fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45 #include <linux/signal.h>
  46
  47 #define CREATE_TRACE_POINTS
  48 #include "ocfs2_trace.h"
  49
  50 #include <cluster/masklog.h>
  51
  52 #include "ocfs2.h"
  53
  54 /* this should be the only file to include a version 1 header */
  55 #include "ocfs1_fs_compat.h"
  56
  57 #include "alloc.h"
  58 #include "aops.h"
  59 #include "blockcheck.h"
  60 #include "dlmglue.h"
  61 #include "export.h"
  62 #include "extent_map.h"
  63 #include "heartbeat.h"
  64 #include "inode.h"
  65 #include "journal.h"
  66 #include "localalloc.h"
  67 #include "namei.h"
  68 #include "slot_map.h"
  69 #include "super.h"
  70 #include "sysfile.h"
  71 #include "uptodate.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78 #include "filecheck.h"
  79
  80 static struct kmem_cache *ocfs2_inode_cachep;
  81 struct kmem_cache *ocfs2_dquot_cachep;
  82 struct kmem_cache *ocfs2_qf_chunk_cachep;
  83
  84 static struct dentry *ocfs2_debugfs_root;
  85
  86 MODULE_AUTHOR("Oracle");
  87 MODULE_LICENSE("GPL");
  88 MODULE_DESCRIPTION("OCFS2 cluster file system");
  89
  90 struct mount_options
  91 {
  92         unsigned long   commit_interval;
  93         unsigned long   mount_opt;
  94         unsigned int    atime_quantum;
  95         signed short    slot;
  96         int             localalloc_opt;
  97         unsigned int    resv_level;
  98         int             dir_resv_level;
  99         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 100 };
 101
 102 static int ocfs2_parse_options(struct super_block *sb, char *options,
 103                                struct mount_options *mopt,
 104                                int is_remount);
 105 static int ocfs2_check_set_options(struct super_block *sb,
 106                                    struct mount_options *options);
 107 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 108 static void ocfs2_put_super(struct super_block *sb);
 109 static int ocfs2_mount_volume(struct super_block *sb);
 110 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 111 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 112 static int ocfs2_initialize_mem_caches(void);
 113 static void ocfs2_free_mem_caches(void);
 114 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 115
 116 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 117
 118 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 119
 120 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 121 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 122 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 123 static int ocfs2_check_volume(struct ocfs2_super *osb);
 124 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 125                                struct buffer_head *bh,
 126                                u32 sectsize,
 127                                struct ocfs2_blockcheck_stats *stats);
 128 static int ocfs2_initialize_super(struct super_block *sb,
 129                                   struct buffer_head *bh,
 130                                   int sector_size,
 131                                   struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_get_sector(struct super_block *sb,
 133                             struct buffer_head **bh,
 134                             int block,
 135                             int sect_size);
 136 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 137 static void ocfs2_free_inode(struct inode *inode);
 138 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 139 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 140 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 141
 142 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 143 {
 144         return OCFS2_I(inode)->i_dquot;
 145 }
 146
 147 static const struct super_operations ocfs2_sops = {
 148         .statfs         = ocfs2_statfs,
 149         .alloc_inode    = ocfs2_alloc_inode,
 150         .free_inode     = ocfs2_free_inode,
 151         .drop_inode     = ocfs2_drop_inode,
 152         .evict_inode    = ocfs2_evict_inode,
 153         .sync_fs        = ocfs2_sync_fs,
 154         .put_super      = ocfs2_put_super,
 155         .remount_fs     = ocfs2_remount,
 156         .show_options   = ocfs2_show_options,
 157         .quota_read     = ocfs2_quota_read,
 158         .quota_write    = ocfs2_quota_write,
 159         .get_dquots     = ocfs2_get_dquots,
 160 };
 161
 162 enum {
 163         Opt_barrier,
 164         Opt_err_panic,
 165         Opt_err_ro,
 166         Opt_intr,
 167         Opt_nointr,
 168         Opt_hb_none,
 169         Opt_hb_local,
 170         Opt_hb_global,
 171         Opt_data_ordered,
 172         Opt_data_writeback,
 173         Opt_atime_quantum,
 174         Opt_slot,
 175         Opt_commit,
 176         Opt_localalloc,
 177         Opt_localflocks,
 178         Opt_stack,
 179         Opt_user_xattr,
 180         Opt_nouser_xattr,
 181         Opt_inode64,
 182         Opt_acl,
 183         Opt_noacl,
 184         Opt_usrquota,
 185         Opt_grpquota,
 186         Opt_coherency_buffered,
 187         Opt_coherency_full,
 188         Opt_resv_level,
 189         Opt_dir_resv_level,
 190         Opt_journal_async_commit,
 191         Opt_err_cont,
 192         Opt_err,
 193 };
 194
 195 static const match_table_t tokens = {
 196         {Opt_barrier, "barrier=%u"},
 197         {Opt_err_panic, "errors=panic"},
 198         {Opt_err_ro, "errors=remount-ro"},
 199         {Opt_intr, "intr"},
 200         {Opt_nointr, "nointr"},
 201         {Opt_hb_none, OCFS2_HB_NONE},
 202         {Opt_hb_local, OCFS2_HB_LOCAL},
 203         {Opt_hb_global, OCFS2_HB_GLOBAL},
 204         {Opt_data_ordered, "data=ordered"},
 205         {Opt_data_writeback, "data=writeback"},
 206         {Opt_atime_quantum, "atime_quantum=%u"},
 207         {Opt_slot, "preferred_slot=%u"},
 208         {Opt_commit, "commit=%u"},
 209         {Opt_localalloc, "localalloc=%d"},
 210         {Opt_localflocks, "localflocks"},
 211         {Opt_stack, "cluster_stack=%s"},
 212         {Opt_user_xattr, "user_xattr"},
 213         {Opt_nouser_xattr, "nouser_xattr"},
 214         {Opt_inode64, "inode64"},
 215         {Opt_acl, "acl"},
 216         {Opt_noacl, "noacl"},
 217         {Opt_usrquota, "usrquota"},
 218         {Opt_grpquota, "grpquota"},
 219         {Opt_coherency_buffered, "coherency=buffered"},
 220         {Opt_coherency_full, "coherency=full"},
 221         {Opt_resv_level, "resv_level=%u"},
 222         {Opt_dir_resv_level, "dir_resv_level=%u"},
 223         {Opt_journal_async_commit, "journal_async_commit"},
 224         {Opt_err_cont, "errors=continue"},
 225         {Opt_err, NULL}
 226 };
 227
 228 #ifdef CONFIG_DEBUG_FS
 229 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 230 {
 231         struct ocfs2_cluster_connection *cconn = osb->cconn;
 232         struct ocfs2_recovery_map *rm = osb->recovery_map;
 233         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 234         int i, out = 0;
 235         unsigned long flags;
 236
 237         out += snprintf(buf + out, len - out,
 238                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 239                         "Device", osb->dev_str, osb->uuid_str,
 240                         osb->fs_generation, osb->vol_label);
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 244                         atomic_read(&osb->vol_state), osb->osb_flags);
 245
 246         out += snprintf(buf + out, len - out,
 247                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 248                         osb->sb->s_blocksize, osb->s_clustersize);
 249
 250         out += snprintf(buf + out, len - out,
 251                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 252                         "ROcompat: 0x%X\n",
 253                         "Features", osb->s_feature_compat,
 254                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 255
 256         out += snprintf(buf + out, len - out,
 257                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 258                         osb->s_mount_opt, osb->s_atime_quantum);
 259
 260         if (cconn) {
 261                 out += snprintf(buf + out, len - out,
 262                                 "%10s => Stack: %s  Name: %*s  "
 263                                 "Version: %d.%d\n", "Cluster",
 264                                 (*osb->osb_cluster_stack == '\0' ?
 265                                  "o2cb" : osb->osb_cluster_stack),
 266                                 cconn->cc_namelen, cconn->cc_name,
 267                                 cconn->cc_version.pv_major,
 268                                 cconn->cc_version.pv_minor);
 269         }
 270
 271         spin_lock_irqsave(&osb->dc_task_lock, flags);
 272         out += snprintf(buf + out, len - out,
 273                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 274                         "WorkSeq: %lu\n", "DownCnvt",
 275                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 276                         osb->blocked_lock_count, osb->dc_wake_sequence,
 277                         osb->dc_work_sequence);
 278         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 279
 280         spin_lock(&osb->osb_lock);
 281         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 282                         "Recovery",
 283                         (osb->recovery_thread_task ?
