fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static const struct super_operations ocfs2_sops = {
 147         .statfs         = ocfs2_statfs,
 148         .alloc_inode    = ocfs2_alloc_inode,
 149         .destroy_inode  = ocfs2_destroy_inode,
 150         .drop_inode     = ocfs2_drop_inode,
 151         .evict_inode    = ocfs2_evict_inode,
 152         .sync_fs        = ocfs2_sync_fs,
 153         .put_super      = ocfs2_put_super,
 154         .remount_fs     = ocfs2_remount,
 155         .show_options   = ocfs2_show_options,
 156         .quota_read     = ocfs2_quota_read,
 157         .quota_write    = ocfs2_quota_write,
 158 };
 159
 160 enum {
 161         Opt_barrier,
 162         Opt_err_panic,
 163         Opt_err_ro,
 164         Opt_intr,
 165         Opt_nointr,
 166         Opt_hb_none,
 167         Opt_hb_local,
 168         Opt_hb_global,
 169         Opt_data_ordered,
 170         Opt_data_writeback,
 171         Opt_atime_quantum,
 172         Opt_slot,
 173         Opt_commit,
 174         Opt_localalloc,
 175         Opt_localflocks,
 176         Opt_stack,
 177         Opt_user_xattr,
 178         Opt_nouser_xattr,
 179         Opt_inode64,
 180         Opt_acl,
 181         Opt_noacl,
 182         Opt_usrquota,
 183         Opt_grpquota,
 184         Opt_coherency_buffered,
 185         Opt_coherency_full,
 186         Opt_resv_level,
 187         Opt_dir_resv_level,
 188         Opt_err,
 189 };
 190
 191 static const match_table_t tokens = {
 192         {Opt_barrier, "barrier=%u"},
 193         {Opt_err_panic, "errors=panic"},
 194         {Opt_err_ro, "errors=remount-ro"},
 195         {Opt_intr, "intr"},
 196         {Opt_nointr, "nointr"},
 197         {Opt_hb_none, OCFS2_HB_NONE},
 198         {Opt_hb_local, OCFS2_HB_LOCAL},
 199         {Opt_hb_global, OCFS2_HB_GLOBAL},
 200         {Opt_data_ordered, "data=ordered"},
 201         {Opt_data_writeback, "data=writeback"},
 202         {Opt_atime_quantum, "atime_quantum=%u"},
 203         {Opt_slot, "preferred_slot=%u"},
 204         {Opt_commit, "commit=%u"},
 205         {Opt_localalloc, "localalloc=%d"},
 206         {Opt_localflocks, "localflocks"},
 207         {Opt_stack, "cluster_stack=%s"},
 208         {Opt_user_xattr, "user_xattr"},
 209         {Opt_nouser_xattr, "nouser_xattr"},
 210         {Opt_inode64, "inode64"},
 211         {Opt_acl, "acl"},
 212         {Opt_noacl, "noacl"},
 213         {Opt_usrquota, "usrquota"},
 214         {Opt_grpquota, "grpquota"},
 215         {Opt_coherency_buffered, "coherency=buffered"},
 216         {Opt_coherency_full, "coherency=full"},
 217         {Opt_resv_level, "resv_level=%u"},
 218         {Opt_dir_resv_level, "dir_resv_level=%u"},
 219         {Opt_err, NULL}
 220 };
 221
 222 #ifdef CONFIG_DEBUG_FS
 223 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 224 {
 225         struct ocfs2_cluster_connection *cconn = osb->cconn;
 226         struct ocfs2_recovery_map *rm = osb->recovery_map;
 227         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 228         int i, out = 0;
 229
 230         out += snprintf(buf + out, len - out,
 231                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 232                         "Device", osb->dev_str, osb->uuid_str,
 233                         osb->fs_generation, osb->vol_label);
 234
 235         out += snprintf(buf + out, len - out,
 236                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 237                         atomic_read(&osb->vol_state), osb->osb_flags);
 238
 239         out += snprintf(buf + out, len - out,
 240                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 241                         osb->sb->s_blocksize, osb->s_clustersize);
 242
 243         out += snprintf(buf + out, len - out,
 244                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 245                         "ROcompat: 0x%X\n",
 246                         "Features", osb->s_feature_compat,
 247                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 251                         osb->s_mount_opt, osb->s_atime_quantum);
 252
 253         if (cconn) {
 254                 out += snprintf(buf + out, len - out,
 255                                 "%10s => Stack: %s  Name: %*s  "
 256                                 "Version: %d.%d\n", "Cluster",
 257                                 (*osb->osb_cluster_stack == '\0' ?
