fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root = NULL;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static const struct super_operations ocfs2_sops = {
 147         .statfs         = ocfs2_statfs,
 148         .alloc_inode    = ocfs2_alloc_inode,
 149         .destroy_inode  = ocfs2_destroy_inode,
 150         .drop_inode     = ocfs2_drop_inode,
 151         .evict_inode    = ocfs2_evict_inode,
 152         .sync_fs        = ocfs2_sync_fs,
 153         .put_super      = ocfs2_put_super,
 154         .remount_fs     = ocfs2_remount,
 155         .show_options   = ocfs2_show_options,
 156         .quota_read     = ocfs2_quota_read,
 157         .quota_write    = ocfs2_quota_write,
 158 };
 159
 160 enum {
 161         Opt_barrier,
 162         Opt_err_panic,
 163         Opt_err_ro,
 164         Opt_intr,
 165         Opt_nointr,
 166         Opt_hb_none,
 167         Opt_hb_local,
 168         Opt_hb_global,
 169         Opt_data_ordered,
 170         Opt_data_writeback,
 171         Opt_atime_quantum,
 172         Opt_slot,
 173         Opt_commit,
 174         Opt_localalloc,
 175         Opt_localflocks,
 176         Opt_stack,
 177         Opt_user_xattr,
 178         Opt_nouser_xattr,
 179         Opt_inode64,
 180         Opt_acl,
 181         Opt_noacl,
 182         Opt_usrquota,
 183         Opt_grpquota,
 184         Opt_coherency_buffered,
 185         Opt_coherency_full,
 186         Opt_resv_level,
 187         Opt_dir_resv_level,
 188         Opt_err,
 189 };
 190
 191 static const match_table_t tokens = {
 192         {Opt_barrier, "barrier=%u"},
 193         {Opt_err_panic, "errors=panic"},
 194         {Opt_err_ro, "errors=remount-ro"},
 195         {Opt_intr, "intr"},
 196         {Opt_nointr, "nointr"},
 197         {Opt_hb_none, OCFS2_HB_NONE},
 198         {Opt_hb_local, OCFS2_HB_LOCAL},
 199         {Opt_hb_global, OCFS2_HB_GLOBAL},
 200         {Opt_data_ordered, "data=ordered"},
 201         {Opt_data_writeback, "data=writeback"},
 202         {Opt_atime_quantum, "atime_quantum=%u"},
 203         {Opt_slot, "preferred_slot=%u"},
 204         {Opt_commit, "commit=%u"},
 205         {Opt_localalloc, "localalloc=%d"},
 206         {Opt_localflocks, "localflocks"},
 207         {Opt_stack, "cluster_stack=%s"},
 208         {Opt_user_xattr, "user_xattr"},
 209         {Opt_nouser_xattr, "nouser_xattr"},
 210         {Opt_inode64, "inode64"},
 211         {Opt_acl, "acl"},
 212         {Opt_noacl, "noacl"},
 213         {Opt_usrquota, "usrquota"},
 214         {Opt_grpquota, "grpquota"},
 215         {Opt_coherency_buffered, "coherency=buffered"},
 216         {Opt_coherency_full, "coherency=full"},
 217         {Opt_resv_level, "resv_level=%u"},
 218         {Opt_dir_resv_level, "dir_resv_level=%u"},
 219         {Opt_err, NULL}
 220 };
 221
 222 #ifdef CONFIG_DEBUG_FS
 223 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 224 {
 225         struct ocfs2_cluster_connection *cconn = osb->cconn;
 226         struct ocfs2_recovery_map *rm = osb->recovery_map;
 227         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 228         int i, out = 0;
 229
 230         out += snprintf(buf + out, len - out,
 231                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 232                         "Device", osb->dev_str, osb->uuid_str,
 233                         osb->fs_generation, osb->vol_label);
 234
 235         out += snprintf(buf + out, len - out,
 236                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 237                         atomic_read(&osb->vol_state), osb->osb_flags);
 238
 239         out += snprintf(buf + out, len - out,
 240                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 241                         osb->sb->s_blocksize, osb->s_clustersize);
 242
 243         out += snprintf(buf + out, len - out,
 244                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 245                         "ROcompat: 0x%X\n",
 246                         "Features", osb->s_feature_compat,
 247                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 251                         osb->s_mount_opt, osb->s_atime_quantum);
 252
 253         if (cconn) {
 254                 out += snprintf(buf + out, len - out,
 255                                 "%10s => Stack: %s  Name: %*s  "
 256                                 "Version: %d.%d\n", "Cluster",
 257                                 (*osb->osb_cluster_stack == '\0' ?
 258                                  "o2cb" : osb->osb_cluster_stack),
 259                                 cconn->cc_namelen, cconn->cc_name,
 260                                 cconn->cc_version.pv_major,
 261                                 cconn->cc_version.pv_minor);
 262         }
 263
 264         spin_lock(&osb->dc_task_lock);
 265         out += snprintf(buf + out, len - out,
 266                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 267                         "WorkSeq: %lu\n", "DownCnvt",
 268                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 269                         osb->blocked_lock_count, osb->dc_wake_sequence,
 270                         osb->dc_work_sequence);
 271         spin_unlock(&osb->dc_task_lock);
 272
 273         spin_lock(&osb->osb_lock);
 274         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 275                         "Recovery",
 276                         (osb->recovery_thread_task ?
 277                          task_pid_nr(osb->recovery_thread_task) : -1));
 278         if (rm->rm_used == 0)
 279                 out += snprintf(buf + out, len - out, " None\n");
 280         else {
 281                 for (i = 0; i < rm->rm_used; i++)
 282                         out += snprintf(buf + out, len - out, " %d",
 283                                         rm->rm_entries[i]);
 284                 out += snprintf(buf + out, len - out, "\n");
 285         }
 286         spin_unlock(&osb->osb_lock);
 287
 288         out += snprintf(buf + out, len - out,
 289                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 290                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 291                         osb->osb_commit_interval);
 292
 293         out += snprintf(buf + out, len - out,
 294                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 295                         "Journal", osb->journal->j_state,
 296                         osb->journal->j_trans_id,
 297                         atomic_read(&osb->journal->j_num_trans));
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 301                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 302                         "Stats",
 303                         atomic_read(&osb->alloc_stats.bitmap_data),
 304                         atomic_read(&osb->alloc_stats.local_data),
 305                         atomic_read(&osb->alloc_stats.bg_allocs),
 306                         atomic_read(&osb->alloc_stats.moves),
 307                         atomic_read(&osb->alloc_stats.bg_extends));
 308
 309         out += snprintf(buf + out, len - out,
 310                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 311                         "Default: %u bits\n",
 312                         "LocalAlloc", osb->local_alloc_state,
 313                         (unsigned long long)osb->la_last_gd,
 314                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 315
 316         spin_lock(&osb->osb_lock);
 317         out += snprintf(buf + out, len - out,
 318                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 319                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 320                         osb->s_inode_steal_slot,
 321                         atomic_read(&osb->s_num_inodes_stolen),
 322                         osb->s_meta_steal_slot,
 323                         atomic_read(&osb->s_num_meta_stolen));
 324         spin_unlock(&osb->osb_lock);
 325
 326         out += snprintf(buf + out, len - out, "OrphanScan => ");
 327         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 328                         os->os_count, os->os_seqno);
 329         out += snprintf(buf + out, len - out, " Last Scan: ");
 330         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 331                 out += snprintf(buf + out, len - out, "Disabled\n");
 332         else
 333                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 334                                 (get_seconds() - os->os_scantime.tv_sec));
 335
 336         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 337                         "Slots", "Num", "RecoGen");
 338         for (i = 0; i < osb->max_slots; ++i) {
 339                 out += snprintf(buf + out, len - out,
 340                                 "%10s  %c %3d  %10d\n",
 341                                 " ",
 342                                 (i == osb->slot_num ? '*' : ' '),
 343                                 i, osb->slot_recovery_generations[i]);
 344         }
 345
 346         return out;
 347 }
 348
 349 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 350 {
 351         struct ocfs2_super *osb = inode->i_private;
 352         char *buf = NULL;
 353
 354         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 355         if (!buf)
 356                 goto bail;
 357
 358         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 359
 360         file->private_data = buf;
 361
 362         return 0;
 363 bail:
 364         return -ENOMEM;
 365 }
 366
 367 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 368 {
 369         kfree(file->private_data);
 370         return 0;
 371 }
 372
 373 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 374                                 size_t nbytes, loff_t *ppos)
 375 {
 376         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 377                                        i_size_read(file->f_mapping->host));
 378 }
 379 #else
 380 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 381 {
 382         return 0;
 383 }
 384 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 385 {
 386         return 0;
 387 }
 388 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 389                                 size_t nbytes, loff_t *ppos)
 390 {
 391         return 0;
 392 }
 393 #endif  /* CONFIG_DEBUG_FS */
 394
 395 static const struct file_operations ocfs2_osb_debug_fops = {
 396         .open =         ocfs2_osb_debug_open,
 397         .release =      ocfs2_debug_release,
 398         .read =         ocfs2_debug_read,
 399         .llseek =       generic_file_llseek,
 400 };
 401
 402 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 403 {
 404         int status;
 405         tid_t target;
 406         struct ocfs2_super *osb = OCFS2_SB(sb);
 407
 408         if (ocfs2_is_hard_readonly(osb))
 409                 return -EROFS;
 410
 411         if (wait) {
 412                 status = ocfs2_flush_truncate_log(osb);
 413                 if (status < 0)
 414                         mlog_errno(status);
 415         } else {
 416                 ocfs2_schedule_truncate_log_flush(osb, 0);
 417         }
 418
 419         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 420                                       &target)) {
 421                 if (wait)
 422                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 423                                              target);
 424         }
 425         return 0;
 426 }
 427
 428 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 429 {
 430         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 431             && (ino == USER_QUOTA_SYSTEM_INODE
 432                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 433                 return 0;
 434         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 435             && (ino == GROUP_QUOTA_SYSTEM_INODE
 436                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 437                 return 0;
 438         return 1;
 439 }
 440
 441 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 442 {
 443         struct inode *new = NULL;
 444         int status = 0;
 445         int i;
 446
 447         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 448         if (IS_ERR(new)) {
 449                 status = PTR_ERR(new);
 450                 mlog_errno(status);
 451                 goto bail;
 452         }
 453         osb->root_inode = new;
 454
 455         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 456         if (IS_ERR(new)) {
 457                 status = PTR_ERR(new);
 458                 mlog_errno(status);
 459                 goto bail;
 460         }
 461         osb->sys_root_inode = new;
 462
 463         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 464              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 465                 if (!ocfs2_need_system_inode(osb, i))
 466                         continue;
