_options(struct super_block *sb,
1268                                    struct mount_options *options)
1269 {
1270         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1271             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1272                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1273                 mlog(ML_ERROR, "User quotas were requested, but this "
1274                      "filesystem does not have the feature enabled.\n");
1275                 return 0;
1276         }
1277         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1278             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1279                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1280                 mlog(ML_ERROR, "Group quotas were requested, but this "
1281                      "filesystem does not have the feature enabled.\n");
1282                 return 0;
1283         }
1284         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1285             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1286                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1287                      "feature is not enabled\n");
1288                 return 0;
1289         }
1290         /* No ACL setting specified? Use XATTR feature... */
1291         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1292                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1293                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1294                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1295                 else
1296                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1297         }
1298         return 1;
1299 }
1300
1301 static int ocfs2_parse_options(struct super_block *sb,
1302                                char *options,
1303                                struct mount_options *mopt,
1304                                int is_remount)
1305 {
1306         int status, user_stack = 0;
1307         char *p;
1308         u32 tmp;
1309
1310         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1311
1312         mopt->commit_interval = 0;
1313         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1314         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1315         mopt->slot = OCFS2_INVALID_SLOT;
1316         mopt->localalloc_opt = -1;
1317         mopt->cluster_stack[0] = '\0';
1318         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1319         mopt->dir_resv_level = -1;
1320
1321         if (!options) {
1322                 status = 1;
1323                 goto bail;
1324         }
1325
1326         while ((p = strsep(&options, ",")) != NULL) {
1327                 int token, option;
1328                 substring_t args[MAX_OPT_ARGS];
1329
1330                 if (!*p)
1331                         continue;
1332
1333                 token = match_token(p, tokens, args);
1334                 switch (token) {
1335                 case Opt_hb_local:
1336                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1337                         break;
1338                 case Opt_hb_none:
1339                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1340                         break;
1341                 case Opt_hb_global:
1342                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1343                         break;
1344                 case Opt_barrier:
1345                         if (match_int(&args[0], &option)) {
1346                                 status = 0;
1347                                 goto bail;
1348                         }
1349                         if (option)
1350                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1351                         else
1352                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1353                         break;
1354                 case Opt_intr:
1355                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1356                         break;
1357                 case Opt_nointr:
1358                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1359                         break;
1360                 case Opt_err_panic:
1361                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1362                         break;
1363                 case Opt_err_ro:
1364                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1365                         break;
1366                 case Opt_data_ordered:
1367                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1368                         break;
1369                 case Opt_data_writeback:
1370                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1371                         break;
1372                 case Opt_user_xattr:
1373                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1374                         break;
1375                 case Opt_nouser_xattr:
1376                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1377                         break;
1378                 case Opt_atime_quantum:
1379                         if (match_int(&args[0], &option)) {
1380                                 status = 0;
1381                                 goto bail;
1382                         }
1383                         if (option >= 0)
1384                                 mopt->atime_quantum = option;
1385                         break;
1386                 case Opt_slot:
1387                         option = 0;
1388                         if (match_int(&args[0], &option)) {
1389                                 status = 0;
1390                                 goto bail;
1391                         }
1392                         if (option)
1393                                 mopt->slot = (s16)option;
1394                         break;
1395                 case Opt_commit:
1396                         option = 0;
1397                         if (match_int(&args[0], &option)) {
1398                                 status = 0;
1399                                 goto bail;
1400                         }
1401                         if (option < 0)
1402                                 return 0;
1403                         if (option == 0)
1404                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1405                         mopt->commit_interval = HZ * option;
1406                         break;
1407                 case Opt_localalloc:
1408                         option = 0;
1409                         if (match_int(&args[0], &option)) {
1410                                 status = 0;
1411                                 goto bail;
1412                         }
1413                         if (option >= 0)
1414                                 mopt->localalloc_opt = option;
1415                         break;
1416                 case Opt_localflocks:
1417                         /*
1418                          * Changing this during remount could race
1419                          * flock() requests, or "unbalance" existing
1420                          * ones (e.g., a lock is taken in one mode but
1421                          * dropped in the other). If users care enough
1422                          * to flip locking modes during remount, we
1423                          * could add a "local" flag to individual
1424                          * flock structures for proper tracking of
1425                          * state.
