fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 147 {
 148         return OCFS2_I(inode)->i_dquot;
 149 }
 150
 151 static const struct super_operations ocfs2_sops = {
 152         .statfs         = ocfs2_statfs,
 153         .alloc_inode    = ocfs2_alloc_inode,
 154         .destroy_inode  = ocfs2_destroy_inode,
 155         .drop_inode     = ocfs2_drop_inode,
 156         .evict_inode    = ocfs2_evict_inode,
 157         .sync_fs        = ocfs2_sync_fs,
 158         .put_super      = ocfs2_put_super,
 159         .remount_fs     = ocfs2_remount,
 160         .show_options   = ocfs2_show_options,
 161         .quota_read     = ocfs2_quota_read,
 162         .quota_write    = ocfs2_quota_write,
 163         .get_dquots     = ocfs2_get_dquots,
 164 };
 165
 166 enum {
 167         Opt_barrier,
 168         Opt_err_panic,
 169         Opt_err_ro,
 170         Opt_intr,
 171         Opt_nointr,
 172         Opt_hb_none,
 173         Opt_hb_local,
 174         Opt_hb_global,
 175         Opt_data_ordered,
 176         Opt_data_writeback,
 177         Opt_atime_quantum,
 178         Opt_slot,
 179         Opt_commit,
 180         Opt_localalloc,
 181         Opt_localflocks,
 182         Opt_stack,
 183         Opt_user_xattr,
 184         Opt_nouser_xattr,
 185         Opt_inode64,
 186         Opt_acl,
 187         Opt_noacl,
 188         Opt_usrquota,
 189         Opt_grpquota,
 190         Opt_coherency_buffered,
 191         Opt_coherency_full,
 192         Opt_resv_level,
 193         Opt_dir_resv_level,
 194         Opt_journal_async_commit,
 195         Opt_err,
 196 };
 197
 198 static const match_table_t tokens = {
 199         {Opt_barrier, "barrier=%u"},
 200         {Opt_err_panic, "errors=panic"},
 201         {Opt_err_ro, "errors=remount-ro"},
 202         {Opt_intr, "intr"},
 203         {Opt_nointr, "nointr"},
 204         {Opt_hb_none, OCFS2_HB_NONE},
 205         {Opt_hb_local, OCFS2_HB_LOCAL},
 206         {Opt_hb_global, OCFS2_HB_GLOBAL},
 207         {Opt_data_ordered, "data=ordered"},
 208         {Opt_data_writeback, "data=writeback"},
 209         {Opt_atime_quantum, "atime_quantum=%u"},
 210         {Opt_slot, "preferred_slot=%u"},
 211         {Opt_commit, "commit=%u"},
 212         {Opt_localalloc, "localalloc=%d"},
 213         {Opt_localflocks, "localflocks"},
 214         {Opt_stack, "cluster_stack=%s"},
 215         {Opt_user_xattr, "user_xattr"},
 216         {Opt_nouser_xattr, "nouser_xattr"},
 217         {Opt_inode64, "inode64"},
 218         {Opt_acl, "acl"},
 219         {Opt_noacl, "noacl"},
 220         {Opt_usrquota, "usrquota"},
 221         {Opt_grpquota, "grpquota"},
 222         {Opt_coherency_buffered, "coherency=buffered"},
 223         {Opt_coherency_full, "coherency=full"},
 224         {Opt_resv_level, "resv_level=%u"},
 225         {Opt_dir_resv_level, "dir_resv_level=%u"},
 226         {Opt_journal_async_commit, "journal_async_commit"},
 227         {Opt_err, NULL}
 228 };
 229
 230 #ifdef CONFIG_DEBUG_FS
 231 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 232 {
 233         struct ocfs2_cluster_connection *cconn = osb->cconn;
 234         struct ocfs2_recovery_map *rm = osb->recovery_map;
 235         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 236         int i, out = 0;
 237
 238         out += snprintf(buf + out, len - out,
 239                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 240                         "Device", osb->dev_str, osb->uuid_str,
 241                         osb->fs_generation, osb->vol_label);
 242
 243         out += snprintf(buf + out, len - out,
 244                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 245                         atomic_read(&osb->vol_state), osb->osb_flags);
 246
 247         out += snprintf(buf + out, len - out,
 248                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 249                         osb->sb->s_blocksize, osb->s_clustersize);
 250
 251         out += snprintf(buf + out, len - out,
 252                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 253                         "ROcompat: 0x%X\n",
 254                         "Features", osb->s_feature_compat,
 255                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 256
 257         out += snprintf(buf + out, len - out,
 258                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 259                         osb->s_mount_opt, osb->s_atime_quantum);
 260
 261         if (cconn) {
 262                 out += snprintf(buf + out, len - out,
 263                                 "%10s => Stack: %s  Name: %*s  "
 264                                 "Version: %d.%d\n", "Cluster",
 265                                 (*osb->osb_cluster_stack == '\0' ?
 266                                  "o2cb" : osb->osb_cluster_stack),
 267                                 cconn->cc_namelen, cconn->cc_name,
 268                                 cconn->cc_version.pv_major,
 269                                 cconn->cc_version.pv_minor);
 270         }
 271
 272         spin_lock(&osb->dc_task_lock);
 273         out += snprintf(buf + out, len - out,
 274                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 275                         "WorkSeq: %lu\n", "DownCnvt",
 276                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 277                         osb->blocked_lock_count, osb->dc_wake_sequence,
 278                         osb->dc_work_sequence);
 279         spin_unlock(&osb->dc_task_lock);
 280
 281         spin_lock(&osb->osb_lock);
 282         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 283                         "Recovery",
 284                         (osb->recovery_thread_task ?
 285                          task_pid_nr(osb->recovery_thread_task) : -1));
 286         if (rm->rm_used == 0)
 287                 out += snprintf(buf + out, len - out, " None\n");
 288         else {
 289                 for (i = 0; i < rm->rm_used; i++)
 290                         out += snprintf(buf + out, len - out, " %d",
 291                                         rm->rm_entries[i]);
 292                 out += snprintf(buf + out, len - out, "\n");
 293         }
 294         spin_unlock(&osb->osb_lock);
 295
 296         out += snprintf(buf + out, len - out,
 297                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 298                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 299                         osb->osb_commit_interval);
 300
 301         out += snprintf(buf + out, len - out,
 302                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 303                         "Journal", osb->journal->j_state,
 304                         osb->journal->j_trans_id,
 305                         atomic_read(&osb->journal->j_num_trans));
 306
 307         out += snprintf(buf + out, len - out,
 308                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 309                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 310                         "Stats",
 311                         atomic_read(&osb->alloc_stats.bitmap_data),
 312                         atomic_read(&osb->alloc_stats.local_data),
 313                         atomic_read(&osb->alloc_stats.bg_allocs),
 314                         atomic_read(&osb->alloc_stats.moves),
 315                         atomic_read(&osb->alloc_stats.bg_extends));
 316
 317         out += snprintf(buf + out, len - out,
 318                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 319                         "Default: %u bits\n",
 320                         "LocalAlloc", osb->local_alloc_state,
 321                         (unsigned long long)osb->la_last_gd,
 322                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 323
 324         spin_lock(&osb->osb_lock);
 325         out += snprintf(buf + out, len - out,
 326                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 327                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 328                         osb->s_inode_steal_slot,
 329                         atomic_read(&osb->s_num_inodes_stolen),
 330                         osb->s_meta_steal_slot,
 331                         atomic_read(&osb->s_num_meta_stolen));
 332         spin_unlock(&osb->osb_lock);
 333
 334         out += snprintf(buf + out, len - out, "OrphanScan => ");
 335         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 336                         os->os_count, os->os_seqno);
 337         out += snprintf(buf + out, len - out, " Last Scan: ");
 338         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 339                 out += snprintf(buf + out, len - out, "Disabled\n");
 340         else
 341                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 342                                 (get_seconds() - os->os_scantime.tv_sec));
 343
 344         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 345                         "Slots", "Num", "RecoGen");
 346         for (i = 0; i < osb->max_slots; ++i) {
 347                 out += snprintf(buf + out, len - out,
 348                                 "%10s  %c %3d  %10d\n",
 349                                 " ",
 350                                 (i == osb->slot_num ? '*' : ' '),
 351                                 i, osb->slot_recovery_generations[i]);
 352         }
 353
 354         return out;
 355 }
 356
 357 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 358 {
 359         struct ocfs2_super *osb = inode->i_private;
 360         char *buf = NULL;
 361
 362         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 363         if (!buf)
 364                 goto bail;
 365
 366         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 367
 368         file->private_data = buf;
 369
 370         return 0;
 371 bail:
 372         return -ENOMEM;
 373 }
 374
 375 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 376 {
 377         kfree(file->private_data);
 378         return 0;
 379 }
 380
 381 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 382                                 size_t nbytes, loff_t *ppos)
 383 {
 384         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 385                                        i_size_read(file->f_mapping->host));
 386 }
 387 #else
 388 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 389 {
 390         return 0;
 391 }
 392 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 393 {
 394         return 0;
 395 }
 396 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 397                                 size_t nbytes, loff_t *ppos)
 398 {
 399         return 0;
 400 }
 401 #endif  /* CONFIG_DEBUG_FS */
 402
 403 static const struct file_operations ocfs2_osb_debug_fops = {
 404         .open =         ocfs2_osb_debug_open,
 405         .release =      ocfs2_debug_release,
 406         .read =         ocfs2_debug_read,
 407         .llseek =       generic_file_llseek,
 408 };
 409
 410 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 411 {
 412         int status;
 413         tid_t target;
 414         struct ocfs2_super *osb = OCFS2_SB(sb);
 415
 416         if (ocfs2_is_hard_readonly(osb))
 417                 return -EROFS;
 418
 419         if (wait) {
 420                 status = ocfs2_flush_truncate_log(osb);
 421                 if (status < 0)
 422                         mlog_errno(status);
 423         } else {
 424                 ocfs2_schedule_truncate_log_flush(osb, 0);
 425         }
 426
 427         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 428                                       &target)) {
 429                 if (wait)
 430                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 431                                              target);
 432         }
 433         return 0;
 434 }
 435
 436 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 437 {
 438         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 439             && (ino == USER_QUOTA_SYSTEM_INODE
 440                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 441                 return 0;
 442         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 443             && (ino == GROUP_QUOTA_SYSTEM_INODE
 444                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 445                 return 0;
 446         return 1;
 447 }
 448
 449 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 450 {
 451         struct inode *new = NULL;
 452         int status = 0;
 453         int i;
 454
 455         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 456         if (IS_ERR(new)) {
 457                 status = PTR_ERR(new);
 458                 mlog_errno(status);
 459                 goto bail;
 460         }
 461         osb->root_inode = new;
 462
 463         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 464         if (IS_ERR(new)) {
 465                 status = PTR_ERR(new);
 466                 mlog_errno(status);
 467                 goto bail;
 468         }
 469         osb->sys_root_inode = new;
 470
 471         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 472              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 473                 if (!ocfs2_need_system_inode(osb, i))
 474                         continue;
