fs/ocfs2/super.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * super.c
   4  *
   5  * load/unload driver, mount/dismount volumes
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/fs.h>
  12 #include <linux/types.h>
  13 #include <linux/slab.h>
  14 #include <linux/highmem.h>
  15 #include <linux/init.h>
  16 #include <linux/random.h>
  17 #include <linux/statfs.h>
  18 #include <linux/moduleparam.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/socket.h>
  21 #include <linux/inet.h>
  22 #include <linux/parser.h>
  23 #include <linux/crc32.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/mount.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/signal.h>
  29
  30 #define CREATE_TRACE_POINTS
  31 #include "ocfs2_trace.h"
  32
  33 #include <cluster/masklog.h>
  34
  35 #include "ocfs2.h"
  36
  37 /* this should be the only file to include a version 1 header */
  38 #include "ocfs1_fs_compat.h"
  39
  40 #include "alloc.h"
  41 #include "aops.h"
  42 #include "blockcheck.h"
  43 #include "dlmglue.h"
  44 #include "export.h"
  45 #include "extent_map.h"
  46 #include "heartbeat.h"
  47 #include "inode.h"
  48 #include "journal.h"
  49 #include "localalloc.h"
  50 #include "namei.h"
  51 #include "slot_map.h"
  52 #include "super.h"
  53 #include "sysfile.h"
  54 #include "uptodate.h"
  55 #include "xattr.h"
  56 #include "quota.h"
  57 #include "refcounttree.h"
  58 #include "suballoc.h"
  59
  60 #include "buffer_head_io.h"
  61 #include "filecheck.h"
  62
  63 static struct kmem_cache *ocfs2_inode_cachep;
  64 struct kmem_cache *ocfs2_dquot_cachep;
  65 struct kmem_cache *ocfs2_qf_chunk_cachep;
  66
  67 static struct dentry *ocfs2_debugfs_root;
  68
  69 MODULE_AUTHOR("Oracle");
  70 MODULE_LICENSE("GPL");
  71 MODULE_DESCRIPTION("OCFS2 cluster file system");
  72
  73 struct mount_options
  74 {
  75         unsigned long   commit_interval;
  76         unsigned long   mount_opt;
  77         unsigned int    atime_quantum;
  78         unsigned short  slot;
  79         int             localalloc_opt;
  80         unsigned int    resv_level;
  81         int             dir_resv_level;
  82         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
  83 };
  84
  85 static int ocfs2_parse_options(struct super_block *sb, char *options,
  86                                struct mount_options *mopt,
  87                                int is_remount);
  88 static int ocfs2_check_set_options(struct super_block *sb,
  89                                    struct mount_options *options);
  90 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
  91 static void ocfs2_put_super(struct super_block *sb);
  92 static int ocfs2_mount_volume(struct super_block *sb);
  93 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
  94 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
  95 static int ocfs2_initialize_mem_caches(void);
  96 static void ocfs2_free_mem_caches(void);
  97 static void ocfs2_delete_osb(struct ocfs2_super *osb);
  98
  99 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 100
 101 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 102
 103 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 104 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 105 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 106 static int ocfs2_check_volume(struct ocfs2_super *osb);
 107 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 108                                struct buffer_head *bh,
 109                                u32 sectsize,
 110                                struct ocfs2_blockcheck_stats *stats);
 111 static int ocfs2_initialize_super(struct super_block *sb,
 112                                   struct buffer_head *bh,
 113                                   int sector_size,
 114                                   struct ocfs2_blockcheck_stats *stats);
 115 static int ocfs2_get_sector(struct super_block *sb,
 116                             struct buffer_head **bh,
 117                             int block,
 118                             int sect_size);
 119 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 120 static void ocfs2_free_inode(struct inode *inode);
 121 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 122 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 123 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 124
 125 static struct dquot __rcu **ocfs2_get_dquots(struct inode *inode)
 126 {
 127         return OCFS2_I(inode)->i_dquot;
 128 }
 129
 130 static const struct super_operations ocfs2_sops = {
 131         .statfs         = ocfs2_statfs,
 132         .alloc_inode    = ocfs2_alloc_inode,
 133         .free_inode     = ocfs2_free_inode,
 134         .drop_inode     = ocfs2_drop_inode,
 135         .evict_inode    = ocfs2_evict_inode,
 136         .sync_fs        = ocfs2_sync_fs,
 137         .put_super      = ocfs2_put_super,
 138         .remount_fs     = ocfs2_remount,
 139         .show_options   = ocfs2_show_options,
 140         .quota_read     = ocfs2_quota_read,
 141         .quota_write    = ocfs2_quota_write,
 142         .get_dquots     = ocfs2_get_dquots,
 143 };
 144
 145 enum {
 146         Opt_barrier,
 147         Opt_err_panic,
 148         Opt_err_ro,
 149         Opt_intr,
 150         Opt_nointr,
 151         Opt_hb_none,
 152         Opt_hb_local,
 153         Opt_hb_global,
 154         Opt_data_ordered,
 155         Opt_data_writeback,
 156         Opt_atime_quantum,
 157         Opt_slot,
 158         Opt_commit,
 159         Opt_localalloc,
 160         Opt_localflocks,
 161         Opt_stack,
 162         Opt_user_xattr,
 163         Opt_nouser_xattr,
 164         Opt_inode64,
 165         Opt_acl,
 166         Opt_noacl,
 167         Opt_usrquota,
 168         Opt_grpquota,
 169         Opt_coherency_buffered,
 170         Opt_coherency_full,
 171         Opt_resv_level,
 172         Opt_dir_resv_level,
 173         Opt_journal_async_commit,
 174         Opt_err_cont,
 175         Opt_err,
 176 };
 177
 178 static const match_table_t tokens = {
 179         {Opt_barrier, "barrier=%u"},
 180         {Opt_err_panic, "errors=panic"},
 181         {Opt_err_ro, "errors=remount-ro"},
 182         {Opt_intr, "intr"},
 183         {Opt_nointr, "nointr"},
 184         {Opt_hb_none, OCFS2_HB_NONE},
 185         {Opt_hb_local, OCFS2_HB_LOCAL},
 186         {Opt_hb_global, OCFS2_HB_GLOBAL},
 187         {Opt_data_ordered, "data=ordered"},
 188         {Opt_data_writeback, "data=writeback"},
 189         {Opt_atime_quantum, "atime_quantum=%u"},
 190         {Opt_slot, "preferred_slot=%u"},
 191         {Opt_commit, "commit=%u"},
 192         {Opt_localalloc, "localalloc=%d"},
 193         {Opt_localflocks, "localflocks"},
 194         {Opt_stack, "cluster_stack=%s"},
 195         {Opt_user_xattr, "user_xattr"},
 196         {Opt_nouser_xattr, "nouser_xattr"},
 197         {Opt_inode64, "inode64"},
 198         {Opt_acl, "acl"},
 199         {Opt_noacl, "noacl"},
 200         {Opt_usrquota, "usrquota"},
 201         {Opt_grpquota, "grpquota"},
 202         {Opt_coherency_buffered, "coherency=buffered"},
 203         {Opt_coherency_full, "coherency=full"},
 204         {Opt_resv_level, "resv_level=%u"},
 205         {Opt_dir_resv_level, "dir_resv_level=%u"},
 206         {Opt_journal_async_commit, "journal_async_commit"},
 207         {Opt_err_cont, "errors=continue"},
 208         {Opt_err, NULL}
 209 };
 210
 211 #ifdef CONFIG_DEBUG_FS
 212 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 213 {
 214         struct ocfs2_cluster_connection *cconn = osb->cconn;
 215         struct ocfs2_recovery_map *rm = osb->recovery_map;
 216         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 217         int i, out = 0;
 218         unsigned long flags;
 219
 220         out += scnprintf(buf + out, len - out,
 221                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 222                         "Device", osb->dev_str, osb->uuid_str,
 223                         osb->fs_generation, osb->vol_label);
 224
 225         out += scnprintf(buf + out, len - out,
 226                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 227                         atomic_read(&osb->vol_state), osb->osb_flags);
 228
 229         out += scnprintf(buf + out, len - out,
 230                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 231                         osb->sb->s_blocksize, osb->s_clustersize);
 232
 233         out += scnprintf(buf + out, len - out,
 234                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 235                         "ROcompat: 0x%X\n",
 236                         "Features", osb->s_feature_compat,
 237                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 238
 239         out += scnprintf(buf + out, len - out,
 240                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 241                         osb->s_mount_opt, osb->s_atime_quantum);
 242
 243         if (cconn) {
 244                 out += scnprintf(buf + out, len - out,
 245                                 "%10s => Stack: %s  Name: %*s  "
 246                                 "Version: %d.%d\n", "Cluster",
 247                                 (*osb->osb_cluster_stack == '\0' ?
 248                                  "o2cb" : osb->osb_cluster_stack),
 249                                 cconn->cc_namelen, cconn->cc_name,
 250                                 cconn->cc_version.pv_major,
 251                                 cconn->cc_version.pv_minor);
 252         }
 253
 254         spin_lock_irqsave(&osb->dc_task_lock, flags);
 255         out += scnprintf(buf + out, len - out,
 256                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 257                         "WorkSeq: %lu\n", "DownCnvt",
 258                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 259                         osb->blocked_lock_count, osb->dc_wake_sequence,
 260                         osb->dc_work_sequence);
 261         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 262
 263         spin_lock(&osb->osb_lock);
 264         out += scnprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 265                         "Recovery",
 266                         (osb->recovery_thread_task ?