1426                          */
1427                         if (!is_remount)
1428                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1429                         break;
1430                 case Opt_stack:
1431                         /* Check both that the option we were passed
1432                          * is of the right length and that it is a proper
1433                          * string of the right length.
1434                          */
1435                         if (((args[0].to - args[0].from) !=
1436                              OCFS2_STACK_LABEL_LEN) ||
1437                             (strnlen(args[0].from,
1438                                      OCFS2_STACK_LABEL_LEN) !=
1439                              OCFS2_STACK_LABEL_LEN)) {
1440                                 mlog(ML_ERROR,
1441                                      "Invalid cluster_stack option\n");
1442                                 status = 0;
1443                                 goto bail;
1444                         }
1445                         memcpy(mopt->cluster_stack, args[0].from,
1446                                OCFS2_STACK_LABEL_LEN);
1447                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1448                         /*
1449                          * Open code the memcmp here as we don't have
1450                          * an osb to pass to
1451                          * ocfs2_userspace_stack().
1452                          */
1453                         if (memcmp(mopt->cluster_stack,
1454                                    OCFS2_CLASSIC_CLUSTER_STACK,
1455                                    OCFS2_STACK_LABEL_LEN))
1456                                 user_stack = 1;
1457                         break;
1458                 case Opt_inode64:
1459                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1460                         break;
1461                 case Opt_usrquota:
1462                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1463                         break;
1464                 case Opt_grpquota:
1465                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1466                         break;
1467                 case Opt_coherency_buffered:
1468                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1469                         break;
1470                 case Opt_coherency_full:
1471                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1472                         break;
1473                 case Opt_acl:
1474                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1475                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1476                         break;
1477                 case Opt_noacl:
1478                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1479                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1480                         break;
1481                 case Opt_resv_level:
1482                         if (is_remount)
1483                                 break;
1484                         if (match_int(&args[0], &option)) {
1485                                 status = 0;
1486                                 goto bail;
1487                         }
1488                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1489                             option < OCFS2_MAX_RESV_LEVEL)
1490                                 mopt->resv_level = option;
1491                         break;
1492                 case Opt_dir_resv_level:
1493                         if (is_remount)
1494                                 break;
1495                         if (match_int(&args[0], &option)) {
1496                                 status = 0;
1497                                 goto bail;
1498                         }
1499                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1500                             option < OCFS2_MAX_RESV_LEVEL)
1501                                 mopt->dir_resv_level = option;
1502                         break;
1503                 default:
1504                         mlog(ML_ERROR,
1505                              "Unrecognized mount option \"%s\" "
1506                              "or missing value\n", p);
1507                         status = 0;
1508                         goto bail;
1509                 }
1510         }
1511
1512         if (user_stack == 0) {
1513                 /* Ensure only one heartbeat mode */
1514                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1515                                          OCFS2_MOUNT_HB_GLOBAL |
1516                                          OCFS2_MOUNT_HB_NONE);
1517                 if (hweight32(tmp) != 1) {
1518                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1519                         status = 0;
1520                         goto bail;
1521                 }
1522         }
1523
1524         status = 1;
1525
1526 bail:
1527         return status;
1528 }
1529
1530 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1531 {
1532         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1533         unsigned long opts = osb->s_mount_opt;