 475                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 476                 if (!new) {
 477                         ocfs2_release_system_inodes(osb);
 478                         status = -EINVAL;
 479                         mlog_errno(status);
 480                         /* FIXME: Should ERROR_RO_FS */
 481                         mlog(ML_ERROR, "Unable to load system inode %d, "
 482                              "possibly corrupt fs?", i);
 483                         goto bail;
 484                 }
 485                 // the array now has one ref, so drop this one
 486                 iput(new);
 487         }
 488
 489 bail:
 490         if (status)
 491                 mlog_errno(status);
 492         return status;
 493 }
 494
 495 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 496 {
 497         struct inode *new = NULL;
 498         int status = 0;
 499         int i;
 500
 501         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 502              i < NUM_SYSTEM_INODES;
 503              i++) {
 504                 if (!ocfs2_need_system_inode(osb, i))
 505                         continue;
 506                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 507                 if (!new) {
 508                         ocfs2_release_system_inodes(osb);
 509                         status = -EINVAL;
 510                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 511                              status, i, osb->slot_num);
 512                         goto bail;
 513                 }
 514                 /* the array now has one ref, so drop this one */
 515                 iput(new);
 516         }
 517
 518 bail:
 519         if (status)
 520                 mlog_errno(status);
 521         return status;
 522 }
 523
 524 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 525 {
 526         int i;
 527         struct inode *inode;
 528
 529         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 530                 inode = osb->global_system_inodes[i];
 531                 if (inode) {
 532                         iput(inode);
 533                         osb->global_system_inodes[i] = NULL;
 534                 }
 535         }
 536
 537         inode = osb->sys_root_inode;
 538         if (inode) {
 539                 iput(inode);
 540                 osb->sys_root_inode = NULL;
 541         }
 542
 543         inode = osb->root_inode;
 544         if (inode) {
 545                 iput(inode);
 546                 osb->root_inode = NULL;
 547         }
 548
 549         if (!osb->local_system_inodes)
 550                 return;
 551
 552         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 553                 if (osb->local_system_inodes[i]) {
 554                         iput(osb->local_system_inodes[i]);
 555                         osb->local_system_inodes[i] = NULL;
 556                 }
 557         }
 558
 559         kfree(osb->local_system_inodes);
 560         osb->local_system_inodes = NULL;
 561 }
 562
 563 /* We're allocating fs objects, use GFP_NOFS */
 564 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 565 {
 566         struct ocfs2_inode_info *oi;
 567
 568         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 569         if (!oi)
 570                 return NULL;
 571
 572         oi->i_sync_tid = 0;
 573         oi->i_datasync_tid = 0;
 574         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 575
 576         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 577         return &oi->vfs_inode;
 578 }
 579
 580 static void ocfs2_i_callback(struct rcu_head *head)
 581 {
 582         struct inode *inode = container_of(head, struct inode, i_rcu);
 583         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 584 }
 585
 586 static void ocfs2_destroy_inode(struct inode *inode)
 587 {
 588         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 589 }
 590
 591 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 592                                                 unsigned int cbits)
 593 {
 594         unsigned int bytes = 1 << cbits;
 595         unsigned int trim = bytes;
 596         unsigned int bitshift = 32;
 597
 598         /*
 599          * i_size and all block offsets in ocfs2 are always 64 bits
 600          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 601          * 64 bit platforms, cluster size will be the limiting factor.
 602          */
 603
 604 #if BITS_PER_LONG == 32
 605 # if defined(CONFIG_LBDAF)
 606         BUILD_BUG_ON(sizeof(sector_t) != 8);
 607         /*
 608          * We might be limited by page cache size.
 609          */
 610         if (bytes > PAGE_CACHE_SIZE) {
 611                 bytes = PAGE_CACHE_SIZE;
 612                 trim = 1;
 613                 /*
 614                  * Shift by 31 here so that we don't get larger than
 615                  * MAX_LFS_FILESIZE
 616                  */
 617                 bitshift = 31;
 618         }
 619 # else
 620         /*
 621          * We are limited by the size of sector_t. Use block size, as
 622          * that's what we expose to the VFS.
 623          */
 624         bytes = 1 << bbits;
 625         trim = 1;
 626         bitshift = 31;
 627 # endif
 628 #endif
 629
 630         /*
 631          * Trim by a whole cluster when we can actually approach the
 632          * on-disk limits. Otherwise we can overflow i_clusters when
 633          * an extent start is at the max offset.
 634          */
 635         return (((unsigned long long)bytes) << bitshift) - trim;
 636 }
 637
 638 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 639 {
 640         int incompat_features;
 641         int ret = 0;
 642         struct mount_options parsed_options;
 643         struct ocfs2_super *osb = OCFS2_SB(sb);
 644         u32 tmp;
 645
 646         sync_filesystem(sb);
 647
 648         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 649             !ocfs2_check_set_options(sb, &parsed_options)) {
 650                 ret = -EINVAL;
 651                 goto out;
 652         }
 653
 654         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 655                 OCFS2_MOUNT_HB_NONE;
 656         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 657                 ret = -EINVAL;
 658                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 659                 goto out;
 660         }
 661
 662         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 663             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 664                 ret = -EINVAL;
 665                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 666                 goto out;
 667         }
 668
 669         /* Probably don't want this on remount; it might
 670          * mess with other nodes */
 671         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 672             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 673                 ret = -EINVAL;
 674                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 675                 goto out;
 676         }
 677
 678         /* We're going to/from readonly mode. */
 679         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 680                 /* Disable quota accounting before remounting RO */
 681                 if (*flags & MS_RDONLY) {
 682                         ret = ocfs2_susp_quotas(osb, 0);
 683                         if (ret < 0)
 684                                 goto out;
 685                 }
 686                 /* Lock here so the check of HARD_RO and the potential
 687                  * setting of SOFT_RO is atomic. */
 688                 spin_lock(&osb->osb_lock);
 689                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 690                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 691                         ret = -EROFS;
 692                         goto unlock_osb;
 693                 }
 694
 695                 if (*flags & MS_RDONLY) {
 696                         sb->s_flags |= MS_RDONLY;
 697                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 698                 } else {
 699                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 700                                 mlog(ML_ERROR, "Cannot remount RDWR "
 701                                      "filesystem due to previous errors.\n");
 702                                 ret = -EROFS;
 703                                 goto unlock_osb;
 704                         }
 705                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 706                         if (incompat_features) {
 707                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 708                                      "of unsupported optional features "
 709                                      "(%x).\n", incompat_features);
 710                                 ret = -EINVAL;
 711                                 goto unlock_osb;
 712                         }
 713                         sb->s_flags &= ~MS_RDONLY;
 714                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 715                 }
 716                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 717 unlock_osb:
 718                 spin_unlock(&osb->osb_lock);
 719                 /* Enable quota accounting after remounting RW */
 720                 if (!ret && !(*flags & MS_RDONLY)) {
 721                         if (sb_any_quota_suspended(sb))
 722                                 ret = ocfs2_susp_quotas(osb, 1);
 723                         else
 724                                 ret = ocfs2_enable_quotas(osb);
 725                         if (ret < 0) {
 726                                 /* Return back changes... */
 727                                 spin_lock(&osb->osb_lock);
 728                                 sb->s_flags |= MS_RDONLY;
 729                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 730                                 spin_unlock(&osb->osb_lock);
 731                                 goto out;
 732                         }
 733                 }
 734         }
 735
 736         if (!ret) {
 737                 /* Only save off the new mount options in case of a successful
 738                  * remount. */
 739                 osb->s_mount_opt = parsed_options.mount_opt;
 740                 osb->s_atime_quantum = parsed_options.atime_quantum;
 741                 osb->preferred_slot = parsed_options.slot;
 742                 if (parsed_options.commit_interval)
 743                         osb->osb_commit_interval = parsed_options.commit_interval;
 744
 745                 if (!ocfs2_is_hard_readonly(osb))
 746                         ocfs2_set_journal_params(osb);
 747
 748                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 749                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 750                                                         MS_POSIXACL : 0);
 751         }
 752 out:
 753         return ret;
 754 }
 755
 756 static int ocfs2_sb_probe(struct super_block *sb,
 757                           struct buffer_head **bh,
 758                           int *sector_size,
 759                           struct ocfs2_blockcheck_stats *stats)
 760 {
 761         int status, tmpstat;
 762         struct ocfs1_vol_disk_hdr *hdr;
 763         struct ocfs2_dinode *di;
 764         int blksize;
 765
 766         *bh = NULL;
 767
 768         /* may be > 512 */
 769         *sector_size = bdev_logical_block_size(sb->s_bdev);
 770         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 771                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 772                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 773                 status = -EINVAL;
 774                 goto bail;
 775         }
 776
 777         /* Can this really happen? */
 778         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 779                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 780
 781         /* check block zero for old format */
 782         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 783         if (status < 0) {
 784                 mlog_errno(status);
 785                 goto bail;
 786         }
 787         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 788         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 789                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 790                      hdr->major_version, hdr->minor_version);
 791                 status = -EINVAL;
 792         }
 793         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 794                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 795                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 796                      hdr->signature);
 797                 status = -EINVAL;
 798         }
 799         brelse(*bh);
 800         *bh = NULL;
 801         if (status < 0) {
 802                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 803                      "upgraded before mounting with ocfs v2\n");
 804                 goto bail;
 805         }
 806
 807         /*
 808          * Now check at magic offset for 512, 1024, 2048, 4096
 809          * blocksizes.  4096 is the maximum blocksize because it is
 810          * the minimum clustersize.