 467                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 468                 if (!new) {
 469                         ocfs2_release_system_inodes(osb);
 470                         status = -EINVAL;
 471                         mlog_errno(status);
 472                         /* FIXME: Should ERROR_RO_FS */
 473                         mlog(ML_ERROR, "Unable to load system inode %d, "
 474                              "possibly corrupt fs?", i);
 475                         goto bail;
 476                 }
 477                 // the array now has one ref, so drop this one
 478                 iput(new);
 479         }
 480
 481 bail:
 482         if (status)
 483                 mlog_errno(status);
 484         return status;
 485 }
 486
 487 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 488 {
 489         struct inode *new = NULL;
 490         int status = 0;
 491         int i;
 492
 493         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 494              i < NUM_SYSTEM_INODES;
 495              i++) {
 496                 if (!ocfs2_need_system_inode(osb, i))
 497                         continue;
 498                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 499                 if (!new) {
 500                         ocfs2_release_system_inodes(osb);
 501                         status = -EINVAL;
 502                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 503                              status, i, osb->slot_num);
 504                         goto bail;
 505                 }
 506                 /* the array now has one ref, so drop this one */
 507                 iput(new);
 508         }
 509
 510 bail:
 511         if (status)
 512                 mlog_errno(status);
 513         return status;
 514 }
 515
 516 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 517 {
 518         int i;
 519         struct inode *inode;
 520
 521         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 522                 inode = osb->global_system_inodes[i];
 523                 if (inode) {
 524                         iput(inode);
 525                         osb->global_system_inodes[i] = NULL;
 526                 }
 527         }
 528
 529         inode = osb->sys_root_inode;
 530         if (inode) {
 531                 iput(inode);
 532                 osb->sys_root_inode = NULL;
 533         }
 534
 535         inode = osb->root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->root_inode = NULL;
 539         }
 540
 541         if (!osb->local_system_inodes)
 542                 return;
 543
 544         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 545                 if (osb->local_system_inodes[i]) {
 546                         iput(osb->local_system_inodes[i]);
 547                         osb->local_system_inodes[i] = NULL;
 548                 }
 549         }
 550
 551         kfree(osb->local_system_inodes);
 552         osb->local_system_inodes = NULL;
 553 }
 554
 555 /* We're allocating fs objects, use GFP_NOFS */
 556 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 557 {
 558         struct ocfs2_inode_info *oi;
 559
 560         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 561         if (!oi)
 562                 return NULL;
 563
 564         oi->i_sync_tid = 0;
 565         oi->i_datasync_tid = 0;
 566
 567         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 568         return &oi->vfs_inode;
 569 }
 570
 571 static void ocfs2_i_callback(struct rcu_head *head)
 572 {
 573         struct inode *inode = container_of(head, struct inode, i_rcu);
 574         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 575 }
 576
 577 static void ocfs2_destroy_inode(struct inode *inode)
 578 {
 579         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 580 }
 581
 582 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 583                                                 unsigned int cbits)
 584 {
 585         unsigned int bytes = 1 << cbits;
 586         unsigned int trim = bytes;
 587         unsigned int bitshift = 32;
 588
 589         /*
 590          * i_size and all block offsets in ocfs2 are always 64 bits
 591          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 592          * 64 bit platforms, cluster size will be the limiting factor.
 593          */
 594
 595 #if BITS_PER_LONG == 32
 596 # if defined(CONFIG_LBDAF)
 597         BUILD_BUG_ON(sizeof(sector_t) != 8);
 598         /*
 599          * We might be limited by page cache size.
 600          */
 601         if (bytes > PAGE_CACHE_SIZE) {
 602                 bytes = PAGE_CACHE_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 # else
 611         /*
 612          * We are limited by the size of sector_t. Use block size, as
 613          * that's what we expose to the VFS.
 614          */
 615         bytes = 1 << bbits;
 616         trim = 1;
 617         bitshift = 31;
 618 # endif
 619 #endif
 620
 621         /*
 622          * Trim by a whole cluster when we can actually approach the
 623          * on-disk limits. Otherwise we can overflow i_clusters when
 624          * an extent start is at the max offset.
 625          */
 626         return (((unsigned long long)bytes) << bitshift) - trim;
 627 }
 628
 629 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 630 {
 631         int incompat_features;
 632         int ret = 0;
 633         struct mount_options parsed_options;
 634         struct ocfs2_super *osb = OCFS2_SB(sb);
 635         u32 tmp;
 636
 637         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 638             !ocfs2_check_set_options(sb, &parsed_options)) {
 639                 ret = -EINVAL;
 640                 goto out;
 641         }
 642
 643         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 644                 OCFS2_MOUNT_HB_NONE;
 645         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 646                 ret = -EINVAL;
 647                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 648                 goto out;
 649         }
 650
 651         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 652             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 653                 ret = -EINVAL;
 654                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 655                 goto out;
 656         }
 657
 658         /* Probably don't want this on remount; it might
 659          * mess with other nodes */
 660         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 661             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 662                 ret = -EINVAL;
 663                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 664                 goto out;
 665         }
 666
 667         /* We're going to/from readonly mode. */
 668         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 669                 /* Disable quota accounting before remounting RO */
 670                 if (*flags & MS_RDONLY) {
 671                         ret = ocfs2_susp_quotas(osb, 0);
 672                         if (ret < 0)
 673                                 goto out;
 674                 }
 675                 /* Lock here so the check of HARD_RO and the potential
 676                  * setting of SOFT_RO is atomic. */
 677                 spin_lock(&osb->osb_lock);
 678                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 679                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 680                         ret = -EROFS;
 681                         goto unlock_osb;
 682                 }
 683
 684                 if (*flags & MS_RDONLY) {
 685                         sb->s_flags |= MS_RDONLY;
 686                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 687                 } else {
 688                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 689                                 mlog(ML_ERROR, "Cannot remount RDWR "
 690                                      "filesystem due to previous errors.\n");
 691                                 ret = -EROFS;
 692                                 goto unlock_osb;
 693                         }
 694                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 695                         if (incompat_features) {
 696                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 697                                      "of unsupported optional features "
 698                                      "(%x).\n", incompat_features);
 699                                 ret = -EINVAL;
 700                                 goto unlock_osb;
 701                         }
 702                         sb->s_flags &= ~MS_RDONLY;
 703                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 704                 }
 705                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 706 unlock_osb:
 707                 spin_unlock(&osb->osb_lock);
 708                 /* Enable quota accounting after remounting RW */
 709                 if (!ret && !(*flags & MS_RDONLY)) {
 710                         if (sb_any_quota_suspended(sb))
 711                                 ret = ocfs2_susp_quotas(osb, 1);
 712                         else
 713                                 ret = ocfs2_enable_quotas(osb);
 714                         if (ret < 0) {
 715                                 /* Return back changes... */
 716                                 spin_lock(&osb->osb_lock);
 717                                 sb->s_flags |= MS_RDONLY;
 718                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 719                                 spin_unlock(&osb->osb_lock);
 720                                 goto out;
 721                         }
 722                 }
 723         }
 724
 725         if (!ret) {
 726                 /* Only save off the new mount options in case of a successful
 727                  * remount. */
 728                 osb->s_mount_opt = parsed_options.mount_opt;
 729                 osb->s_atime_quantum = parsed_options.atime_quantum;
 730                 osb->preferred_slot = parsed_options.slot;
 731                 if (parsed_options.commit_interval)
 732                         osb->osb_commit_interval = parsed_options.commit_interval;
 733
 734                 if (!ocfs2_is_hard_readonly(osb))
 735                         ocfs2_set_journal_params(osb);
 736
 737                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 738                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 739                                                         MS_POSIXACL : 0);
 740         }
 741 out:
 742         return ret;
 743 }
 744
 745 static int ocfs2_sb_probe(struct super_block *sb,
 746                           struct buffer_head **bh,
 747                           int *sector_size,
 748                           struct ocfs2_blockcheck_stats *stats)
 749 {
 750         int status, tmpstat;
 751         struct ocfs1_vol_disk_hdr *hdr;
 752         struct ocfs2_dinode *di;
 753         int blksize;
 754
 755         *bh = NULL;
 756
 757         /* may be > 512 */
 758         *sector_size = bdev_logical_block_size(sb->s_bdev);
 759         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 760                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 761                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 762                 status = -EINVAL;
 763                 goto bail;
 764         }
 765
 766         /* Can this really happen? */
 767         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 768                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 769
 770         /* check block zero for old format */
 771         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 772         if (status < 0) {
 773                 mlog_errno(status);
 774                 goto bail;
 775         }
 776         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 777         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 778                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 779                      hdr->major_version, hdr->minor_version);
 780                 status = -EINVAL;
 781         }
 782         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 783                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 784                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 785                      hdr->signature);
 786                 status = -EINVAL;
 787         }
 788         brelse(*bh);
 789         *bh = NULL;
 790         if (status < 0) {
 791                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 792                      "upgraded before mounting with ocfs v2\n");
 793                 goto bail;
 794         }
 795
 796         /*
 797          * Now check at magic offset for 512, 1024, 2048, 4096
 798          * blocksizes.  4096 is the maximum blocksize because it is
 799          * the minimum clustersize.