 258                                  "o2cb" : osb->osb_cluster_stack),
 259                                 cconn->cc_namelen, cconn->cc_name,
 260                                 cconn->cc_version.pv_major,
 261                                 cconn->cc_version.pv_minor);
 262         }
 263
 264         spin_lock(&osb->dc_task_lock);
 265         out += snprintf(buf + out, len - out,
 266                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 267                         "WorkSeq: %lu\n", "DownCnvt",
 268                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 269                         osb->blocked_lock_count, osb->dc_wake_sequence,
 270                         osb->dc_work_sequence);
 271         spin_unlock(&osb->dc_task_lock);
 272
 273         spin_lock(&osb->osb_lock);
 274         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 275                         "Recovery",
 276                         (osb->recovery_thread_task ?
 277                          task_pid_nr(osb->recovery_thread_task) : -1));
 278         if (rm->rm_used == 0)
 279                 out += snprintf(buf + out, len - out, " None\n");
 280         else {
 281                 for (i = 0; i < rm->rm_used; i++)
 282                         out += snprintf(buf + out, len - out, " %d",
 283                                         rm->rm_entries[i]);
 284                 out += snprintf(buf + out, len - out, "\n");
 285         }
 286         spin_unlock(&osb->osb_lock);
 287
 288         out += snprintf(buf + out, len - out,
 289                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 290                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 291                         osb->osb_commit_interval);
 292
 293         out += snprintf(buf + out, len - out,
 294                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 295                         "Journal", osb->journal->j_state,
 296                         osb->journal->j_trans_id,
 297                         atomic_read(&osb->journal->j_num_trans));
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 301                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 302                         "Stats",
 303                         atomic_read(&osb->alloc_stats.bitmap_data),
 304                         atomic_read(&osb->alloc_stats.local_data),
 305                         atomic_read(&osb->alloc_stats.bg_allocs),
 306                         atomic_read(&osb->alloc_stats.moves),
 307                         atomic_read(&osb->alloc_stats.bg_extends));
 308
 309         out += snprintf(buf + out, len - out,
 310                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 311                         "Default: %u bits\n",
 312                         "LocalAlloc", osb->local_alloc_state,
 313                         (unsigned long long)osb->la_last_gd,
 314                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 315
 316         spin_lock(&osb->osb_lock);
 317         out += snprintf(buf + out, len - out,
 318                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 319                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 320                         osb->s_inode_steal_slot,
 321                         atomic_read(&osb->s_num_inodes_stolen),
 322                         osb->s_meta_steal_slot,
 323                         atomic_read(&osb->s_num_meta_stolen));
 324         spin_unlock(&osb->osb_lock);
 325
 326         out += snprintf(buf + out, len - out, "OrphanScan => ");
 327         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 328                         os->os_count, os->os_seqno);
 329         out += snprintf(buf + out, len - out, " Last Scan: ");
 330         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 331                 out += snprintf(buf + out, len - out, "Disabled\n");
 332         else
 333                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 334                                 (get_seconds() - os->os_scantime.tv_sec));
 335
 336         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 337                         "Slots", "Num", "RecoGen");
 338         for (i = 0; i < osb->max_slots; ++i) {
 339                 out += snprintf(buf + out, len - out,
 340                                 "%10s  %c %3d  %10d\n",
 341                                 " ",
 342                                 (i == osb->slot_num ? '*' : ' '),
 343                                 i, osb->slot_recovery_generations[i]);
 344         }
 345
 346         return out;
 347 }
 348
 349 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 350 {
 351         struct ocfs2_super *osb = inode->i_private;
 352         char *buf = NULL;
 353
 354         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 355         if (!buf)
 356                 goto bail;
 357
 358         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 359
 360         file->private_data = buf;
 361
 362         return 0;
 363 bail:
 364         return -ENOMEM;
 365 }
 366
 367 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 368 {
 369         kfree(file->private_data);
 370         return 0;
 371 }
 372
 373 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 374                                 size_t nbytes, loff_t *ppos)
 375 {
 376         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 377                                        i_size_read(file->f_mapping->host));
 378 }
 379 #else
 380 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 381 {
 382         return 0;
 383 }
 384 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 385 {
 386         return 0;
 387 }
 388 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 389                                 size_t nbytes, loff_t *ppos)
 390 {
 391         return 0;
 392 }
 393 #endif  /* CONFIG_DEBUG_FS */
 394
 395 static const struct file_operations ocfs2_osb_debug_fops = {
 396         .open =         ocfs2_osb_debug_open,
 397         .release =      ocfs2_debug_release,
 398         .read =         ocfs2_debug_read,
 399         .llseek =       generic_file_llseek,
 400 };
 401
 402 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 403 {
 404         int status;
 405         tid_t target;
 406         struct ocfs2_super *osb = OCFS2_SB(sb);
 407
 408         if (ocfs2_is_hard_readonly(osb))
 409                 return -EROFS;