 267                          task_pid_nr(osb->recovery_thread_task) : -1));
 268         if (rm->rm_used == 0)
 269                 out += scnprintf(buf + out, len - out, " None\n");
 270         else {
 271                 for (i = 0; i < rm->rm_used; i++)
 272                         out += scnprintf(buf + out, len - out, " %d",
 273                                         rm->rm_entries[i]);
 274                 out += scnprintf(buf + out, len - out, "\n");
 275         }
 276         spin_unlock(&osb->osb_lock);
 277
 278         out += scnprintf(buf + out, len - out,
 279                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 280                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 281                         osb->osb_commit_interval);
 282
 283         out += scnprintf(buf + out, len - out,
 284                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 285                         "Journal", osb->journal->j_state,
 286                         osb->journal->j_trans_id,
 287                         atomic_read(&osb->journal->j_num_trans));
 288
 289         out += scnprintf(buf + out, len - out,
 290                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 291                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 292                         "Stats",
 293                         atomic_read(&osb->alloc_stats.bitmap_data),
 294                         atomic_read(&osb->alloc_stats.local_data),
 295                         atomic_read(&osb->alloc_stats.bg_allocs),
 296                         atomic_read(&osb->alloc_stats.moves),
 297                         atomic_read(&osb->alloc_stats.bg_extends));
 298
 299         out += scnprintf(buf + out, len - out,
 300                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 301                         "Default: %u bits\n",
 302                         "LocalAlloc", osb->local_alloc_state,
 303                         (unsigned long long)osb->la_last_gd,
 304                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 305
 306         spin_lock(&osb->osb_lock);
 307         out += scnprintf(buf + out, len - out,
 308                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 309                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 310                         osb->s_inode_steal_slot,
 311                         atomic_read(&osb->s_num_inodes_stolen),
 312                         osb->s_meta_steal_slot,
 313                         atomic_read(&osb->s_num_meta_stolen));
 314         spin_unlock(&osb->osb_lock);
 315
 316         out += scnprintf(buf + out, len - out, "OrphanScan => ");
 317         out += scnprintf(buf + out, len - out, "Local: %u  Global: %u ",
 318                         os->os_count, os->os_seqno);
 319         out += scnprintf(buf + out, len - out, " Last Scan: ");
 320         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 321                 out += scnprintf(buf + out, len - out, "Disabled\n");
 322         else
 323                 out += scnprintf(buf + out, len - out, "%lu seconds ago\n",
 324                                 (unsigned long)(ktime_get_seconds() - os->os_scantime));
 325
 326         out += scnprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 327                         "Slots", "Num", "RecoGen");
 328         for (i = 0; i < osb->max_slots; ++i) {
 329                 out += scnprintf(buf + out, len - out,
 330                                 "%10s  %c %3d  %10d\n",
 331                                 " ",
 332                                 (i == osb->slot_num ? '*' : ' '),
 333                                 i, osb->slot_recovery_generations[i]);
 334         }
 335
 336         return out;
 337 }
 338
 339 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 340 {
 341         struct ocfs2_super *osb = inode->i_private;
 342         char *buf = NULL;
 343
 344         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 345         if (!buf)
 346                 goto bail;
 347
 348         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 349
 350         file->private_data = buf;
 351
 352         return 0;
 353 bail:
 354         return -ENOMEM;
 355 }
 356
 357 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 358 {
 359         kfree(file->private_data);
 360         return 0;
 361 }
 362
 363 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 364                                 size_t nbytes, loff_t *ppos)
 365 {
 366         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 367                                        i_size_read(file->f_mapping->host));
 368 }
 369 #else
 370 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 371 {
 372         return 0;
 373 }
 374 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 375 {
 376         return 0;
 377 }
 378 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 379                                 size_t nbytes, loff_t *ppos)
 380 {
 381         return 0;
 382 }
 383 #endif  /* CONFIG_DEBUG_FS */
 384
 385 static const struct file_operations ocfs2_osb_debug_fops = {
 386         .open =         ocfs2_osb_debug_open,
 387         .release =      ocfs2_debug_release,
 388         .read =         ocfs2_debug_read,
 389         .llseek =       generic_file_llseek,
 390 };
 391
 392 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 393 {
 394         int status;
 395         tid_t target;
 396         struct ocfs2_super *osb = OCFS2_SB(sb);
 397
 398         if (ocfs2_is_hard_readonly(osb))
 399                 return -EROFS;
 400
 401         if (wait) {
 402                 status = ocfs2_flush_truncate_log(osb);
 403                 if (status < 0)
 404                         mlog_errno(status);
 405         } else {
 406                 ocfs2_schedule_truncate_log_flush(osb, 0);
 407         }
 408
 409         if (jbd2_journal_start_commit(osb->journal->j_journal,
 410                                       &target)) {
 411                 if (wait)
 412                         jbd2_log_wait_commit(osb->journal->j_journal,
 413                                              target);
 414         }
 415         return 0;
 416 }
 417
 418 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 419 {
 420         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 421             && (ino == USER_QUOTA_SYSTEM_INODE
 422                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 423                 return 0;
 424         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 425             && (ino == GROUP_QUOTA_SYSTEM_INODE
 426                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 427                 return 0;
 428         return 1;
 429 }
 430
 431 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 432 {
 433         struct inode *new = NULL;
 434         int status = 0;
 435         int i;
 436
 437         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 438         if (IS_ERR(new)) {
 439                 status = PTR_ERR(new);
 440                 mlog_errno(status);
 441                 goto bail;
 442         }
 443         osb->root_inode = new;
 444
 445         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 446         if (IS_ERR(new)) {
 447                 status = PTR_ERR(new);
 448                 mlog_errno(status);
 449                 goto bail;
 450         }
 451         osb->sys_root_inode = new;
 452
 453         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 454              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 455                 if (!ocfs2_need_system_inode(osb, i))
 456                         continue;
 457                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 458                 if (!new) {
 459                         ocfs2_release_system_inodes(osb);
 460                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 461                         mlog_errno(status);
 462                         mlog(ML_ERROR, "Unable to load system inode %d, "
 463                              "possibly corrupt fs?", i);
 464                         goto bail;
 465                 }
 466                 // the array now has one ref, so drop this one
 467                 iput(new);
 468         }
 469
 470 bail:
 471         if (status)
 472                 mlog_errno(status);
 473         return status;
 474 }
 475
 476 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 477 {
 478         struct inode *new = NULL;
 479         int status = 0;
 480         int i;
 481
 482         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 483              i < NUM_SYSTEM_INODES;
 484              i++) {
 485                 if (!ocfs2_need_system_inode(osb, i))
 486                         continue;
 487                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 488                 if (!new) {
 489                         ocfs2_release_system_inodes(osb);
 490                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 491                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 492                              status, i, osb->slot_num);
 493                         goto bail;
 494                 }
 495                 /* the array now has one ref, so drop this one */
 496                 iput(new);
 497         }
 498
 499 bail:
 500         if (status)
 501                 mlog_errno(status);
 502         return status;
 503 }
 504
 505 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 506 {
 507         int i;
 508         struct inode *inode;
 509
 510         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 511                 inode = osb->global_system_inodes[i];
 512                 if (inode) {
 513                         iput(inode);
 514                         osb->global_system_inodes[i] = NULL;
 515                 }
 516         }
 517
 518         inode = osb->sys_root_inode;
 519         if (inode) {
 520                 iput(inode);
 521                 osb->sys_root_inode = NULL;
 522         }
 523
 524         inode = osb->root_inode;
 525         if (inode) {
 526                 iput(inode);
 527                 osb->root_inode = NULL;
 528         }
 529
 530         if (!osb->local_system_inodes)
 531                 return;
 532
 533         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 534                 if (osb->local_system_inodes[i]) {
 535                         iput(osb->local_system_inodes[i]);
 536                         osb->local_system_inodes[i] = NULL;
 537                 }
 538         }
 539
 540         kfree(osb->local_system_inodes);
 541         osb->local_system_inodes = NULL;
 542 }
 543
 544 /* We're allocating fs objects, use GFP_NOFS */
 545 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 546 {
 547         struct ocfs2_inode_info *oi;
 548
 549         oi = alloc_inode_sb(sb, ocfs2_inode_cachep, GFP_NOFS);
 550         if (!oi)
 551                 return NULL;
 552
 553         oi->i_sync_tid = 0;
 554         oi->i_datasync_tid = 0;
 555         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 556
 557         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 558         return &oi->vfs_inode;
 559 }
 560
 561 static void ocfs2_free_inode(struct inode *inode)
 562 {
 563         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 564 }
 565
 566 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 567                                                 unsigned int cbits)
 568 {
 569         unsigned int bytes = 1 << cbits;
 570         unsigned int trim = bytes;
 571         unsigned int bitshift = 32;
 572
 573         /*
 574          * i_size and all block offsets in ocfs2 are always 64 bits
 575          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 576          * 64 bit platforms, cluster size will be the limiting factor.
 577          */
 578
 579 #if BITS_PER_LONG == 32
 580         BUILD_BUG_ON(sizeof(sector_t) != 8);
 581         /*
 582          * We might be limited by page cache size.
 583          */
 584         if (bytes > PAGE_SIZE) {
 585                 bytes = PAGE_SIZE;
 586                 trim = 1;
 587                 /*
 588                  * Shift by 31 here so that we don't get larger than
 589                  * MAX_LFS_FILESIZE
 590                  */
 591                 bitshift = 31;
 592         }
 593 #endif
 594
 595         /*
 596          * Trim by a whole cluster when we can actually approach the
 597          * on-disk limits. Otherwise we can overflow i_clusters when
 598          * an extent start is at the max offset.