 811          */
 812         status = -EINVAL;
 813         for (blksize = *sector_size;
 814              blksize <= OCFS2_MAX_BLOCKSIZE;
 815              blksize <<= 1) {
 816                 tmpstat = ocfs2_get_sector(sb, bh,
 817                                            OCFS2_SUPER_BLOCK_BLKNO,
 818                                            blksize);
 819                 if (tmpstat < 0) {
 820                         status = tmpstat;
 821                         mlog_errno(status);
 822                         break;
 823                 }
 824                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 825                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 826                 spin_lock_init(&stats->b_lock);
 827                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 828                 if (tmpstat < 0) {
 829                         brelse(*bh);
 830                         *bh = NULL;
 831                 }
 832                 if (tmpstat != -EAGAIN) {
 833                         status = tmpstat;
 834                         break;
 835                 }
 836         }
 837
 838 bail:
 839         return status;
 840 }
 841
 842 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 843 {
 844         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 845
 846         if (osb->s_mount_opt & hb_enabled) {
 847                 if (ocfs2_mount_local(osb)) {
 848                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 849                              "mounted device.\n");
 850                         return -EINVAL;
 851                 }
 852                 if (ocfs2_userspace_stack(osb)) {
 853                         mlog(ML_ERROR, "Userspace stack expected, but "
 854                              "o2cb heartbeat arguments passed to mount\n");
 855                         return -EINVAL;
 856                 }
 857                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 858                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 859                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 860                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 861                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 862                         return -EINVAL;
 863                 }
 864         }
 865
 866         if (!(osb->s_mount_opt & hb_enabled)) {
 867                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 868                     !ocfs2_userspace_stack(osb)) {
 869                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 870                              "a read-write clustered device.\n");
 871                         return -EINVAL;
 872                 }
 873         }
 874
 875         return 0;
 876 }
 877
 878 /*
 879  * If we're using a userspace stack, mount should have passed
 880  * a name that matches the disk.  If not, mount should not
 881  * have passed a stack.
 882  */
 883 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 884                                         struct mount_options *mopt)
 885 {
 886         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 887                 mlog(ML_ERROR,
 888                      "cluster stack passed to mount, but this filesystem "
 889                      "does not support it\n");
 890                 return -EINVAL;
 891         }
 892
 893         if (ocfs2_userspace_stack(osb) &&
 894             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 895                     OCFS2_STACK_LABEL_LEN)) {
 896                 mlog(ML_ERROR,
 897                      "cluster stack passed to mount (\"%s\") does not "
 898                      "match the filesystem (\"%s\")\n",
 899                      mopt->cluster_stack,
 900                      osb->osb_cluster_stack);
 901                 return -EINVAL;
 902         }
 903
 904         return 0;
 905 }
 906
 907 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 908 {
 909         int type;
 910         struct super_block *sb = osb->sb;
 911         unsigned int feature[OCFS2_MAXQUOTAS] = {
 912                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 913                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 914         int status = 0;
 915
 916         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 917                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 918                         continue;
 919                 if (unsuspend)
 920                         status = dquot_resume(sb, type);
 921                 else {
 922                         struct ocfs2_mem_dqinfo *oinfo;
 923
 924                         /* Cancel periodic syncing before suspending */
 925                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 926                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 927                         status = dquot_suspend(sb, type);
 928                 }
 929                 if (status < 0)
 930                         break;
 931         }
 932         if (status < 0)
 933                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 934                      "remount (error = %d).\n", status);
 935         return status;
 936 }
 937
 938 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 939 {
 940         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 941         struct super_block *sb = osb->sb;
 942         unsigned int feature[OCFS2_MAXQUOTAS] = {
 943                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 944                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 945         unsigned int ino[OCFS2_MAXQUOTAS] = {
 946                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 947                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 948         int status;
 949         int type;
 950
 951         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 952         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 953                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 954                         continue;
 955                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 956                                                         osb->slot_num);
 957                 if (!inode[type]) {
 958                         status = -ENOENT;
 959                         goto out_quota_off;
 960                 }
 961                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 962                                       DQUOT_USAGE_ENABLED);
 963                 if (status < 0)
 964                         goto out_quota_off;
 965         }
 966
 967         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 968                 iput(inode[type]);
 969         return 0;
 970 out_quota_off:
 971         ocfs2_disable_quotas(osb);
 972         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 973                 iput(inode[type]);
 974         mlog_errno(status);
 975         return status;
 976 }
 977
 978 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 979 {
 980         int type;
 981         struct inode *inode;
 982         struct super_block *sb = osb->sb;
 983         struct ocfs2_mem_dqinfo *oinfo;
 984
 985         /* We mostly ignore errors in this function because there's not much
 986          * we can do when we see them */
 987         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 988                 if (!sb_has_quota_loaded(sb, type))
 989                         continue;
 990                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 991                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 992                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 993                 inode = igrab(sb->s_dquot.files[type]);
 994                 /* Turn off quotas. This will remove all dquot structures from
 995                  * memory and so they will be automatically synced to global
 996                  * quota files */
 997                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 998                                         DQUOT_LIMITS_ENABLED);
 999                 if (!inode)
1000                         continue;
1001                 iput(inode);
1002         }
1003 }
1004
1005 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1006 {
1007         struct dentry *root;
1008         int status, sector_size;
1009         struct mount_options parsed_options;
1010         struct inode *inode = NULL;
1011         struct ocfs2_super *osb = NULL;
1012         struct buffer_head *bh = NULL;
1013         char nodestr[12];
1014         struct ocfs2_blockcheck_stats stats;
1015
1016         trace_ocfs2_fill_super(sb, data, silent);
1017
1018         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1019                 status = -EINVAL;
1020                 goto read_super_error;
1021         }
1022
1023         /* probe for superblock */
1024         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1025         if (status < 0) {
1026                 mlog(ML_ERROR, "superblock probe failed!\n");
1027                 goto read_super_error;
1028         }
1029
1030         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1031         osb = OCFS2_SB(sb);
1032         if (status < 0) {
1033                 mlog_errno(status);
1034                 goto read_super_error;
1035         }
1036         brelse(bh);
1037         bh = NULL;
1038
1039         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1040                 status = -EINVAL;
1041                 goto read_super_error;
1042         }
1043         osb->s_mount_opt = parsed_options.mount_opt;
1044         osb->s_atime_quantum = parsed_options.atime_quantum;
1045         osb->preferred_slot = parsed_options.slot;
1046         osb->osb_commit_interval = parsed_options.commit_interval;
1047
1048         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1049         osb->osb_resv_level = parsed_options.resv_level;
1050         osb->osb_dir_resv_level = parsed_options.resv_level;
1051         if (parsed_options.dir_resv_level == -1)
1052                 osb->osb_dir_resv_level = parsed_options.resv_level;
1053         else
1054                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1055
1056         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1057         if (status)
1058                 goto read_super_error;
1059
1060         sb->s_magic = OCFS2_SUPER_MAGIC;
1061
1062         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1063                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1064
1065         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1066          * heartbeat=none */
1067         if (bdev_read_only(sb->s_bdev)) {
1068                 if (!(sb->s_flags & MS_RDONLY)) {
1069                         status = -EACCES;
1070                         mlog(ML_ERROR, "Readonly device detected but readonly "
1071                              "mount was not specified.\n");
1072                         goto read_super_error;
1073                 }
1074
1075                 /* You should not be able to start a local heartbeat
1076                  * on a readonly device. */
1077                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1078                         status = -EROFS;
1079                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1080                              "device.\n");
1081                         goto read_super_error;
1082                 }
1083
1084                 status = ocfs2_check_journals_nolocks(osb);
1085                 if (status < 0) {
1086                         if (status == -EROFS)
1087                                 mlog(ML_ERROR, "Recovery required on readonly "
1088                                      "file system, but write access is "
1089                                      "unavailable.\n");
1090                         else
1091                                 mlog_errno(status);
1092                         goto read_super_error;
1093                 }
1094
1095                 ocfs2_set_ro_flag(osb, 1);
1096
1097                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1098                        "Cluster services will not be used for this mount. "
1099                        "Recovery will be skipped.\n", osb->dev_str);
1100         }
1101
1102         if (!ocfs2_is_hard_readonly(osb)) {
1103                 if (sb->s_flags & MS_RDONLY)
1104                         ocfs2_set_ro_flag(osb, 0);
1105         }
1106
1107         status = ocfs2_verify_heartbeat(osb);
1108         if (status < 0) {
1109                 mlog_errno(status);
1110                 goto read_super_error;
1111         }
1112