1534         unsigned int local_alloc_megs;
1535
1536         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1537                 seq_printf(s, ",_netdev");
1538                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1539                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1540                 else
1541                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1542         } else
1543                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1544
1545         if (opts & OCFS2_MOUNT_NOINTR)
1546                 seq_printf(s, ",nointr");
1547
1548         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1549                 seq_printf(s, ",data=writeback");
1550         else
1551                 seq_printf(s, ",data=ordered");
1552
1553         if (opts & OCFS2_MOUNT_BARRIER)
1554                 seq_printf(s, ",barrier=1");
1555
1556         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1557                 seq_printf(s, ",errors=panic");
1558         else
1559                 seq_printf(s, ",errors=remount-ro");
1560
1561         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1562                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1563
1564         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1565
1566         if (osb->osb_commit_interval)
1567                 seq_printf(s, ",commit=%u",
1568                            (unsigned) (osb->osb_commit_interval / HZ));
1569
1570         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1571         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1572                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1573
1574         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1575                 seq_printf(s, ",localflocks,");
1576
1577         if (osb->osb_cluster_stack[0])
1578                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1579                            osb->osb_cluster_stack);
1580         if (opts & OCFS2_MOUNT_USRQUOTA)
1581                 seq_printf(s, ",usrquota");
1582         if (opts & OCFS2_MOUNT_GRPQUOTA)
1583                 seq_printf(s, ",grpquota");
1584
1585         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1586                 seq_printf(s, ",coherency=buffered");
1587         else
1588                 seq_printf(s, ",coherency=full");
1589
1590         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1591                 seq_printf(s, ",nouser_xattr");
1592         else
1593                 seq_printf(s, ",user_xattr");
1594
1595         if (opts & OCFS2_MOUNT_INODE64)
1596                 seq_printf(s, ",inode64");
1597
1598         if (opts & OCFS2_MOUNT_POSIX_ACL)
1599                 seq_printf(s, ",acl");
1600         else
1601                 seq_printf(s, ",noacl");
1602
1603         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1604                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1605
1606         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1607                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1608
1609         return 0;
1610 }
1611
1612 static int __init ocfs2_init(void)
1613 {
1614         int status;
1615
1616         status = init_ocfs2_uptodate_cache();
1617         if (status < 0)
1618                 goto out1;
1619
1620         status = ocfs2_initialize_mem_caches();
1621         if (status < 0)
1622                 goto out2;
1623
1624         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1625         if (!ocfs2_wq) {
1626                 status = -ENOMEM;
1627                 goto out3;
1628         }
1629
1630         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1631         if (!ocfs2_debugfs_root) {
1632                 status = -ENOMEM;
1633                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1634                 goto out4;
1635         }
1636
1637         ocfs2_set_locking_protocol();
1638
1639         status = register_quota_format(&ocfs2_quota_format);
1640         if (status < 0)
1641                 goto out4;
1642         status = register_filesystem(&ocfs2_fs_type);
1643         if (!status)
1644                 return 0;
1645
1646         unregister_quota_format(&ocfs2_quota_format);
1647 out4:
1648         destroy_workqueue(ocfs2_wq);
1649         debugfs_remove(ocfs2_debugfs_root);
1650 out3:
1651         ocfs2_free_mem_caches();
1652 out2:
1653         exit_ocfs2_uptodate_cache();
1654 out1:
1655         mlog_errno(status);
1656         return status;
1657 }
1658
1659 static void __exit ocfs2_exit(void)
1660 {
1661         if (ocfs2_wq) {
1662                 flush_workqueue(ocfs2_wq);
1663                 destroy_workqueue(ocfs2_wq);
1664         }
1665
1666         unregister_quota_format(&ocfs2_quota_format);
1667
1668         debugfs_remove(ocfs2_debugfs_root);
1669
1670         ocfs2_free_mem_caches();
1671
1672         unregister_filesystem(&ocfs2_fs_type);
1673
1674         exit_ocfs2_uptodate_cache();
1675 }
1676
1677 static void ocfs2_put_super(struct super_block *sb)
1678 {
1679         trace_ocfs2_put_super(sb);
1680
1681         ocfs2_sync_blockdev(sb);
1682         ocfs2_dismount_volume(sb, 0);
1683 }