 800          */
 801         status = -EINVAL;
 802         for (blksize = *sector_size;
 803              blksize <= OCFS2_MAX_BLOCKSIZE;
 804              blksize <<= 1) {
 805                 tmpstat = ocfs2_get_sector(sb, bh,
 806                                            OCFS2_SUPER_BLOCK_BLKNO,
 807                                            blksize);
 808                 if (tmpstat < 0) {
 809                         status = tmpstat;
 810                         mlog_errno(status);
 811                         break;
 812                 }
 813                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 814                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 815                 spin_lock_init(&stats->b_lock);
 816                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 817                 if (tmpstat < 0) {
 818                         brelse(*bh);
 819                         *bh = NULL;
 820                 }
 821                 if (tmpstat != -EAGAIN) {
 822                         status = tmpstat;
 823                         break;
 824                 }
 825         }
 826
 827 bail:
 828         return status;
 829 }
 830
 831 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 832 {
 833         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 834
 835         if (osb->s_mount_opt & hb_enabled) {
 836                 if (ocfs2_mount_local(osb)) {
 837                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 838                              "mounted device.\n");
 839                         return -EINVAL;
 840                 }
 841                 if (ocfs2_userspace_stack(osb)) {
 842                         mlog(ML_ERROR, "Userspace stack expected, but "
 843                              "o2cb heartbeat arguments passed to mount\n");
 844                         return -EINVAL;
 845                 }
 846                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 847                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 848                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 849                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 850                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 851                         return -EINVAL;
 852                 }
 853         }
 854
 855         if (!(osb->s_mount_opt & hb_enabled)) {
 856                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 857                     !ocfs2_userspace_stack(osb)) {
 858                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 859                              "a read-write clustered device.\n");
 860                         return -EINVAL;
 861                 }
 862         }
 863
 864         return 0;
 865 }
 866
 867 /*
 868  * If we're using a userspace stack, mount should have passed
 869  * a name that matches the disk.  If not, mount should not
 870  * have passed a stack.
 871  */
 872 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 873                                         struct mount_options *mopt)
 874 {
 875         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 876                 mlog(ML_ERROR,
 877                      "cluster stack passed to mount, but this filesystem "
 878                      "does not support it\n");
 879                 return -EINVAL;
 880         }
 881
 882         if (ocfs2_userspace_stack(osb) &&
 883             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 884                     OCFS2_STACK_LABEL_LEN)) {
 885                 mlog(ML_ERROR,
 886                      "cluster stack passed to mount (\"%s\") does not "
 887                      "match the filesystem (\"%s\")\n",
 888                      mopt->cluster_stack,
 889                      osb->osb_cluster_stack);
 890                 return -EINVAL;
 891         }
 892
 893         return 0;
 894 }
 895
 896 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 897 {
 898         int type;
 899         struct super_block *sb = osb->sb;
 900         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 901                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 902         int status = 0;
 903
 904         for (type = 0; type < MAXQUOTAS; type++) {
 905                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 906                         continue;
 907                 if (unsuspend)
 908                         status = dquot_resume(sb, type);
 909                 else {
 910                         struct ocfs2_mem_dqinfo *oinfo;
 911
 912                         /* Cancel periodic syncing before suspending */
 913                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 914                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 915                         status = dquot_suspend(sb, type);
 916                 }
 917                 if (status < 0)
 918                         break;
 919         }
 920         if (status < 0)
 921                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 922                      "remount (error = %d).\n", status);
 923         return status;
 924 }
 925
 926 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 927 {
 928         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 929         struct super_block *sb = osb->sb;
 930         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 931                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 932         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 933                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 934         int status;
 935         int type;
 936
 937         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 938         for (type = 0; type < MAXQUOTAS; type++) {
 939                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 940                         continue;
 941                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 942                                                         osb->slot_num);
 943                 if (!inode[type]) {
 944                         status = -ENOENT;
 945                         goto out_quota_off;
 946                 }
 947                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 948                                       DQUOT_USAGE_ENABLED);
 949                 if (status < 0)
 950                         goto out_quota_off;
 951         }
 952
 953         for (type = 0; type < MAXQUOTAS; type++)
 954                 iput(inode[type]);
 955         return 0;
 956 out_quota_off:
 957         ocfs2_disable_quotas(osb);
 958         for (type = 0; type < MAXQUOTAS; type++)
 959                 iput(inode[type]);
 960         mlog_errno(status);
 961         return status;
 962 }
 963
 964 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 965 {
 966         int type;
 967         struct inode *inode;
 968         struct super_block *sb = osb->sb;
 969         struct ocfs2_mem_dqinfo *oinfo;
 970
 971         /* We mostly ignore errors in this function because there's not much
 972          * we can do when we see them */
 973         for (type = 0; type < MAXQUOTAS; type++) {
 974                 if (!sb_has_quota_loaded(sb, type))
 975                         continue;
 976                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 977                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 978                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 979                 inode = igrab(sb->s_dquot.files[type]);
 980                 /* Turn off quotas. This will remove all dquot structures from
 981                  * memory and so they will be automatically synced to global
 982                  * quota files */
 983                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 984                                         DQUOT_LIMITS_ENABLED);
 985                 if (!inode)
 986                         continue;
 987                 iput(inode);
 988         }
 989 }
 990
 991 /* Handle quota on quotactl */
 992 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 993 {
 994         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 995                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 996
 997         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 998                 return -EINVAL;
 999
1000         return dquot_enable(sb_dqopt(sb)->files[type], type,
1001                             format_id, DQUOT_LIMITS_ENABLED);
1002 }
1003
1004 /* Handle quota off quotactl */
1005 static int ocfs2_quota_off(struct super_block *sb, int type)
1006 {
1007         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1008 }
1009
1010 static const struct quotactl_ops ocfs2_quotactl_ops = {
1011         .quota_on_meta  = ocfs2_quota_on,
1012         .quota_off      = ocfs2_quota_off,
1013         .quota_sync     = dquot_quota_sync,
1014         .get_info       = dquot_get_dqinfo,
1015         .set_info       = dquot_set_dqinfo,
1016         .get_dqblk      = dquot_get_dqblk,
1017         .set_dqblk      = dquot_set_dqblk,
1018 };
1019
1020 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1021 {
1022         struct dentry *root;
1023         int status, sector_size;
1024         struct mount_options parsed_options;
1025         struct inode *inode = NULL;
1026         struct ocfs2_super *osb = NULL;
1027         struct buffer_head *bh = NULL;
1028         char nodestr[12];
1029         struct ocfs2_blockcheck_stats stats;
1030
1031         trace_ocfs2_fill_super(sb, data, silent);
1032
1033         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1034                 status = -EINVAL;
1035                 goto read_super_error;
1036         }
1037
1038         /* probe for superblock */
1039         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1040         if (status < 0) {
1041                 mlog(ML_ERROR, "superblock probe failed!\n");
1042                 goto read_super_error;
1043         }
1044
1045         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1046         osb = OCFS2_SB(sb);
1047         if (status < 0) {
1048                 mlog_errno(status);
1049                 goto read_super_error;
1050         }
1051         brelse(bh);
1052         bh = NULL;
1053
1054         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1055                 status = -EINVAL;
1056                 goto read_super_error;
1057         }
1058         osb->s_mount_opt = parsed_options.mount_opt;
1059         osb->s_atime_quantum = parsed_options.atime_quantum;
1060         osb->preferred_slot = parsed_options.slot;
1061         osb->osb_commit_interval = parsed_options.commit_interval;
1062
1063         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1064         osb->osb_resv_level = parsed_options.resv_level;
1065         osb->osb_dir_resv_level = parsed_options.resv_level;
1066         if (parsed_options.dir_resv_level == -1)
1067                 osb->osb_dir_resv_level = parsed_options.resv_level;
1068         else
1069                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1070
1071         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1072         if (status)
1073                 goto read_super_error;
1074
1075         sb->s_magic = OCFS2_SUPER_MAGIC;
1076
1077         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1078                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1079
1080         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1081          * heartbeat=none */
1082         if (bdev_read_only(sb->s_bdev)) {
1083                 if (!(sb->s_flags & MS_RDONLY)) {
1084                         status = -EACCES;
1085                         mlog(ML_ERROR, "Readonly device detected but readonly "
1086                              "mount was not specified.\n");
1087                         goto read_super_error;
1088                 }
1089
1090                 /* You should not be able to start a local heartbeat
1091                  * on a readonly device. */
1092                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1093                         status = -EROFS;
1094                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1095                              "device.\n");
1096                         goto read_super_error;
1097                 }
1098
1099                 status = ocfs2_check_journals_nolocks(osb);
1100                 if (status < 0) {
1101                         if (status == -EROFS)
1102                                 mlog(ML_ERROR, "Recovery required on readonly "
1103                                      "file system, but write access is "
1104                                      "unavailable.\n");
1105                         else
1106                                 mlog_errno(status);
1107                         goto read_super_error;
1108                 }
1109
1110                 ocfs2_set_ro_flag(osb, 1);
1111
1112                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1113                        "Cluster services will not be used for this mount. "
1114                        "Recovery will be skipped.\n", osb->dev_str);
1115         }
1116
1117         if (!ocfs2_is_hard_readonly(osb)) {