 599          */
 600         return (((unsigned long long)bytes) << bitshift) - trim;
 601 }
 602
 603 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 604 {
 605         int incompat_features;
 606         int ret = 0;
 607         struct mount_options parsed_options;
 608         struct ocfs2_super *osb = OCFS2_SB(sb);
 609         u32 tmp;
 610
 611         sync_filesystem(sb);
 612
 613         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 614             !ocfs2_check_set_options(sb, &parsed_options)) {
 615                 ret = -EINVAL;
 616                 goto out;
 617         }
 618
 619         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 620                 OCFS2_MOUNT_HB_NONE;
 621         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 622                 ret = -EINVAL;
 623                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 624                 goto out;
 625         }
 626
 627         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 628             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 629                 ret = -EINVAL;
 630                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 631                 goto out;
 632         }
 633
 634         /* Probably don't want this on remount; it might
 635          * mess with other nodes */
 636         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 637             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 638                 ret = -EINVAL;
 639                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 640                 goto out;
 641         }
 642
 643         /* We're going to/from readonly mode. */
 644         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
 645                 /* Disable quota accounting before remounting RO */
 646                 if (*flags & SB_RDONLY) {
 647                         ret = ocfs2_susp_quotas(osb, 0);
 648                         if (ret < 0)
 649                                 goto out;
 650                 }
 651                 /* Lock here so the check of HARD_RO and the potential
 652                  * setting of SOFT_RO is atomic. */
 653                 spin_lock(&osb->osb_lock);
 654                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 655                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 656                         ret = -EROFS;
 657                         goto unlock_osb;
 658                 }
 659
 660                 if (*flags & SB_RDONLY) {
 661                         sb->s_flags |= SB_RDONLY;
 662                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 663                 } else {
 664                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 665                                 mlog(ML_ERROR, "Cannot remount RDWR "
 666                                      "filesystem due to previous errors.\n");
 667                                 ret = -EROFS;
 668                                 goto unlock_osb;
 669                         }
 670                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 671                         if (incompat_features) {
 672                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 673                                      "of unsupported optional features "
 674                                      "(%x).\n", incompat_features);
 675                                 ret = -EINVAL;
 676                                 goto unlock_osb;
 677                         }
 678                         sb->s_flags &= ~SB_RDONLY;
 679                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 680                 }
 681                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 682 unlock_osb:
 683                 spin_unlock(&osb->osb_lock);
 684                 /* Enable quota accounting after remounting RW */
 685                 if (!ret && !(*flags & SB_RDONLY)) {
 686                         if (sb_any_quota_suspended(sb))
 687                                 ret = ocfs2_susp_quotas(osb, 1);
 688                         else
 689                                 ret = ocfs2_enable_quotas(osb);
 690                         if (ret < 0) {
 691                                 /* Return back changes... */
 692                                 spin_lock(&osb->osb_lock);
 693                                 sb->s_flags |= SB_RDONLY;
 694                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 695                                 spin_unlock(&osb->osb_lock);
 696                                 goto out;
 697                         }
 698                 }
 699         }
 700
 701         if (!ret) {
 702                 /* Only save off the new mount options in case of a successful
 703                  * remount. */
 704                 osb->s_mount_opt = parsed_options.mount_opt;
 705                 osb->s_atime_quantum = parsed_options.atime_quantum;
 706                 osb->preferred_slot = parsed_options.slot;
 707                 if (parsed_options.commit_interval)
 708                         osb->osb_commit_interval = parsed_options.commit_interval;
 709
 710                 if (!ocfs2_is_hard_readonly(osb))
 711                         ocfs2_set_journal_params(osb);
 712
 713                 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
 714                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 715                                                         SB_POSIXACL : 0);
 716         }
 717 out:
 718         return ret;
 719 }
 720
 721 static int ocfs2_sb_probe(struct super_block *sb,
 722                           struct buffer_head **bh,
 723                           int *sector_size,
 724                           struct ocfs2_blockcheck_stats *stats)
 725 {
 726         int status, tmpstat;
 727         struct ocfs1_vol_disk_hdr *hdr;
 728         struct ocfs2_dinode *di;
 729         int blksize;
 730
 731         *bh = NULL;
 732
 733         /* may be > 512 */
 734         *sector_size = bdev_logical_block_size(sb->s_bdev);
 735         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 736                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 737                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 738                 status = -EINVAL;
 739                 goto bail;
 740         }
 741
 742         /* Can this really happen? */
 743         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 744                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 745
 746         /* check block zero for old format */
 747         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 748         if (status < 0) {
 749                 mlog_errno(status);
 750                 goto bail;
 751         }
 752         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 753         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 754                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 755                      hdr->major_version, hdr->minor_version);
 756                 status = -EINVAL;
 757         }
 758         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 759                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 760                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 761                      hdr->signature);
 762                 status = -EINVAL;
 763         }
 764         brelse(*bh);
 765         *bh = NULL;
 766         if (status < 0) {
 767                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 768                      "upgraded before mounting with ocfs v2\n");
 769                 goto bail;
 770         }
 771
 772         /*
 773          * Now check at magic offset for 512, 1024, 2048, 4096
 774          * blocksizes.  4096 is the maximum blocksize because it is
 775          * the minimum clustersize.
 776          */
 777         status = -EINVAL;
 778         for (blksize = *sector_size;
 779              blksize <= OCFS2_MAX_BLOCKSIZE;
 780              blksize <<= 1) {
 781                 tmpstat = ocfs2_get_sector(sb, bh,
 782                                            OCFS2_SUPER_BLOCK_BLKNO,
 783                                            blksize);
 784                 if (tmpstat < 0) {
 785                         status = tmpstat;
 786                         mlog_errno(status);
 787                         break;
 788                 }
 789                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 790                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 791                 spin_lock_init(&stats->b_lock);
 792                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 793                 if (tmpstat < 0) {
 794                         brelse(*bh);
 795                         *bh = NULL;
 796                 }
 797                 if (tmpstat != -EAGAIN) {
 798                         status = tmpstat;
 799                         break;
 800                 }
 801         }
 802
 803 bail:
 804         return status;
 805 }
 806
 807 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 808 {
 809         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 810
 811         if (osb->s_mount_opt & hb_enabled) {
 812                 if (ocfs2_mount_local(osb)) {
 813                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 814                              "mounted device.\n");
 815                         return -EINVAL;
 816                 }
 817                 if (ocfs2_userspace_stack(osb)) {
 818                         mlog(ML_ERROR, "Userspace stack expected, but "
 819                              "o2cb heartbeat arguments passed to mount\n");
 820                         return -EINVAL;
 821                 }
 822                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 823                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 824                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 825                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 826                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 827                         return -EINVAL;
 828                 }
 829         }
 830
 831         if (!(osb->s_mount_opt & hb_enabled)) {
 832                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 833                     !ocfs2_userspace_stack(osb)) {
 834                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 835                              "a read-write clustered device.\n");
 836                         return -EINVAL;
 837                 }
 838         }
 839
 840         return 0;
 841 }
 842
 843 /*
 844  * If we're using a userspace stack, mount should have passed
 845  * a name that matches the disk.  If not, mount should not
 846  * have passed a stack.
 847  */
 848 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 849                                         struct mount_options *mopt)
 850 {
 851         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 852                 mlog(ML_ERROR,
 853                      "cluster stack passed to mount, but this filesystem "
 854                      "does not support it\n");
 855                 return -EINVAL;
 856         }
 857
 858         if (ocfs2_userspace_stack(osb) &&
 859             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 860                     OCFS2_STACK_LABEL_LEN)) {
 861                 mlog(ML_ERROR,
 862                      "cluster stack passed to mount (\"%s\") does not "
 863                      "match the filesystem (\"%s\")\n",
 864                      mopt->cluster_stack,
 865                      osb->osb_cluster_stack);
 866                 return -EINVAL;
 867         }
 868
 869         return 0;
 870 }
 871
 872 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 873 {
 874         int type;
 875         struct super_block *sb = osb->sb;
 876         unsigned int feature[OCFS2_MAXQUOTAS] = {
 877                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 878                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 879         int status = 0;
 880
 881         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 882                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 883                         continue;
 884                 if (unsuspend)
 885                         status = dquot_resume(sb, type);
 886                 else {
 887                         struct ocfs2_mem_dqinfo *oinfo;
 888
 889                         /* Cancel periodic syncing before suspending */
 890                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 891                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 892                         status = dquot_suspend(sb, type);
 893                 }
 894                 if (status < 0)
 895                         break;
 896         }
 897         if (status < 0)
 898                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 899                      "remount (error = %d).\n", status);
 900         return status;
 901 }
 902
 903 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 904 {
 905         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 906         struct super_block *sb = osb->sb;
 907         unsigned int feature[OCFS2_MAXQUOTAS] = {
 908                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 909                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 910         unsigned int ino[OCFS2_MAXQUOTAS] = {
 911                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 912                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 913         int status;
 914         int type;
 915
 916         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 917         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 918                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 919                         continue;
 920                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 921                                                         osb->slot_num);
 922                 if (!inode[type]) {
 923                         status = -ENOENT;
 924                         goto out_quota_off;
 925                 }
 926                 status = dquot_load_quota_inode(inode[type], type, QFMT_OCFS2,
 927                                                 DQUOT_USAGE_ENABLED);
 928                 if (status < 0)
 929                         goto out_quota_off;
 930         }
 931
 932         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 933                 iput(inode[type]);
 934         return 0;
 935 out_quota_off:
 936         ocfs2_disable_quotas(osb);
 937         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 938                 iput(inode[type]);
 939         mlog_errno(status);
 940         return status;
 941 }
 942
 943 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 944 {
 945         int type;
 946         struct inode *inode;
 947         struct super_block *sb = osb->sb;
 948         struct ocfs2_mem_dqinfo *oinfo;
 949
 950         /* We mostly ignore errors in this function because there's not much
 951          * we can do when we see them */
 952         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 953                 if (!sb_has_quota_loaded(sb, type))
 954                         continue;
 955                 if (!sb_has_quota_suspended(sb, type)) {
 956                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 957                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 958                 }
 959                 inode = igrab(sb->s_dquot.files[type]);
 960                 /* Turn off quotas. This will remove all dquot structures from
 961                  * memory and so they will be automatically synced to global
 962                  * quota files */
 963                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 964                                         DQUOT_LIMITS_ENABLED);
 965                 iput(inode);
 966         }
 967 }
 968
 969 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 970 {
 971         struct dentry *root;
 972         int status, sector_size;
 973         struct mount_options parsed_options;
 974         struct inode *inode = NULL;
 975         struct ocfs2_super *osb = NULL;
 976         struct buffer_head *bh = NULL;
 977         char nodestr[12];
 978         struct ocfs2_blockcheck_stats stats;
 979