1684
1685 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1686 {
1687         struct ocfs2_super *osb;
1688         u32 numbits, freebits;
1689         int status;
1690         struct ocfs2_dinode *bm_lock;
1691         struct buffer_head *bh = NULL;
1692         struct inode *inode = NULL;
1693
1694         trace_ocfs2_statfs(dentry->d_sb, buf);
1695
1696         osb = OCFS2_SB(dentry->d_sb);
1697
1698         inode = ocfs2_get_system_file_inode(osb,
1699                                             GLOBAL_BITMAP_SYSTEM_INODE,
1700                                             OCFS2_INVALID_SLOT);
1701         if (!inode) {
1702                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1703                 status = -EIO;
1704                 goto bail;
1705         }
1706
1707         status = ocfs2_inode_lock(inode, &bh, 0);
1708         if (status < 0) {
1709                 mlog_errno(status);
1710                 goto bail;
1711         }
1712
1713         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1714
1715         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1716         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1717
1718         buf->f_type = OCFS2_SUPER_MAGIC;
1719         buf->f_bsize = dentry->d_sb->s_blocksize;
1720         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1721         buf->f_blocks = ((sector_t) numbits) *
1722                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1723         buf->f_bfree = ((sector_t) freebits) *
1724                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1725         buf->f_bavail = buf->f_bfree;
1726         buf->f_files = numbits;
1727         buf->f_ffree = freebits;
1728         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1729                                 & 0xFFFFFFFFUL;
1730         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1731                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1732
1733         brelse(bh);
1734
1735         ocfs2_inode_unlock(inode, 0);
1736         status = 0;
1737 bail:
1738         if (inode)
1739                 iput(inode);
1740
1741         if (status)
1742                 mlog_errno(status);
1743
1744         return status;
1745 }
1746
1747 static void ocfs2_inode_init_once(void *data)
1748 {
1749         struct ocfs2_inode_info *oi = data;
1750
1751         oi->ip_flags = 0;
1752         oi->ip_open_count = 0;
1753         spin_lock_init(&oi->ip_lock);
1754         ocfs2_extent_map_init(&oi->vfs_inode);
1755         INIT_LIST_HEAD(&oi->ip_io_markers);
1756         oi->ip_dir_start_lookup = 0;
1757         mutex_init(&oi->ip_unaligned_aio);
1758         init_rwsem(&oi->ip_alloc_sem);
1759         init_rwsem(&oi->ip_xattr_sem);
1760         mutex_init(&oi->ip_io_mutex);
1761
1762         oi->ip_blkno = 0ULL;
1763         oi->ip_clusters = 0;
1764
1765         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1766
1767         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1768         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1769         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1770
1771         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1772                                   &ocfs2_inode_caching_ops);
1773
1774         inode_init_once(&oi->vfs_inode);
1775 }
1776
1777 static int ocfs2_initialize_mem_caches(void)
1778 {
1779         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1780                                        sizeof(struct ocfs2_inode_info),
1781                                        0,
1782                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1783                                                 SLAB_MEM_SPREAD),
1784                                        ocfs2_inode_init_once);
1785         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1786                                         sizeof(struct ocfs2_dquot),
1787                                         0,
1788                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1789                                                 SLAB_MEM_SPREAD),
1790                                         NULL);
1791         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1792                                         sizeof(struct ocfs2_quota_chunk),
1793                                         0,
1794                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1795                                         NULL);
1796         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1797             !ocfs2_qf_chunk_cachep) {
1798                 if (ocfs2_inode_cachep)
1799                         kmem_cache_destroy(ocfs2_inode_cachep);
1800                 if (ocfs2_dquot_cachep)
1801                         kmem_cache_destroy(ocfs2_dquot_cachep);
1802                 if (ocfs2_qf_chunk_cachep)
1803                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1804                 return -ENOMEM;
1805         }
1806
1807         return 0;
1808 }
1809
1810 static void ocfs2_free_mem_caches(void)
1811 {
1812         /*
1813          * Make sure all delayed rcu free inodes are flushed before we
1814          * destroy cache.