 284                          task_pid_nr(osb->recovery_thread_task) : -1));
 285         if (rm->rm_used == 0)
 286                 out += snprintf(buf + out, len - out, " None\n");
 287         else {
 288                 for (i = 0; i < rm->rm_used; i++)
 289                         out += snprintf(buf + out, len - out, " %d",
 290                                         rm->rm_entries[i]);
 291                 out += snprintf(buf + out, len - out, "\n");
 292         }
 293         spin_unlock(&osb->osb_lock);
 294
 295         out += snprintf(buf + out, len - out,
 296                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 297                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 298                         osb->osb_commit_interval);
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 302                         "Journal", osb->journal->j_state,
 303                         osb->journal->j_trans_id,
 304                         atomic_read(&osb->journal->j_num_trans));
 305
 306         out += snprintf(buf + out, len - out,
 307                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 308                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 309                         "Stats",
 310                         atomic_read(&osb->alloc_stats.bitmap_data),
 311                         atomic_read(&osb->alloc_stats.local_data),
 312                         atomic_read(&osb->alloc_stats.bg_allocs),
 313                         atomic_read(&osb->alloc_stats.moves),
 314                         atomic_read(&osb->alloc_stats.bg_extends));
 315
 316         out += snprintf(buf + out, len - out,
 317                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 318                         "Default: %u bits\n",
 319                         "LocalAlloc", osb->local_alloc_state,
 320                         (unsigned long long)osb->la_last_gd,
 321                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 322
 323         spin_lock(&osb->osb_lock);
 324         out += snprintf(buf + out, len - out,
 325                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 326                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 327                         osb->s_inode_steal_slot,
 328                         atomic_read(&osb->s_num_inodes_stolen),
 329                         osb->s_meta_steal_slot,
 330                         atomic_read(&osb->s_num_meta_stolen));
 331         spin_unlock(&osb->osb_lock);
 332
 333         out += snprintf(buf + out, len - out, "OrphanScan => ");
 334         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 335                         os->os_count, os->os_seqno);
 336         out += snprintf(buf + out, len - out, " Last Scan: ");
 337         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 338                 out += snprintf(buf + out, len - out, "Disabled\n");
 339         else
 340                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 341                                 (unsigned long)(ktime_get_seconds() - os->os_scantime));
 342
 343         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 344                         "Slots", "Num", "RecoGen");
 345         for (i = 0; i < osb->max_slots; ++i) {
 346                 out += snprintf(buf + out, len - out,
 347                                 "%10s  %c %3d  %10d\n",
 348                                 " ",
 349                                 (i == osb->slot_num ? '*' : ' '),
 350                                 i, osb->slot_recovery_generations[i]);
 351         }
 352
 353         return out;
 354 }
 355
 356 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 357 {
 358         struct ocfs2_super *osb = inode->i_private;
 359         char *buf = NULL;
 360
 361         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 362         if (!buf)
 363                 goto bail;
 364
 365         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 366
 367         file->private_data = buf;
 368
 369         return 0;
 370 bail:
 371         return -ENOMEM;
 372 }
 373
 374 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 375 {
 376         kfree(file->private_data);
 377         return 0;
 378 }
 379
 380 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 381                                 size_t nbytes, loff_t *ppos)
 382 {
 383         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 384                                        i_size_read(file->f_mapping->host));
 385 }
 386 #else
 387 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 388 {
 389         return 0;
 390 }
 391 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 392 {
 393         return 0;
 394 }
 395 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 396                                 size_t nbytes, loff_t *ppos)
 397 {
 398         return 0;
 399 }
 400 #endif  /* CONFIG_DEBUG_FS */
 401
 402 static const struct file_operations ocfs2_osb_debug_fops = {
 403         .open =         ocfs2_osb_debug_open,
 404         .release =      ocfs2_debug_release,
 405         .read =         ocfs2_debug_read,
 406         .llseek =       generic_file_llseek,
 407 };
 408
 409 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 410 {
 411         int status;
 412         tid_t target;
 413         struct ocfs2_super *osb = OCFS2_SB(sb);
 414
 415         if (ocfs2_is_hard_readonly(osb))
 416                 return -EROFS;
 417
 418         if (wait) {
 419                 status = ocfs2_flush_truncate_log(osb);
 420                 if (status < 0)
 421                         mlog_errno(status);
 422         } else {
 423                 ocfs2_schedule_truncate_log_flush(osb, 0);
 424         }
 425
 426         if (jbd2_journal_start_commit(osb->journal->j_journal,
 427                                       &target)) {
 428                 if (wait)
 429                         jbd2_log_wait_commit(osb->journal->j_journal,
 430                                              target);
 431         }
 432         return 0;
 433 }
 434
 435 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 436 {
 437         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 438             && (ino == USER_QUOTA_SYSTEM_INODE
 439                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 440                 return 0;
 441         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 442             && (ino == GROUP_QUOTA_SYSTEM_INODE
 443                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 444                 return 0;
 445         return 1;
 446 }
 447
 448 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 449 {
 450         struct inode *new = NULL;
 451         int status = 0;
 452         int i;
 453
 454         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 455         if (IS_ERR(new)) {
 456                 status = PTR_ERR(new);
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460         osb->root_inode = new;
 461
 462         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 463         if (IS_ERR(new)) {
 464                 status = PTR_ERR(new);
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468         osb->sys_root_inode = new;
 469
 470         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 471              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 472                 if (!ocfs2_need_system_inode(osb, i))
 473                         continue;
 474                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 475                 if (!new) {
 476                         ocfs2_release_system_inodes(osb);
 477                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 478                         mlog_errno(status);
 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 = ocfs2_is_soft_readonly(osb) ? -EROFS : -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_free_inode(struct inode *inode)
 579 {
 580         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 581 }
 582
 583 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 584                                                 unsigned int cbits)
 585 {
 586         unsigned int bytes = 1 << cbits;
 587         unsigned int trim = bytes;
 588         unsigned int bitshift = 32;
 589
 590         /*
 591          * i_size and all block offsets in ocfs2 are always 64 bits
 592          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 593          * 64 bit platforms, cluster size will be the limiting factor.
 594          */
 595
 596 #if BITS_PER_LONG == 32
 597         BUILD_BUG_ON(sizeof(sector_t) != 8);
 598         /*
 599          * We might be limited by page cache size.
 600          */
 601         if (bytes > PAGE_SIZE) {
 602                 bytes = PAGE_SIZE;
 603                 trim = 1;
 604                 /*
 605                  * Shift by 31 here so that we don't get larger than
 606                  * MAX_LFS_FILESIZE
 607                  */
 608                 bitshift = 31;
 609         }
 610 #endif
 611
 612         /*
 613          * Trim by a whole cluster when we can actually approach the
 614          * on-disk limits. Otherwise we can overflow i_clusters when
 615          * an extent start is at the max offset.
 616          */
 617         return (((unsigned long long)bytes) << bitshift) - trim;
 618 }
 619
 620 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 621 {
 622         int incompat_features;
 623         int ret = 0;
 624         struct mount_options parsed_options;
 625         struct ocfs2_super *osb = OCFS2_SB(sb);
 626         u32 tmp;
 627
 628         sync_filesystem(sb);
 629
 630         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 631             !ocfs2_check_set_options(sb, &parsed_options)) {
 632                 ret = -EINVAL;
 633                 goto out;
 634         }
 635
 636         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 637                 OCFS2_MOUNT_HB_NONE;
 638         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 639                 ret = -EINVAL;
 640                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 641                 goto out;
 642         }
 643
 644         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 645             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 646                 ret = -EINVAL;
 647                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 648                 goto out;
 649         }
 650
 651         /* Probably don't want this on remount; it might
 652          * mess with other nodes */
 653         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 654             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 657                 goto out;
 658         }
 659
 660         /* We're going to/from readonly mode. */
 661         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
 662                 /* Disable quota accounting before remounting RO */
 663                 if (*flags & SB_RDONLY) {
 664                         ret = ocfs2_susp_quotas(osb, 0);
 665                         if (ret < 0)
 666                                 goto out;
 667                 }
 668                 /* Lock here so the check of HARD_RO and the potential
 669                  * setting of SOFT_RO is atomic. */
 670                 spin_lock(&osb->osb_lock);
 671                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 672                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 673                         ret = -EROFS;
 674                         goto unlock_osb;
 675                 }
 676
 677                 if (*flags & SB_RDONLY) {
 678                         sb->s_flags |= SB_RDONLY;
 679                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 680                 } else {
 681                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 682                                 mlog(ML_ERROR, "Cannot remount RDWR "
 683                                      "filesystem due to previous errors.\n");
 684                                 ret = -EROFS;
 685                                 goto unlock_osb;
 686                         }
 687                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 688                         if (incompat_features) {
 689                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 690                                      "of unsupported optional features "
 691                                      "(%x).\n", incompat_features);
 692                                 ret = -EINVAL;
 693                                 goto unlock_osb;
 694                         }
 695                         sb->s_flags &= ~SB_RDONLY;
 696                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 697                 }
 698                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 699 unlock_osb:
 700                 spin_unlock(&osb->osb_lock);
 701                 /* Enable quota accounting after remounting RW */
 702                 if (!ret && !(*flags & SB_RDONLY)) {
 703                         if (sb_any_quota_suspended(sb))
 704                                 ret = ocfs2_susp_quotas(osb, 1);
 705                         else
 706                                 ret = ocfs2_enable_quotas(osb);
 707                         if (ret < 0) {
 708                                 /* Return back changes... */
 709                                 spin_lock(&osb->osb_lock);
 710                                 sb->s_flags |= SB_RDONLY;
 711                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 712                                 spin_unlock(&osb->osb_lock);
 713                                 goto out;
 714                         }
 715                 }
 716         }
 717
 718         if (!ret) {
 719                 /* Only save off the new mount options in case of a successful
 720                  * remount. */
 721                 osb->s_mount_opt = parsed_options.mount_opt;
 722                 osb->s_atime_quantum = parsed_options.atime_quantum;
 723                 osb->preferred_slot = parsed_options.slot;
 724                 if (parsed_options.commit_interval)
 725                         osb->osb_commit_interval = parsed_options.commit_interval;
 726
 727                 if (!ocfs2_is_hard_readonly(osb))
 728                         ocfs2_set_journal_params(osb);
 729
 730                 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
 731                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 732                                                         SB_POSIXACL : 0);
 733         }
 734 out:
 735         return ret;
 736 }
 737
 738 static int ocfs2_sb_probe(struct super_block *sb,
 739                           struct buffer_head **bh,
 740                           int *sector_size,
 741                           struct ocfs2_blockcheck_stats *stats)
 742 {
 743         int status, tmpstat;
 744         struct ocfs1_vol_disk_hdr *hdr;
 745         struct ocfs2_dinode *di;
 746         int blksize;
 747
 748         *bh = NULL;
 749
 750         /* may be > 512 */
 751         *sector_size = bdev_logical_block_size(sb->s_bdev);
 752         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 753                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 754                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 755                 status = -EINVAL;
 756                 goto bail;
 757         }
 758
 759         /* Can this really happen? */
 760         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 761                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 762
 763         /* check block zero for old format */
 764         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 765         if (status < 0) {
 766                 mlog_errno(status);
 767                 goto bail;
 768         }
 769         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 770         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 771                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 772                      hdr->major_version, hdr->minor_version);
 773                 status = -EINVAL;
 774         }
 775         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 776                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 777                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 778                      hdr->signature);
 779                 status = -EINVAL;
 780         }
 781         brelse(*bh);
 782         *bh = NULL;
 783         if (status < 0) {
 784                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 785                      "upgraded before mounting with ocfs v2\n");
 786                 goto bail;
 787         }
 788
 789         /*
 790          * Now check at magic offset for 512, 1024, 2048, 4096
 791          * blocksizes.  4096 is the maximum blocksize because it is
 792          * the minimum clustersize.