1118                 if (sb->s_flags & MS_RDONLY)
1119                         ocfs2_set_ro_flag(osb, 0);
1120         }
1121
1122         status = ocfs2_verify_heartbeat(osb);
1123         if (status < 0) {
1124                 mlog_errno(status);
1125                 goto read_super_error;
1126         }
1127
1128         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1129                                                  ocfs2_debugfs_root);
1130         if (!osb->osb_debug_root) {
1131                 status = -EINVAL;
1132                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1133                 goto read_super_error;
1134         }
1135
1136         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1137                                             osb->osb_debug_root,
1138                                             osb,
1139                                             &ocfs2_osb_debug_fops);
1140         if (!osb->osb_ctxt) {
1141                 status = -EINVAL;
1142                 mlog_errno(status);
1143                 goto read_super_error;
1144         }
1145
1146         if (ocfs2_meta_ecc(osb)) {
1147                 status = ocfs2_blockcheck_stats_debugfs_install(
1148                                                 &osb->osb_ecc_stats,
1149                                                 osb->osb_debug_root);
1150                 if (status) {
1151                         mlog(ML_ERROR,
1152                              "Unable to create blockcheck statistics "
1153                              "files\n");
1154                         goto read_super_error;
1155                 }
1156         }
1157
1158         status = ocfs2_mount_volume(sb);
1159         if (status < 0)
1160                 goto read_super_error;
1161
1162         if (osb->root_inode)
1163                 inode = igrab(osb->root_inode);
1164
1165         if (!inode) {
1166                 status = -EIO;
1167                 mlog_errno(status);
1168                 goto read_super_error;
1169         }
1170
1171         root = d_make_root(inode);
1172         if (!root) {
1173                 status = -ENOMEM;
1174                 mlog_errno(status);
1175                 goto read_super_error;
1176         }
1177
1178         sb->s_root = root;
1179
1180         ocfs2_complete_mount_recovery(osb);
1181
1182         if (ocfs2_mount_local(osb))
1183                 snprintf(nodestr, sizeof(nodestr), "local");
1184         else
1185                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1186
1187         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1188                "with %s data mode.\n",
1189                osb->dev_str, nodestr, osb->slot_num,
1190                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1191                "ordered");
1192
1193         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1194         wake_up(&osb->osb_mount_event);
1195
1196         /* Now we can initialize quotas because we can afford to wait
1197          * for cluster locks recovery now. That also means that truncation
1198          * log recovery can happen but that waits for proper quota setup */
1199         if (!(sb->s_flags & MS_RDONLY)) {
1200                 status = ocfs2_enable_quotas(osb);
1201                 if (status < 0) {
1202                         /* We have to err-out specially here because
1203                          * s_root is already set */
1204                         mlog_errno(status);
1205                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1206                         wake_up(&osb->osb_mount_event);
1207                         return status;
1208                 }
1209         }
1210
1211         ocfs2_complete_quota_recovery(osb);
1212
1213         /* Now we wake up again for processes waiting for quotas */
1214         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1215         wake_up(&osb->osb_mount_event);
1216
1217         /* Start this when the mount is almost sure of being successful */
1218         ocfs2_orphan_scan_start(osb);
1219
1220         return status;
1221
1222 read_super_error:
1223         brelse(bh);
1224
1225         if (osb) {
1226                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1227                 wake_up(&osb->osb_mount_event);
1228                 ocfs2_dismount_volume(sb, 1);
1229         }
1230
1231         if (status)
1232                 mlog_errno(status);
1233         return status;
1234 }
1235
1236 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1237                         int flags,
1238                         const char *dev_name,
1239                         void *data)
1240 {
1241         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1242 }
1243
1244 static void ocfs2_kill_sb(struct super_block *sb)
1245 {
1246         struct ocfs2_super *osb = OCFS2_SB(sb);
1247
1248         /* Failed mount? */
1249         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1250                 goto out;
1251
1252         /* Prevent further queueing of inode drop events */
1253         spin_lock(&dentry_list_lock);
1254         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1255         spin_unlock(&dentry_list_lock);
1256         /* Wait for work to finish and/or remove it */
1257         cancel_work_sync(&osb->dentry_lock_work);
1258 out:
1259         kill_block_super(sb);
1260 }
1261
1262 static struct file_system_type ocfs2_fs_type = {
1263         .owner          = THIS_MODULE,
1264         .name           = "ocfs2",
1265         .mount          = ocfs2_mount,
1266         .kill_sb        = ocfs2_kill_sb,
1267
1268         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1269         .next           = NULL
1270 };
1271 MODULE_ALIAS_FS("ocfs2");
1272
1273 static int ocfs2_check_set_options(struct super_block *sb,
1274                                    struct mount_options *options)
1275 {
1276         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1277             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1278                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1279                 mlog(ML_ERROR, "User quotas were requested, but this "
1280                      "filesystem does not have the feature enabled.\n");
1281                 return 0;
1282         }
1283         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1284             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1285                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1286                 mlog(ML_ERROR, "Group quotas were requested, but this "
1287                      "filesystem does not have the feature enabled.\n");
1288                 return 0;
1289         }
1290         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1291             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1292                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1293                      "feature is not enabled\n");
1294                 return 0;
1295         }
1296         /* No ACL setting specified? Use XATTR feature... */
1297         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1298                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1299                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1300                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1301                 else
1302                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1303         }
1304         return 1;
1305 }
1306
1307 static int ocfs2_parse_options(struct super_block *sb,
1308                                char *options,
1309                                struct mount_options *mopt,
1310                                int is_remount)
1311 {
1312         int status, user_stack = 0;
1313         char *p;
1314         u32 tmp;
1315
1316         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1317
1318         mopt->commit_interval = 0;
1319         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1320         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1321         mopt->slot = OCFS2_INVALID_SLOT;
1322         mopt->localalloc_opt = -1;
1323         mopt->cluster_stack[0] = '\0';
1324         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1325         mopt->dir_resv_level = -1;
1326
1327         if (!options) {
1328                 status = 1;
1329                 goto bail;
1330         }
1331
1332         while ((p = strsep(&options, ",")) != NULL) {
1333                 int token, option;
1334                 substring_t args[MAX_OPT_ARGS];
1335
1336                 if (!*p)
1337                         continue;
1338
1339                 token = match_token(p, tokens, args);
1340                 switch (token) {
1341                 case Opt_hb_local:
1342                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1343                         break;
1344                 case Opt_hb_none:
1345                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1346                         break;
1347                 case Opt_hb_global:
1348                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1349                         break;
1350                 case Opt_barrier:
1351                         if (match_int(&args[0], &option)) {
1352                                 status = 0;
1353                                 goto bail;
1354                         }
1355                         if (option)
1356                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1357                         else
1358                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1359                         break;
1360                 case Opt_intr:
1361                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1362                         break;
1363                 case Opt_nointr:
1364                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1365                         break;
1366                 case Opt_err_panic:
1367                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1368                         break;
1369                 case Opt_err_ro:
1370                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1371                         break;
1372                 case Opt_data_ordered:
1373                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1374                         break;
1375                 case Opt_data_writeback:
1376                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1377                         break;
1378                 case Opt_user_xattr:
1379                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1380                         break;
1381                 case Opt_nouser_xattr:
1382                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1383                         break;
1384                 case Opt_atime_quantum:
1385                         if (match_int(&args[0], &option)) {
1386                                 status = 0;
1387                                 goto bail;
1388                         }
1389                         if (option >= 0)
1390                                 mopt->atime_quantum = option;
1391                         break;
1392                 case Opt_slot:
1393                         option = 0;
1394                         if (match_int(&args[0], &option)) {
1395                                 status = 0;
1396                                 goto bail;
1397                         }
1398                         if (option)
1399                                 mopt->slot = (s16)option;
1400                         break;
1401                 case Opt_commit:
1402                         option = 0;
1403                         if (match_int(&args[0], &option)) {
1404                                 status = 0;
1405                                 goto bail;
1406                         }
1407                         if (option < 0)
1408                                 return 0;
1409                         if (option == 0)
1410                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1411                         mopt->commit_interval = HZ * option;
1412                         break;
1413                 case Opt_localalloc:
1414                         option = 0;
1415                         if (match_int(&args[0], &option)) {
1416                                 status = 0;
1417                                 goto bail;
1418                         }
1419                         if (option >= 0)
1420                                 mopt->localalloc_opt = option;
1421                         break;
1422                 case Opt_localflocks:
1423                         /*
1424                          * Changing this during remount could race
1425                          * flock() requests, or "unbalance" existing
1426                          * ones (e.g., a lock is taken in one mode but
1427                          * dropped in the other). If users care enough
1428                          * to flip locking modes during remount, we
1429                          * could add a "local" flag to individual
1430                          * flock structures for proper tracking of
1431                          * state.