1815          */
1816         rcu_barrier();
1817         if (ocfs2_inode_cachep)
1818                 kmem_cache_destroy(ocfs2_inode_cachep);
1819         ocfs2_inode_cachep = NULL;
1820
1821         if (ocfs2_dquot_cachep)
1822                 kmem_cache_destroy(ocfs2_dquot_cachep);
1823         ocfs2_dquot_cachep = NULL;
1824
1825         if (ocfs2_qf_chunk_cachep)
1826                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1827         ocfs2_qf_chunk_cachep = NULL;
1828 }
1829
1830 static int ocfs2_get_sector(struct super_block *sb,
1831                             struct buffer_head **bh,
1832                             int block,
1833                             int sect_size)
1834 {
1835         if (!sb_set_blocksize(sb, sect_size)) {
1836                 mlog(ML_ERROR, "unable to set blocksize\n");
1837                 return -EIO;
1838         }
1839
1840         *bh = sb_getblk(sb, block);
1841         if (!*bh) {
1842                 mlog_errno(-ENOMEM);
1843                 return -ENOMEM;
1844         }
1845         lock_buffer(*bh);
1846         if (!buffer_dirty(*bh))
1847                 clear_buffer_uptodate(*bh);
1848         unlock_buffer(*bh);
1849         ll_rw_block(READ, 1, bh);
1850         wait_on_buffer(*bh);
1851         if (!buffer_uptodate(*bh)) {
1852                 mlog_errno(-EIO);
1853                 brelse(*bh);
1854                 *bh = NULL;
1855                 return -EIO;
1856         }
1857
1858         return 0;
1859 }
1860
1861 static int ocfs2_mount_volume(struct super_block *sb)
1862 {
1863         int status = 0;
1864         int unlock_super = 0;
1865         struct ocfs2_super *osb = OCFS2_SB(sb);
1866
1867         if (ocfs2_is_hard_readonly(osb))
1868                 goto leave;
1869
1870         status = ocfs2_dlm_init(osb);
1871         if (status < 0) {
1872                 mlog_errno(status);
1873                 goto leave;
1874         }
1875
1876         status = ocfs2_super_lock(osb, 1);
1877         if (status < 0) {
1878                 mlog_errno(status);
1879                 goto leave;
1880         }
1881         unlock_super = 1;
1882
1883         /* This will load up the node map and add ourselves to it. */
1884         status = ocfs2_find_slot(osb);
1885         if (status < 0) {
1886                 mlog_errno(status);
1887                 goto leave;
1888         }
1889
1890         /* load all node-local system inodes */
1891         status = ocfs2_init_local_system_inodes(osb);
1892         if (status < 0) {
1893                 mlog_errno(status);
1894                 goto leave;
1895         }
1896
1897         status = ocfs2_check_volume(osb);
1898         if (status < 0) {
1899                 mlog_errno(status);
1900                 goto leave;
1901         }
1902
1903         status = ocfs2_truncate_log_init(osb);
1904         if (status < 0)
1905                 mlog_errno(status);
1906
1907 leave:
1908         if (unlock_super)
1909                 ocfs2_super_unlock(osb, 1);
1910
1911         return status;
1912 }
1913
1914 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1915 {
1916         int tmp, hangup_needed = 0;
1917         struct ocfs2_super *osb = NULL;
1918         char nodestr[12];
1919
1920         trace_ocfs2_dismount_volume(sb);
1921
1922         BUG_ON(!sb);
1923         osb = OCFS2_SB(sb);
1924         BUG_ON(!osb);
1925
1926         debugfs_remove(osb->osb_ctxt);
1927
1928         /* Orphan scan should be stopped as early as possible */
1929         ocfs2_orphan_scan_stop(osb);
1930
1931         ocfs2_disable_quotas(osb);
1932
1933         /* All dquots should be freed by now */
1934         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1935         /* Wait for worker to be done with the work structure in osb */
1936         cancel_work_sync(&osb->dquot_drop_work);
1937
1938         ocfs2_shutdown_local_alloc(osb);
1939
1940         ocfs2_truncate_log_shutdown(osb);
1941
1942         /* This will disable recovery and flush any recovery work. */
1943         ocfs2_recovery_exit(osb);
1944
1945         ocfs2_journal_shutdown(osb);
1946
1947         ocfs2_sync_blockdev(sb);
1948
1949         ocfs2_purge_refcount_trees(osb);
1950
1951         /* No cluster connection means we've failed during mount, so skip
1952          * all the steps which depended on that to complete. */
1953         if (osb->cconn) {
1954                 tmp = ocfs2_super_lock(osb, 1);
1955                 if (tmp < 0) {
1956                         mlog_errno(tmp);
1957                         return;
1958                 }
1959         }
1960
1961         if (osb->slot_num != OCFS2_INVALID_SLOT)
1962                 ocfs2_put_slot(osb);
1963
1964         if (osb->cconn)
1965                 ocfs2_super_unlock(osb, 1);
1966
1967         ocfs2_release_system_inodes(osb);
1968
1969         /*
1970          * If we're dismounting due to mount error, mount.ocfs2 will clean
1971          * up heartbeat.  If we're a local mount, there is no heartbeat.