 793          */
 794         status = -EINVAL;
 795         for (blksize = *sector_size;
 796              blksize <= OCFS2_MAX_BLOCKSIZE;
 797              blksize <<= 1) {
 798                 tmpstat = ocfs2_get_sector(sb, bh,
 799                                            OCFS2_SUPER_BLOCK_BLKNO,
 800                                            blksize);
 801                 if (tmpstat < 0) {
 802                         status = tmpstat;
 803                         mlog_errno(status);
 804                         break;
 805                 }
 806                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 807                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 808                 spin_lock_init(&stats->b_lock);
 809                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 810                 if (tmpstat < 0) {
 811                         brelse(*bh);
 812                         *bh = NULL;
 813                 }
 814                 if (tmpstat != -EAGAIN) {
 815                         status = tmpstat;
 816                         break;
 817                 }
 818         }
 819
 820 bail:
 821         return status;
 822 }
 823
 824 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 825 {
 826         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 827
 828         if (osb->s_mount_opt & hb_enabled) {
 829                 if (ocfs2_mount_local(osb)) {
 830                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 831                              "mounted device.\n");
 832                         return -EINVAL;
 833                 }
 834                 if (ocfs2_userspace_stack(osb)) {
 835                         mlog(ML_ERROR, "Userspace stack expected, but "
 836                              "o2cb heartbeat arguments passed to mount\n");
 837                         return -EINVAL;
 838                 }
 839                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 840                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 841                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 842                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 843                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 844                         return -EINVAL;
 845                 }
 846         }
 847
 848         if (!(osb->s_mount_opt & hb_enabled)) {
 849                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 850                     !ocfs2_userspace_stack(osb)) {
 851                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 852                              "a read-write clustered device.\n");
 853                         return -EINVAL;
 854                 }
 855         }
 856
 857         return 0;
 858 }
 859
 860 /*
 861  * If we're using a userspace stack, mount should have passed
 862  * a name that matches the disk.  If not, mount should not
 863  * have passed a stack.
 864  */
 865 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 866                                         struct mount_options *mopt)
 867 {
 868         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 869                 mlog(ML_ERROR,
 870                      "cluster stack passed to mount, but this filesystem "
 871                      "does not support it\n");
 872                 return -EINVAL;
 873         }
 874
 875         if (ocfs2_userspace_stack(osb) &&
 876             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 877                     OCFS2_STACK_LABEL_LEN)) {
 878                 mlog(ML_ERROR,
 879                      "cluster stack passed to mount (\"%s\") does not "
 880                      "match the filesystem (\"%s\")\n",
 881                      mopt->cluster_stack,
 882                      osb->osb_cluster_stack);
 883                 return -EINVAL;
 884         }
 885
 886         return 0;
 887 }
 888
 889 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 890 {
 891         int type;
 892         struct super_block *sb = osb->sb;
 893         unsigned int feature[OCFS2_MAXQUOTAS] = {
 894                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 895                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 896         int status = 0;
 897
 898         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 899                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 900                         continue;
 901                 if (unsuspend)
 902                         status = dquot_resume(sb, type);
 903                 else {
 904                         struct ocfs2_mem_dqinfo *oinfo;
 905
 906                         /* Cancel periodic syncing before suspending */
 907                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 908                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 909                         status = dquot_suspend(sb, type);
 910                 }
 911                 if (status < 0)
 912                         break;
 913         }
 914         if (status < 0)
 915                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 916                      "remount (error = %d).\n", status);
 917         return status;
 918 }
 919
 920 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 921 {
 922         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 923         struct super_block *sb = osb->sb;
 924         unsigned int feature[OCFS2_MAXQUOTAS] = {
 925                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 926                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 927         unsigned int ino[OCFS2_MAXQUOTAS] = {
 928                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 929                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 930         int status;
 931         int type;
 932
 933         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 934         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 935                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 936                         continue;
 937                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 938                                                         osb->slot_num);
 939                 if (!inode[type]) {
 940                         status = -ENOENT;
 941                         goto out_quota_off;
 942                 }
 943                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 944                                       DQUOT_USAGE_ENABLED);
 945                 if (status < 0)
 946                         goto out_quota_off;
 947         }
 948
 949         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 950                 iput(inode[type]);
 951         return 0;
 952 out_quota_off:
 953         ocfs2_disable_quotas(osb);
 954         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 955                 iput(inode[type]);
 956         mlog_errno(status);
 957         return status;
 958 }
 959
 960 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 961 {
 962         int type;
 963         struct inode *inode;
 964         struct super_block *sb = osb->sb;
 965         struct ocfs2_mem_dqinfo *oinfo;
 966
 967         /* We mostly ignore errors in this function because there's not much
 968          * we can do when we see them */
 969         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 970                 if (!sb_has_quota_loaded(sb, type))
 971                         continue;
 972                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 973                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 974                 inode = igrab(sb->s_dquot.files[type]);
 975                 /* Turn off quotas. This will remove all dquot structures from
 976                  * memory and so they will be automatically synced to global
 977                  * quota files */
 978                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 979                                         DQUOT_LIMITS_ENABLED);
 980                 if (!inode)
 981                         continue;
 982                 iput(inode);
 983         }
 984 }
 985
 986 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 987 {
 988         struct dentry *root;
 989         int status, sector_size;
 990         struct mount_options parsed_options;
 991         struct inode *inode = NULL;
 992         struct ocfs2_super *osb = NULL;
 993         struct buffer_head *bh = NULL;
 994         char nodestr[12];
 995         struct ocfs2_blockcheck_stats stats;
 996
 997         trace_ocfs2_fill_super(sb, data, silent);
 998
 999         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1000                 status = -EINVAL;
1001                 goto read_super_error;
1002         }
1003
1004         /* probe for superblock */
1005         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1006         if (status < 0) {
1007                 mlog(ML_ERROR, "superblock probe failed!\n");
1008                 goto read_super_error;
1009         }
1010
1011         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1012         osb = OCFS2_SB(sb);
1013         if (status < 0) {
1014                 mlog_errno(status);
1015                 goto read_super_error;
1016         }
1017         brelse(bh);
1018         bh = NULL;
1019
1020         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1021                 status = -EINVAL;
1022                 goto read_super_error;
1023         }
1024         osb->s_mount_opt = parsed_options.mount_opt;
1025         osb->s_atime_quantum = parsed_options.atime_quantum;
1026         osb->preferred_slot = parsed_options.slot;
1027         osb->osb_commit_interval = parsed_options.commit_interval;
1028
1029         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1030         osb->osb_resv_level = parsed_options.resv_level;
1031         osb->osb_dir_resv_level = parsed_options.resv_level;
1032         if (parsed_options.dir_resv_level == -1)
1033                 osb->osb_dir_resv_level = parsed_options.resv_level;
1034         else
1035                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1036
1037         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1038         if (status)
1039                 goto read_super_error;
1040
1041         sb->s_magic = OCFS2_SUPER_MAGIC;
1042
1043         sb->s_flags = (sb->s_flags & ~(SB_POSIXACL | SB_NOSEC)) |
1044                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? SB_POSIXACL : 0);
1045
1046         /* Hard readonly mode only if: bdev_read_only, SB_RDONLY,
1047          * heartbeat=none */
1048         if (bdev_read_only(sb->s_bdev)) {
1049                 if (!sb_rdonly(sb)) {
1050                         status = -EACCES;
1051                         mlog(ML_ERROR, "Readonly device detected but readonly "
1052                              "mount was not specified.\n");
1053                         goto read_super_error;
1054                 }
1055
1056                 /* You should not be able to start a local heartbeat
1057                  * on a readonly device. */
1058                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1059                         status = -EROFS;
1060                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1061                              "device.\n");
1062                         goto read_super_error;
1063                 }
1064
1065                 status = ocfs2_check_journals_nolocks(osb);
1066                 if (status < 0) {
1067                         if (status == -EROFS)
1068                                 mlog(ML_ERROR, "Recovery required on readonly "
1069                                      "file system, but write access is "
1070                                      "unavailable.\n");
1071                         else
1072                                 mlog_errno(status);
1073                         goto read_super_error;
1074                 }
1075
1076                 ocfs2_set_ro_flag(osb, 1);
1077
1078                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1079                        "Cluster services will not be used for this mount. "
1080                        "Recovery will be skipped.\n", osb->dev_str);
1081         }
1082
1083         if (!ocfs2_is_hard_readonly(osb)) {
1084                 if (sb_rdonly(sb))
1085                         ocfs2_set_ro_flag(osb, 0);
1086         }
1087
1088         status = ocfs2_verify_heartbeat(osb);
1089         if (status < 0) {
1090                 mlog_errno(status);
1091                 goto read_super_error;
1092         }
1093
1094         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1095                                                  ocfs2_debugfs_root);
1096         if (!osb->osb_debug_root) {
1097                 status = -EINVAL;
1098                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1099                 goto read_super_error;
1100         }
1101
1102         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1103                                             osb->osb_debug_root,