1113         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1114                                                  ocfs2_debugfs_root);
1115         if (IS_ERR_OR_NULL(osb->osb_debug_root)) {
1116                 status = -EINVAL;
1117                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1118                 goto read_super_error;
1119         }
1120
1121         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1122                                             osb->osb_debug_root,
1123                                             osb,
1124                                             &ocfs2_osb_debug_fops);
1125         if (IS_ERR_OR_NULL(osb->osb_ctxt)) {
1126                 status = -EINVAL;
1127                 mlog_errno(status);
1128                 goto read_super_error;
1129         }
1130
1131         if (ocfs2_meta_ecc(osb)) {
1132                 status = ocfs2_blockcheck_stats_debugfs_install(
1133                                                 &osb->osb_ecc_stats,
1134                                                 osb->osb_debug_root);
1135                 if (status) {
1136                         mlog(ML_ERROR,
1137                              "Unable to create blockcheck statistics "
1138                              "files\n");
1139                         goto read_super_error;
1140                 }
1141         }
1142
1143         status = ocfs2_mount_volume(sb);
1144         if (status < 0)
1145                 goto read_super_error;
1146
1147         if (osb->root_inode)
1148                 inode = igrab(osb->root_inode);
1149
1150         if (!inode) {
1151                 status = -EIO;
1152                 mlog_errno(status);
1153                 goto read_super_error;
1154         }
1155
1156         root = d_make_root(inode);
1157         if (!root) {
1158                 status = -ENOMEM;
1159                 mlog_errno(status);
1160                 goto read_super_error;
1161         }
1162
1163         sb->s_root = root;
1164
1165         ocfs2_complete_mount_recovery(osb);
1166
1167         if (ocfs2_mount_local(osb))
1168                 snprintf(nodestr, sizeof(nodestr), "local");
1169         else
1170                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1171
1172         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1173                "with %s data mode.\n",
1174                osb->dev_str, nodestr, osb->slot_num,
1175                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1176                "ordered");
1177
1178         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1179         wake_up(&osb->osb_mount_event);
1180
1181         /* Now we can initialize quotas because we can afford to wait
1182          * for cluster locks recovery now. That also means that truncation
1183          * log recovery can happen but that waits for proper quota setup */
1184         if (!(sb->s_flags & MS_RDONLY)) {
1185                 status = ocfs2_enable_quotas(osb);
1186                 if (status < 0) {
1187                         /* We have to err-out specially here because
1188                          * s_root is already set */
1189                         mlog_errno(status);
1190                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1191                         wake_up(&osb->osb_mount_event);
1192                         return status;
1193                 }
1194         }
1195
1196         ocfs2_complete_quota_recovery(osb);
1197
1198         /* Now we wake up again for processes waiting for quotas */
1199         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1200         wake_up(&osb->osb_mount_event);
1201
1202         /* Start this when the mount is almost sure of being successful */
1203         ocfs2_orphan_scan_start(osb);
1204
1205         return status;
1206
1207 read_super_error:
1208         brelse(bh);
1209
1210         if (osb) {
1211                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1212                 wake_up(&osb->osb_mount_event);
1213                 ocfs2_dismount_volume(sb, 1);
1214         }
1215
1216         if (status)
1217                 mlog_errno(status);
1218         return status;
1219 }
1220
1221 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1222                         int flags,
1223                         const char *dev_name,
1224                         void *data)
1225 {
1226         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1227 }
1228
1229 static struct file_system_type ocfs2_fs_type = {
1230         .owner          = THIS_MODULE,
1231         .name           = "ocfs2",
1232         .mount          = ocfs2_mount,
1233         .kill_sb        = kill_block_super,
1234         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1235         .next           = NULL
1236 };
1237 MODULE_ALIAS_FS("ocfs2");
1238
1239 static int ocfs2_check_set_options(struct super_block *sb,
1240                                    struct mount_options *options)
1241 {
1242         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1243             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1244                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1245                 mlog(ML_ERROR, "User quotas were requested, but this "
1246                      "filesystem does not have the feature enabled.\n");
1247                 return 0;
1248         }
1249         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1250             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1251                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1252                 mlog(ML_ERROR, "Group quotas were requested, but this "
1253                      "filesystem does not have the feature enabled.\n");
1254                 return 0;
1255         }
1256         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1257             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1258                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1259                      "feature is not enabled\n");
1260                 return 0;
1261         }
1262         /* No ACL setting specified? Use XATTR feature... */
1263         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1264                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1265                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1266                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1267                 else
1268                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1269         }
1270         return 1;
1271 }
1272
1273 static int ocfs2_parse_options(struct super_block *sb,
1274                                char *options,
1275                                struct mount_options *mopt,
1276                                int is_remount)
1277 {
1278         int status, user_stack = 0;
1279         char *p;
1280         u32 tmp;
1281
1282         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1283
1284         mopt->commit_interval = 0;
1285         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1286         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1287         mopt->slot = OCFS2_INVALID_SLOT;
1288         mopt->localalloc_opt = -1;
1289         mopt->cluster_stack[0] = '\0';
1290         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1291         mopt->dir_resv_level = -1;
1292
1293         if (!options) {
1294                 status = 1;
1295                 goto bail;
1296         }
1297
1298         while ((p = strsep(&options, ",")) != NULL) {
1299                 int token, option;
1300                 substring_t args[MAX_OPT_ARGS];
1301
1302                 if (!*p)
1303                         continue;
1304
1305                 token = match_token(p, tokens, args);
1306                 switch (token) {
1307                 case Opt_hb_local:
1308                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1309                         break;
1310                 case Opt_hb_none:
1311                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1312                         break;
1313                 case Opt_hb_global:
1314                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1315                         break;
1316                 case Opt_barrier:
1317                         if (match_int(&args[0], &option)) {
1318                                 status = 0;
1319                                 goto bail;
1320                         }
1321                         if (option)
1322                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1323                         else
1324                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1325                         break;
1326                 case Opt_intr:
1327                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1328                         break;
1329                 case Opt_nointr:
1330                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1331                         break;
1332                 case Opt_err_panic:
1333                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1334                         break;
1335                 case Opt_err_ro:
1336                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1337                         break;
1338                 case Opt_data_ordered:
1339                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1340                         break;
1341                 case Opt_data_writeback:
1342                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1343                         break;
1344                 case Opt_user_xattr:
1345                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1346                         break;
1347                 case Opt_nouser_xattr:
1348                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1349                         break;
1350                 case Opt_atime_quantum:
1351                         if (match_int(&args[0], &option)) {
1352                                 status = 0;
1353                                 goto bail;
1354                         }
1355                         if (option >= 0)
1356                                 mopt->atime_quantum = option;
1357                         break;
1358                 case Opt_slot:
1359                         option = 0;
1360                         if (match_int(&args[0], &option)) {
1361                                 status = 0;
1362                                 goto bail;
1363                         }
1364                         if (option)
1365                                 mopt->slot = (s16)option;
1366                         break;
1367                 case Opt_commit:
1368                         option = 0;
1369                         if (match_int(&args[0], &option)) {
1370                                 status = 0;
1371                                 goto bail;
1372                         }
1373                         if (option < 0)
1374                                 return 0;
1375                         if (option == 0)
1376                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1377                         mopt->commit_interval = HZ * option;
1378                         break;
1379                 case Opt_localalloc:
1380                         option = 0;
1381                         if (match_int(&args[0], &option)) {
1382                                 status = 0;
1383                                 goto bail;
1384                         }
1385                         if (option >= 0)
1386                                 mopt->localalloc_opt = option;
1387                         break;
1388                 case Opt_localflocks:
1389                         /*
1390                          * Changing this during remount could race
1391                          * flock() requests, or "unbalance" existing
1392                          * ones (e.g., a lock is taken in one mode but
1393                          * dropped in the other). If users care enough
1394                          * to flip locking modes during remount, we
1395                          * could add a "local" flag to individual
1396                          * flock structures for proper tracking of
1397                          * state.
1398                          */
1399                         if (!is_remount)
1400                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1401                         break;
1402                 case Opt_stack:
1403                         /* Check both that the option we were passed
1404                          * is of the right length and that it is a proper
1405                          * string of the right length.
