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