1104                                             osb,
1105                                             &ocfs2_osb_debug_fops);
1106         if (!osb->osb_ctxt) {
1107                 status = -EINVAL;
1108                 mlog_errno(status);
1109                 goto read_super_error;
1110         }
1111
1112         if (ocfs2_meta_ecc(osb)) {
1113                 status = ocfs2_blockcheck_stats_debugfs_install(
1114                                                 &osb->osb_ecc_stats,
1115                                                 osb->osb_debug_root);
1116                 if (status) {
1117                         mlog(ML_ERROR,
1118                              "Unable to create blockcheck statistics "
1119                              "files\n");
1120                         goto read_super_error;
1121                 }
1122         }
1123
1124         status = ocfs2_mount_volume(sb);
1125         if (status < 0)
1126                 goto read_super_error;
1127
1128         if (osb->root_inode)
1129                 inode = igrab(osb->root_inode);
1130
1131         if (!inode) {
1132                 status = -EIO;
1133                 mlog_errno(status);
1134                 goto read_super_error;
1135         }
1136
1137         root = d_make_root(inode);
1138         if (!root) {
1139                 status = -ENOMEM;
1140                 mlog_errno(status);
1141                 goto read_super_error;
1142         }
1143
1144         sb->s_root = root;
1145
1146         ocfs2_complete_mount_recovery(osb);
1147
1148         osb->osb_dev_kset = kset_create_and_add(sb->s_id, NULL,
1149                                                 &ocfs2_kset->kobj);
1150         if (!osb->osb_dev_kset) {
1151                 status = -ENOMEM;
1152                 mlog(ML_ERROR, "Unable to create device kset %s.\n", sb->s_id);
1153                 goto read_super_error;
1154         }
1155
1156         /* Create filecheck sysfs related directories/files at
1157          * /sys/fs/ocfs2/<devname>/filecheck */
1158         if (ocfs2_filecheck_create_sysfs(osb)) {
1159                 status = -ENOMEM;
1160                 mlog(ML_ERROR, "Unable to create filecheck sysfs directory at "
1161                         "/sys/fs/ocfs2/%s/filecheck.\n", sb->s_id);
1162                 goto read_super_error;
1163         }
1164
1165         if (ocfs2_mount_local(osb))
1166                 snprintf(nodestr, sizeof(nodestr), "local");
1167         else
1168                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1169
1170         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1171                "with %s data mode.\n",
1172                osb->dev_str, nodestr, osb->slot_num,
1173                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1174                "ordered");
1175
1176         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1177         wake_up(&osb->osb_mount_event);
1178
1179         /* Now we can initialize quotas because we can afford to wait
1180          * for cluster locks recovery now. That also means that truncation
1181          * log recovery can happen but that waits for proper quota setup */
1182         if (!sb_rdonly(sb)) {
1183                 status = ocfs2_enable_quotas(osb);
1184                 if (status < 0) {
1185                         /* We have to err-out specially here because
1186                          * s_root is already set */
1187                         mlog_errno(status);
1188                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1189                         wake_up(&osb->osb_mount_event);
1190                         return status;
1191                 }
1192         }
1193
1194         ocfs2_complete_quota_recovery(osb);
1195
1196         /* Now we wake up again for processes waiting for quotas */
1197         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1198         wake_up(&osb->osb_mount_event);
1199
1200         /* Start this when the mount is almost sure of being successful */
1201         ocfs2_orphan_scan_start(osb);
1202
1203         return status;
1204
1205 read_super_error:
1206         brelse(bh);
1207
1208         if (status)
1209                 mlog_errno(status);
1210
1211         if (osb) {
1212                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1213                 wake_up(&osb->osb_mount_event);
1214                 ocfs2_dismount_volume(sb, 1);
1215         }
1216
1217         return status;
1218 }
1219
1220 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1221                         int flags,
1222                         const char *dev_name,
1223                         void *data)
1224 {
1225         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1226 }
1227
1228 static struct file_system_type ocfs2_fs_type = {
1229         .owner          = THIS_MODULE,
1230         .name           = "ocfs2",
1231         .mount          = ocfs2_mount,
1232         .kill_sb        = kill_block_super,
1233         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1234         .next           = NULL
1235 };
1236 MODULE_ALIAS_FS("ocfs2");
1237
1238 static int ocfs2_check_set_options(struct super_block *sb,
1239                                    struct mount_options *options)
1240 {
1241         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1242             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1243                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1244                 mlog(ML_ERROR, "User quotas were requested, but this "
1245                      "filesystem does not have the feature enabled.\n");
1246                 return 0;
1247         }
1248         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1249             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1250                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1251                 mlog(ML_ERROR, "Group quotas were requested, but this "
1252                      "filesystem does not have the feature enabled.\n");
1253                 return 0;
1254         }
1255         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1256             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1257                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1258                      "feature is not enabled\n");
1259                 return 0;
1260         }
1261         /* No ACL setting specified? Use XATTR feature... */
1262         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1263                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1264                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1265                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1266                 else
1267                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1268         }
1269         return 1;
1270 }
1271
1272 static int ocfs2_parse_options(struct super_block *sb,
1273                                char *options,
1274                                struct mount_options *mopt,
1275                                int is_remount)
1276 {
1277         int status, user_stack = 0;
1278         char *p;
1279         u32 tmp;
1280         int token, option;
1281         substring_t args[MAX_OPT_ARGS];
1282
1283         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1284
1285         mopt->commit_interval = 0;
1286         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1287         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1288         mopt->slot = OCFS2_INVALID_SLOT;
1289         mopt->localalloc_opt = -1;
1290         mopt->cluster_stack[0] = '\0';
1291         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1292         mopt->dir_resv_level = -1;
1293
1294         if (!options) {
1295                 status = 1;
1296                 goto bail;
1297         }
1298
1299         while ((p = strsep(&options, ",")) != NULL) {
1300                 if (!*p)
1301                         continue;
1302
1303                 token = match_token(p, tokens, args);
1304                 switch (token) {
1305                 case Opt_hb_local:
1306                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1307                         break;
1308                 case Opt_hb_none:
1309                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1310                         break;
1311                 case Opt_hb_global:
1312                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1313                         break;
1314                 case Opt_barrier:
1315                         if (match_int(&args[0], &option)) {
1316                                 status = 0;
1317                                 goto bail;
1318                         }
1319                         if (option)
1320                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1321                         else
1322                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1323                         break;
1324                 case Opt_intr:
1325                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1326                         break;
1327                 case Opt_nointr:
1328                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1329                         break;
1330                 case Opt_err_panic:
1331                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1332                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1333                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1334                         break;
1335                 case Opt_err_ro:
1336                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1337                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1338                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1339                         break;
1340                 case Opt_err_cont:
1341                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1342                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1343                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1344                         break;
1345                 case Opt_data_ordered:
1346                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1347                         break;
1348                 case Opt_data_writeback:
1349                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1350                         break;
1351                 case Opt_user_xattr:
1352                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1353                         break;
1354                 case Opt_nouser_xattr:
1355                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1356                         break;
1357                 case Opt_atime_quantum:
1358                         if (match_int(&args[0], &option)) {
1359                                 status = 0;
1360                                 goto bail;
1361                         }
1362                         if (option >= 0)
1363                                 mopt->atime_quantum = option;
1364                         break;
1365                 case Opt_slot:
1366                         if (match_int(&args[0], &option)) {
1367                                 status = 0;
1368                                 goto bail;
1369                         }
1370                         if (option)
1371                                 mopt->slot = (s16)option;
1372                         break;
1373                 case Opt_commit:
1374                         if (match_int(&args[0], &option)) {
1375                                 status = 0;
1376                                 goto bail;
1377                         }
1378                         if (option < 0)
1379                                 return 0;
1380                         if (option == 0)
1381                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1382                         mopt->commit_interval = HZ * option;
1383                         break;
1384                 case Opt_localalloc:
1385                         if (match_int(&args[0], &option)) {
1386                                 status = 0;
1387                                 goto bail;
1388                         }
1389                         if (option >= 0)
1390                                 mopt->localalloc_opt = option;
1391                         break;
1392                 case Opt_localflocks:
1393                         /*
1394                          * Changing this during remount could race
1395                          * flock() requests, or "unbalance" existing
1396                          * ones (e.g., a lock is taken in one mode but
1397                          * dropped in the other). If users care enough
1398                          * to flip locking modes during remount, we
1399                          * could add a "local" flag to individual
1400                          * flock structures for proper tracking of
1401                          * state.