1432                          */
1433                         if (!is_remount)
1434                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1435                         break;
1436                 case Opt_stack:
1437                         /* Check both that the option we were passed
1438                          * is of the right length and that it is a proper
1439                          * string of the right length.
1440                          */
1441                         if (((args[0].to - args[0].from) !=
1442                              OCFS2_STACK_LABEL_LEN) ||
1443                             (strnlen(args[0].from,
1444                                      OCFS2_STACK_LABEL_LEN) !=
1445                              OCFS2_STACK_LABEL_LEN)) {
1446                                 mlog(ML_ERROR,
1447                                      "Invalid cluster_stack option\n");
1448                                 status = 0;
1449                                 goto bail;
1450                         }
1451                         memcpy(mopt->cluster_stack, args[0].from,
1452                                OCFS2_STACK_LABEL_LEN);
1453                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1454                         /*
1455                          * Open code the memcmp here as we don't have
1456                          * an osb to pass to
1457                          * ocfs2_userspace_stack().
1458                          */
1459                         if (memcmp(mopt->cluster_stack,
1460                                    OCFS2_CLASSIC_CLUSTER_STACK,
1461                                    OCFS2_STACK_LABEL_LEN))
1462                                 user_stack = 1;
1463                         break;
1464                 case Opt_inode64:
1465                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1466                         break;
1467                 case Opt_usrquota:
1468                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1469                         break;
1470                 case Opt_grpquota:
1471                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1472                         break;
1473                 case Opt_coherency_buffered:
1474                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1475                         break;
1476                 case Opt_coherency_full:
1477                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1478                         break;
1479                 case Opt_acl:
1480                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1481                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1482                         break;
1483                 case Opt_noacl:
1484                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1485                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1486                         break;
1487                 case Opt_resv_level:
1488                         if (is_remount)
1489                                 break;
1490                         if (match_int(&args[0], &option)) {
1491                                 status = 0;
1492                                 goto bail;
1493                         }
1494                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1495                             option < OCFS2_MAX_RESV_LEVEL)
1496                                 mopt->resv_level = option;
1497                         break;
1498                 case Opt_dir_resv_level:
1499                         if (is_remount)
1500                                 break;
1501                         if (match_int(&args[0], &option)) {
1502                                 status = 0;
1503                                 goto bail;
1504                         }
1505                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1506                             option < OCFS2_MAX_RESV_LEVEL)
1507                                 mopt->dir_resv_level = option;
1508                         break;
1509                 default:
1510                         mlog(ML_ERROR,
1511                              "Unrecognized mount option \"%s\" "
1512                              "or missing value\n", p);
1513                         status = 0;
1514                         goto bail;
1515                 }
1516         }
1517
1518         if (user_stack == 0) {
1519                 /* Ensure only one heartbeat mode */
1520                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1521                                          OCFS2_MOUNT_HB_GLOBAL |
1522                                          OCFS2_MOUNT_HB_NONE);
1523                 if (hweight32(tmp) != 1) {
1524                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1525                         status = 0;
1526                         goto bail;
1527                 }
1528         }
1529
1530         status = 1;
1531
1532 bail:
1533         return status;
1534 }
1535
1536 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1537 {
1538         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1539         unsigned long opts = osb->s_mount_opt;
1540         unsigned int local_alloc_megs;
1541
1542         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1543                 seq_printf(s, ",_netdev");
1544                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1545                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1546                 else
1547                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1548         } else
1549                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1550
1551         if (opts & OCFS2_MOUNT_NOINTR)
1552                 seq_printf(s, ",nointr");
1553
1554         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1555                 seq_printf(s, ",data=writeback");
1556         else
1557                 seq_printf(s, ",data=ordered");
1558
1559         if (opts & OCFS2_MOUNT_BARRIER)
1560                 seq_printf(s, ",barrier=1");
1561
1562         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1563                 seq_printf(s, ",errors=panic");
1564         else
1565                 seq_printf(s, ",errors=remount-ro");
1566
1567         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1568                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1569
1570         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1571
1572         if (osb->osb_commit_interval)
1573                 seq_printf(s, ",commit=%u",
1574                            (unsigned) (osb->osb_commit_interval / HZ));
1575
1576         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1577         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1578                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1579
1580         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1581                 seq_printf(s, ",localflocks,");
1582
1583         if (osb->osb_cluster_stack[0])
1584                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1585                            osb->osb_cluster_stack);
1586         if (opts & OCFS2_MOUNT_USRQUOTA)
1587                 seq_printf(s, ",usrquota");
1588         if (opts & OCFS2_MOUNT_GRPQUOTA)
1589                 seq_printf(s, ",grpquota");
1590
1591         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1592                 seq_printf(s, ",coherency=buffered");
1593         else
1594                 seq_printf(s, ",coherency=full");
1595
1596         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1597                 seq_printf(s, ",nouser_xattr");
1598         else
1599                 seq_printf(s, ",user_xattr");
1600
1601         if (opts & OCFS2_MOUNT_INODE64)
1602                 seq_printf(s, ",inode64");
1603
1604         if (opts & OCFS2_MOUNT_POSIX_ACL)
1605                 seq_printf(s, ",acl");
1606         else
1607                 seq_printf(s, ",noacl");
1608
1609         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1610                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1611
1612         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1613                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1614
1615         return 0;
1616 }
1617
1618 static int __init ocfs2_init(void)
1619 {
1620         int status;
1621
1622         status = init_ocfs2_uptodate_cache();
1623         if (status < 0)
1624                 goto out1;
1625
1626         status = ocfs2_initialize_mem_caches();
1627         if (status < 0)
1628                 goto out2;
1629
1630         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1631         if (!ocfs2_wq) {
1632                 status = -ENOMEM;
1633                 goto out3;
1634         }
1635
1636         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1637         if (!ocfs2_debugfs_root) {
1638                 status = -EFAULT;
1639                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1640         }
1641
1642         ocfs2_set_locking_protocol();
1643
1644         status = register_quota_format(&ocfs2_quota_format);
1645         if (status < 0)
1646                 goto out4;
1647         status = register_filesystem(&ocfs2_fs_type);
1648         if (!status)
1649                 return 0;
1650
1651         unregister_quota_format(&ocfs2_quota_format);
1652 out4:
1653         destroy_workqueue(ocfs2_wq);
1654         debugfs_remove(ocfs2_debugfs_root);
1655 out3:
1656         ocfs2_free_mem_caches();
1657 out2:
1658         exit_ocfs2_uptodate_cache();
1659 out1:
1660         mlog_errno(status);
1661         return status;
1662 }
1663
1664 static void __exit ocfs2_exit(void)
1665 {
1666         if (ocfs2_wq) {
1667                 flush_workqueue(ocfs2_wq);
1668                 destroy_workqueue(ocfs2_wq);
1669         }
1670
1671         unregister_quota_format(&ocfs2_quota_format);
1672
1673         debugfs_remove(ocfs2_debugfs_root);
1674
1675         ocfs2_free_mem_caches();
1676
1677         unregister_filesystem(&ocfs2_fs_type);
1678
1679         exit_ocfs2_uptodate_cache();
1680 }
1681
1682 static void ocfs2_put_super(struct super_block *sb)
1683 {
1684         trace_ocfs2_put_super(sb);
1685
1686         ocfs2_sync_blockdev(sb);
1687         ocfs2_dismount_volume(sb, 0);
1688 }
1689
1690 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1691 {