1972          * If we failed before we got a uuid_str yet, we can't stop
1973          * heartbeat.  Otherwise, do it.
1974          */
1975         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1976             !ocfs2_is_hard_readonly(osb))
1977                 hangup_needed = 1;
1978
1979         if (osb->cconn)
1980                 ocfs2_dlm_shutdown(osb, hangup_needed);
1981
1982         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1983         debugfs_remove(osb->osb_debug_root);
1984
1985         if (hangup_needed)
1986                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1987
1988         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1989
1990         if (ocfs2_mount_local(osb))
1991                 snprintf(nodestr, sizeof(nodestr), "local");
1992         else
1993                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1994
1995         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1996                osb->dev_str, nodestr);
1997
1998         ocfs2_delete_osb(osb);
1999         kfree(osb);
2000         sb->s_dev = 0;
2001         sb->s_fs_info = NULL;
2002 }
2003
2004 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2005                                 unsigned uuid_bytes)
2006 {
2007         int i, ret;
2008         char *ptr;
2009
2010         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2011
2012         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2013         if (osb->uuid_str == NULL)
2014                 return -ENOMEM;
2015
2016         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2017                 /* print with null */
2018                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2019                 if (ret != 2) /* drop super cleans up */
2020                         return -EINVAL;
2021                 /* then only advance past the last char */
2022                 ptr += 2;
2023         }
2024
2025         return 0;
2026 }
2027
2028 /* Make sure entire volume is addressable by our journal.  Requires
2029    osb_clusters_at_boot to be valid and for the journal to have been
2030    initialized by ocfs2_journal_init(). */
2031 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2032 {
2033         int status = 0;
2034         u64 max_block =
2035                 ocfs2_clusters_to_blocks(osb->sb,
2036                                          osb->osb_clusters_at_boot) - 1;
2037
2038         /* 32-bit block number is always OK. */
2039         if (max_block <= (u32)~0ULL)
2040                 goto out;
2041
2042         /* Volume is "huge", so see if our journal is new enough to
2043            support it. */
2044         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2045                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2046               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2047                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2048                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2049                      "Enable the 'block64' journal option with tunefs.ocfs2");
2050                 status = -EFBIG;
2051                 goto out;
2052         }
2053
2054  out:
2055         return status;
2056 }
2057
2058 static int ocfs2_initialize_super(struct super_block *sb,
2059                                   struct buffer_head *bh,
2060                                   int sector_size,
2061                                   struct ocfs2_blockcheck_stats *stats)
2062 {
2063         int status;
2064         int i, cbits, bbits;
2065         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2066         struct inode *inode = NULL;
2067         struct ocfs2_journal *journal;
2068         struct ocfs2_super *osb;
2069         u64 total_blocks;
2070
2071         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2072         if (!osb) {
2073                 status = -ENOMEM;
2074                 mlog_errno(status);
2075                 goto bail;
2076         }
2077
2078         sb->s_fs_info = osb;
2079         sb->s_op = &ocfs2_sops;
2080         sb->s_d_op = &ocfs2_dentry_ops;
2081         sb->s_export_op = &ocfs2_export_ops;
2082         sb->s_qcop = &ocfs2_quotactl_ops;
2083         sb->dq_op = &ocfs2_quota_operations;
2084         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2085         sb->s_xattr = ocfs2_xattr_handlers;
2086         sb->s_time_gran = 1;
2087         sb->s_flags |= MS_NOATIME;
2088         /* this is needed to support O_LARGEFILE */
2089         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2090         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2091         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2092
2093         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2094
2095         for (i = 0; i < 3; i++)
2096                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2097         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2098
2099         osb->sb = sb;
2100         /* Save off for ocfs2_rw_direct */
2101         osb->s_sectsize_bits = blksize_bits(sector_size);
2102         BUG_ON(!osb->s_sectsize_bits);
2103
2104         spin_lock_init(&osb->dc_task_lock);
2105         init_waitqueue_head(&osb->dc_event);
2106         osb->dc_work_sequence = 0;
2107         osb->dc_wake_sequence = 0;
2108         INIT_LIST_HEAD(&osb->blocked_lock_list);
2109         osb->blocked_lock_count = 0;
2110         spin_lock_init(&osb->osb_lock);
2111         spin_lock_init(&osb->osb_xattr_lock);
2112         ocfs2_init_steal_slots(osb);
2113
2114         mutex_init(&osb->system_file_mutex);
2115
2116         atomic_set(&osb->alloc_stats.moves, 0);
2117         atomic_set(&osb->alloc_stats.local_data, 0);