1402                          */
1403                         if (!is_remount)
1404                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1405                         break;
1406                 case Opt_stack:
1407                         /* Check both that the option we were passed
1408                          * is of the right length and that it is a proper
1409                          * string of the right length.
1410                          */
1411                         if (((args[0].to - args[0].from) !=
1412                              OCFS2_STACK_LABEL_LEN) ||
1413                             (strnlen(args[0].from,
1414                                      OCFS2_STACK_LABEL_LEN) !=
1415                              OCFS2_STACK_LABEL_LEN)) {
1416                                 mlog(ML_ERROR,
1417                                      "Invalid cluster_stack option\n");
1418                                 status = 0;
1419                                 goto bail;
1420                         }
1421                         memcpy(mopt->cluster_stack, args[0].from,
1422                                OCFS2_STACK_LABEL_LEN);
1423                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1424                         /*
1425                          * Open code the memcmp here as we don't have
1426                          * an osb to pass to
1427                          * ocfs2_userspace_stack().
1428                          */
1429                         if (memcmp(mopt->cluster_stack,
1430                                    OCFS2_CLASSIC_CLUSTER_STACK,
1431                                    OCFS2_STACK_LABEL_LEN))
1432                                 user_stack = 1;
1433                         break;
1434                 case Opt_inode64:
1435                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1436                         break;
1437                 case Opt_usrquota:
1438                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1439                         break;
1440                 case Opt_grpquota:
1441                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1442                         break;
1443                 case Opt_coherency_buffered:
1444                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1445                         break;
1446                 case Opt_coherency_full:
1447                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1448                         break;
1449                 case Opt_acl:
1450                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1451                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1452                         break;
1453                 case Opt_noacl:
1454                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1455                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1456                         break;
1457                 case Opt_resv_level:
1458                         if (is_remount)
1459                                 break;
1460                         if (match_int(&args[0], &option)) {
1461                                 status = 0;
1462                                 goto bail;
1463                         }
1464                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1465                             option < OCFS2_MAX_RESV_LEVEL)
1466                                 mopt->resv_level = option;
1467                         break;
1468                 case Opt_dir_resv_level:
1469                         if (is_remount)
1470                                 break;
1471                         if (match_int(&args[0], &option)) {
1472                                 status = 0;
1473                                 goto bail;
1474                         }
1475                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1476                             option < OCFS2_MAX_RESV_LEVEL)
1477                                 mopt->dir_resv_level = option;
1478                         break;
1479                 case Opt_journal_async_commit:
1480                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1481                         break;
1482                 default:
1483                         mlog(ML_ERROR,
1484                              "Unrecognized mount option \"%s\" "
1485                              "or missing value\n", p);
1486                         status = 0;
1487                         goto bail;
1488                 }
1489         }
1490
1491         if (user_stack == 0) {
1492                 /* Ensure only one heartbeat mode */
1493                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1494                                          OCFS2_MOUNT_HB_GLOBAL |
1495                                          OCFS2_MOUNT_HB_NONE);
1496                 if (hweight32(tmp) != 1) {
1497                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1498                         status = 0;
1499                         goto bail;
1500                 }
1501         }
1502
1503         status = 1;
1504
1505 bail:
1506         return status;
1507 }
1508
1509 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1510 {
1511         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1512         unsigned long opts = osb->s_mount_opt;
1513         unsigned int local_alloc_megs;
1514
1515         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1516                 seq_printf(s, ",_netdev");
1517                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1518                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1519                 else
1520                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1521         } else
1522                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1523
1524         if (opts & OCFS2_MOUNT_NOINTR)
1525                 seq_printf(s, ",nointr");
1526
1527         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1528                 seq_printf(s, ",data=writeback");
1529         else
1530                 seq_printf(s, ",data=ordered");
1531
1532         if (opts & OCFS2_MOUNT_BARRIER)
1533                 seq_printf(s, ",barrier=1");
1534
1535         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1536                 seq_printf(s, ",errors=panic");
1537         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1538                 seq_printf(s, ",errors=continue");
1539         else
1540                 seq_printf(s, ",errors=remount-ro");
1541
1542         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1543                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1544
1545         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1546
1547         if (osb->osb_commit_interval)
1548                 seq_printf(s, ",commit=%u",
1549                            (unsigned) (osb->osb_commit_interval / HZ));
1550
1551         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1552         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1553                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1554
1555         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1556                 seq_printf(s, ",localflocks,");
1557
1558         if (osb->osb_cluster_stack[0])
1559                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1560                                   OCFS2_STACK_LABEL_LEN);
1561         if (opts & OCFS2_MOUNT_USRQUOTA)
1562                 seq_printf(s, ",usrquota");
1563         if (opts & OCFS2_MOUNT_GRPQUOTA)
1564                 seq_printf(s, ",grpquota");
1565
1566         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1567                 seq_printf(s, ",coherency=buffered");
1568         else
1569                 seq_printf(s, ",coherency=full");
1570
1571         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1572                 seq_printf(s, ",nouser_xattr");
1573         else
1574                 seq_printf(s, ",user_xattr");
1575
1576         if (opts & OCFS2_MOUNT_INODE64)
1577                 seq_printf(s, ",inode64");
1578
1579         if (opts & OCFS2_MOUNT_POSIX_ACL)
1580                 seq_printf(s, ",acl");
1581         else
1582                 seq_printf(s, ",noacl");
1583
1584         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1585                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1586
1587         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1588                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1589
1590         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1591                 seq_printf(s, ",journal_async_commit");
1592
1593         return 0;
1594 }
1595
1596 static int __init ocfs2_init(void)
1597 {
1598         int status;
1599
1600         status = init_ocfs2_uptodate_cache();
1601         if (status < 0)
1602                 goto out1;
1603
1604         status = ocfs2_initialize_mem_caches();
1605         if (status < 0)
1606                 goto out2;
1607
1608         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1609         if (!ocfs2_debugfs_root) {
1610                 status = -ENOMEM;
1611                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1612                 goto out3;
1613         }
1614
1615         ocfs2_set_locking_protocol();
1616
1617         status = register_quota_format(&ocfs2_quota_format);
1618         if (status < 0)
1619                 goto out3;
1620         status = register_filesystem(&ocfs2_fs_type);
1621         if (!status)
1622                 return 0;
1623
1624         unregister_quota_format(&ocfs2_quota_format);
1625 out3:
1626         debugfs_remove(ocfs2_debugfs_root);
1627         ocfs2_free_mem_caches();
1628 out2:
1629         exit_ocfs2_uptodate_cache();
1630 out1:
1631         mlog_errno(status);
1632         return status;
1633 }
1634
1635 static void __exit ocfs2_exit(void)
1636 {
1637         unregister_quota_format(&ocfs2_quota_format);
1638
1639         debugfs_remove(ocfs2_debugfs_root);
1640
1641         ocfs2_free_mem_caches();
1642
1643         unregister_filesystem(&ocfs2_fs_type);
1644
1645         exit_ocfs2_uptodate_cache();
1646 }
1647
1648 static void ocfs2_put_super(struct super_block *sb)
1649 {
1650         trace_ocfs2_put_super(sb);
1651
1652         ocfs2_sync_blockdev(sb);
1653         ocfs2_dismount_volume(sb, 0);
1654 }
1655
1656 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1657 {
1658         struct ocfs2_super *osb;
1659         u32 numbits, freebits;
1660         int status;
1661         struct ocfs2_dinode *bm_lock;
1662         struct buffer_head *bh = NULL;
1663         struct inode *inode = NULL;
1664
1665         trace_ocfs2_statfs(dentry->d_sb, buf);
1666
1667         osb = OCFS2_SB(dentry->d_sb);
1668
1669         inode = ocfs2_get_system_file_inode(osb,
1670                                             GLOBAL_BITMAP_SYSTEM_INODE,
1671                                             OCFS2_INVALID_SLOT);
1672         if (!inode) {
1673                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1674                 status = -EIO;
1675                 goto bail;
1676         }
1677
1678         status = ocfs2_inode_lock(inode, &bh, 0);
1679         if (status < 0) {
1680                 mlog_errno(status);
1681                 goto bail;
1682         }
1683
1684         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1685
1686         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1687         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1688
1689         buf->f_type = OCFS2_SUPER_MAGIC;
1690         buf->f_bsize = dentry->d_sb->s_blocksize;
1691         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1692         buf->f_blocks = ((sector_t) numbits) *
1693                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1694         buf->f_bfree = ((sector_t) freebits) *
1695                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1696         buf->f_bavail = buf->f_bfree;
1697         buf->f_files = numbits;
1698         buf->f_ffree = freebits;
1699         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1700                                 & 0xFFFFFFFFUL;
1701         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1702                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1703
1704         brelse(bh);
1705
1706         ocfs2_inode_unlock(inode, 0);
1707         status = 0;
1708 bail:
1709         iput(inode);
1710
1711         if (status)
1712                 mlog_errno(status);
1713
1714         return status;
1715 }
1716
1717 static void ocfs2_inode_init_once(void *data)
1718 {
1719         struct ocfs2_inode_info *oi = data;
1720
1721         oi->ip_flags = 0;
1722         oi->ip_open_count = 0;
1723         spin_lock_init(&oi->ip_lock);
1724         ocfs2_extent_map_init(&oi->vfs_inode);
1725         INIT_LIST_HEAD(&oi->ip_io_markers);
1726         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1727         oi->ip_dir_start_lookup = 0;
1728         init_rwsem(&oi->ip_alloc_sem);
1729         init_rwsem(&oi->ip_xattr_sem);
1730         mutex_init(&oi->ip_io_mutex);
1731
1732         oi->ip_blkno = 0ULL;
1733         oi->ip_clusters = 0;
1734
1735         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1736
1737         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1738         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1739         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1740
1741         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1742                                   &ocfs2_inode_caching_ops);
1743
1744         inode_init_once(&oi->vfs_inode);
1745 }
1746
1747 static int ocfs2_initialize_mem_caches(void)
1748 {
1749         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1750                                        sizeof(struct ocfs2_inode_info),
1751                                        0,
1752                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1753                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1754                                        ocfs2_inode_init_once);
1755         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1756                                         sizeof(struct ocfs2_dquot),
1757                                         0,
1758                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1759                                                 SLAB_MEM_SPREAD),
1760                                         NULL);
1761         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1762                                         sizeof(struct ocfs2_quota_chunk),
1763                                         0,
1764                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1765                                         NULL);
1766         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1767             !ocfs2_qf_chunk_cachep) {
1768                 kmem_cache_destroy(ocfs2_inode_cachep);
1769                 kmem_cache_destroy(ocfs2_dquot_cachep);
1770                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1771                 return -ENOMEM;
1772         }
1773
1774         return 0;
1775 }
1776
1777 static void ocfs2_free_mem_caches(void)
1778 {
1779         /*
1780          * Make sure all delayed rcu free inodes are flushed before we
1781          * destroy cache.