 410
 411         if (wait) {
 412                 status = ocfs2_flush_truncate_log(osb);
 413                 if (status < 0)
 414                         mlog_errno(status);
 415         } else {
 416                 ocfs2_schedule_truncate_log_flush(osb, 0);
 417         }
 418
 419         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 420                                       &target)) {
 421                 if (wait)
 422                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 423                                              target);
 424         }
 425         return 0;
 426 }
 427
 428 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 429 {
 430         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 431             && (ino == USER_QUOTA_SYSTEM_INODE
 432                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 433                 return 0;
 434         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 435             && (ino == GROUP_QUOTA_SYSTEM_INODE
 436                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 437                 return 0;
 438         return 1;
 439 }
 440
 441 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 442 {
 443         struct inode *new = NULL;
 444         int status = 0;
 445         int i;
 446
 447         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 448         if (IS_ERR(new)) {
 449                 status = PTR_ERR(new);
 450                 mlog_errno(status);
 451                 goto bail;
 452         }
 453         osb->root_inode = new;
 454
 455         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 456         if (IS_ERR(new)) {
 457                 status = PTR_ERR(new);
 458                 mlog_errno(status);
 459                 goto bail;
 460         }
 461         osb->sys_root_inode = new;
 462
 463         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 464              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 465                 if (!ocfs2_need_system_inode(osb, i))
 466                         continue;
 467                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 468                 if (!new) {
 469                         ocfs2_release_system_inodes(osb);
 470                         status = -EINVAL;
 471                         mlog_errno(status);
 472                         /* FIXME: Should ERROR_RO_FS */
 473                         mlog(ML_ERROR, "Unable to load system inode %d, "
 474                              "possibly corrupt fs?", i);
 475                         goto bail;
 476                 }
 477                 // the array now has one ref, so drop this one
 478                 iput(new);
 479         }
 480
 481 bail:
 482         if (status)
 483                 mlog_errno(status);
 484         return status;
 485 }
 486
 487 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 488 {
 489         struct inode *new = NULL;
 490         int status = 0;
 491         int i;
 492
 493         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 494              i < NUM_SYSTEM_INODES;
 495              i++) {
 496                 if (!ocfs2_need_system_inode(osb, i))
 497                         continue;
 498                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 499                 if (!new) {
 500                         ocfs2_release_system_inodes(osb);
 501                         status = -EINVAL;
 502                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 503                              status, i, osb->slot_num);
 504                         goto bail;
 505                 }
 506                 /* the array now has one ref, so drop this one */
 507                 iput(new);
 508         }
 509
 510 bail:
 511         if (status)
 512                 mlog_errno(status);
 513         return status;
 514 }
 515
 516 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 517 {
 518         int i;
 519         struct inode *inode;
 520
 521         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 522                 inode = osb->global_system_inodes[i];
 523                 if (inode) {
 524                         iput(inode);
 525                         osb->global_system_inodes[i] = NULL;
 526                 }
 527         }
 528
 529         inode = osb->sys_root_inode;
 530         if (inode) {
 531                 iput(inode);
 532                 osb->sys_root_inode = NULL;
 533         }
 534
 535         inode = osb->root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->root_inode = NULL;
 539         }
 540
 541         if (!osb->local_system_inodes)
 542                 return;
 543
 544         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 545                 if (osb->local_system_inodes[i]) {
 546                         iput(osb->local_system_inodes[i]);
 547                         osb->local_system_inodes[i] = NULL;
 548                 }
 549         }
 550
 551         kfree(osb->local_system_inodes);
 552         osb->local_system_inodes = NULL;
 553 }
 554
 555 /* We're allocating fs objects, use GFP_NOFS */
 556 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 557 {
 558         struct ocfs2_inode_info *oi;
 559
 560         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 561         if (!oi)
 562                 return NULL;
 563
 564         oi->i_sync_tid = 0;
 565         oi->i_datasync_tid = 0;
 566
 567         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 568         return &oi->vfs_inode;
 569 }
 570
 571 static void ocfs2_i_callback(struct rcu_head *head)
 572 {
 573         struct inode *inode = container_of(head, struct inode, i_rcu);
 574         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 575 }
 576
 577 static void ocfs2_destroy_inode(struct inode *inode)
 578 {
 579         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 580 }
 581
 582 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 583                                                 unsigned int cbits)
 584 {
 585         unsigned int bytes = 1 << cbits;
 586         unsigned int trim = bytes;
 587         unsigned int bitshift = 32;
 588
 589         /*
 590          * i_size and all block offsets in ocfs2 are always 64 bits
 591          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 592          * 64 bit platforms, cluster size will be the limiting factor.