2118         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2119         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2120         atomic_set(&osb->alloc_stats.bg_extends, 0);
2121
2122         /* Copy the blockcheck stats from the superblock probe */
2123         osb->osb_ecc_stats = *stats;
2124
2125         ocfs2_init_node_maps(osb);
2126
2127         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2128                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2129
2130         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2131         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2132                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2133                      osb->max_slots);
2134                 status = -EINVAL;
2135                 goto bail;
2136         }
2137
2138         ocfs2_orphan_scan_init(osb);
2139
2140         status = ocfs2_recovery_init(osb);
2141         if (status) {
2142                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2143                 mlog_errno(status);
2144                 goto bail;
2145         }
2146
2147         init_waitqueue_head(&osb->checkpoint_event);
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                 strlcpy(osb->osb_cluster_stack,
2221                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2222                        OCFS2_STACK_LABEL_LEN + 1);
2223                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2224                         mlog(ML_ERROR,
2225                              "couldn't mount because of an invalid "
2226                              "cluster stack label (%s) \n",
2227                              osb->osb_cluster_stack);
2228                         status = -EINVAL;
2229                         goto bail;
2230                 }
2231                 strlcpy(osb->osb_cluster_name,
2232                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2233                         OCFS2_CLUSTER_NAME_LEN + 1);
2234         } else {
2235                 /* The empty string is identical with classic tools that
2236                  * don't know about s_cluster_info. */
2237                 osb->osb_cluster_stack[0] = '\0';
2238         }
2239
2240         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2241
2242         /* FIXME
2243          * This should be done in ocfs2_journal_init(), but unknown
2244          * ordering issues will cause the filesystem to crash.
2245          * If anyone wants to figure out what part of the code
2246          * refers to osb->journal before ocfs2_journal_init() is run,
2247          * be my guest.
2248          */
2249         /* initialize our journal structure */
2250
2251         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2252         if (!journal) {
2253                 mlog(ML_ERROR, "unable to alloc journal\n");
2254                 status = -ENOMEM;
2255                 goto bail;
2256         }
2257         osb->journal = journal;
2258         journal->j_osb = osb;
2259
2260         atomic_set(&journal->j_num_trans, 0);
2261         init_rwsem(&journal->j_trans_barrier);
2262         init_waitqueue_head(&journal->j_checkpointed);
2263         spin_lock_init(&journal->j_lock);
2264         journal->j_trans_id = (unsigned long) 1;
2265         INIT_LIST_HEAD(&journal->j_la_cleanups);
2266         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2267         journal->j_state = OCFS2_JOURNAL_FREE;
2268
2269         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2270         init_llist_head(&osb->dquot_drop_list);
2271
2272         /* get some pseudo constants for clustersize bits */
2273         osb->s_clustersize_bits =
2274                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2275         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2276
2277         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2278             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2279                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2280                      osb->s_clustersize);
2281                 status = -EINVAL;
2282                 goto bail;
2283         }
2284
2285         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2286                                                 le32_to_cpu(di->i_clusters));
2287
2288         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2289                                            total_blocks);
2290         if (status) {
2291                 mlog(ML_ERROR, "Volume too large "
2292                      "to mount safely on this system");
2293                 status = -EFBIG;
2294                 goto bail;
2295         }
2296
2297         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2298                                  sizeof(di->id2.i_super.s_uuid))) {
2299                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2300                 status = -ENOMEM;
2301                 goto bail;
2302         }
2303
2304         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2305                 OCFS2_MAX_VOL_LABEL_LEN);
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 *)&di->id2.i_super.s_uuid, 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                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2466                        "unmounted cleanly, recovering it.\n", osb->dev_str);
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         kfree(local_alloc);
2520
2521         if (status)
2522                 mlog_errno(status);
2523         return status;
2524 }
2525
2526 /*
2527  * The routine gets called from dismount or close whenever a dismount on
2528  * volume is requested and the osb open count becomes 1.