1782          */
1783         rcu_barrier();
1784         kmem_cache_destroy(ocfs2_inode_cachep);
1785         ocfs2_inode_cachep = NULL;
1786
1787         kmem_cache_destroy(ocfs2_dquot_cachep);
1788         ocfs2_dquot_cachep = NULL;
1789
1790         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1791         ocfs2_qf_chunk_cachep = NULL;
1792 }
1793
1794 static int ocfs2_get_sector(struct super_block *sb,
1795                             struct buffer_head **bh,
1796                             int block,
1797                             int sect_size)
1798 {
1799         if (!sb_set_blocksize(sb, sect_size)) {
1800                 mlog(ML_ERROR, "unable to set blocksize\n");
1801                 return -EIO;
1802         }
1803
1804         *bh = sb_getblk(sb, block);
1805         if (!*bh) {
1806                 mlog_errno(-ENOMEM);
1807                 return -ENOMEM;
1808         }
1809         lock_buffer(*bh);
1810         if (!buffer_dirty(*bh))
1811                 clear_buffer_uptodate(*bh);
1812         unlock_buffer(*bh);
1813         ll_rw_block(REQ_OP_READ, 0, 1, bh);
1814         wait_on_buffer(*bh);
1815         if (!buffer_uptodate(*bh)) {
1816                 mlog_errno(-EIO);
1817                 brelse(*bh);
1818                 *bh = NULL;
1819                 return -EIO;
1820         }
1821
1822         return 0;
1823 }
1824
1825 static int ocfs2_mount_volume(struct super_block *sb)
1826 {
1827         int status = 0;
1828         int unlock_super = 0;
1829         struct ocfs2_super *osb = OCFS2_SB(sb);
1830
1831         if (ocfs2_is_hard_readonly(osb))
1832                 goto leave;
1833
1834         mutex_init(&osb->obs_trim_fs_mutex);
1835
1836         status = ocfs2_dlm_init(osb);
1837         if (status < 0) {
1838                 mlog_errno(status);
1839                 if (status == -EBADR && ocfs2_userspace_stack(osb))
1840                         mlog(ML_ERROR, "couldn't mount because cluster name on"
1841                         " disk does not match the running cluster name.\n");
1842                 goto leave;
1843         }
1844
1845         status = ocfs2_super_lock(osb, 1);
1846         if (status < 0) {
1847                 mlog_errno(status);
1848                 goto leave;
1849         }
1850         unlock_super = 1;
1851
1852         /* This will load up the node map and add ourselves to it. */
1853         status = ocfs2_find_slot(osb);
1854         if (status < 0) {
1855                 mlog_errno(status);
1856                 goto leave;
1857         }
1858
1859         /* load all node-local system inodes */
1860         status = ocfs2_init_local_system_inodes(osb);
1861         if (status < 0) {
1862                 mlog_errno(status);
1863                 goto leave;
1864         }
1865
1866         status = ocfs2_check_volume(osb);
1867         if (status < 0) {
1868                 mlog_errno(status);
1869                 goto leave;
1870         }
1871
1872         status = ocfs2_truncate_log_init(osb);
1873         if (status < 0)
1874                 mlog_errno(status);
1875
1876 leave:
1877         if (unlock_super)
1878                 ocfs2_super_unlock(osb, 1);
1879
1880         return status;
1881 }
1882
1883 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1884 {
1885         int tmp, hangup_needed = 0;
1886         struct ocfs2_super *osb = NULL;
1887         char nodestr[12];
1888
1889         trace_ocfs2_dismount_volume(sb);
1890
1891         BUG_ON(!sb);
1892         osb = OCFS2_SB(sb);
1893         BUG_ON(!osb);
1894
1895         /* Remove file check sysfs related directores/files,
1896          * and wait for the pending file check operations */
1897         ocfs2_filecheck_remove_sysfs(osb);
1898
1899         kset_unregister(osb->osb_dev_kset);
1900
1901         debugfs_remove(osb->osb_ctxt);
1902
1903         /* Orphan scan should be stopped as early as possible */
1904         ocfs2_orphan_scan_stop(osb);
1905
1906         ocfs2_disable_quotas(osb);
1907
1908         /* All dquots should be freed by now */
1909         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1910         /* Wait for worker to be done with the work structure in osb */
1911         cancel_work_sync(&osb->dquot_drop_work);
1912
1913         ocfs2_shutdown_local_alloc(osb);
1914
1915         ocfs2_truncate_log_shutdown(osb);
1916
1917         /* This will disable recovery and flush any recovery work. */
1918         ocfs2_recovery_exit(osb);
1919
1920         ocfs2_journal_shutdown(osb);
1921
1922         ocfs2_sync_blockdev(sb);
1923
1924         ocfs2_purge_refcount_trees(osb);
1925
1926         /* No cluster connection means we've failed during mount, so skip
1927          * all the steps which depended on that to complete. */
1928         if (osb->cconn) {
1929                 tmp = ocfs2_super_lock(osb, 1);
1930                 if (tmp < 0) {
1931                         mlog_errno(tmp);
1932                         return;
1933                 }
1934         }
1935
1936         if (osb->slot_num != OCFS2_INVALID_SLOT)
1937                 ocfs2_put_slot(osb);
1938
1939         if (osb->cconn)
1940                 ocfs2_super_unlock(osb, 1);
1941
1942         ocfs2_release_system_inodes(osb);
1943
1944         /*
1945          * If we're dismounting due to mount error, mount.ocfs2 will clean
1946          * up heartbeat.  If we're a local mount, there is no heartbeat.
1947          * If we failed before we got a uuid_str yet, we can't stop
1948          * heartbeat.  Otherwise, do it.