1406                          */
1407                         if (((args[0].to - args[0].from) !=
1408                              OCFS2_STACK_LABEL_LEN) ||
1409                             (strnlen(args[0].from,
1410                                      OCFS2_STACK_LABEL_LEN) !=
1411                              OCFS2_STACK_LABEL_LEN)) {
1412                                 mlog(ML_ERROR,
1413                                      "Invalid cluster_stack option\n");
1414                                 status = 0;
1415                                 goto bail;
1416                         }
1417                         memcpy(mopt->cluster_stack, args[0].from,
1418                                OCFS2_STACK_LABEL_LEN);
1419                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1420                         /*
1421                          * Open code the memcmp here as we don't have
1422                          * an osb to pass to
1423                          * ocfs2_userspace_stack().
1424                          */
1425                         if (memcmp(mopt->cluster_stack,
1426                                    OCFS2_CLASSIC_CLUSTER_STACK,
1427                                    OCFS2_STACK_LABEL_LEN))
1428                                 user_stack = 1;
1429                         break;
1430                 case Opt_inode64:
1431                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1432                         break;
1433                 case Opt_usrquota:
1434                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1435                         break;
1436                 case Opt_grpquota:
1437                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1438                         break;
1439                 case Opt_coherency_buffered:
1440                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1441                         break;
1442                 case Opt_coherency_full:
1443                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1444                         break;
1445                 case Opt_acl:
1446                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1447                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1448                         break;
1449                 case Opt_noacl:
1450                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1451                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1452                         break;
1453                 case Opt_resv_level:
1454                         if (is_remount)
1455                                 break;
1456                         if (match_int(&args[0], &option)) {
1457                                 status = 0;
1458                                 goto bail;
1459                         }
1460                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1461                             option < OCFS2_MAX_RESV_LEVEL)
1462                                 mopt->resv_level = option;
1463                         break;
1464                 case Opt_dir_resv_level:
1465                         if (is_remount)
1466                                 break;
1467                         if (match_int(&args[0], &option)) {
1468                                 status = 0;
1469                                 goto bail;
1470                         }
1471                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1472                             option < OCFS2_MAX_RESV_LEVEL)
1473                                 mopt->dir_resv_level = option;
1474                         break;
1475                 case Opt_journal_async_commit:
1476                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1477                         break;
1478                 default:
1479                         mlog(ML_ERROR,
1480                              "Unrecognized mount option \"%s\" "
1481                              "or missing value\n", p);
1482                         status = 0;
1483                         goto bail;
1484                 }
1485         }
1486
1487         if (user_stack == 0) {
1488                 /* Ensure only one heartbeat mode */
1489                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1490                                          OCFS2_MOUNT_HB_GLOBAL |
1491                                          OCFS2_MOUNT_HB_NONE);
1492                 if (hweight32(tmp) != 1) {
1493                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1494                         status = 0;
1495                         goto bail;
1496                 }
1497         }
1498
1499         status = 1;
1500
1501 bail:
1502         return status;
1503 }
1504
1505 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1506 {
1507         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1508         unsigned long opts = osb->s_mount_opt;
1509         unsigned int local_alloc_megs;
1510
1511         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1512                 seq_printf(s, ",_netdev");
1513                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1514                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1515                 else
1516                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1517         } else
1518                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1519
1520         if (opts & OCFS2_MOUNT_NOINTR)
1521                 seq_printf(s, ",nointr");
1522
1523         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1524                 seq_printf(s, ",data=writeback");
1525         else
1526                 seq_printf(s, ",data=ordered");
1527
1528         if (opts & OCFS2_MOUNT_BARRIER)
1529                 seq_printf(s, ",barrier=1");
1530
1531         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1532                 seq_printf(s, ",errors=panic");
1533         else
1534                 seq_printf(s, ",errors=remount-ro");
1535
1536         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1537                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1538
1539         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1540
1541         if (osb->osb_commit_interval)
1542                 seq_printf(s, ",commit=%u",
1543                            (unsigned) (osb->osb_commit_interval / HZ));
1544
1545         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1546         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1547                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1548
1549         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1550                 seq_printf(s, ",localflocks,");
1551
1552         if (osb->osb_cluster_stack[0])
1553                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1554                            osb->osb_cluster_stack);
1555         if (opts & OCFS2_MOUNT_USRQUOTA)
1556                 seq_printf(s, ",usrquota");
1557         if (opts & OCFS2_MOUNT_GRPQUOTA)
1558                 seq_printf(s, ",grpquota");
1559
1560         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1561                 seq_printf(s, ",coherency=buffered");
1562         else
1563                 seq_printf(s, ",coherency=full");
1564
1565         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1566                 seq_printf(s, ",nouser_xattr");
1567         else
1568                 seq_printf(s, ",user_xattr");
1569
1570         if (opts & OCFS2_MOUNT_INODE64)
1571                 seq_printf(s, ",inode64");
1572
1573         if (opts & OCFS2_MOUNT_POSIX_ACL)
1574                 seq_printf(s, ",acl");
1575         else
1576                 seq_printf(s, ",noacl");
1577
1578         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1579                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1580
1581         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1582                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1583
1584         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1585                 seq_printf(s, ",journal_async_commit");
1586
1587         return 0;
1588 }
1589
1590 static int __init ocfs2_init(void)
1591 {
1592         int status;
1593
1594         status = init_ocfs2_uptodate_cache();
1595         if (status < 0)
1596                 goto out1;
1597
1598         status = ocfs2_initialize_mem_caches();
1599         if (status < 0)
1600                 goto out2;
1601
1602         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1603         if (!ocfs2_wq) {
1604                 status = -ENOMEM;
1605                 goto out3;
1606         }
1607
1608         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1609         if (IS_ERR_OR_NULL(ocfs2_debugfs_root)) {
1610                 status = ocfs2_debugfs_root ?
1611                         PTR_ERR(ocfs2_debugfs_root) : -ENOMEM;
1612                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1613                 goto out4;
1614         }
1615
1616         ocfs2_set_locking_protocol();
1617
1618         status = register_quota_format(&ocfs2_quota_format);
1619         if (status < 0)
1620                 goto out4;
1621         status = register_filesystem(&ocfs2_fs_type);
1622         if (!status)
1623                 return 0;
1624
1625         unregister_quota_format(&ocfs2_quota_format);
1626 out4:
1627         destroy_workqueue(ocfs2_wq);
1628         debugfs_remove(ocfs2_debugfs_root);
1629 out3:
1630         ocfs2_free_mem_caches();
1631 out2:
1632         exit_ocfs2_uptodate_cache();
1633 out1:
1634         mlog_errno(status);
1635         return status;
1636 }
1637
1638 static void __exit ocfs2_exit(void)
1639 {
1640         if (ocfs2_wq) {
1641                 flush_workqueue(ocfs2_wq);
1642                 destroy_workqueue(ocfs2_wq);
1643         }
1644
1645         unregister_quota_format(&ocfs2_quota_format);
1646
1647         debugfs_remove(ocfs2_debugfs_root);
1648
1649         ocfs2_free_mem_caches();
1650
1651         unregister_filesystem(&ocfs2_fs_type);
1652
1653         exit_ocfs2_uptodate_cache();
1654 }
1655
1656 static void ocfs2_put_super(struct super_block *sb)
1657 {
1658         trace_ocfs2_put_super(sb);
1659
1660         ocfs2_sync_blockdev(sb);
1661         ocfs2_dismount_volume(sb, 0);
1662 }
1663
1664 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1665 {
1666         struct ocfs2_super *osb;
1667         u32 numbits, freebits;
1668         int status;
1669         struct ocfs2_dinode *bm_lock;
1670         struct buffer_head *bh = NULL;
1671         struct inode *inode = NULL;
1672
1673         trace_ocfs2_statfs(dentry->d_sb, buf);
1674
1675         osb = OCFS2_SB(dentry->d_sb);
1676
1677         inode = ocfs2_get_system_file_inode(osb,
1678                                             GLOBAL_BITMAP_SYSTEM_INODE,
1679                                             OCFS2_INVALID_SLOT);
1680         if (!inode) {
1681                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1682                 status = -EIO;
1683                 goto bail;
1684         }
1685
1686         status = ocfs2_inode_lock(inode, &bh, 0);
1687         if (status < 0) {
1688                 mlog_errno(status);
1689                 goto bail;
1690         }
1691
1692         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1693
1694         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1695         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1696
1697         buf->f_type = OCFS2_SUPER_MAGIC;
1698         buf->f_bsize = dentry->d_sb->s_blocksize;
1699         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1700         buf->f_blocks = ((sector_t) numbits) *
1701                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1702         buf->f_bfree = ((sector_t) freebits) *
1703                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1704         buf->f_bavail = buf->f_bfree;
1705         buf->f_files = numbits;
1706         buf->f_ffree = freebits;
1707         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1708                                 & 0xFFFFFFFFUL;
1709         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1710                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1711
1712         brelse(bh);
1713
1714         ocfs2_inode_unlock(inode, 0);
1715         status = 0;
1716 bail:
1717         if (inode)
1718                 iput(inode);
1719
1720         if (status)
1721                 mlog_errno(status);
1722
1723         return status;
1724 }
1725
1726 static void ocfs2_inode_init_once(void *data)
1727 {
1728         struct ocfs2_inode_info *oi = data;
1729
1730         oi->ip_flags = 0;
1731         oi->ip_open_count = 0;
1732         spin_lock_init(&oi->ip_lock);
1733         ocfs2_extent_map_init(&oi->vfs_inode);
1734         INIT_LIST_HEAD(&oi->ip_io_markers);
1735         oi->ip_dir_start_lookup = 0;
1736         mutex_init(&oi->ip_unaligned_aio);
1737         init_rwsem(&oi->ip_alloc_sem);
1738         init_rwsem(&oi->ip_xattr_sem);
1739         mutex_init(&oi->ip_io_mutex);
1740
1741         oi->ip_blkno = 0ULL;
1742         oi->ip_clusters = 0;
1743
1744         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1745
1746         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1747         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1748         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1749
1750         init_waitqueue_head(&oi->append_dio_wq);
1751
1752         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1753                                   &ocfs2_inode_caching_ops);
1754
1755         inode_init_once(&oi->vfs_inode);
1756 }
1757
1758 static int ocfs2_initialize_mem_caches(void)
1759 {
1760         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1761                                        sizeof(struct ocfs2_inode_info),
1762                                        0,
1763                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1764                                                 SLAB_MEM_SPREAD),
1765                                        ocfs2_inode_init_once);
1766         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1767                                         sizeof(struct ocfs2_dquot),
1768                                         0,
1769                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1770                                                 SLAB_MEM_SPREAD),
1771                                         NULL);
1772         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1773                                         sizeof(struct ocfs2_quota_chunk),
1774                                         0,
1775                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1776                                         NULL);
1777         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1778             !ocfs2_qf_chunk_cachep) {
1779                 if (ocfs2_inode_cachep)
1780                         kmem_cache_destroy(ocfs2_inode_cachep);
1781                 if (ocfs2_dquot_cachep)
1782                         kmem_cache_destroy(ocfs2_dquot_cachep);
1783                 if (ocfs2_qf_chunk_cachep)
1784                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1785                 return -ENOMEM;
1786         }
1787
1788         return 0;
1789 }
1790
1791 static void ocfs2_free_mem_caches(void)
1792 {
1793         /*
1794          * Make sure all delayed rcu free inodes are flushed before we
1795          * destroy cache.