 980         trace_ocfs2_fill_super(sb, data, silent);
 981
 982         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 983                 status = -EINVAL;
 984                 goto out;
 985         }
 986
 987         /* probe for superblock */
 988         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
 989         if (status < 0) {
 990                 mlog(ML_ERROR, "superblock probe failed!\n");
 991                 goto out;
 992         }
 993
 994         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
 995         brelse(bh);
 996         bh = NULL;
 997         if (status < 0)
 998                 goto out;
 999
1000         osb = OCFS2_SB(sb);
1001
1002         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1003                 status = -EINVAL;
1004                 goto out_super;
1005         }
1006         osb->s_mount_opt = parsed_options.mount_opt;
1007         osb->s_atime_quantum = parsed_options.atime_quantum;
1008         osb->preferred_slot = parsed_options.slot;
1009         osb->osb_commit_interval = parsed_options.commit_interval;
1010
1011         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1012         osb->osb_resv_level = parsed_options.resv_level;
1013         osb->osb_dir_resv_level = parsed_options.resv_level;
1014         if (parsed_options.dir_resv_level == -1)
1015                 osb->osb_dir_resv_level = parsed_options.resv_level;
1016         else
1017                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1018
1019         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1020         if (status)
1021                 goto out_super;
1022
1023         sb->s_magic = OCFS2_SUPER_MAGIC;
1024
1025         sb->s_flags = (sb->s_flags & ~(SB_POSIXACL | SB_NOSEC)) |
1026                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? SB_POSIXACL : 0);
1027
1028         /* Hard readonly mode only if: bdev_read_only, SB_RDONLY,
1029          * heartbeat=none */
1030         if (bdev_read_only(sb->s_bdev)) {
1031                 if (!sb_rdonly(sb)) {
1032                         status = -EACCES;
1033                         mlog(ML_ERROR, "Readonly device detected but readonly "
1034                              "mount was not specified.\n");
1035                         goto out_super;
1036                 }
1037
1038                 /* You should not be able to start a local heartbeat
1039                  * on a readonly device. */
1040                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1041                         status = -EROFS;
1042                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1043                              "device.\n");
1044                         goto out_super;
1045                 }
1046
1047                 status = ocfs2_check_journals_nolocks(osb);
1048                 if (status < 0) {
1049                         if (status == -EROFS)
1050                                 mlog(ML_ERROR, "Recovery required on readonly "
1051                                      "file system, but write access is "
1052                                      "unavailable.\n");
1053                         goto out_super;
1054                 }
1055
1056                 ocfs2_set_ro_flag(osb, 1);
1057
1058                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1059                        "Cluster services will not be used for this mount. "
1060                        "Recovery will be skipped.\n", osb->dev_str);
1061         }
1062
1063         if (!ocfs2_is_hard_readonly(osb)) {
1064                 if (sb_rdonly(sb))
1065                         ocfs2_set_ro_flag(osb, 0);
1066         }
1067
1068         status = ocfs2_verify_heartbeat(osb);
1069         if (status < 0)
1070                 goto out_super;
1071
1072         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1073                                                  ocfs2_debugfs_root);
1074
1075         debugfs_create_file("fs_state", S_IFREG|S_IRUSR, osb->osb_debug_root,
1076                             osb, &ocfs2_osb_debug_fops);
1077
1078         if (ocfs2_meta_ecc(osb))
1079                 ocfs2_blockcheck_stats_debugfs_install( &osb->osb_ecc_stats,
1080                                                         osb->osb_debug_root);
1081
1082         status = ocfs2_mount_volume(sb);
1083         if (status < 0)
1084                 goto out_debugfs;
1085
1086         if (osb->root_inode)
1087                 inode = igrab(osb->root_inode);
1088
1089         if (!inode) {
1090                 status = -EIO;
1091                 goto out_dismount;
1092         }
1093
1094         osb->osb_dev_kset = kset_create_and_add(sb->s_id, NULL,
1095                                                 &ocfs2_kset->kobj);
1096         if (!osb->osb_dev_kset) {
1097                 status = -ENOMEM;
1098                 mlog(ML_ERROR, "Unable to create device kset %s.\n", sb->s_id);
1099                 goto out_dismount;
1100         }
1101
1102         /* Create filecheck sysfs related directories/files at
1103          * /sys/fs/ocfs2/<devname>/filecheck */
1104         if (ocfs2_filecheck_create_sysfs(osb)) {
1105                 status = -ENOMEM;
1106                 mlog(ML_ERROR, "Unable to create filecheck sysfs directory at "
1107                         "/sys/fs/ocfs2/%s/filecheck.\n", sb->s_id);
1108                 goto out_dismount;
1109         }
1110
1111         root = d_make_root(inode);
1112         if (!root) {
1113                 status = -ENOMEM;
1114                 goto out_dismount;
1115         }
1116
1117         sb->s_root = root;
1118
1119         ocfs2_complete_mount_recovery(osb);
1120
1121         if (ocfs2_mount_local(osb))
1122                 snprintf(nodestr, sizeof(nodestr), "local");
1123         else
1124                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1125
1126         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1127                "with %s data mode.\n",
1128                osb->dev_str, nodestr, osb->slot_num,
1129                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1130                "ordered");
1131
1132         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1133         wake_up(&osb->osb_mount_event);
1134
1135         /* Now we can initialize quotas because we can afford to wait
1136          * for cluster locks recovery now. That also means that truncation
1137          * log recovery can happen but that waits for proper quota setup */
1138         if (!sb_rdonly(sb)) {
1139                 status = ocfs2_enable_quotas(osb);
1140                 if (status < 0) {
1141                         /* We have to err-out specially here because
1142                          * s_root is already set */
1143                         mlog_errno(status);
1144                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1145                         wake_up(&osb->osb_mount_event);
1146                         return status;
1147                 }
1148         }
1149
1150         ocfs2_complete_quota_recovery(osb);
1151
1152         /* Now we wake up again for processes waiting for quotas */
1153         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1154         wake_up(&osb->osb_mount_event);
1155
1156         /* Start this when the mount is almost sure of being successful */
1157         ocfs2_orphan_scan_start(osb);
1158
1159         return status;
1160
1161 out_dismount:
1162         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1163         wake_up(&osb->osb_mount_event);
1164         ocfs2_free_replay_slots(osb);
1165         ocfs2_dismount_volume(sb, 1);
1166         goto out;
1167
1168 out_debugfs:
1169         debugfs_remove_recursive(osb->osb_debug_root);
1170 out_super:
1171         ocfs2_release_system_inodes(osb);
1172         kfree(osb->recovery_map);
1173         ocfs2_delete_osb(osb);
1174         kfree(osb);
1175 out:
1176         mlog_errno(status);
1177
1178         return status;
1179 }
1180
1181 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1182                         int flags,
1183                         const char *dev_name,
1184                         void *data)
1185 {
1186         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1187 }
1188
1189 static struct file_system_type ocfs2_fs_type = {
1190         .owner          = THIS_MODULE,
1191         .name           = "ocfs2",
1192         .mount          = ocfs2_mount,
1193         .kill_sb        = kill_block_super,
1194         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1195         .next           = NULL
1196 };
1197 MODULE_ALIAS_FS("ocfs2");
1198
1199 static int ocfs2_check_set_options(struct super_block *sb,
1200                                    struct mount_options *options)
1201 {
1202         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1203             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1204                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1205                 mlog(ML_ERROR, "User quotas were requested, but this "
1206                      "filesystem does not have the feature enabled.\n");
1207                 return 0;
1208         }
1209         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1210             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1211                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1212                 mlog(ML_ERROR, "Group quotas were requested, but this "
1213                      "filesystem does not have the feature enabled.\n");
1214                 return 0;
1215         }
1216         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1217             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1218                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1219                      "feature is not enabled\n");
1220                 return 0;
1221         }
1222         /* No ACL setting specified? Use XATTR feature... */
1223         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1224                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1225                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1226                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1227                 else
1228                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1229         }
1230         return 1;
1231 }
1232
1233 static int ocfs2_parse_options(struct super_block *sb,
1234                                char *options,
1235                                struct mount_options *mopt,
1236                                int is_remount)
1237 {
1238         int status, user_stack = 0;
1239         char *p;
1240         u32 tmp;
1241         int token, option;
1242         substring_t args[MAX_OPT_ARGS];
1243
1244         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1245
1246         mopt->commit_interval = 0;
1247         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1248         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1249         mopt->slot = OCFS2_INVALID_SLOT;
1250         mopt->localalloc_opt = -1;
1251         mopt->cluster_stack[0] = '\0';
1252         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1253         mopt->dir_resv_level = -1;
1254
1255         if (!options) {
1256                 status = 1;
1257                 goto bail;
1258         }
1259
1260         while ((p = strsep(&options, ",")) != NULL) {
1261                 if (!*p)
1262                         continue;
1263
1264                 token = match_token(p, tokens, args);
1265                 switch (token) {
1266                 case Opt_hb_local:
1267                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1268                         break;
1269                 case Opt_hb_none:
1270                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1271                         break;
1272                 case Opt_hb_global:
1273                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1274                         break;
1275                 case Opt_barrier:
1276                         if (match_int(&args[0], &option)) {
1277                                 status = 0;
1278                                 goto bail;
1279                         }
1280                         if (option)
1281                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1282                         else
1283                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1284                         break;
1285                 case Opt_intr:
1286                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1287                         break;
1288                 case Opt_nointr:
1289                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1290                         break;
1291                 case Opt_err_panic:
1292                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1293                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1294                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1295                         break;
1296                 case Opt_err_ro:
1297                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1298                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1299                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1300                         break;
1301                 case Opt_err_cont:
1302                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1303                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1304                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1305                         break;
1306                 case Opt_data_ordered:
1307                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1308                         break;
1309                 case Opt_data_writeback:
1310                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1311                         break;
1312                 case Opt_user_xattr:
1313                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1314                         break;
1315                 case Opt_nouser_xattr:
1316                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1317                         break;
1318                 case Opt_atime_quantum:
1319                         if (match_int(&args[0], &option)) {
1320                                 status = 0;
1321                                 goto bail;
1322                         }
1323                         if (option >= 0)
1324                                 mopt->atime_quantum = option;
1325                         break;
1326                 case Opt_slot:
1327                         if (match_int(&args[0], &option)) {
1328                                 status = 0;
1329                                 goto bail;
1330                         }
1331                         if (option)
1332                                 mopt->slot = (u16)option;
1333                         break;
1334                 case Opt_commit:
1335                         if (match_int(&args[0], &option)) {
1336                                 status = 0;
1337                                 goto bail;
1338                         }
1339                         if (option < 0)
1340                                 return 0;
1341                         if (option == 0)
1342                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1343                         mopt->commit_interval = HZ * option;
1344                         break;
1345                 case Opt_localalloc:
1346                         if (match_int(&args[0], &option)) {
1347                                 status = 0;
1348                                 goto bail;
1349                         }
1350                         if (option >= 0)
1351                                 mopt->localalloc_opt = option;
1352                         break;
1353                 case Opt_localflocks:
1354                         /*
1355                          * Changing this during remount could race
1356                          * flock() requests, or "unbalance" existing
1357                          * ones (e.g., a lock is taken in one mode but
1358                          * dropped in the other). If users care enough
1359                          * to flip locking modes during remount, we
1360                          * could add a "local" flag to individual
1361                          * flock structures for proper tracking of
1362                          * state.