1949          */
1950         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1951             !ocfs2_is_hard_readonly(osb))
1952                 hangup_needed = 1;
1953
1954         if (osb->cconn)
1955                 ocfs2_dlm_shutdown(osb, hangup_needed);
1956
1957         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1958         debugfs_remove(osb->osb_debug_root);
1959
1960         if (hangup_needed)
1961                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1962
1963         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1964
1965         if (ocfs2_mount_local(osb))
1966                 snprintf(nodestr, sizeof(nodestr), "local");
1967         else
1968                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1969
1970         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1971                osb->dev_str, nodestr);
1972
1973         ocfs2_delete_osb(osb);
1974         kfree(osb);
1975         sb->s_dev = 0;
1976         sb->s_fs_info = NULL;
1977 }
1978
1979 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1980                                 unsigned uuid_bytes)
1981 {
1982         int i, ret;
1983         char *ptr;
1984
1985         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1986
1987         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1988         if (osb->uuid_str == NULL)
1989                 return -ENOMEM;
1990
1991         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1992                 /* print with null */
1993                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1994                 if (ret != 2) /* drop super cleans up */
1995                         return -EINVAL;
1996                 /* then only advance past the last char */
1997                 ptr += 2;
1998         }
1999
2000         return 0;
2001 }
2002
2003 /* Make sure entire volume is addressable by our journal.  Requires
2004    osb_clusters_at_boot to be valid and for the journal to have been
2005    initialized by ocfs2_journal_init(). */
2006 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2007 {
2008         int status = 0;
2009         u64 max_block =
2010                 ocfs2_clusters_to_blocks(osb->sb,
2011                                          osb->osb_clusters_at_boot) - 1;
2012
2013         /* 32-bit block number is always OK. */
2014         if (max_block <= (u32)~0ULL)
2015                 goto out;
2016
2017         /* Volume is "huge", so see if our journal is new enough to
2018            support it. */
2019         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2020                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2021               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2022                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2023                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2024                      "Enable the 'block64' journal option with tunefs.ocfs2");
2025                 status = -EFBIG;
2026                 goto out;
2027         }
2028
2029  out:
2030         return status;
2031 }
2032
2033 static int ocfs2_initialize_super(struct super_block *sb,
2034                                   struct buffer_head *bh,
2035                                   int sector_size,
2036                                   struct ocfs2_blockcheck_stats *stats)
2037 {
2038         int status;
2039         int i, cbits, bbits;
2040         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2041         struct inode *inode = NULL;
2042         struct ocfs2_journal *journal;
2043         struct ocfs2_super *osb;
2044         u64 total_blocks;
2045
2046         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2047         if (!osb) {
2048                 status = -ENOMEM;
2049                 mlog_errno(status);
2050                 goto bail;
2051         }
2052
2053         sb->s_fs_info = osb;
2054         sb->s_op = &ocfs2_sops;
2055         sb->s_d_op = &ocfs2_dentry_ops;
2056         sb->s_export_op = &ocfs2_export_ops;
2057         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2058         sb->dq_op = &ocfs2_quota_operations;
2059         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2060         sb->s_xattr = ocfs2_xattr_handlers;
2061         sb->s_time_gran = 1;
2062         sb->s_flags |= SB_NOATIME;
2063         /* this is needed to support O_LARGEFILE */
2064         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2065         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2066         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2067         memcpy(&sb->s_uuid, di->id2.i_super.s_uuid,
2068                sizeof(di->id2.i_super.s_uuid));
2069
2070         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2071
2072         for (i = 0; i < 3; i++)
2073                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2074         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2075
2076         osb->sb = sb;
2077         osb->s_sectsize_bits = blksize_bits(sector_size);
2078         BUG_ON(!osb->s_sectsize_bits);
2079
2080         spin_lock_init(&osb->dc_task_lock);
2081         init_waitqueue_head(&osb->dc_event);
2082         osb->dc_work_sequence = 0;
2083         osb->dc_wake_sequence = 0;
2084         INIT_LIST_HEAD(&osb->blocked_lock_list);
2085         osb->blocked_lock_count = 0;
2086         spin_lock_init(&osb->osb_lock);
2087         spin_lock_init(&osb->osb_xattr_lock);
2088         ocfs2_init_steal_slots(osb);
2089
2090         mutex_init(&osb->system_file_mutex);
2091
2092         atomic_set(&osb->alloc_stats.moves, 0);
2093         atomic_set(&osb->alloc_stats.local_data, 0);
2094         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2095         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2096         atomic_set(&osb->alloc_stats.bg_extends, 0);
2097
2098         /* Copy the blockcheck stats from the superblock probe */
2099         osb->osb_ecc_stats = *stats;
2100
2101         ocfs2_init_node_maps(osb);
2102
2103         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2104                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2105
2106         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2107         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2108                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2109                      osb->max_slots);
2110                 status = -EINVAL;
2111                 goto bail;
2112         }
2113
2114         ocfs2_orphan_scan_init(osb);
2115
2116         status = ocfs2_recovery_init(osb);
2117         if (status) {
2118                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2119                 mlog_errno(status);
2120                 goto bail;
2121         }
2122
2123         init_waitqueue_head(&osb->checkpoint_event);
2124
2125         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2126
2127         osb->slot_num = OCFS2_INVALID_SLOT;
2128
2129         osb->s_xattr_inline_size = le16_to_cpu(
2130                                         di->id2.i_super.s_xattr_inline_size);
2131
2132         osb->local_alloc_state = OCFS2_LA_UNUSED;
2133         osb->local_alloc_bh = NULL;
2134         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2135
2136         init_waitqueue_head(&osb->osb_mount_event);
2137
2138         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2139         if (status) {
2140                 mlog_errno(status);
2141                 goto bail;
2142         }
2143
2144         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2145         if (!osb->vol_label) {
2146                 mlog(ML_ERROR, "unable to alloc vol label\n");
2147                 status = -ENOMEM;
2148                 goto bail;
2149         }
2150
2151         osb->slot_recovery_generations =
2152                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2153                         GFP_KERNEL);
2154         if (!osb->slot_recovery_generations) {
2155                 status = -ENOMEM;
2156                 mlog_errno(status);
2157                 goto bail;
2158         }
2159
2160         init_waitqueue_head(&osb->osb_wipe_event);
2161         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2162                                         sizeof(*osb->osb_orphan_wipes),
2163                                         GFP_KERNEL);
2164         if (!osb->osb_orphan_wipes) {
2165                 status = -ENOMEM;
2166                 mlog_errno(status);
2167                 goto bail;
2168         }
2169
2170         osb->osb_rf_lock_tree = RB_ROOT;
2171
2172         osb->s_feature_compat =
2173                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2174         osb->s_feature_ro_compat =
2175                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2176         osb->s_feature_incompat =
2177                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2178
2179         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2180                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2181                      "optional features (%x).\n", i);
2182                 status = -EINVAL;
2183                 goto bail;
2184         }
2185         if (!sb_rdonly(osb->sb) && (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2186                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2187                      "unsupported optional features (%x).\n", i);
2188                 status = -EINVAL;
2189                 goto bail;
2190         }
2191
2192         if (ocfs2_clusterinfo_valid(osb)) {
2193                 osb->osb_stackflags =
2194                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2195                 strlcpy(osb->osb_cluster_stack,
2196                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2197                        OCFS2_STACK_LABEL_LEN + 1);
2198                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2199                         mlog(ML_ERROR,
2200                              "couldn't mount because of an invalid "
2201                              "cluster stack label (%s) \n",
2202                              osb->osb_cluster_stack);
2203                         status = -EINVAL;
2204                         goto bail;
2205                 }
2206                 strlcpy(osb->osb_cluster_name,
2207                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2208                         OCFS2_CLUSTER_NAME_LEN + 1);
2209         } else {
2210                 /* The empty string is identical with classic tools that
2211                  * don't know about s_cluster_info. */
2212                 osb->osb_cluster_stack[0] = '\0';
2213         }
2214
2215         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2216
2217         /* FIXME
2218          * This should be done in ocfs2_journal_init(), but unknown
2219          * ordering issues will cause the filesystem to crash.
2220          * If anyone wants to figure out what part of the code
2221          * refers to osb->journal before ocfs2_journal_init() is run,
2222          * be my guest.
2223          */
2224         /* initialize our journal structure */
2225
2226         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2227         if (!journal) {
2228                 mlog(ML_ERROR, "unable to alloc journal\n");
2229                 status = -ENOMEM;
2230                 goto bail;
2231         }
2232         osb->journal = journal;
2233         journal->j_osb = osb;
2234
2235         atomic_set(&journal->j_num_trans, 0);
2236         init_rwsem(&journal->j_trans_barrier);
2237         init_waitqueue_head(&journal->j_checkpointed);
2238         spin_lock_init(&journal->j_lock);
2239         journal->j_trans_id = (unsigned long) 1;
2240         INIT_LIST_HEAD(&journal->j_la_cleanups);
2241         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2242         journal->j_state = OCFS2_JOURNAL_FREE;
2243
2244         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2245         init_llist_head(&osb->dquot_drop_list);
2246
2247         /* get some pseudo constants for clustersize bits */
2248         osb->s_clustersize_bits =
2249                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2250         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2251
2252         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2253             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2254                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2255                      osb->s_clustersize);
2256                 status = -EINVAL;
2257                 goto bail;
2258         }
2259
2260         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2261                                                 le32_to_cpu(di->i_clusters));
2262
2263         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2264                                            total_blocks);
2265         if (status) {
2266                 mlog(ML_ERROR, "Volume too large "
2267                      "to mount safely on this system");
2268                 status = -EFBIG;
2269                 goto bail;
2270         }
2271
2272         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2273                                  sizeof(di->id2.i_super.s_uuid))) {
2274                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2275                 status = -ENOMEM;
2276                 goto bail;
2277         }
2278
2279         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2280                 OCFS2_MAX_VOL_LABEL_LEN);
2281         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2282         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2283         osb->first_cluster_group_blkno =
2284                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2285         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2286         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2287         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2288                                      (unsigned long long)osb->root_blkno,
2289                                      (unsigned long long)osb->system_dir_blkno,
2290                                      osb->s_clustersize_bits);
2291
2292         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2293         if (!osb->osb_dlm_debug) {
2294                 status = -ENOMEM;
2295                 mlog_errno(status);
2296                 goto bail;
2297         }
2298