1692         struct ocfs2_super *osb;
1693         u32 numbits, freebits;
1694         int status;
1695         struct ocfs2_dinode *bm_lock;
1696         struct buffer_head *bh = NULL;
1697         struct inode *inode = NULL;
1698
1699         trace_ocfs2_statfs(dentry->d_sb, buf);
1700
1701         osb = OCFS2_SB(dentry->d_sb);
1702
1703         inode = ocfs2_get_system_file_inode(osb,
1704                                             GLOBAL_BITMAP_SYSTEM_INODE,
1705                                             OCFS2_INVALID_SLOT);
1706         if (!inode) {
1707                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1708                 status = -EIO;
1709                 goto bail;
1710         }
1711
1712         status = ocfs2_inode_lock(inode, &bh, 0);
1713         if (status < 0) {
1714                 mlog_errno(status);
1715                 goto bail;
1716         }
1717
1718         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1719
1720         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1721         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1722
1723         buf->f_type = OCFS2_SUPER_MAGIC;
1724         buf->f_bsize = dentry->d_sb->s_blocksize;
1725         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1726         buf->f_blocks = ((sector_t) numbits) *
1727                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1728         buf->f_bfree = ((sector_t) freebits) *
1729                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1730         buf->f_bavail = buf->f_bfree;
1731         buf->f_files = numbits;
1732         buf->f_ffree = freebits;
1733         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1734                                 & 0xFFFFFFFFUL;
1735         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1736                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1737
1738         brelse(bh);
1739
1740         ocfs2_inode_unlock(inode, 0);
1741         status = 0;
1742 bail:
1743         if (inode)
1744                 iput(inode);
1745
1746         if (status)
1747                 mlog_errno(status);
1748
1749         return status;
1750 }
1751
1752 static void ocfs2_inode_init_once(void *data)
1753 {
1754         struct ocfs2_inode_info *oi = data;
1755
1756         oi->ip_flags = 0;
1757         oi->ip_open_count = 0;
1758         spin_lock_init(&oi->ip_lock);
1759         ocfs2_extent_map_init(&oi->vfs_inode);
1760         INIT_LIST_HEAD(&oi->ip_io_markers);
1761         oi->ip_dir_start_lookup = 0;
1762         mutex_init(&oi->ip_unaligned_aio);
1763         init_rwsem(&oi->ip_alloc_sem);
1764         init_rwsem(&oi->ip_xattr_sem);
1765         mutex_init(&oi->ip_io_mutex);
1766
1767         oi->ip_blkno = 0ULL;
1768         oi->ip_clusters = 0;
1769
1770         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1771
1772         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1773         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1774         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1775
1776         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1777                                   &ocfs2_inode_caching_ops);
1778
1779         inode_init_once(&oi->vfs_inode);
1780 }
1781
1782 static int ocfs2_initialize_mem_caches(void)
1783 {
1784         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1785                                        sizeof(struct ocfs2_inode_info),
1786                                        0,
1787                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1788                                                 SLAB_MEM_SPREAD),
1789                                        ocfs2_inode_init_once);
1790         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1791                                         sizeof(struct ocfs2_dquot),
1792                                         0,
1793                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1794                                                 SLAB_MEM_SPREAD),
1795                                         NULL);
1796         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1797                                         sizeof(struct ocfs2_quota_chunk),
1798                                         0,
1799                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1800                                         NULL);
1801         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1802             !ocfs2_qf_chunk_cachep) {
1803                 if (ocfs2_inode_cachep)
1804                         kmem_cache_destroy(ocfs2_inode_cachep);
1805                 if (ocfs2_dquot_cachep)
1806                         kmem_cache_destroy(ocfs2_dquot_cachep);
1807                 if (ocfs2_qf_chunk_cachep)
1808                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1809                 return -ENOMEM;
1810         }
1811
1812         return 0;
1813 }
1814
1815 static void ocfs2_free_mem_caches(void)
1816 {
1817         /*
1818          * Make sure all delayed rcu free inodes are flushed before we
1819          * destroy cache.
1820          */
1821         rcu_barrier();
1822         if (ocfs2_inode_cachep)
1823                 kmem_cache_destroy(ocfs2_inode_cachep);
1824         ocfs2_inode_cachep = NULL;
1825
1826         if (ocfs2_dquot_cachep)
1827                 kmem_cache_destroy(ocfs2_dquot_cachep);
1828         ocfs2_dquot_cachep = NULL;
1829
1830         if (ocfs2_qf_chunk_cachep)
1831                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1832         ocfs2_qf_chunk_cachep = NULL;
1833 }
1834
1835 static int ocfs2_get_sector(struct super_block *sb,
1836                             struct buffer_head **bh,
1837                             int block,
1838                             int sect_size)
1839 {
1840         if (!sb_set_blocksize(sb, sect_size)) {
1841                 mlog(ML_ERROR, "unable to set blocksize\n");
1842                 return -EIO;
1843         }
1844
1845         *bh = sb_getblk(sb, block);
1846         if (!*bh) {
1847                 mlog_errno(-ENOMEM);
1848                 return -ENOMEM;
1849         }
1850         lock_buffer(*bh);
1851         if (!buffer_dirty(*bh))
1852                 clear_buffer_uptodate(*bh);
1853         unlock_buffer(*bh);
1854         ll_rw_block(READ, 1, bh);
1855         wait_on_buffer(*bh);
1856         if (!buffer_uptodate(*bh)) {
1857                 mlog_errno(-EIO);
1858                 brelse(*bh);
1859                 *bh = NULL;
1860                 return -EIO;
1861         }
1862
1863         return 0;
1864 }
1865
1866 static int ocfs2_mount_volume(struct super_block *sb)
1867 {
1868         int status = 0;
1869         int unlock_super = 0;
1870         struct ocfs2_super *osb = OCFS2_SB(sb);
1871
1872         if (ocfs2_is_hard_readonly(osb))
1873                 goto leave;
1874
1875         status = ocfs2_dlm_init(osb);
1876         if (status < 0) {
1877                 mlog_errno(status);
1878                 goto leave;
1879         }
1880
1881         status = ocfs2_super_lock(osb, 1);
1882         if (status < 0) {
1883                 mlog_errno(status);
1884                 goto leave;
1885         }
1886         unlock_super = 1;
1887
1888         /* This will load up the node map and add ourselves to it. */
1889         status = ocfs2_find_slot(osb);
1890         if (status < 0) {
1891                 mlog_errno(status);
1892                 goto leave;
1893         }
1894
1895         /* load all node-local system inodes */
1896         status = ocfs2_init_local_system_inodes(osb);
1897         if (status < 0) {
1898                 mlog_errno(status);
1899                 goto leave;
1900         }
1901
1902         status = ocfs2_check_volume(osb);
1903         if (status < 0) {
1904                 mlog_errno(status);
1905                 goto leave;
1906         }
1907
1908         status = ocfs2_truncate_log_init(osb);
1909         if (status < 0)
1910                 mlog_errno(status);
1911
1912 leave:
1913         if (unlock_super)
1914                 ocfs2_super_unlock(osb, 1);
1915
1916         return status;
1917 }
1918
1919 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1920 {
1921         int tmp, hangup_needed = 0;
1922         struct ocfs2_super *osb = NULL;
1923         char nodestr[12];
1924
1925         trace_ocfs2_dismount_volume(sb);
1926
1927         BUG_ON(!sb);
1928         osb = OCFS2_SB(sb);
1929         BUG_ON(!osb);
1930
1931         debugfs_remove(osb->osb_ctxt);
1932
1933         /*
1934          * Flush inode dropping work queue so that deletes are
1935          * performed while the filesystem is still working
1936          */
1937         ocfs2_drop_all_dl_inodes(osb);
1938
1939         /* Orphan scan should be stopped as early as possible */
1940         ocfs2_orphan_scan_stop(osb);
1941
1942         ocfs2_disable_quotas(osb);
1943
1944         ocfs2_shutdown_local_alloc(osb);
1945
1946         /* This will disable recovery and flush any recovery work. */
1947         ocfs2_recovery_exit(osb);
1948
1949         /*
1950          * During dismount, when it recovers another node it will call
1951          * ocfs2_recover_orphans and queue delayed work osb_truncate_log_wq.
1952          */
1953         ocfs2_truncate_log_shutdown(osb);
1954
1955         ocfs2_journal_shutdown(osb);
1956
1957         ocfs2_sync_blockdev(sb);
1958
1959         ocfs2_purge_refcount_trees(osb);
1960
1961         /* No cluster connection means we've failed during mount, so skip
1962          * all the steps which depended on that to complete. */
1963         if (osb->cconn) {
1964                 tmp = ocfs2_super_lock(osb, 1);
1965                 if (tmp < 0) {
1966                         mlog_errno(tmp);
1967                         return;
1968                 }
1969         }
1970
1971         if (osb->slot_num != OCFS2_INVALID_SLOT)
1972                 ocfs2_put_slot(osb);
1973
1974         if (osb->cconn)
1975                 ocfs2_super_unlock(osb, 1);
1976
1977         ocfs2_release_system_inodes(osb);
1978
1979         /*
1980          * If we're dismounting due to mount error, mount.ocfs2 will clean
1981          * up heartbeat.  If we're a local mount, there is no heartbeat.
1982          * If we failed before we got a uuid_str yet, we can't stop
1983          * heartbeat.  Otherwise, do it.