1363                          */
1364                         if (!is_remount)
1365                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1366                         break;
1367                 case Opt_stack:
1368                         /* Check both that the option we were passed
1369                          * is of the right length and that it is a proper
1370                          * string of the right length.
1371                          */
1372                         if (((args[0].to - args[0].from) !=
1373                              OCFS2_STACK_LABEL_LEN) ||
1374                             (strnlen(args[0].from,
1375                                      OCFS2_STACK_LABEL_LEN) !=
1376                              OCFS2_STACK_LABEL_LEN)) {
1377                                 mlog(ML_ERROR,
1378                                      "Invalid cluster_stack option\n");
1379                                 status = 0;
1380                                 goto bail;
1381                         }
1382                         memcpy(mopt->cluster_stack, args[0].from,
1383                                OCFS2_STACK_LABEL_LEN);
1384                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1385                         /*
1386                          * Open code the memcmp here as we don't have
1387                          * an osb to pass to
1388                          * ocfs2_userspace_stack().
1389                          */
1390                         if (memcmp(mopt->cluster_stack,
1391                                    OCFS2_CLASSIC_CLUSTER_STACK,
1392                                    OCFS2_STACK_LABEL_LEN))
1393                                 user_stack = 1;
1394                         break;
1395                 case Opt_inode64:
1396                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1397                         break;
1398                 case Opt_usrquota:
1399                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1400                         break;
1401                 case Opt_grpquota:
1402                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1403                         break;
1404                 case Opt_coherency_buffered:
1405                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1406                         break;
1407                 case Opt_coherency_full:
1408                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1409                         break;
1410                 case Opt_acl:
1411                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1412                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1413                         break;
1414                 case Opt_noacl:
1415                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1416                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1417                         break;
1418                 case Opt_resv_level:
1419                         if (is_remount)
1420                                 break;
1421                         if (match_int(&args[0], &option)) {
1422                                 status = 0;
1423                                 goto bail;
1424                         }
1425                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1426                             option < OCFS2_MAX_RESV_LEVEL)
1427                                 mopt->resv_level = option;
1428                         break;
1429                 case Opt_dir_resv_level:
1430                         if (is_remount)
1431                                 break;
1432                         if (match_int(&args[0], &option)) {
1433                                 status = 0;
1434                                 goto bail;
1435                         }
1436                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1437                             option < OCFS2_MAX_RESV_LEVEL)
1438                                 mopt->dir_resv_level = option;
1439                         break;
1440                 case Opt_journal_async_commit:
1441                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1442                         break;
1443                 default:
1444                         mlog(ML_ERROR,
1445                              "Unrecognized mount option \"%s\" "
1446                              "or missing value\n", p);
1447                         status = 0;
1448                         goto bail;
1449                 }
1450         }
1451
1452         if (user_stack == 0) {
1453                 /* Ensure only one heartbeat mode */
1454                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1455                                          OCFS2_MOUNT_HB_GLOBAL |
1456                                          OCFS2_MOUNT_HB_NONE);
1457                 if (hweight32(tmp) != 1) {
1458                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1459                         status = 0;
1460                         goto bail;
1461                 }
1462         }
1463
1464         status = 1;
1465
1466 bail:
1467         return status;
1468 }
1469
1470 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1471 {
1472         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1473         unsigned long opts = osb->s_mount_opt;
1474         unsigned int local_alloc_megs;
1475
1476         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1477                 seq_printf(s, ",_netdev");
1478                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1479                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1480                 else
1481                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1482         } else
1483                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1484
1485         if (opts & OCFS2_MOUNT_NOINTR)
1486                 seq_printf(s, ",nointr");
1487
1488         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1489                 seq_printf(s, ",data=writeback");
1490         else
1491                 seq_printf(s, ",data=ordered");
1492
1493         if (opts & OCFS2_MOUNT_BARRIER)
1494                 seq_printf(s, ",barrier=1");
1495
1496         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1497                 seq_printf(s, ",errors=panic");
1498         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1499                 seq_printf(s, ",errors=continue");
1500         else
1501                 seq_printf(s, ",errors=remount-ro");
1502
1503         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1504                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1505
1506         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1507
1508         if (osb->osb_commit_interval)
1509                 seq_printf(s, ",commit=%u",
1510                            (unsigned) (osb->osb_commit_interval / HZ));
1511
1512         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1513         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1514                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1515
1516         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1517                 seq_printf(s, ",localflocks,");
1518
1519         if (osb->osb_cluster_stack[0])
1520                 seq_show_option(s, "cluster_stack", osb->osb_cluster_stack);
1521         if (opts & OCFS2_MOUNT_USRQUOTA)
1522                 seq_printf(s, ",usrquota");
1523         if (opts & OCFS2_MOUNT_GRPQUOTA)
1524                 seq_printf(s, ",grpquota");
1525
1526         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1527                 seq_printf(s, ",coherency=buffered");
1528         else
1529                 seq_printf(s, ",coherency=full");
1530
1531         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1532                 seq_printf(s, ",nouser_xattr");
1533         else
1534                 seq_printf(s, ",user_xattr");
1535
1536         if (opts & OCFS2_MOUNT_INODE64)
1537                 seq_printf(s, ",inode64");
1538
1539         if (opts & OCFS2_MOUNT_POSIX_ACL)
1540                 seq_printf(s, ",acl");
1541         else
1542                 seq_printf(s, ",noacl");
1543
1544         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1545                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1546
1547         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1548                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1549
1550         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1551                 seq_printf(s, ",journal_async_commit");
1552
1553         return 0;
1554 }
1555
1556 static int __init ocfs2_init(void)
1557 {
1558         int status;
1559
1560         status = init_ocfs2_uptodate_cache();
1561         if (status < 0)
1562                 goto out1;
1563
1564         status = ocfs2_initialize_mem_caches();
1565         if (status < 0)
1566                 goto out2;
1567
1568         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1569
1570         ocfs2_set_locking_protocol();
1571
1572         status = register_quota_format(&ocfs2_quota_format);
1573         if (status < 0)
1574                 goto out3;
1575         status = register_filesystem(&ocfs2_fs_type);
1576         if (!status)
1577                 return 0;
1578
1579         unregister_quota_format(&ocfs2_quota_format);
1580 out3:
1581         debugfs_remove(ocfs2_debugfs_root);
1582         ocfs2_free_mem_caches();
1583 out2:
1584         exit_ocfs2_uptodate_cache();
1585 out1:
1586         mlog_errno(status);
1587         return status;
1588 }
1589
1590 static void __exit ocfs2_exit(void)
1591 {
1592         unregister_quota_format(&ocfs2_quota_format);
1593
1594         debugfs_remove(ocfs2_debugfs_root);
1595
1596         ocfs2_free_mem_caches();
1597
1598         unregister_filesystem(&ocfs2_fs_type);
1599
1600         exit_ocfs2_uptodate_cache();
1601 }
1602
1603 static void ocfs2_put_super(struct super_block *sb)
1604 {
1605         trace_ocfs2_put_super(sb);
1606
1607         ocfs2_sync_blockdev(sb);
1608         ocfs2_dismount_volume(sb, 0);
1609 }
1610
1611 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1612 {
1613         struct ocfs2_super *osb;
1614         u32 numbits, freebits;
1615         int status;
1616         struct ocfs2_dinode *bm_lock;
1617         struct buffer_head *bh = NULL;
1618         struct inode *inode = NULL;
1619
1620         trace_ocfs2_statfs(dentry->d_sb, buf);
1621
1622         osb = OCFS2_SB(dentry->d_sb);
1623
1624         inode = ocfs2_get_system_file_inode(osb,
1625                                             GLOBAL_BITMAP_SYSTEM_INODE,
1626                                             OCFS2_INVALID_SLOT);
1627         if (!inode) {
1628                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1629                 status = -EIO;
1630                 goto bail;
1631         }
1632
1633         status = ocfs2_inode_lock(inode, &bh, 0);
1634         if (status < 0) {
1635                 mlog_errno(status);
1636                 goto bail;
1637         }
1638
1639         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1640
1641         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1642         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1643
1644         buf->f_type = OCFS2_SUPER_MAGIC;
1645         buf->f_bsize = dentry->d_sb->s_blocksize;
1646         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1647         buf->f_blocks = ((sector_t) numbits) *
1648                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1649         buf->f_bfree = ((sector_t) freebits) *
1650                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1651         buf->f_bavail = buf->f_bfree;
1652         buf->f_files = numbits;
1653         buf->f_ffree = freebits;
1654         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1655                                 & 0xFFFFFFFFUL;
1656         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1657                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1658
1659         brelse(bh);
1660
1661         ocfs2_inode_unlock(inode, 0);
1662         status = 0;
1663 bail:
1664         iput(inode);
1665
1666         if (status)
1667                 mlog_errno(status);
1668
1669         return status;
1670 }
1671
1672 static void ocfs2_inode_init_once(void *data)
1673 {
1674         struct ocfs2_inode_info *oi = data;
1675
1676         oi->ip_flags = 0;
1677         oi->ip_open_count = 0;
1678         spin_lock_init(&oi->ip_lock);
1679         ocfs2_extent_map_init(&oi->vfs_inode);
1680         INIT_LIST_HEAD(&oi->ip_io_markers);
1681         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1682         oi->ip_dir_start_lookup = 0;
1683         init_rwsem(&oi->ip_alloc_sem);
1684         init_rwsem(&oi->ip_xattr_sem);
1685         mutex_init(&oi->ip_io_mutex);
1686
1687         oi->ip_blkno = 0ULL;
1688         oi->ip_clusters = 0;
1689         oi->ip_next_orphan = NULL;
1690
1691         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1692
1693         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1694         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1695         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1696
1697         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1698                                   &ocfs2_inode_caching_ops);
1699
1700         inode_init_once(&oi->vfs_inode);
1701 }
1702
1703 static int ocfs2_initialize_mem_caches(void)
1704 {
1705         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1706                                        sizeof(struct ocfs2_inode_info),
1707                                        0,
1708                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1709                                                 SLAB_ACCOUNT),
1710                                        ocfs2_inode_init_once);
1711         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1712                                         sizeof(struct ocfs2_dquot),
1713                                         0,
1714                                         SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1715                                         NULL);
1716         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1717                                         sizeof(struct ocfs2_quota_chunk),
1718                                         0,
1719                                         SLAB_RECLAIM_ACCOUNT,
1720                                         NULL);
1721         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1722             !ocfs2_qf_chunk_cachep) {
1723                 kmem_cache_destroy(ocfs2_inode_cachep);
1724                 kmem_cache_destroy(ocfs2_dquot_cachep);
1725                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1726                 return -ENOMEM;
1727         }
1728
1729         return 0;
1730 }
1731
1732 static void ocfs2_free_mem_caches(void)
1733 {
1734         /*
1735          * Make sure all delayed rcu free inodes are flushed before we
1736          * destroy cache.