 593          */
 594
 595 #if BITS_PER_LONG == 32
 596 # if defined(CONFIG_LBDAF)
 597         BUILD_BUG_ON(sizeof(sector_t) != 8);
 598         /*
 599          * We might be limited by page cache size.
 600          */
 601         if (bytes > PAGE_CACHE_SIZE) {
 602                 bytes = PAGE_CACHE_SIZE;
 603                 trim = 1;
 604                 /*
 605                  * Shift by 31 here so that we don't get larger than
 606                  * MAX_LFS_FILESIZE
 607                  */
 608                 bitshift = 31;
 609         }
 610 # else
 611         /*
 612          * We are limited by the size of sector_t. Use block size, as
 613          * that's what we expose to the VFS.
 614          */
 615         bytes = 1 << bbits;
 616         trim = 1;
 617         bitshift = 31;
 618 # endif
 619 #endif
 620
 621         /*
 622          * Trim by a whole cluster when we can actually approach the
 623          * on-disk limits. Otherwise we can overflow i_clusters when
 624          * an extent start is at the max offset.
 625          */
 626         return (((unsigned long long)bytes) << bitshift) - trim;
 627 }
 628
 629 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 630 {
 631         int incompat_features;
 632         int ret = 0;
 633         struct mount_options parsed_options;
 634         struct ocfs2_super *osb = OCFS2_SB(sb);
 635         u32 tmp;
 636
 637         sync_filesystem(sb);
 638
 639         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 640             !ocfs2_check_set_options(sb, &parsed_options)) {
 641                 ret = -EINVAL;
 642                 goto out;
 643         }
 644
 645         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 646                 OCFS2_MOUNT_HB_NONE;
 647         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 648                 ret = -EINVAL;
 649                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 650                 goto out;
 651         }
 652
 653         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 654             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 657                 goto out;
 658         }
 659
 660         /* Probably don't want this on remount; it might
 661          * mess with other nodes */
 662         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 663             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 664                 ret = -EINVAL;
 665                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 666                 goto out;
 667         }
 668
 669         /* We're going to/from readonly mode. */
 670         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 671                 /* Disable quota accounting before remounting RO */
 672                 if (*flags & MS_RDONLY) {
 673                         ret = ocfs2_susp_quotas(osb, 0);
 674                         if (ret < 0)
 675                                 goto out;
 676                 }
 677                 /* Lock here so the check of HARD_RO and the potential
 678                  * setting of SOFT_RO is atomic. */
 679                 spin_lock(&osb->osb_lock);
 680                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 681                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 682                         ret = -EROFS;
 683                         goto unlock_osb;
 684                 }
 685
 686                 if (*flags & MS_RDONLY) {
 687                         sb->s_flags |= MS_RDONLY;
 688                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 689                 } else {
 690                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 691                                 mlog(ML_ERROR, "Cannot remount RDWR "
 692                                      "filesystem due to previous errors.\n");
 693                                 ret = -EROFS;
 694                                 goto unlock_osb;
 695                         }
 696                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 697                         if (incompat_features) {
 698                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 699                                      "of unsupported optional features "
 700                                      "(%x).\n", incompat_features);
 701                                 ret = -EINVAL;
 702                                 goto unlock_osb;
 703                         }
 704                         sb->s_flags &= ~MS_RDONLY;
 705                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 706                 }
 707                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 708 unlock_osb:
 709                 spin_unlock(&osb->osb_lock);
 710                 /* Enable quota accounting after remounting RW */
 711                 if (!ret && !(*flags & MS_RDONLY)) {
 712                         if (sb_any_quota_suspended(sb))
 713                                 ret = ocfs2_susp_quotas(osb, 1);
 714                         else
 715                                 ret = ocfs2_enable_quotas(osb);
 716                         if (ret < 0) {
 717                                 /* Return back changes... */
 718                                 spin_lock(&osb->osb_lock);
 719                                 sb->s_flags |= MS_RDONLY;
 720                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 721                                 spin_unlock(&osb->osb_lock);
 722                                 goto out;
 723                         }
 724                 }
 725         }
 726
 727         if (!ret) {
 728                 /* Only save off the new mount options in case of a successful
 729                  * remount. */
 730                 osb->s_mount_opt = parsed_options.mount_opt;
 731                 osb->s_atime_quantum = parsed_options.atime_quantum;