1796          */
1797         rcu_barrier();
1798         if (ocfs2_inode_cachep)
1799                 kmem_cache_destroy(ocfs2_inode_cachep);
1800         ocfs2_inode_cachep = NULL;
1801
1802         if (ocfs2_dquot_cachep)
1803                 kmem_cache_destroy(ocfs2_dquot_cachep);
1804         ocfs2_dquot_cachep = NULL;
1805
1806         if (ocfs2_qf_chunk_cachep)
1807                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1808         ocfs2_qf_chunk_cachep = NULL;
1809 }
1810
1811 static int ocfs2_get_sector(struct super_block *sb,
1812                             struct buffer_head **bh,
1813                             int block,
1814                             int sect_size)
1815 {
1816         if (!sb_set_blocksize(sb, sect_size)) {
1817                 mlog(ML_ERROR, "unable to set blocksize\n");
1818                 return -EIO;
1819         }
1820
1821         *bh = sb_getblk(sb, block);
1822         if (!*bh) {
1823                 mlog_errno(-ENOMEM);
1824                 return -ENOMEM;
1825         }
1826         lock_buffer(*bh);
1827         if (!buffer_dirty(*bh))
1828                 clear_buffer_uptodate(*bh);
1829         unlock_buffer(*bh);
1830         ll_rw_block(READ, 1, bh);
1831         wait_on_buffer(*bh);
1832         if (!buffer_uptodate(*bh)) {
1833                 mlog_errno(-EIO);
1834                 brelse(*bh);
1835                 *bh = NULL;
1836                 return -EIO;
1837         }
1838
1839         return 0;
1840 }
1841
1842 static int ocfs2_mount_volume(struct super_block *sb)
1843 {
1844         int status = 0;
1845         int unlock_super = 0;
1846         struct ocfs2_super *osb = OCFS2_SB(sb);
1847
1848         if (ocfs2_is_hard_readonly(osb))
1849                 goto leave;
1850
1851         status = ocfs2_dlm_init(osb);
1852         if (status < 0) {
1853                 mlog_errno(status);
1854                 goto leave;
1855         }
1856
1857         status = ocfs2_super_lock(osb, 1);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto leave;
1861         }
1862         unlock_super = 1;
1863
1864         /* This will load up the node map and add ourselves to it. */
1865         status = ocfs2_find_slot(osb);
1866         if (status < 0) {
1867                 mlog_errno(status);
1868                 goto leave;
1869         }
1870
1871         /* load all node-local system inodes */
1872         status = ocfs2_init_local_system_inodes(osb);
1873         if (status < 0) {
1874                 mlog_errno(status);
1875                 goto leave;
1876         }
1877
1878         status = ocfs2_check_volume(osb);
1879         if (status < 0) {
1880                 mlog_errno(status);
1881                 goto leave;
1882         }
1883
1884         status = ocfs2_truncate_log_init(osb);
1885         if (status < 0)
1886                 mlog_errno(status);
1887
1888 leave:
1889         if (unlock_super)
1890                 ocfs2_super_unlock(osb, 1);
1891
1892         return status;
1893 }
1894
1895 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1896 {
1897         int tmp, hangup_needed = 0;
1898         struct ocfs2_super *osb = NULL;
1899         char nodestr[12];
1900
1901         trace_ocfs2_dismount_volume(sb);
1902
1903         BUG_ON(!sb);
1904         osb = OCFS2_SB(sb);
1905         BUG_ON(!osb);
1906
1907         debugfs_remove(osb->osb_ctxt);
1908
1909         /* Orphan scan should be stopped as early as possible */
1910         ocfs2_orphan_scan_stop(osb);
1911
1912         ocfs2_disable_quotas(osb);
1913
1914         /* All dquots should be freed by now */
1915         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1916         /* Wait for worker to be done with the work structure in osb */
1917         cancel_work_sync(&osb->dquot_drop_work);
1918
1919         ocfs2_shutdown_local_alloc(osb);
1920
1921         ocfs2_truncate_log_shutdown(osb);
1922
1923         /* This will disable recovery and flush any recovery work. */
1924         ocfs2_recovery_exit(osb);
1925
1926         ocfs2_journal_shutdown(osb);
1927
1928         ocfs2_sync_blockdev(sb);
1929
1930         ocfs2_purge_refcount_trees(osb);
1931
1932         /* No cluster connection means we've failed during mount, so skip
1933          * all the steps which depended on that to complete. */
1934         if (osb->cconn) {
1935                 tmp = ocfs2_super_lock(osb, 1);
1936                 if (tmp < 0) {
1937                         mlog_errno(tmp);
1938                         return;
1939                 }
1940         }
1941
1942         if (osb->slot_num != OCFS2_INVALID_SLOT)
1943                 ocfs2_put_slot(osb);
1944
1945         if (osb->cconn)
1946                 ocfs2_super_unlock(osb, 1);
1947
1948         ocfs2_release_system_inodes(osb);
1949
1950         /*
1951          * If we're dismounting due to mount error, mount.ocfs2 will clean
1952          * up heartbeat.  If we're a local mount, there is no heartbeat.
1953          * If we failed before we got a uuid_str yet, we can't stop
1954          * heartbeat.  Otherwise, do it.