1737          */
1738         rcu_barrier();
1739         kmem_cache_destroy(ocfs2_inode_cachep);
1740         ocfs2_inode_cachep = NULL;
1741
1742         kmem_cache_destroy(ocfs2_dquot_cachep);
1743         ocfs2_dquot_cachep = NULL;
1744
1745         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1746         ocfs2_qf_chunk_cachep = NULL;
1747 }
1748
1749 static int ocfs2_get_sector(struct super_block *sb,
1750                             struct buffer_head **bh,
1751                             int block,
1752                             int sect_size)
1753 {
1754         if (!sb_set_blocksize(sb, sect_size)) {
1755                 mlog(ML_ERROR, "unable to set blocksize\n");
1756                 return -EIO;
1757         }
1758
1759         *bh = sb_getblk(sb, block);
1760         if (!*bh) {
1761                 mlog_errno(-ENOMEM);
1762                 return -ENOMEM;
1763         }
1764         lock_buffer(*bh);
1765         if (!buffer_dirty(*bh))
1766                 clear_buffer_uptodate(*bh);
1767         unlock_buffer(*bh);
1768         if (bh_read(*bh, 0) < 0) {
1769                 mlog_errno(-EIO);
1770                 brelse(*bh);
1771                 *bh = NULL;
1772                 return -EIO;
1773         }
1774
1775         return 0;
1776 }
1777
1778 static int ocfs2_mount_volume(struct super_block *sb)
1779 {
1780         int status = 0;
1781         struct ocfs2_super *osb = OCFS2_SB(sb);
1782
1783         if (ocfs2_is_hard_readonly(osb))
1784                 goto out;
1785
1786         mutex_init(&osb->obs_trim_fs_mutex);
1787
1788         status = ocfs2_dlm_init(osb);
1789         if (status < 0) {
1790                 mlog_errno(status);
1791                 if (status == -EBADR && ocfs2_userspace_stack(osb))
1792                         mlog(ML_ERROR, "couldn't mount because cluster name on"
1793                         " disk does not match the running cluster name.\n");
1794                 goto out;
1795         }
1796
1797         status = ocfs2_super_lock(osb, 1);
1798         if (status < 0) {
1799                 mlog_errno(status);
1800                 goto out_dlm;
1801         }
1802
1803         /* This will load up the node map and add ourselves to it. */
1804         status = ocfs2_find_slot(osb);
1805         if (status < 0) {
1806                 mlog_errno(status);
1807                 goto out_super_lock;
1808         }
1809
1810         /* load all node-local system inodes */
1811         status = ocfs2_init_local_system_inodes(osb);
1812         if (status < 0) {
1813                 mlog_errno(status);
1814                 goto out_super_lock;
1815         }
1816
1817         status = ocfs2_check_volume(osb);
1818         if (status < 0) {
1819                 mlog_errno(status);
1820                 goto out_system_inodes;
1821         }
1822
1823         status = ocfs2_truncate_log_init(osb);
1824         if (status < 0) {
1825                 mlog_errno(status);
1826                 goto out_check_volume;
1827         }
1828
1829         ocfs2_super_unlock(osb, 1);
1830         return 0;
1831
1832 out_check_volume:
1833         ocfs2_free_replay_slots(osb);
1834 out_system_inodes:
1835         if (osb->local_alloc_state == OCFS2_LA_ENABLED)
1836                 ocfs2_shutdown_local_alloc(osb);
1837         ocfs2_release_system_inodes(osb);
1838         /* before journal shutdown, we should release slot_info */
1839         ocfs2_free_slot_info(osb);
1840         ocfs2_journal_shutdown(osb);
1841 out_super_lock:
1842         ocfs2_super_unlock(osb, 1);
1843 out_dlm:
1844         ocfs2_dlm_shutdown(osb, 0);
1845 out:
1846         return status;
1847 }
1848
1849 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1850 {
1851         int tmp, hangup_needed = 0;
1852         struct ocfs2_super *osb = NULL;
1853         char nodestr[12];
1854
1855         trace_ocfs2_dismount_volume(sb);
1856
1857         BUG_ON(!sb);
1858         osb = OCFS2_SB(sb);
1859         BUG_ON(!osb);
1860
1861         /* Remove file check sysfs related directores/files,
1862          * and wait for the pending file check operations */
1863         ocfs2_filecheck_remove_sysfs(osb);
1864
1865         kset_unregister(osb->osb_dev_kset);
1866
1867         /* Orphan scan should be stopped as early as possible */
1868         ocfs2_orphan_scan_stop(osb);
1869
1870         ocfs2_disable_quotas(osb);
1871
1872         /* All dquots should be freed by now */
1873         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1874         /* Wait for worker to be done with the work structure in osb */
1875         cancel_work_sync(&osb->dquot_drop_work);
1876
1877         ocfs2_shutdown_local_alloc(osb);
1878
1879         ocfs2_truncate_log_shutdown(osb);
1880
1881         /* This will disable recovery and flush any recovery work. */
1882         ocfs2_recovery_exit(osb);
1883
1884         ocfs2_sync_blockdev(sb);
1885
1886         ocfs2_purge_refcount_trees(osb);
1887
1888         /* No cluster connection means we've failed during mount, so skip
1889          * all the steps which depended on that to complete. */
1890         if (osb->cconn) {
1891                 tmp = ocfs2_super_lock(osb, 1);
1892                 if (tmp < 0) {
1893                         mlog_errno(tmp);
1894                         return;
1895                 }
1896         }
1897
1898         if (osb->slot_num != OCFS2_INVALID_SLOT)
1899                 ocfs2_put_slot(osb);
1900
1901         if (osb->cconn)
1902                 ocfs2_super_unlock(osb, 1);
1903
1904         ocfs2_release_system_inodes(osb);
1905
1906         ocfs2_journal_shutdown(osb);
1907
1908         /*
1909          * If we're dismounting due to mount error, mount.ocfs2 will clean
1910          * up heartbeat.  If we're a local mount, there is no heartbeat.
1911          * If we failed before we got a uuid_str yet, we can't stop
1912          * heartbeat.  Otherwise, do it.