 732                 osb->preferred_slot = parsed_options.slot;
 733                 if (parsed_options.commit_interval)
 734                         osb->osb_commit_interval = parsed_options.commit_interval;
 735
 736                 if (!ocfs2_is_hard_readonly(osb))
 737                         ocfs2_set_journal_params(osb);
 738
 739                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 740                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 741                                                         MS_POSIXACL : 0);
 742         }
 743 out:
 744         return ret;
 745 }
 746
 747 static int ocfs2_sb_probe(struct super_block *sb,
 748                           struct buffer_head **bh,
 749                           int *sector_size,
 750                           struct ocfs2_blockcheck_stats *stats)
 751 {
 752         int status, tmpstat;
 753         struct ocfs1_vol_disk_hdr *hdr;
 754         struct ocfs2_dinode *di;
 755         int blksize;
 756
 757         *bh = NULL;
 758
 759         /* may be > 512 */
 760         *sector_size = bdev_logical_block_size(sb->s_bdev);
 761         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 762                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 763                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 764                 status = -EINVAL;
 765                 goto bail;
 766         }
 767
 768         /* Can this really happen? */
 769         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 770                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 771
 772         /* check block zero for old format */
 773         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 774         if (status < 0) {
 775                 mlog_errno(status);
 776                 goto bail;
 777         }
 778         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 779         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 780                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 781                      hdr->major_version, hdr->minor_version);
 782                 status = -EINVAL;
 783         }
 784         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 785                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 786                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 787                      hdr->signature);
 788                 status = -EINVAL;
 789         }
 790         brelse(*bh);
 791         *bh = NULL;
 792         if (status < 0) {
 793                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 794                      "upgraded before mounting with ocfs v2\n");
 795                 goto bail;
 796         }
 797
 798         /*
 799          * Now check at magic offset for 512, 1024, 2048, 4096
 800          * blocksizes.  4096 is the maximum blocksize because it is
 801          * the minimum clustersize.
 802          */
 803         status = -EINVAL;
 804         for (blksize = *sector_size;
 805              blksize <= OCFS2_MAX_BLOCKSIZE;
 806              blksize <<= 1) {
 807                 tmpstat = ocfs2_get_sector(sb, bh,
 808                                            OCFS2_SUPER_BLOCK_BLKNO,
 809                                            blksize);
 810                 if (tmpstat < 0) {
 811                         status = tmpstat;
 812                         mlog_errno(status);
 813                         break;
 814                 }
 815                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 816                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 817                 spin_lock_init(&stats->b_lock);
 818                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 819                 if (tmpstat < 0) {
 820                         brelse(*bh);
 821                         *bh = NULL;
 822                 }
 823                 if (tmpstat != -EAGAIN) {
 824                         status = tmpstat;
 825                         break;
 826                 }
 827         }
 828
 829 bail:
 830         return status;
 831 }
 832
 833 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 834 {
 835         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 836
 837         if (osb->s_mount_opt & hb_enabled) {
 838                 if (ocfs2_mount_local(osb)) {
 839                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 840                              "mounted device.\n");
 841                         return -EINVAL;
 842                 }
 843                 if (ocfs2_userspace_stack(osb)) {
 844                         mlog(ML_ERROR, "Userspace stack expected, but "
 845                              "o2cb heartbeat arguments passed to mount\n");
 846                         return -EINVAL;
 847                 }
 848                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 849                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 850                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 851                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 852                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 853                         return -EINVAL;
 854                 }
 855         }
 856
 857         if (!(osb->s_mount_opt & hb_enabled)) {
 858                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 859                     !ocfs2_userspace_stack(osb)) {
 860                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 861                              "a read-write clustered device.\n");
 862                         return -EINVAL;
 863                 }
 864         }
 865
 866         return 0;
 867 }
 868
 869 /*
 870  * If we're using a userspace stack, mount should have passed
 871  * a name that matches the disk.  If not, mount should not
 872  * have passed a stack.