2299         atomic_set(&osb->vol_state, VOLUME_INIT);
2300
2301         /* load root, system_dir, and all global system inodes */
2302         status = ocfs2_init_global_system_inodes(osb);
2303         if (status < 0) {
2304                 mlog_errno(status);
2305                 goto bail;
2306         }
2307
2308         /*
2309          * global bitmap
2310          */
2311         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2312                                             OCFS2_INVALID_SLOT);
2313         if (!inode) {
2314                 status = -EINVAL;
2315                 mlog_errno(status);
2316                 goto bail;
2317         }
2318
2319         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2320         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2321         iput(inode);
2322
2323         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2324                                  osb->s_feature_incompat) * 8;
2325
2326         status = ocfs2_init_slot_info(osb);
2327         if (status < 0) {
2328                 mlog_errno(status);
2329                 goto bail;
2330         }
2331         cleancache_init_shared_fs(sb);
2332
2333         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2334         if (!osb->ocfs2_wq) {
2335                 status = -ENOMEM;
2336                 mlog_errno(status);
2337         }
2338
2339 bail:
2340         return status;
2341 }
2342
2343 /*
2344  * will return: -EAGAIN if it is ok to keep searching for superblocks
2345  *              -EINVAL if there is a bad superblock
2346  *              0 on success
2347  */
2348 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2349                                struct buffer_head *bh,
2350                                u32 blksz,
2351                                struct ocfs2_blockcheck_stats *stats)
2352 {
2353         int status = -EAGAIN;
2354
2355         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2356                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2357                 /* We have to do a raw check of the feature here */
2358                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2359                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2360                         status = ocfs2_block_check_validate(bh->b_data,
2361                                                             bh->b_size,
2362                                                             &di->i_check,
2363                                                             stats);
2364                         if (status)
2365                                 goto out;
2366                 }
2367                 status = -EINVAL;
2368                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2369                         mlog(ML_ERROR, "found superblock with incorrect block "
2370                              "size: found %u, should be %u\n",
2371                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2372                                blksz);
2373                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2374                            OCFS2_MAJOR_REV_LEVEL ||
2375                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2376                            OCFS2_MINOR_REV_LEVEL) {
2377                         mlog(ML_ERROR, "found superblock with bad version: "
2378                              "found %u.%u, should be %u.%u\n",
2379                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2380                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2381                              OCFS2_MAJOR_REV_LEVEL,
2382                              OCFS2_MINOR_REV_LEVEL);
2383                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2384                         mlog(ML_ERROR, "bad block number on superblock: "
2385                              "found %llu, should be %llu\n",
2386                              (unsigned long long)le64_to_cpu(di->i_blkno),
2387                              (unsigned long long)bh->b_blocknr);
2388                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2389                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2390                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2391                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2392                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2393                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2394                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2395                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2396                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2397                         mlog(ML_ERROR,
2398                              "Superblock slots found greater than file system "
2399                              "maximum: found %u, max %u\n",
2400                              le16_to_cpu(di->id2.i_super.s_max_slots),
2401                              OCFS2_MAX_SLOTS);
2402                 } else {
2403                         /* found it! */
2404                         status = 0;
2405                 }
2406         }
2407
2408 out:
2409         if (status && status != -EAGAIN)
2410                 mlog_errno(status);
2411         return status;
2412 }
2413
2414 static int ocfs2_check_volume(struct ocfs2_super *osb)
2415 {
2416         int status;
2417         int dirty;
2418         int local;
2419         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2420                                                   * recover
2421                                                   * ourselves. */
2422
2423         /* Init our journal object. */
2424         status = ocfs2_journal_init(osb->journal, &dirty);
2425         if (status < 0) {
2426                 mlog(ML_ERROR, "Could not initialize journal!\n");
2427                 goto finally;
2428         }
2429
2430         /* Now that journal has been initialized, check to make sure
2431            entire volume is addressable. */
2432         status = ocfs2_journal_addressable(osb);
2433         if (status)
2434                 goto finally;
2435
2436         /* If the journal was unmounted cleanly then we don't want to
2437          * recover anything. Otherwise, journal_load will do that
2438          * dirty work for us :) */
2439         if (!dirty) {
2440                 status = ocfs2_journal_wipe(osb->journal, 0);
2441                 if (status < 0) {
2442                         mlog_errno(status);
2443                         goto finally;
2444                 }
2445         } else {
2446                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2447                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2448         }
2449
2450         local = ocfs2_mount_local(osb);
2451
2452         /* will play back anything left in the journal. */
2453         status = ocfs2_journal_load(osb->journal, local, dirty);
2454         if (status < 0) {
2455                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2456                 goto finally;
2457         }
2458
2459         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2460                 jbd2_journal_set_features(osb->journal->j_journal,
2461                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2462                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2463         else
2464                 jbd2_journal_clear_features(osb->journal->j_journal,
2465                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2466                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2467
2468         if (dirty) {
2469                 /* recover my local alloc if we didn't unmount cleanly. */
2470                 status = ocfs2_begin_local_alloc_recovery(osb,
2471                                                           osb->slot_num,
2472                                                           &local_alloc);
2473                 if (status < 0) {
2474                         mlog_errno(status);
2475                         goto finally;
2476                 }
2477                 /* we complete the recovery process after we've marked
2478                  * ourselves as mounted. */
2479         }
2480
2481         status = ocfs2_load_local_alloc(osb);
2482         if (status < 0) {
2483                 mlog_errno(status);
2484                 goto finally;
2485         }
2486
2487         if (dirty) {
2488                 /* Recovery will be completed after we've mounted the
2489                  * rest of the volume. */
2490                 osb->local_alloc_copy = local_alloc;
2491                 local_alloc = NULL;
2492         }
2493
2494         /* go through each journal, trylock it and if you get the
2495          * lock, and it's marked as dirty, set the bit in the recover
2496          * map and launch a recovery thread for it. */
2497         status = ocfs2_mark_dead_nodes(osb);
2498         if (status < 0) {
2499                 mlog_errno(status);
2500                 goto finally;
2501         }
2502
2503         status = ocfs2_compute_replay_slots(osb);
2504         if (status < 0)
2505                 mlog_errno(status);
2506
2507 finally:
2508         kfree(local_alloc);
2509
2510         if (status)
2511                 mlog_errno(status);
2512         return status;
2513 }
2514
2515 /*
2516  * The routine gets called from dismount or close whenever a dismount on
2517  * volume is requested and the osb open count becomes 1.
2518  * It will remove the osb from the global list and also free up all the
2519  * initialized resources and fileobject.
2520  */
2521 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2522 {
2523         /* This function assumes that the caller has the main osb resource */
2524
2525         /* ocfs2_initializer_super have already created this workqueue */
2526         if (osb->ocfs2_wq)
2527                 destroy_workqueue(osb->ocfs2_wq);
2528
2529         ocfs2_free_slot_info(osb);
2530
2531         kfree(osb->osb_orphan_wipes);
2532         kfree(osb->slot_recovery_generations);
2533         /* FIXME
2534          * This belongs in journal shutdown, but because we have to
2535          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2536          * we free it here.
2537          */
2538         kfree(osb->journal);
2539         kfree(osb->local_alloc_copy);
2540         kfree(osb->uuid_str);
2541         kfree(osb->vol_label);
2542         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2543         memset(osb, 0, sizeof(struct ocfs2_super));
2544 }
2545
2546 /* Depending on the mount option passed, perform one of the following:
2547  * Put OCFS2 into a readonly state (default)
2548  * Return EIO so that only the process errs
2549  * Fix the error as if fsck.ocfs2 -y
2550  * panic
2551  */
2552 static int ocfs2_handle_error(struct super_block *sb)
2553 {
2554         struct ocfs2_super *osb = OCFS2_SB(sb);
2555         int rv = 0;
2556
2557         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2558         pr_crit("On-disk corruption discovered. "
2559                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2560
2561         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2562                 panic("OCFS2: (device %s): panic forced after error\n",
2563                       sb->s_id);
2564         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2565                 pr_crit("OCFS2: Returning error to the calling process.\n");
2566                 rv = -EIO;
2567         } else { /* default option */
2568                 rv = -EROFS;
2569                 if (sb_rdonly(sb) && (ocfs2_is_soft_readonly(osb) || ocfs2_is_hard_readonly(osb)))
2570                         return rv;
2571
2572                 pr_crit("OCFS2: File system is now read-only.\n");
2573                 sb->s_flags |= SB_RDONLY;
2574                 ocfs2_set_ro_flag(osb, 0);
2575         }
2576
2577         return rv;
2578 }
2579
2580 int __ocfs2_error(struct super_block *sb, const char *function,
2581                   const char *fmt, ...)
2582 {
2583         struct va_format vaf;
2584         va_list args;
2585
2586         va_start(args, fmt);
2587         vaf.fmt = fmt;
2588         vaf.va = &args;
2589
2590         /* Not using mlog here because we want to show the actual
2591          * function the error came from. */
2592         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2593                sb->s_id, function, &vaf);
2594
2595         va_end(args);
2596
2597         return ocfs2_handle_error(sb);
2598 }
2599
2600 /* Handle critical errors. This is intentionally more drastic than
2601  * ocfs2_handle_error, so we only use for things like journal errors,
2602  * etc. */
2603 void __ocfs2_abort(struct super_block *sb, const char *function,
2604                    const char *fmt, ...)
2605 {
2606         struct va_format vaf;
2607         va_list args;
2608
2609         va_start(args, fmt);
2610
2611         vaf.fmt = fmt;
2612         vaf.va = &args;
2613
2614         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2615                sb->s_id, function, &vaf);
2616
2617         va_end(args);
2618
2619         /* We don't have the cluster support yet to go straight to
2620          * hard readonly in here. Until then, we want to keep
2621          * ocfs2_abort() so that we can at least mark critical
2622          * errors.
2623          *
2624          * TODO: This should abort the journal and alert other nodes
2625          * that our slot needs recovery. */
2626
2627         /* Force a panic(). This stinks, but it's better than letting
2628          * things continue without having a proper hard readonly
2629          * here. */
2630         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2631                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2632         ocfs2_handle_error(sb);
2633 }
2634
2635 /*
2636  * Void signal blockers, because in-kernel sigprocmask() only fails
2637  * when SIG_* is wrong.
2638  */
2639 void ocfs2_block_signals(sigset_t *oldset)
2640 {
2641         int rc;
2642         sigset_t blocked;
2643
2644         sigfillset(&blocked);
2645         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2646         BUG_ON(rc);
2647 }
2648
2649 void ocfs2_unblock_signals(sigset_t *oldset)
2650 {
2651         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2652         BUG_ON(rc);
2653 }
2654
2655 module_init(ocfs2_init);
2656 module_exit(ocfs2_exit);