2529  * It will remove the osb from the global list and also free up all the
2530  * initialized resources and fileobject.
2531  */
2532 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2533 {
2534         /* This function assumes that the caller has the main osb resource */
2535
2536         ocfs2_free_slot_info(osb);
2537
2538         kfree(osb->osb_orphan_wipes);
2539         kfree(osb->slot_recovery_generations);
2540         /* FIXME
2541          * This belongs in journal shutdown, but because we have to
2542          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2543          * we free it here.
2544          */
2545         kfree(osb->journal);
2546         kfree(osb->local_alloc_copy);
2547         kfree(osb->uuid_str);
2548         kfree(osb->vol_label);
2549         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2550         memset(osb, 0, sizeof(struct ocfs2_super));
2551 }
2552
2553 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2554  * panic(). We do not support continue-on-error operation. */
2555 static void ocfs2_handle_error(struct super_block *sb)
2556 {
2557         struct ocfs2_super *osb = OCFS2_SB(sb);
2558
2559         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2560                 panic("OCFS2: (device %s): panic forced after error\n",
2561                       sb->s_id);
2562
2563         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2564
2565         if (sb->s_flags & MS_RDONLY &&
2566             (ocfs2_is_soft_readonly(osb) ||
2567              ocfs2_is_hard_readonly(osb)))
2568                 return;
2569
2570         printk(KERN_CRIT "File system is now read-only due to the potential "
2571                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2572                "system is unmounted.\n");
2573         sb->s_flags |= MS_RDONLY;
2574         ocfs2_set_ro_flag(osb, 0);
2575 }
2576
2577 static char error_buf[1024];
2578
2579 void __ocfs2_error(struct super_block *sb,
2580                    const char *function,
2581                    const char *fmt, ...)
2582 {
2583         va_list args;
2584
2585         va_start(args, fmt);
2586         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2587         va_end(args);
2588
2589         /* Not using mlog here because we want to show the actual
2590          * function the error came from. */
2591         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2592                sb->s_id, function, error_buf);
2593
2594         ocfs2_handle_error(sb);
2595 }
2596
2597 /* Handle critical errors. This is intentionally more drastic than
2598  * ocfs2_handle_error, so we only use for things like journal errors,
2599  * etc. */
2600 void __ocfs2_abort(struct super_block* sb,
2601                    const char *function,
2602                    const char *fmt, ...)
2603 {
2604         va_list args;
2605
2606         va_start(args, fmt);
2607         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2608         va_end(args);
2609
2610         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2611                sb->s_id, function, error_buf);
2612
2613         /* We don't have the cluster support yet to go straight to
2614          * hard readonly in here. Until then, we want to keep
2615          * ocfs2_abort() so that we can at least mark critical
2616          * errors.
2617          *
2618          * TODO: This should abort the journal and alert other nodes
2619          * that our slot needs recovery. */
2620
2621         /* Force a panic(). This stinks, but it's better than letting
2622          * things continue without having a proper hard readonly
2623          * here. */
2624         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2625                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2626         ocfs2_handle_error(sb);
2627 }
2628
2629 /*
2630  * Void signal blockers, because in-kernel sigprocmask() only fails
2631  * when SIG_* is wrong.
2632  */
2633 void ocfs2_block_signals(sigset_t *oldset)
2634 {
2635         int rc;
2636         sigset_t blocked;
2637
2638         sigfillset(&blocked);
2639         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2640         BUG_ON(rc);
2641 }
2642
2643 void ocfs2_unblock_signals(sigset_t *oldset)
2644 {
2645         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2646         BUG_ON(rc);
2647 }
2648
2649 module_init(ocfs2_init);
2650 module_exit(ocfs2_exit);