1913          */
1914         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1915             !ocfs2_is_hard_readonly(osb))
1916                 hangup_needed = 1;
1917
1918         ocfs2_dlm_shutdown(osb, hangup_needed);
1919
1920         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1921         debugfs_remove_recursive(osb->osb_debug_root);
1922
1923         if (hangup_needed)
1924                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1925
1926         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1927
1928         if (ocfs2_mount_local(osb))
1929                 snprintf(nodestr, sizeof(nodestr), "local");
1930         else
1931                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1932
1933         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1934                osb->dev_str, nodestr);
1935
1936         ocfs2_delete_osb(osb);
1937         kfree(osb);
1938         sb->s_dev = 0;
1939         sb->s_fs_info = NULL;
1940 }
1941
1942 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1943                                 unsigned uuid_bytes)
1944 {
1945         int i, ret;
1946         char *ptr;
1947
1948         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1949
1950         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1951         if (osb->uuid_str == NULL)
1952                 return -ENOMEM;
1953
1954         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1955                 /* print with null */
1956                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1957                 if (ret != 2) /* drop super cleans up */
1958                         return -EINVAL;
1959                 /* then only advance past the last char */
1960                 ptr += 2;
1961         }
1962
1963         return 0;
1964 }
1965
1966 /* Make sure entire volume is addressable by our journal.  Requires
1967    osb_clusters_at_boot to be valid and for the journal to have been
1968    initialized by ocfs2_journal_init(). */
1969 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
1970 {
1971         int status = 0;
1972         u64 max_block =
1973                 ocfs2_clusters_to_blocks(osb->sb,
1974                                          osb->osb_clusters_at_boot) - 1;
1975
1976         /* 32-bit block number is always OK. */
1977         if (max_block <= (u32)~0ULL)
1978                 goto out;
1979
1980         /* Volume is "huge", so see if our journal is new enough to
1981            support it. */
1982         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
1983                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
1984               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
1985                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
1986                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
1987                      "Enable the 'block64' journal option with tunefs.ocfs2");
1988                 status = -EFBIG;
1989                 goto out;
1990         }
1991
1992  out:
1993         return status;
1994 }
1995
1996 static int ocfs2_initialize_super(struct super_block *sb,
1997                                   struct buffer_head *bh,
1998                                   int sector_size,
1999                                   struct ocfs2_blockcheck_stats *stats)
2000 {
2001         int status;
2002         int i, cbits, bbits;
2003         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2004         struct inode *inode = NULL;
2005         struct ocfs2_super *osb;
2006         u64 total_blocks;
2007
2008         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2009         if (!osb) {
2010                 status = -ENOMEM;
2011                 mlog_errno(status);
2012                 goto out;
2013         }
2014
2015         sb->s_fs_info = osb;
2016         sb->s_op = &ocfs2_sops;
2017         sb->s_d_op = &ocfs2_dentry_ops;
2018         sb->s_export_op = &ocfs2_export_ops;
2019         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2020         sb->dq_op = &ocfs2_quota_operations;
2021         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2022         sb->s_xattr = ocfs2_xattr_handlers;
2023         sb->s_time_gran = 1;
2024         sb->s_flags |= SB_NOATIME;
2025         /* this is needed to support O_LARGEFILE */
2026         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2027         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2028         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2029         super_set_uuid(sb, di->id2.i_super.s_uuid,
2030                        sizeof(di->id2.i_super.s_uuid));
2031
2032         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2033
2034         for (i = 0; i < 3; i++)
2035                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2036         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2037
2038         osb->sb = sb;
2039         osb->s_sectsize_bits = blksize_bits(sector_size);
2040         BUG_ON(!osb->s_sectsize_bits);
2041
2042         spin_lock_init(&osb->dc_task_lock);
2043         init_waitqueue_head(&osb->dc_event);
2044         osb->dc_work_sequence = 0;
2045         osb->dc_wake_sequence = 0;
2046         INIT_LIST_HEAD(&osb->blocked_lock_list);
2047         osb->blocked_lock_count = 0;
2048         spin_lock_init(&osb->osb_lock);
2049         spin_lock_init(&osb->osb_xattr_lock);
2050         ocfs2_init_steal_slots(osb);
2051
2052         mutex_init(&osb->system_file_mutex);
2053
2054         atomic_set(&osb->alloc_stats.moves, 0);
2055         atomic_set(&osb->alloc_stats.local_data, 0);
2056         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2057         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2058         atomic_set(&osb->alloc_stats.bg_extends, 0);
2059
2060         /* Copy the blockcheck stats from the superblock probe */
2061         osb->osb_ecc_stats = *stats;
2062
2063         ocfs2_init_node_maps(osb);
2064
2065         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2066                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2067
2068         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2069         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2070                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2071                      osb->max_slots);
2072                 status = -EINVAL;
2073                 goto out;
2074         }
2075
2076         ocfs2_orphan_scan_init(osb);
2077
2078         status = ocfs2_recovery_init(osb);
2079         if (status) {
2080                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2081                 mlog_errno(status);
2082                 goto out;
2083         }
2084
2085         init_waitqueue_head(&osb->checkpoint_event);
2086
2087         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2088
2089         osb->slot_num = OCFS2_INVALID_SLOT;
2090
2091         osb->s_xattr_inline_size = le16_to_cpu(
2092                                         di->id2.i_super.s_xattr_inline_size);
2093
2094         osb->local_alloc_state = OCFS2_LA_UNUSED;
2095         osb->local_alloc_bh = NULL;
2096         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2097
2098         init_waitqueue_head(&osb->osb_mount_event);
2099
2100         ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2101
2102         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2103         if (!osb->vol_label) {
2104                 mlog(ML_ERROR, "unable to alloc vol label\n");
2105                 status = -ENOMEM;
2106                 goto out_recovery_map;
2107         }
2108
2109         osb->slot_recovery_generations =
2110                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2111                         GFP_KERNEL);
2112         if (!osb->slot_recovery_generations) {
2113                 status = -ENOMEM;
2114                 mlog_errno(status);
2115                 goto out_vol_label;
2116         }
2117
2118         init_waitqueue_head(&osb->osb_wipe_event);
2119         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2120                                         sizeof(*osb->osb_orphan_wipes),
2121                                         GFP_KERNEL);
2122         if (!osb->osb_orphan_wipes) {
2123                 status = -ENOMEM;
2124                 mlog_errno(status);
2125                 goto out_slot_recovery_gen;
2126         }
2127
2128         osb->osb_rf_lock_tree = RB_ROOT;
2129
2130         osb->s_feature_compat =
2131                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2132         osb->s_feature_ro_compat =
2133                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2134         osb->s_feature_incompat =
2135                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2136
2137         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2138                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2139                      "optional features (%x).\n", i);
2140                 status = -EINVAL;
2141                 goto out_orphan_wipes;
2142         }
2143         if (!sb_rdonly(osb->sb) && (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2144                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2145                      "unsupported optional features (%x).\n", i);
2146                 status = -EINVAL;
2147                 goto out_orphan_wipes;
2148         }
2149
2150         if (ocfs2_clusterinfo_valid(osb)) {
2151                 /*
2152                  * ci_stack and ci_cluster in ocfs2_cluster_info may not be null
2153                  * terminated, so make sure no overflow happens here by using
2154                  * memcpy. Destination strings will always be null terminated
2155                  * because osb is allocated using kzalloc.
2156                  */
2157                 osb->osb_stackflags =
2158                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2159                 memcpy(osb->osb_cluster_stack,
2160                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2161                        OCFS2_STACK_LABEL_LEN);
2162                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2163                         mlog(ML_ERROR,
2164                              "couldn't mount because of an invalid "
2165                              "cluster stack label (%s) \n",
2166                              osb->osb_cluster_stack);
2167                         status = -EINVAL;
2168                         goto out_orphan_wipes;
2169                 }
2170                 memcpy(osb->osb_cluster_name,
2171                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2172                         OCFS2_CLUSTER_NAME_LEN);
2173         } else {
2174                 /* The empty string is identical with classic tools that
2175                  * don't know about s_cluster_info. */
2176                 osb->osb_cluster_stack[0] = '\0';
2177         }
2178
2179         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2180
2181         /*
2182          * FIXME
2183          * This should be done in ocfs2_journal_init(), but any inode
2184          * writes back operation will cause the filesystem to crash.
2185          */
2186         status = ocfs2_journal_alloc(osb);
2187         if (status < 0)
2188                 goto out_orphan_wipes;
2189
2190         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2191         init_llist_head(&osb->dquot_drop_list);
2192
2193         /* get some pseudo constants for clustersize bits */
2194         osb->s_clustersize_bits =
2195                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2196         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2197
2198         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2199             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2200                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2201                      osb->s_clustersize);
2202                 status = -EINVAL;
2203                 goto out_journal;
2204         }
2205
2206         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2207                                                 le32_to_cpu(di->i_clusters));
2208
2209         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2210                                            total_blocks);
2211         if (status) {
2212                 mlog(ML_ERROR, "Volume too large "
2213                      "to mount safely on this system");
2214                 status = -EFBIG;
2215                 goto out_journal;
2216         }
2217
2218         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2219                                  sizeof(di->id2.i_super.s_uuid))) {
2220                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2221                 status = -ENOMEM;
2222                 goto out_journal;
2223         }
2224
2225         strscpy(osb->vol_label, di->id2.i_super.s_label,
2226                 OCFS2_MAX_VOL_LABEL_LEN);
2227         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2228         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2229         osb->first_cluster_group_blkno =
2230                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2231         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2232         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2233         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2234                                      (unsigned long long)osb->root_blkno,
2235                                      (unsigned long long)osb->system_dir_blkno,
2236                                      osb->s_clustersize_bits);
2237
2238         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2239         if (!osb->osb_dlm_debug) {
2240                 status = -ENOMEM;
2241                 mlog_errno(status);
2242                 goto out_uuid_str;
2243         }
2244
2245         atomic_set(&osb->vol_state, VOLUME_INIT);
2246
2247         /* load root, system_dir, and all global system inodes */
2248         status = ocfs2_init_global_system_inodes(osb);
2249         if (status < 0) {
2250                 mlog_errno(status);
2251                 goto out_dlm_out;
2252         }
2253
2254         /*
2255          * global bitmap
2256          */
2257         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2258                                             OCFS2_INVALID_SLOT);
2259         if (!inode) {
2260                 status = -EINVAL;
2261                 mlog_errno(status);