 873  */
 874 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 875                                         struct mount_options *mopt)
 876 {
 877         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 878                 mlog(ML_ERROR,
 879                      "cluster stack passed to mount, but this filesystem "
 880                      "does not support it\n");
 881                 return -EINVAL;
 882         }
 883
 884         if (ocfs2_userspace_stack(osb) &&
 885             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 886                     OCFS2_STACK_LABEL_LEN)) {
 887                 mlog(ML_ERROR,
 888                      "cluster stack passed to mount (\"%s\") does not "
 889                      "match the filesystem (\"%s\")\n",
 890                      mopt->cluster_stack,
 891                      osb->osb_cluster_stack);
 892                 return -EINVAL;
 893         }
 894
 895         return 0;
 896 }
 897
 898 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 899 {
 900         int type;
 901         struct super_block *sb = osb->sb;
 902         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 903                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 904         int status = 0;
 905
 906         for (type = 0; type < MAXQUOTAS; type++) {
 907                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 908                         continue;
 909                 if (unsuspend)
 910                         status = dquot_resume(sb, type);
 911                 else {
 912                         struct ocfs2_mem_dqinfo *oinfo;
 913
 914                         /* Cancel periodic syncing before suspending */
 915                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 916                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 917                         status = dquot_suspend(sb, type);
 918                 }
 919                 if (status < 0)
 920                         break;
 921         }
 922         if (status < 0)
 923                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 924                      "remount (error = %d).\n", status);
 925         return status;
 926 }
 927
 928 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 929 {
 930         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 931         struct super_block *sb = osb->sb;
 932         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 933                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 934         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 935                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 936         int status;
 937         int type;
 938
 939         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 940         for (type = 0; type < MAXQUOTAS; type++) {
 941                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 942                         continue;
 943                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 944                                                         osb->slot_num);
 945                 if (!inode[type]) {
 946                         status = -ENOENT;
 947                         goto out_quota_off;
 948                 }
 949                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 950                                       DQUOT_USAGE_ENABLED);
 951                 if (status < 0)
 952                         goto out_quota_off;
 953         }
 954
 955         for (type = 0; type < MAXQUOTAS; type++)
 956                 iput(inode[type]);
 957         return 0;
 958 out_quota_off:
 959         ocfs2_disable_quotas(osb);
 960         for (type = 0; type < MAXQUOTAS; type++)
 961                 iput(inode[type]);
 962         mlog_errno(status);
 963         return status;
 964 }
 965
 966 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 967 {
 968         int type;
 969         struct inode *inode;
 970         struct super_block *sb = osb->sb;
 971         struct ocfs2_mem_dqinfo *oinfo;
 972
 973         /* We mostly ignore errors in this function because there's not much
 974          * we can do when we see them */
 975         for (type = 0; type < MAXQUOTAS; type++) {
 976                 if (!sb_has_quota_loaded(sb, type))
 977                         continue;
 978                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 979                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 980                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 981                 inode = igrab(sb->s_dquot.files[type]);
 982                 /* Turn off quotas. This will remove all dquot structures from
 983                  * memory and so they will be automatically synced to global
 984                  * quota files */
 985                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 986                                         DQUOT_LIMITS_ENABLED);
 987                 if (!inode)
 988                         continue;
 989                 iput(inode);
 990         }
 991 }
 992
 993 /* Handle quota on quotactl */
 994 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 995 {
 996         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 997                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 998
 999         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
1000                 return -EINVAL;
1001
1002         return dquot_enable(sb_dqopt(sb)->files[type], type,
1003                             format_id, DQUOT_LIMITS_ENABLED);
1004 }
1005
1006 /* Handle quota off quotactl */
1007 static int ocfs2_quota_off(struct super_block *sb, int type)
1008 {