1984          */
1985         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1986             !ocfs2_is_hard_readonly(osb))
1987                 hangup_needed = 1;
1988
1989         if (osb->cconn)
1990                 ocfs2_dlm_shutdown(osb, hangup_needed);
1991
1992         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1993         debugfs_remove(osb->osb_debug_root);
1994
1995         if (hangup_needed)
1996                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1997
1998         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1999
2000         if (ocfs2_mount_local(osb))
2001                 snprintf(nodestr, sizeof(nodestr), "local");
2002         else
2003                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2004
2005         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2006                osb->dev_str, nodestr);
2007
2008         ocfs2_delete_osb(osb);
2009         kfree(osb);
2010         sb->s_dev = 0;
2011         sb->s_fs_info = NULL;
2012 }
2013
2014 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2015                                 unsigned uuid_bytes)
2016 {
2017         int i, ret;
2018         char *ptr;
2019
2020         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2021
2022         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2023         if (osb->uuid_str == NULL)
2024                 return -ENOMEM;
2025
2026         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2027                 /* print with null */
2028                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2029                 if (ret != 2) /* drop super cleans up */
2030                         return -EINVAL;
2031                 /* then only advance past the last char */
2032                 ptr += 2;
2033         }
2034
2035         return 0;
2036 }
2037
2038 /* Make sure entire volume is addressable by our journal.  Requires
2039    osb_clusters_at_boot to be valid and for the journal to have been
2040    initialized by ocfs2_journal_init(). */
2041 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2042 {
2043         int status = 0;
2044         u64 max_block =
2045                 ocfs2_clusters_to_blocks(osb->sb,
2046                                          osb->osb_clusters_at_boot) - 1;
2047
2048         /* 32-bit block number is always OK. */
2049         if (max_block <= (u32)~0ULL)
2050                 goto out;
2051
2052         /* Volume is "huge", so see if our journal is new enough to
2053            support it. */
2054         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2055                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2056               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2057                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2058                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2059                      "Enable the 'block64' journal option with tunefs.ocfs2");
2060                 status = -EFBIG;
2061                 goto out;
2062         }
2063
2064  out:
2065         return status;
2066 }
2067
2068 static int ocfs2_initialize_super(struct super_block *sb,
2069                                   struct buffer_head *bh,
2070                                   int sector_size,
2071                                   struct ocfs2_blockcheck_stats *stats)
2072 {
2073         int status;
2074         int i, cbits, bbits;
2075         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2076         struct inode *inode = NULL;
2077         struct ocfs2_journal *journal;
2078         struct ocfs2_super *osb;
2079         u64 total_blocks;
2080
2081         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2082         if (!osb) {
2083                 status = -ENOMEM;
2084                 mlog_errno(status);
2085                 goto bail;
2086         }
2087
2088         sb->s_fs_info = osb;
2089         sb->s_op = &ocfs2_sops;
2090         sb->s_d_op = &ocfs2_dentry_ops;
2091         sb->s_export_op = &ocfs2_export_ops;
2092         sb->s_qcop = &ocfs2_quotactl_ops;
2093         sb->dq_op = &ocfs2_quota_operations;
2094         sb->s_xattr = ocfs2_xattr_handlers;
2095         sb->s_time_gran = 1;
2096         sb->s_flags |= MS_NOATIME;
2097         /* this is needed to support O_LARGEFILE */
2098         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2099         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2100         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2101
2102         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2103
2104         for (i = 0; i < 3; i++)
2105                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2106         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2107
2108         osb->sb = sb;
2109         /* Save off for ocfs2_rw_direct */
2110         osb->s_sectsize_bits = blksize_bits(sector_size);
2111         BUG_ON(!osb->s_sectsize_bits);
2112
2113         spin_lock_init(&osb->dc_task_lock);
2114         init_waitqueue_head(&osb->dc_event);
2115         osb->dc_work_sequence = 0;
2116         osb->dc_wake_sequence = 0;
2117         INIT_LIST_HEAD(&osb->blocked_lock_list);
2118         osb->blocked_lock_count = 0;
2119         spin_lock_init(&osb->osb_lock);
2120         spin_lock_init(&osb->osb_xattr_lock);
2121         ocfs2_init_steal_slots(osb);
2122
2123         atomic_set(&osb->alloc_stats.moves, 0);
2124         atomic_set(&osb->alloc_stats.local_data, 0);
2125         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2126         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2127         atomic_set(&osb->alloc_stats.bg_extends, 0);
2128
2129         /* Copy the blockcheck stats from the superblock probe */
2130         osb->osb_ecc_stats = *stats;
2131
2132         ocfs2_init_node_maps(osb);
2133
2134         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2135                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2136
2137         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2138         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2139                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2140                      osb->max_slots);
2141                 status = -EINVAL;
2142                 goto bail;
2143         }
2144
2145         ocfs2_orphan_scan_init(osb);
2146
2147         status = ocfs2_recovery_init(osb);
2148         if (status) {
2149                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2150                 mlog_errno(status);
2151                 goto bail;
2152         }
2153
2154         init_waitqueue_head(&osb->checkpoint_event);
2155
2156         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2157
2158         osb->slot_num = OCFS2_INVALID_SLOT;
2159
2160         osb->s_xattr_inline_size = le16_to_cpu(
2161                                         di->id2.i_super.s_xattr_inline_size);
2162
2163         osb->local_alloc_state = OCFS2_LA_UNUSED;
2164         osb->local_alloc_bh = NULL;
2165         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2166
2167         init_waitqueue_head(&osb->osb_mount_event);
2168
2169         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2170         if (status) {
2171                 mlog_errno(status);
2172                 goto bail;
2173         }
2174
2175         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2176         if (!osb->vol_label) {
2177                 mlog(ML_ERROR, "unable to alloc vol label\n");
2178                 status = -ENOMEM;
2179                 goto bail;
2180         }
2181
2182         osb->slot_recovery_generations =
2183                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2184                         GFP_KERNEL);
2185         if (!osb->slot_recovery_generations) {
2186                 status = -ENOMEM;
2187                 mlog_errno(status);
2188                 goto bail;
2189         }
2190
2191         init_waitqueue_head(&osb->osb_wipe_event);
2192         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2193                                         sizeof(*osb->osb_orphan_wipes),
2194                                         GFP_KERNEL);
2195         if (!osb->osb_orphan_wipes) {
2196                 status = -ENOMEM;
2197                 mlog_errno(status);
2198                 goto bail;
2199         }
2200
2201         osb->osb_rf_lock_tree = RB_ROOT;
2202
2203         osb->s_feature_compat =
2204                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2205         osb->s_feature_ro_compat =
2206                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2207         osb->s_feature_incompat =
2208                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2209
2210         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2211                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2212                      "optional features (%x).\n", i);
2213                 status = -EINVAL;
2214                 goto bail;
2215         }
2216         if (!(osb->sb->s_flags & MS_RDONLY) &&
2217             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2218                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2219                      "unsupported optional features (%x).\n", i);
2220                 status = -EINVAL;
2221                 goto bail;
2222         }
2223
2224         if (ocfs2_clusterinfo_valid(osb)) {
2225                 osb->osb_stackflags =
2226                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2227                 strlcpy(osb->osb_cluster_stack,
2228                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2229                        OCFS2_STACK_LABEL_LEN + 1);
2230                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2231                         mlog(ML_ERROR,
2232                              "couldn't mount because of an invalid "
2233                              "cluster stack label (%s) \n",
2234                              osb->osb_cluster_stack);
2235                         status = -EINVAL;
2236                         goto bail;
2237                 }
2238                 strlcpy(osb->osb_cluster_name,
2239                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2240                         OCFS2_CLUSTER_NAME_LEN + 1);
2241         } else {
2242                 /* The empty string is identical with classic tools that
2243                  * don't know about s_cluster_info. */
2244                 osb->osb_cluster_stack[0] = '\0';
2245         }
2246
2247         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2248
2249         /* FIXME
2250          * This should be done in ocfs2_journal_init(), but unknown
2251          * ordering issues will cause the filesystem to crash.
2252          * If anyone wants to figure out what part of the code
2253          * refers to osb->journal before ocfs2_journal_init() is run,
2254          * be my guest.