2262                 goto out_system_inodes;
2263         }
2264
2265         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2266         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2267         iput(inode);
2268
2269         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2270                                  osb->s_feature_incompat) * 8;
2271
2272         status = ocfs2_init_slot_info(osb);
2273         if (status < 0) {
2274                 mlog_errno(status);
2275                 goto out_system_inodes;
2276         }
2277
2278         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2279         if (!osb->ocfs2_wq) {
2280                 status = -ENOMEM;
2281                 mlog_errno(status);
2282                 goto out_slot_info;
2283         }
2284
2285         return status;
2286
2287 out_slot_info:
2288         ocfs2_free_slot_info(osb);
2289 out_system_inodes:
2290         ocfs2_release_system_inodes(osb);
2291 out_dlm_out:
2292         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2293 out_uuid_str:
2294         kfree(osb->uuid_str);
2295 out_journal:
2296         kfree(osb->journal);
2297 out_orphan_wipes:
2298         kfree(osb->osb_orphan_wipes);
2299 out_slot_recovery_gen:
2300         kfree(osb->slot_recovery_generations);
2301 out_vol_label:
2302         kfree(osb->vol_label);
2303 out_recovery_map:
2304         kfree(osb->recovery_map);
2305 out:
2306         kfree(osb);
2307         sb->s_fs_info = NULL;
2308         return status;
2309 }
2310
2311 /*
2312  * will return: -EAGAIN if it is ok to keep searching for superblocks
2313  *              -EINVAL if there is a bad superblock
2314  *              0 on success
2315  */
2316 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2317                                struct buffer_head *bh,
2318                                u32 blksz,
2319                                struct ocfs2_blockcheck_stats *stats)
2320 {
2321         int status = -EAGAIN;
2322
2323         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2324                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2325                 /* We have to do a raw check of the feature here */
2326                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2327                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2328                         status = ocfs2_block_check_validate(bh->b_data,
2329                                                             bh->b_size,
2330                                                             &di->i_check,
2331                                                             stats);
2332                         if (status)
2333                                 goto out;
2334                 }
2335                 status = -EINVAL;
2336                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2337                         mlog(ML_ERROR, "found superblock with incorrect block "
2338                              "size: found %u, should be %u\n",
2339                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2340                                blksz);
2341                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2342                            OCFS2_MAJOR_REV_LEVEL ||
2343                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2344                            OCFS2_MINOR_REV_LEVEL) {
2345                         mlog(ML_ERROR, "found superblock with bad version: "
2346                              "found %u.%u, should be %u.%u\n",
2347                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2348                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2349                              OCFS2_MAJOR_REV_LEVEL,
2350                              OCFS2_MINOR_REV_LEVEL);
2351                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2352                         mlog(ML_ERROR, "bad block number on superblock: "
2353                              "found %llu, should be %llu\n",
2354                              (unsigned long long)le64_to_cpu(di->i_blkno),
2355                              (unsigned long long)bh->b_blocknr);
2356                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2357                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2358                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2359                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2360                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2361                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2362                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2363                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2364                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2365                         mlog(ML_ERROR,
2366                              "Superblock slots found greater than file system "
2367                              "maximum: found %u, max %u\n",
2368                              le16_to_cpu(di->id2.i_super.s_max_slots),
2369                              OCFS2_MAX_SLOTS);
2370                 } else {
2371                         /* found it! */
2372                         status = 0;
2373                 }
2374         }
2375
2376 out:
2377         if (status && status != -EAGAIN)
2378                 mlog_errno(status);
2379         return status;
2380 }
2381
2382 static int ocfs2_check_volume(struct ocfs2_super *osb)
2383 {
2384         int status;
2385         int dirty;
2386         int local;
2387         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2388                                                   * recover
2389                                                   * ourselves. */
2390
2391         /* Init our journal object. */
2392         status = ocfs2_journal_init(osb, &dirty);
2393         if (status < 0) {
2394                 mlog(ML_ERROR, "Could not initialize journal!\n");
2395                 goto finally;
2396         }
2397
2398         /* Now that journal has been initialized, check to make sure
2399            entire volume is addressable. */
2400         status = ocfs2_journal_addressable(osb);
2401         if (status)
2402                 goto finally;
2403
2404         /* If the journal was unmounted cleanly then we don't want to
2405          * recover anything. Otherwise, journal_load will do that
2406          * dirty work for us :) */
2407         if (!dirty) {
2408                 status = ocfs2_journal_wipe(osb->journal, 0);
2409                 if (status < 0) {
2410                         mlog_errno(status);
2411                         goto finally;
2412                 }
2413         } else {
2414                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2415                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2416         }
2417
2418         local = ocfs2_mount_local(osb);
2419
2420         /* will play back anything left in the journal. */
2421         status = ocfs2_journal_load(osb->journal, local, dirty);
2422         if (status < 0) {
2423                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2424                 goto finally;
2425         }
2426
2427         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2428                 jbd2_journal_set_features(osb->journal->j_journal,
2429                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2430                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2431         else
2432                 jbd2_journal_clear_features(osb->journal->j_journal,
2433                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2434                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2435
2436         if (dirty) {
2437                 /* recover my local alloc if we didn't unmount cleanly. */
2438                 status = ocfs2_begin_local_alloc_recovery(osb,
2439                                                           osb->slot_num,
2440                                                           &local_alloc);
2441                 if (status < 0) {
2442                         mlog_errno(status);
2443                         goto finally;
2444                 }
2445                 /* we complete the recovery process after we've marked
2446                  * ourselves as mounted. */
2447         }
2448
2449         status = ocfs2_load_local_alloc(osb);
2450         if (status < 0) {
2451                 mlog_errno(status);
2452                 goto finally;
2453         }
2454
2455         if (dirty) {
2456                 /* Recovery will be completed after we've mounted the
2457                  * rest of the volume. */
2458                 osb->local_alloc_copy = local_alloc;
2459                 local_alloc = NULL;
2460         }
2461
2462         /* go through each journal, trylock it and if you get the
2463          * lock, and it's marked as dirty, set the bit in the recover
2464          * map and launch a recovery thread for it. */
2465         status = ocfs2_mark_dead_nodes(osb);
2466         if (status < 0) {
2467                 mlog_errno(status);
2468                 goto finally;
2469         }
2470
2471         status = ocfs2_compute_replay_slots(osb);
2472         if (status < 0)
2473                 mlog_errno(status);
2474
2475 finally:
2476         kfree(local_alloc);
2477
2478         if (status)
2479                 mlog_errno(status);
2480         return status;
2481 }
2482
2483 /*
2484  * The routine gets called from dismount or close whenever a dismount on
2485  * volume is requested and the osb open count becomes 1.
2486  * It will remove the osb from the global list and also free up all the
2487  * initialized resources and fileobject.
2488  */
2489 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2490 {
2491         /* This function assumes that the caller has the main osb resource */
2492
2493         /* ocfs2_initializer_super have already created this workqueue */
2494         if (osb->ocfs2_wq)
2495                 destroy_workqueue(osb->ocfs2_wq);
2496
2497         ocfs2_free_slot_info(osb);
2498
2499         kfree(osb->osb_orphan_wipes);
2500         kfree(osb->slot_recovery_generations);
2501         /* FIXME
2502          * This belongs in journal shutdown, but because we have to
2503          * allocate osb->journal at the middle of ocfs2_initialize_super(),
2504          * we free it here.
2505          */
2506         kfree(osb->journal);
2507         kfree(osb->local_alloc_copy);
2508         kfree(osb->uuid_str);
2509         kfree(osb->vol_label);
2510         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2511         memset(osb, 0, sizeof(struct ocfs2_super));
2512 }
2513
2514 /* Depending on the mount option passed, perform one of the following:
2515  * Put OCFS2 into a readonly state (default)
2516  * Return EIO so that only the process errs
2517  * Fix the error as if fsck.ocfs2 -y
2518  * panic
2519  */
2520 static int ocfs2_handle_error(struct super_block *sb)
2521 {
2522         struct ocfs2_super *osb = OCFS2_SB(sb);
2523         int rv = 0;
2524
2525         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2526         pr_crit("On-disk corruption discovered. "
2527                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2528
2529         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2530                 panic("OCFS2: (device %s): panic forced after error\n",
2531                       sb->s_id);
2532         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2533                 pr_crit("OCFS2: Returning error to the calling process.\n");
2534                 rv = -EIO;
2535         } else { /* default option */
2536                 rv = -EROFS;
2537                 if (sb_rdonly(sb) && (ocfs2_is_soft_readonly(osb) || ocfs2_is_hard_readonly(osb)))
2538                         return rv;
2539
2540                 pr_crit("OCFS2: File system is now read-only.\n");
2541                 sb->s_flags |= SB_RDONLY;
2542                 ocfs2_set_ro_flag(osb, 0);
2543         }
2544
2545         return rv;
2546 }
2547
2548 int __ocfs2_error(struct super_block *sb, const char *function,
2549                   const char *fmt, ...)
2550 {
2551         struct va_format vaf;
2552         va_list args;
2553
2554         va_start(args, fmt);
2555         vaf.fmt = fmt;
2556         vaf.va = &args;
2557
2558         /* Not using mlog here because we want to show the actual
2559          * function the error came from. */
2560         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2561                sb->s_id, function, &vaf);
2562
2563         va_end(args);
2564
2565         return ocfs2_handle_error(sb);
2566 }
2567
2568 /* Handle critical errors. This is intentionally more drastic than
2569  * ocfs2_handle_error, so we only use for things like journal errors,
2570  * etc. */
2571 void __ocfs2_abort(struct super_block *sb, const char *function,
2572                    const char *fmt, ...)
2573 {
2574         struct va_format vaf;
2575         va_list args;
2576
2577         va_start(args, fmt);
2578
2579         vaf.fmt = fmt;
2580         vaf.va = &args;
2581
2582         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2583                sb->s_id, function, &vaf);
2584
2585         va_end(args);
2586
2587         /* We don't have the cluster support yet to go straight to
2588          * hard readonly in here. Until then, we want to keep
2589          * ocfs2_abort() so that we can at least mark critical
2590          * errors.
2591          *
2592          * TODO: This should abort the journal and alert other nodes
2593          * that our slot needs recovery. */
2594
2595         /* Force a panic(). This stinks, but it's better than letting
2596          * things continue without having a proper hard readonly
2597          * here. */
2598         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2599                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2600         ocfs2_handle_error(sb);
2601 }
2602
2603 /*
2604  * Void signal blockers, because in-kernel sigprocmask() only fails
2605  * when SIG_* is wrong.
2606  */
2607 void ocfs2_block_signals(sigset_t *oldset)
2608 {
2609         int rc;
2610         sigset_t blocked;
2611
2612         sigfillset(&blocked);
2613         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2614         BUG_ON(rc);
2615 }
2616
2617 void ocfs2_unblock_signals(sigset_t *oldset)
2618 {
2619         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2620         BUG_ON(rc);
2621 }
2622
2623 module_init(ocfs2_init);
2624 module_exit(ocfs2_exit);