1955          */
1956         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1957             !ocfs2_is_hard_readonly(osb))
1958                 hangup_needed = 1;
1959
1960         if (osb->cconn)
1961                 ocfs2_dlm_shutdown(osb, hangup_needed);
1962
1963         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1964         debugfs_remove(osb->osb_debug_root);
1965
1966         if (hangup_needed)
1967                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1968
1969         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1970
1971         if (ocfs2_mount_local(osb))
1972                 snprintf(nodestr, sizeof(nodestr), "local");
1973         else
1974                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1975
1976         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1977                osb->dev_str, nodestr);
1978
1979         ocfs2_delete_osb(osb);
1980         kfree(osb);
1981         sb->s_dev = 0;
1982         sb->s_fs_info = NULL;
1983 }
1984
1985 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1986                                 unsigned uuid_bytes)
1987 {
1988         int i, ret;
1989         char *ptr;
1990
1991         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1992
1993         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1994         if (osb->uuid_str == NULL)
1995                 return -ENOMEM;
1996
1997         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1998                 /* print with null */
1999                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2000                 if (ret != 2) /* drop super cleans up */
2001                         return -EINVAL;
2002                 /* then only advance past the last char */
2003                 ptr += 2;
2004         }
2005
2006         return 0;
2007 }
2008
2009 /* Make sure entire volume is addressable by our journal.  Requires
2010    osb_clusters_at_boot to be valid and for the journal to have been
2011    initialized by ocfs2_journal_init(). */
2012 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2013 {
2014         int status = 0;
2015         u64 max_block =
2016                 ocfs2_clusters_to_blocks(osb->sb,
2017                                          osb->osb_clusters_at_boot) - 1;
2018
2019         /* 32-bit block number is always OK. */
2020         if (max_block <= (u32)~0ULL)
2021                 goto out;
2022
2023         /* Volume is "huge", so see if our journal is new enough to
2024            support it. */
2025         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2026                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2027               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2028                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2029                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2030                      "Enable the 'block64' journal option with tunefs.ocfs2");
2031                 status = -EFBIG;
2032                 goto out;
2033         }
2034
2035  out:
2036         return status;
2037 }
2038
2039 static int ocfs2_initialize_super(struct super_block *sb,
2040                                   struct buffer_head *bh,
2041                                   int sector_size,
2042                                   struct ocfs2_blockcheck_stats *stats)
2043 {
2044         int status;
2045         int i, cbits, bbits;
2046         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2047         struct inode *inode = NULL;
2048         struct ocfs2_journal *journal;
2049         struct ocfs2_super *osb;
2050         u64 total_blocks;
2051
2052         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2053         if (!osb) {
2054                 status = -ENOMEM;
2055                 mlog_errno(status);
2056                 goto bail;
2057         }
2058
2059         sb->s_fs_info = osb;
2060         sb->s_op = &ocfs2_sops;
2061         sb->s_d_op = &ocfs2_dentry_ops;
2062         sb->s_export_op = &ocfs2_export_ops;
2063         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2064         sb->dq_op = &ocfs2_quota_operations;
2065         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2066         sb->s_xattr = ocfs2_xattr_handlers;
2067         sb->s_time_gran = 1;
2068         sb->s_flags |= MS_NOATIME;
2069         /* this is needed to support O_LARGEFILE */
2070         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2071         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2072         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2073         memcpy(sb->s_uuid, di->id2.i_super.s_uuid,
2074                sizeof(di->id2.i_super.s_uuid));
2075
2076         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2077
2078         for (i = 0; i < 3; i++)
2079                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2080         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2081
2082         osb->sb = sb;
2083         /* Save off for ocfs2_rw_direct */
2084         osb->s_sectsize_bits = blksize_bits(sector_size);
2085         BUG_ON(!osb->s_sectsize_bits);
2086
2087         spin_lock_init(&osb->dc_task_lock);
2088         init_waitqueue_head(&osb->dc_event);
2089         osb->dc_work_sequence = 0;
2090         osb->dc_wake_sequence = 0;
2091         INIT_LIST_HEAD(&osb->blocked_lock_list);
2092         osb->blocked_lock_count = 0;
2093         spin_lock_init(&osb->osb_lock);
2094         spin_lock_init(&osb->osb_xattr_lock);
2095         ocfs2_init_steal_slots(osb);
2096
2097         mutex_init(&osb->system_file_mutex);
2098
2099         atomic_set(&osb->alloc_stats.moves, 0);
2100         atomic_set(&osb->alloc_stats.local_data, 0);
2101         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2102         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2103         atomic_set(&osb->alloc_stats.bg_extends, 0);
2104
2105         /* Copy the blockcheck stats from the superblock probe */
2106         osb->osb_ecc_stats = *stats;
2107
2108         ocfs2_init_node_maps(osb);
2109
2110         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2111                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2112
2113         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2114         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2115                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2116                      osb->max_slots);
2117                 status = -EINVAL;
2118                 goto bail;
2119         }
2120
2121         ocfs2_orphan_scan_init(osb);
2122
2123         status = ocfs2_recovery_init(osb);
2124         if (status) {
2125                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2126                 mlog_errno(status);
2127                 goto bail;
2128         }
2129
2130         init_waitqueue_head(&osb->checkpoint_event);
2131
2132         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2133
2134         osb->slot_num = OCFS2_INVALID_SLOT;
2135
2136         osb->s_xattr_inline_size = le16_to_cpu(
2137                                         di->id2.i_super.s_xattr_inline_size);
2138
2139         osb->local_alloc_state = OCFS2_LA_UNUSED;
2140         osb->local_alloc_bh = NULL;
2141         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2142
2143         init_waitqueue_head(&osb->osb_mount_event);
2144
2145         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2146         if (status) {
2147                 mlog_errno(status);
2148                 goto bail;
2149         }
2150
2151         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2152         if (!osb->vol_label) {
2153                 mlog(ML_ERROR, "unable to alloc vol label\n");
2154                 status = -ENOMEM;
2155                 goto bail;
2156         }
2157
2158         osb->slot_recovery_generations =
2159                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2160                         GFP_KERNEL);
2161         if (!osb->slot_recovery_generations) {
2162                 status = -ENOMEM;
2163                 mlog_errno(status);
2164                 goto bail;
2165         }
2166
2167         init_waitqueue_head(&osb->osb_wipe_event);
2168         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2169                                         sizeof(*osb->osb_orphan_wipes),
2170                                         GFP_KERNEL);
2171         if (!osb->osb_orphan_wipes) {
2172                 status = -ENOMEM;
2173                 mlog_errno(status);
2174                 goto bail;
2175         }
2176
2177         osb->osb_rf_lock_tree = RB_ROOT;
2178
2179         osb->s_feature_compat =
2180                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2181         osb->s_feature_ro_compat =
2182                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2183         osb->s_feature_incompat =
2184                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2185
2186         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2187                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2188                      "optional features (%x).\n", i);
2189                 status = -EINVAL;
2190                 goto bail;
2191         }
2192         if (!(osb->sb->s_flags & MS_RDONLY) &&
2193             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2194                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2195                      "unsupported optional features (%x).\n", i);
2196                 status = -EINVAL;
2197                 goto bail;
2198         }
2199
2200         if (ocfs2_clusterinfo_valid(osb)) {
2201                 osb->osb_stackflags =
2202                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2203                 strlcpy(osb->osb_cluster_stack,
2204                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2205                        OCFS2_STACK_LABEL_LEN + 1);
2206                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2207                         mlog(ML_ERROR,
2208                              "couldn't mount because of an invalid "
2209                              "cluster stack label (%s) \n",
2210                              osb->osb_cluster_stack);
2211                         status = -EINVAL;
2212                         goto bail;
2213                 }
2214                 strlcpy(osb->osb_cluster_name,
2215                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2216                         OCFS2_CLUSTER_NAME_LEN + 1);
2217         } else {
2218                 /* The empty string is identical with classic tools that
2219                  * don't know about s_cluster_info. */
2220                 osb->osb_cluster_stack[0] = '\0';
2221         }
2222
2223         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2224
2225         /* FIXME
2226          * This should be done in ocfs2_journal_init(), but unknown
2227          * ordering issues will cause the filesystem to crash.
2228          * If anyone wants to figure out what part of the code
2229          * refers to osb->journal before ocfs2_journal_init() is run,
2230          * be my guest.
2231          */
2232         /* initialize our journal structure */
2233
2234         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2235         if (!journal) {
2236                 mlog(ML_ERROR, "unable to alloc journal\n");
2237                 status = -ENOMEM;
2238                 goto bail;
2239         }
2240         osb->journal = journal;
2241         journal->j_osb = osb;
2242
2243         atomic_set(&journal->j_num_trans, 0);
2244         init_rwsem(&journal->j_trans_barrier);
2245         init_waitqueue_head(&journal->j_checkpointed);
2246         spin_lock_init(&journal->j_lock);
2247         journal->j_trans_id = (unsigned long) 1;
2248         INIT_LIST_HEAD(&journal->j_la_cleanups);
2249         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2250         journal->j_state = OCFS2_JOURNAL_FREE;
2251
2252         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2253         init_llist_head(&osb->dquot_drop_list);
2254
2255         /* get some pseudo constants for clustersize bits */
2256         osb->s_clustersize_bits =
2257                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2258         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2259
2260         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2261             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2262                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2263                      osb->s_clustersize);
2264                 status = -EINVAL;
2265                 goto bail;
2266         }
2267
2268         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2269                                                 le32_to_cpu(di->i_clusters));
2270
2271         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2272                                            total_blocks);
2273         if (status) {
2274                 mlog(ML_ERROR, "Volume too large "
2275                      "to mount safely on this system");
2276                 status = -EFBIG;
2277                 goto bail;
2278         }
2279
2280         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2281                                  sizeof(di->id2.i_super.s_uuid))) {
2282                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2283                 status = -ENOMEM;
2284                 goto bail;
2285         }
2286
2287         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2288                 OCFS2_MAX_VOL_LABEL_LEN);
2289         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2290         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2291         osb->first_cluster_group_blkno =
2292                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2293         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2294         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2295         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2296                                      (unsigned long long)osb->root_blkno,