2255          */
2256         /* initialize our journal structure */
2257
2258         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2259         if (!journal) {
2260                 mlog(ML_ERROR, "unable to alloc journal\n");
2261                 status = -ENOMEM;
2262                 goto bail;
2263         }
2264         osb->journal = journal;
2265         journal->j_osb = osb;
2266
2267         atomic_set(&journal->j_num_trans, 0);
2268         init_rwsem(&journal->j_trans_barrier);
2269         init_waitqueue_head(&journal->j_checkpointed);
2270         spin_lock_init(&journal->j_lock);
2271         journal->j_trans_id = (unsigned long) 1;
2272         INIT_LIST_HEAD(&journal->j_la_cleanups);
2273         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2274         journal->j_state = OCFS2_JOURNAL_FREE;
2275
2276         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2277         osb->dentry_lock_list = NULL;
2278
2279         /* get some pseudo constants for clustersize bits */
2280         osb->s_clustersize_bits =
2281                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2282         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2283
2284         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2285             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2286                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2287                      osb->s_clustersize);
2288                 status = -EINVAL;
2289                 goto bail;
2290         }
2291
2292         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2293                                                 le32_to_cpu(di->i_clusters));
2294
2295         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2296                                            total_blocks);
2297         if (status) {
2298                 mlog(ML_ERROR, "Volume too large "
2299                      "to mount safely on this system");
2300                 status = -EFBIG;
2301                 goto bail;
2302         }
2303
2304         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2305                                  sizeof(di->id2.i_super.s_uuid))) {
2306                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2307                 status = -ENOMEM;
2308                 goto bail;
2309         }
2310
2311         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2312         osb->vol_label[63] = '\0';
2313         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2314         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2315         osb->first_cluster_group_blkno =
2316                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2317         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2318         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2319         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2320                                      (unsigned long long)osb->root_blkno,
2321                                      (unsigned long long)osb->system_dir_blkno,
2322                                      osb->s_clustersize_bits);
2323
2324         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2325         if (!osb->osb_dlm_debug) {
2326                 status = -ENOMEM;
2327                 mlog_errno(status);
2328                 goto bail;
2329         }
2330
2331         atomic_set(&osb->vol_state, VOLUME_INIT);
2332
2333         /* load root, system_dir, and all global system inodes */
2334         status = ocfs2_init_global_system_inodes(osb);
2335         if (status < 0) {
2336                 mlog_errno(status);
2337                 goto bail;
2338         }
2339
2340         /*
2341          * global bitmap
2342          */
2343         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2344                                             OCFS2_INVALID_SLOT);
2345         if (!inode) {
2346                 status = -EINVAL;
2347                 mlog_errno(status);
2348                 goto bail;
2349         }
2350
2351         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2352         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2353         iput(inode);
2354
2355         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2356                                  osb->s_feature_incompat) * 8;
2357
2358         status = ocfs2_init_slot_info(osb);
2359         if (status < 0) {
2360                 mlog_errno(status);
2361                 goto bail;
2362         }
2363         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2364
2365 bail:
2366         return status;
2367 }
2368
2369 /*
2370  * will return: -EAGAIN if it is ok to keep searching for superblocks
2371  *              -EINVAL if there is a bad superblock
2372  *              0 on success
2373  */
2374 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2375                                struct buffer_head *bh,
2376                                u32 blksz,
2377                                struct ocfs2_blockcheck_stats *stats)
2378 {
2379         int status = -EAGAIN;
2380
2381         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2382                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2383                 /* We have to do a raw check of the feature here */
2384                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2385                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2386                         status = ocfs2_block_check_validate(bh->b_data,
2387                                                             bh->b_size,
2388                                                             &di->i_check,
2389                                                             stats);
2390                         if (status)
2391                                 goto out;
2392                 }
2393                 status = -EINVAL;
2394                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2395                         mlog(ML_ERROR, "found superblock with incorrect block "
2396                              "size: found %u, should be %u\n",
2397                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2398                                blksz);
2399                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2400                            OCFS2_MAJOR_REV_LEVEL ||
2401                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2402                            OCFS2_MINOR_REV_LEVEL) {
2403                         mlog(ML_ERROR, "found superblock with bad version: "
2404                              "found %u.%u, should be %u.%u\n",
2405                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2406                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2407                              OCFS2_MAJOR_REV_LEVEL,
2408                              OCFS2_MINOR_REV_LEVEL);
2409                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2410                         mlog(ML_ERROR, "bad block number on superblock: "
2411                              "found %llu, should be %llu\n",
2412                              (unsigned long long)le64_to_cpu(di->i_blkno),
2413                              (unsigned long long)bh->b_blocknr);
2414                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2415                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2416                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2417                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2418                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2419                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2420                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2421                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2422                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2423                         mlog(ML_ERROR,
2424                              "Superblock slots found greater than file system "
2425                              "maximum: found %u, max %u\n",
2426                              le16_to_cpu(di->id2.i_super.s_max_slots),
2427                              OCFS2_MAX_SLOTS);
2428                 } else {
2429                         /* found it! */
2430                         status = 0;
2431                 }
2432         }
2433
2434 out:
2435         if (status && status != -EAGAIN)
2436                 mlog_errno(status);
2437         return status;
2438 }
2439
2440 static int ocfs2_check_volume(struct ocfs2_super *osb)
2441 {
2442         int status;
2443         int dirty;
2444         int local;
2445         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2446                                                   * recover
2447                                                   * ourselves. */
2448
2449         /* Init our journal object. */
2450         status = ocfs2_journal_init(osb->journal, &dirty);
2451         if (status < 0) {
2452                 mlog(ML_ERROR, "Could not initialize journal!\n");
2453                 goto finally;
2454         }
2455
2456         /* Now that journal has been initialized, check to make sure
2457            entire volume is addressable. */
2458         status = ocfs2_journal_addressable(osb);
2459         if (status)
2460                 goto finally;
2461
2462         /* If the journal was unmounted cleanly then we don't want to
2463          * recover anything. Otherwise, journal_load will do that
2464          * dirty work for us :) */
2465         if (!dirty) {
2466                 status = ocfs2_journal_wipe(osb->journal, 0);
2467                 if (status < 0) {
2468                         mlog_errno(status);
2469                         goto finally;
2470                 }
2471         } else {
2472                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2473                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2474         }
2475
2476         local = ocfs2_mount_local(osb);
2477
2478         /* will play back anything left in the journal. */
2479         status = ocfs2_journal_load(osb->journal, local, dirty);
2480         if (status < 0) {
2481                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2482                 goto finally;
2483         }
2484
2485         if (dirty) {
2486                 /* recover my local alloc if we didn't unmount cleanly. */
2487                 status = ocfs2_begin_local_alloc_recovery(osb,
2488                                                           osb->slot_num,
2489                                                           &local_alloc);
2490                 if (status < 0) {
2491                         mlog_errno(status);
2492                         goto finally;
2493                 }
2494                 /* we complete the recovery process after we've marked
2495                  * ourselves as mounted. */
2496         }
2497
2498         status = ocfs2_load_local_alloc(osb);
2499         if (status < 0) {
2500                 mlog_errno(status);
2501                 goto finally;
2502         }
2503
2504         if (dirty) {
2505                 /* Recovery will be completed after we've mounted the
2506                  * rest of the volume. */
2507                 osb->dirty = 1;
2508                 osb->local_alloc_copy = local_alloc;
2509                 local_alloc = NULL;
2510         }
2511
2512         /* go through each journal, trylock it and if you get the
2513          * lock, and it's marked as dirty, set the bit in the recover
2514          * map and launch a recovery thread for it. */
2515         status = ocfs2_mark_dead_nodes(osb);
2516         if (status < 0) {
2517                 mlog_errno(status);
2518                 goto finally;
2519         }
2520
2521         status = ocfs2_compute_replay_slots(osb);
2522         if (status < 0)
2523                 mlog_errno(status);
2524
2525 finally:
2526         kfree(local_alloc);
2527
2528         if (status)
2529                 mlog_errno(status);
2530         return status;
2531 }
2532
2533 /*
2534  * The routine gets called from dismount or close whenever a dismount on
2535  * volume is requested and the osb open count becomes 1.
2536  * It will remove the osb from the global list and also free up all the
2537  * initialized resources and fileobject.
2538  */
2539 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2540 {
2541         /* This function assumes that the caller has the main osb resource */
2542
2543         ocfs2_free_slot_info(osb);
2544
2545         kfree(osb->osb_orphan_wipes);
2546         kfree(osb->slot_recovery_generations);
2547         /* FIXME
2548          * This belongs in journal shutdown, but because we have to
2549          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2550          * we free it here.
2551          */
2552         kfree(osb->journal);
2553         kfree(osb->local_alloc_copy);
2554         kfree(osb->uuid_str);
2555         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2556         memset(osb, 0, sizeof(struct ocfs2_super));
2557 }
2558
2559 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2560  * panic(). We do not support continue-on-error operation. */
2561 static void ocfs2_handle_error(struct super_block *sb)
2562 {
2563         struct ocfs2_super *osb = OCFS2_SB(sb);
2564
2565         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2566                 panic("OCFS2: (device %s): panic forced after error\n",
2567                       sb->s_id);
2568
2569         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2570
2571         if (sb->s_flags & MS_RDONLY &&
2572             (ocfs2_is_soft_readonly(osb) ||
2573              ocfs2_is_hard_readonly(osb)))
2574                 return;
2575
2576         printk(KERN_CRIT "File system is now read-only due to the potential "
2577                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2578                "system is unmounted.\n");
2579         sb->s_flags |= MS_RDONLY;
2580         ocfs2_set_ro_flag(osb, 0);
2581 }
2582
2583 static char error_buf[1024];
2584
2585 void __ocfs2_error(struct super_block *sb,
2586                    const char *function,
2587                    const char *fmt, ...)
2588 {
2589         va_list args;
2590
2591         va_start(args, fmt);
2592         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2593         va_end(args);
2594
2595         /* Not using mlog here because we want to show the actual
2596          * function the error came from. */
2597         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2598                sb->s_id, function, error_buf);
2599
2600         ocfs2_handle_error(sb);
2601 }
2602
2603 /* Handle critical errors. This is intentionally more drastic than
2604  * ocfs2_handle_error, so we only use for things like journal errors,
2605  * etc. */
2606 void __ocfs2_abort(struct super_block* sb,
2607                    const char *function,
2608                    const char *fmt, ...)
2609 {
2610         va_list args;
2611
2612         va_start(args, fmt);
2613         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2614         va_end(args);
2615
2616         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2617                sb->s_id, function, error_buf);
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);