1009         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1010 }
1011
1012 static const struct quotactl_ops ocfs2_quotactl_ops = {
1013         .quota_on_meta  = ocfs2_quota_on,
1014         .quota_off      = ocfs2_quota_off,
1015         .quota_sync     = dquot_quota_sync,
1016         .get_info       = dquot_get_dqinfo,
1017         .set_info       = dquot_set_dqinfo,
1018         .get_dqblk      = dquot_get_dqblk,
1019         .set_dqblk      = dquot_set_dqblk,
1020 };
1021
1022 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1023 {
1024         struct dentry *root;
1025         int status, sector_size;
1026         struct mount_options parsed_options;
1027         struct inode *inode = NULL;
1028         struct ocfs2_super *osb = NULL;
1029         struct buffer_head *bh = NULL;
1030         char nodestr[12];
1031         struct ocfs2_blockcheck_stats stats;
1032
1033         trace_ocfs2_fill_super(sb, data, silent);
1034
1035         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1036                 status = -EINVAL;
1037                 goto read_super_error;
1038         }
1039
1040         /* probe for superblock */
1041         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1042         if (status < 0) {
1043                 mlog(ML_ERROR, "superblock probe failed!\n");
1044                 goto read_super_error;
1045         }
1046
1047         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1048         osb = OCFS2_SB(sb);
1049         if (status < 0) {
1050                 mlog_errno(status);
1051                 goto read_super_error;
1052         }
1053         brelse(bh);
1054         bh = NULL;
1055
1056         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1057                 status = -EINVAL;
1058                 goto read_super_error;
1059         }
1060         osb->s_mount_opt = parsed_options.mount_opt;
1061         osb->s_atime_quantum = parsed_options.atime_quantum;
1062         osb->preferred_slot = parsed_options.slot;
1063         osb->osb_commit_interval = parsed_options.commit_interval;
1064
1065         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1066         osb->osb_resv_level = parsed_options.resv_level;
1067         osb->osb_dir_resv_level = parsed_options.resv_level;
1068         if (parsed_options.dir_resv_level == -1)
1069                 osb->osb_dir_resv_level = parsed_options.resv_level;
1070         else
1071                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1072
1073         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1074         if (status)
1075                 goto read_super_error;
1076
1077         sb->s_magic = OCFS2_SUPER_MAGIC;
1078
1079         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1080                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1081
1082         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1083          * heartbeat=none */
1084         if (bdev_read_only(sb->s_bdev)) {
1085                 if (!(sb->s_flags & MS_RDONLY)) {
1086                         status = -EACCES;
1087                         mlog(ML_ERROR, "Readonly device detected but readonly "
1088                              "mount was not specified.\n");
1089                         goto read_super_error;
1090                 }
1091
1092                 /* You should not be able to start a local heartbeat
1093                  * on a readonly device. */
1094                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1095                         status = -EROFS;
1096                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1097                              "device.\n");
1098                         goto read_super_error;
1099                 }
1100
1101                 status = ocfs2_check_journals_nolocks(osb);
1102                 if (status < 0) {
1103                         if (status == -EROFS)
1104                                 mlog(ML_ERROR, "Recovery required on readonly "
1105                                      "file system, but write access is "
1106                                      "unavailable.\n");
1107                         else
1108                                 mlog_errno(status);
1109                         goto read_super_error;
1110                 }
1111
1112                 ocfs2_set_ro_flag(osb, 1);
1113
1114                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1115                        "Cluster services will not be used for this mount. "
1116                        "Recovery will be skipped.\n", osb->dev_str);
1117         }
1118
1119         if (!ocfs2_is_hard_readonly(osb)) {
1120                 if (sb->s_flags & MS_RDONLY)
1121                         ocfs2_set_ro_flag(osb, 0);
1122         }
1123
1124         status = ocfs2_verify_heartbeat(osb);
1125         if (status < 0) {
1126                 mlog_errno(status);
1127                 goto read_super_error;
1128         }
1129
1130         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1131                                                  ocfs2_debugfs_root);
1132         if (!osb->osb_debug_root) {
1133                 status = -EINVAL;
1134                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1135                 goto read_super_error;
1136         }
1137
1138         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1139                                             osb->osb_debug_root,
1140                                             osb,
1141                                             &ocfs2_osb_debug_fops);
1142         if (!osb->osb_ctxt) {
1143                 status = -EINVAL;
1144                 mlog_errno(status);
1145                 goto read_super_error;
1146         }
1147
1148         if (ocfs2_meta_ecc(osb)) {