2297                                      (unsigned long long)osb->system_dir_blkno,
2298                                      osb->s_clustersize_bits);
2299
2300         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2301         if (!osb->osb_dlm_debug) {
2302                 status = -ENOMEM;
2303                 mlog_errno(status);
2304                 goto bail;
2305         }
2306
2307         atomic_set(&osb->vol_state, VOLUME_INIT);
2308
2309         /* load root, system_dir, and all global system inodes */
2310         status = ocfs2_init_global_system_inodes(osb);
2311         if (status < 0) {
2312                 mlog_errno(status);
2313                 goto bail;
2314         }
2315
2316         /*
2317          * global bitmap
2318          */
2319         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2320                                             OCFS2_INVALID_SLOT);
2321         if (!inode) {
2322                 status = -EINVAL;
2323                 mlog_errno(status);
2324                 goto bail;
2325         }
2326
2327         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2328         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2329         iput(inode);
2330
2331         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2332                                  osb->s_feature_incompat) * 8;
2333
2334         status = ocfs2_init_slot_info(osb);
2335         if (status < 0) {
2336                 mlog_errno(status);
2337                 goto bail;
2338         }
2339         cleancache_init_shared_fs(sb);
2340
2341 bail:
2342         return status;
2343 }
2344
2345 /*
2346  * will return: -EAGAIN if it is ok to keep searching for superblocks
2347  *              -EINVAL if there is a bad superblock
2348  *              0 on success
2349  */
2350 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2351                                struct buffer_head *bh,
2352                                u32 blksz,
2353                                struct ocfs2_blockcheck_stats *stats)
2354 {
2355         int status = -EAGAIN;
2356
2357         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2358                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2359                 /* We have to do a raw check of the feature here */
2360                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2361                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2362                         status = ocfs2_block_check_validate(bh->b_data,
2363                                                             bh->b_size,
2364                                                             &di->i_check,
2365                                                             stats);
2366                         if (status)
2367                                 goto out;
2368                 }
2369                 status = -EINVAL;
2370                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2371                         mlog(ML_ERROR, "found superblock with incorrect block "
2372                              "size: found %u, should be %u\n",
2373                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2374                                blksz);
2375                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2376                            OCFS2_MAJOR_REV_LEVEL ||
2377                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2378                            OCFS2_MINOR_REV_LEVEL) {
2379                         mlog(ML_ERROR, "found superblock with bad version: "
2380                              "found %u.%u, should be %u.%u\n",
2381                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2382                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2383                              OCFS2_MAJOR_REV_LEVEL,
2384                              OCFS2_MINOR_REV_LEVEL);
2385                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2386                         mlog(ML_ERROR, "bad block number on superblock: "
2387                              "found %llu, should be %llu\n",
2388                              (unsigned long long)le64_to_cpu(di->i_blkno),
2389                              (unsigned long long)bh->b_blocknr);
2390                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2391                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2392                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2393                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2394                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2395                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2396                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2397                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2398                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2399                         mlog(ML_ERROR,
2400                              "Superblock slots found greater than file system "
2401                              "maximum: found %u, max %u\n",
2402                              le16_to_cpu(di->id2.i_super.s_max_slots),
2403                              OCFS2_MAX_SLOTS);
2404                 } else {
2405                         /* found it! */
2406                         status = 0;
2407                 }
2408         }
2409
2410 out:
2411         if (status && status != -EAGAIN)
2412                 mlog_errno(status);
2413         return status;
2414 }
2415
2416 static int ocfs2_check_volume(struct ocfs2_super *osb)
2417 {
2418         int status;
2419         int dirty;
2420         int local;
2421         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2422                                                   * recover
2423                                                   * ourselves. */
2424
2425         /* Init our journal object. */
2426         status = ocfs2_journal_init(osb->journal, &dirty);
2427         if (status < 0) {
2428                 mlog(ML_ERROR, "Could not initialize journal!\n");
2429                 goto finally;
2430         }
2431
2432         /* Now that journal has been initialized, check to make sure
2433            entire volume is addressable. */
2434         status = ocfs2_journal_addressable(osb);
2435         if (status)
2436                 goto finally;
2437
2438         /* If the journal was unmounted cleanly then we don't want to
2439          * recover anything. Otherwise, journal_load will do that
2440          * dirty work for us :) */
2441         if (!dirty) {
2442                 status = ocfs2_journal_wipe(osb->journal, 0);
2443                 if (status < 0) {
2444                         mlog_errno(status);
2445                         goto finally;
2446                 }
2447         } else {
2448                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2449                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2450         }
2451
2452         local = ocfs2_mount_local(osb);
2453
2454         /* will play back anything left in the journal. */
2455         status = ocfs2_journal_load(osb->journal, local, dirty);
2456         if (status < 0) {
2457                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2458                 goto finally;
2459         }
2460
2461         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2462                 jbd2_journal_set_features(osb->journal->j_journal,
2463                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2464                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2465         else
2466                 jbd2_journal_clear_features(osb->journal->j_journal,
2467                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2468                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2469
2470         if (dirty) {
2471                 /* recover my local alloc if we didn't unmount cleanly. */
2472                 status = ocfs2_begin_local_alloc_recovery(osb,
2473                                                           osb->slot_num,
2474                                                           &local_alloc);
2475                 if (status < 0) {
2476                         mlog_errno(status);
2477                         goto finally;
2478                 }
2479                 /* we complete the recovery process after we've marked
2480                  * ourselves as mounted. */
2481         }
2482
2483         status = ocfs2_load_local_alloc(osb);
2484         if (status < 0) {
2485                 mlog_errno(status);
2486                 goto finally;
2487         }
2488
2489         if (dirty) {
2490                 /* Recovery will be completed after we've mounted the
2491                  * rest of the volume. */
2492                 osb->dirty = 1;
2493                 osb->local_alloc_copy = local_alloc;
2494                 local_alloc = NULL;
2495         }
2496
2497         /* go through each journal, trylock it and if you get the
2498          * lock, and it's marked as dirty, set the bit in the recover
2499          * map and launch a recovery thread for it. */
2500         status = ocfs2_mark_dead_nodes(osb);
2501         if (status < 0) {
2502                 mlog_errno(status);
2503                 goto finally;
2504         }
2505
2506         status = ocfs2_compute_replay_slots(osb);
2507         if (status < 0)
2508                 mlog_errno(status);
2509
2510 finally:
2511         kfree(local_alloc);
2512
2513         if (status)
2514                 mlog_errno(status);
2515         return status;
2516 }
2517
2518 /*
2519  * The routine gets called from dismount or close whenever a dismount on
2520  * volume is requested and the osb open count becomes 1.
2521  * It will remove the osb from the global list and also free up all the
2522  * initialized resources and fileobject.
2523  */
2524 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2525 {
2526         /* This function assumes that the caller has the main osb resource */
2527
2528         ocfs2_free_slot_info(osb);
2529
2530         kfree(osb->osb_orphan_wipes);
2531         kfree(osb->slot_recovery_generations);
2532         /* FIXME
2533          * This belongs in journal shutdown, but because we have to
2534          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2535          * we free it here.
2536          */
2537         kfree(osb->journal);
2538         kfree(osb->local_alloc_copy);
2539         kfree(osb->uuid_str);
2540         kfree(osb->vol_label);
2541         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2542         memset(osb, 0, sizeof(struct ocfs2_super));
2543 }
2544
2545 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2546  * panic(). We do not support continue-on-error operation. */
2547 static void ocfs2_handle_error(struct super_block *sb)
2548 {
2549         struct ocfs2_super *osb = OCFS2_SB(sb);
2550
2551         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2552                 panic("OCFS2: (device %s): panic forced after error\n",
2553                       sb->s_id);
2554
2555         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2556
2557         if (sb->s_flags & MS_RDONLY &&
2558             (ocfs2_is_soft_readonly(osb) ||
2559              ocfs2_is_hard_readonly(osb)))
2560                 return;
2561
2562         printk(KERN_CRIT "File system is now read-only due to the potential "
2563                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2564                "system is unmounted.\n");
2565         sb->s_flags |= MS_RDONLY;
2566         ocfs2_set_ro_flag(osb, 0);
2567 }
2568
2569 void __ocfs2_error(struct super_block *sb, const char *function,
2570                   const char *fmt, ...)
2571 {
2572         struct va_format vaf;
2573         va_list args;
2574
2575         va_start(args, fmt);
2576         vaf.fmt = fmt;
2577         vaf.va = &args;
2578
2579         /* Not using mlog here because we want to show the actual
2580          * function the error came from. */
2581         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV\n",
2582                sb->s_id, function, &vaf);
2583
2584         va_end(args);
2585
2586         ocfs2_handle_error(sb);
2587 }
2588
2589 /* Handle critical errors. This is intentionally more drastic than
2590  * ocfs2_handle_error, so we only use for things like journal errors,
2591  * etc. */
2592 void __ocfs2_abort(struct super_block *sb, const char *function,
2593                    const char *fmt, ...)
2594 {
2595         struct va_format vaf;
2596         va_list args;
2597
2598         va_start(args, fmt);
2599
2600         vaf.fmt = fmt;
2601         vaf.va = &args;
2602
2603         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV\n",
2604                sb->s_id, function, &vaf);
2605
2606         va_end(args);
2607
2608         /* We don't have the cluster support yet to go straight to
2609          * hard readonly in here. Until then, we want to keep
2610          * ocfs2_abort() so that we can at least mark critical
2611          * errors.
2612          *
2613          * TODO: This should abort the journal and alert other nodes
2614          * that our slot needs recovery. */
2615
2616         /* Force a panic(). This stinks, but it's better than letting
2617          * things continue without having a proper hard readonly
2618          * here. */
2619         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2620                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2621         ocfs2_handle_error(sb);
2622 }
2623
2624 /*
2625  * Void signal blockers, because in-kernel sigprocmask() only fails
2626  * when SIG_* is wrong.
2627  */
2628 void ocfs2_block_signals(sigset_t *oldset)
2629 {
2630         int rc;
2631         sigset_t blocked;
2632
2633         sigfillset(&blocked);
2634         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2635         BUG_ON(rc);
2636 }
2637
2638 void ocfs2_unblock_signals(sigset_t *oldset)
2639 {
2640         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2641         BUG_ON(rc);
2642 }
2643
2644 module_init(ocfs2_init);
2645 module_exit(ocfs2_exit);