1149                 status = ocfs2_blockcheck_stats_debugfs_install(
1150                                                 &osb->osb_ecc_stats,
1151                                                 osb->osb_debug_root);
1152                 if (status) {
1153                         mlog(ML_ERROR,
1154                              "Unable to create blockcheck statistics "
1155                              "files\n");
1156                         goto read_super_error;
1157                 }
1158         }
1159
1160         status = ocfs2_mount_volume(sb);
1161         if (status < 0)
1162                 goto read_super_error;
1163
1164         if (osb->root_inode)
1165                 inode = igrab(osb->root_inode);
1166
1167         if (!inode) {
1168                 status = -EIO;
1169                 mlog_errno(status);
1170                 goto read_super_error;
1171         }
1172
1173         root = d_make_root(inode);
1174         if (!root) {
1175                 status = -ENOMEM;
1176                 mlog_errno(status);
1177                 goto read_super_error;
1178         }
1179
1180         sb->s_root = root;
1181
1182         ocfs2_complete_mount_recovery(osb);
1183
1184         if (ocfs2_mount_local(osb))
1185                 snprintf(nodestr, sizeof(nodestr), "local");
1186         else
1187                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1188
1189         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1190                "with %s data mode.\n",
1191                osb->dev_str, nodestr, osb->slot_num,
1192                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1193                "ordered");
1194
1195         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1196         wake_up(&osb->osb_mount_event);
1197
1198         /* Now we can initialize quotas because we can afford to wait
1199          * for cluster locks recovery now. That also means that truncation
1200          * log recovery can happen but that waits for proper quota setup */
1201         if (!(sb->s_flags & MS_RDONLY)) {
1202                 status = ocfs2_enable_quotas(osb);
1203                 if (status < 0) {
1204                         /* We have to err-out specially here because
1205                          * s_root is already set */
1206                         mlog_errno(status);
1207                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1208                         wake_up(&osb->osb_mount_event);
1209                         return status;
1210                 }
1211         }
1212
1213         ocfs2_complete_quota_recovery(osb);
1214
1215         /* Now we wake up again for processes waiting for quotas */
1216         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1217         wake_up(&osb->osb_mount_event);
1218
1219         /* Start this when the mount is almost sure of being successful */
1220         ocfs2_orphan_scan_start(osb);
1221
1222         return status;
1223
1224 read_super_error:
1225         brelse(bh);
1226
1227         if (osb) {
1228                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1229                 wake_up(&osb->osb_mount_event);
1230                 ocfs2_dismount_volume(sb, 1);
1231         }
1232
1233         if (status)
1234                 mlog_errno(status);
1235         return status;
1236 }
1237
1238 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1239                         int flags,
1240                         const char *dev_name,
1241                         void *data)
1242 {
1243         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1244 }
1245
1246 static struct file_system_type ocfs2_fs_type = {
1247         .owner          = THIS_MODULE,
1248         .name           = "ocfs2",
1249         .mount          = ocfs2_mount,
1250         .kill_sb        = kill_block_super,
1251         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1252         .next           = NULL
1253 };
1254 MODULE_ALIAS_FS("ocfs2");
1255
1256 static int ocfs2_check_set_options(struct super_block *sb,
1257                                    struct mount_options *options)
1258 {
1259         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1260             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1261                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1262                 mlog(ML_ERROR, "User quotas were requested, but this "
1263                      "filesystem does not have the feature enabled.\n");
1264                 return 0;
1265         }
1266         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1267             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1268                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1269                 mlog(ML_ERROR, "Group quotas were requested, but this "
1270                      "filesystem does not have the feature enabled.\n");
1271                 return 0;
1272         }
1273         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1274             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1275                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1276                      "feature is not enabled\n");
1277                 return 0;
1278         }
1279         /* No ACL setting specified? Use XATTR feature... */
1280         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1281                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1282                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1283                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1284                 else
1285                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1286         }
1287         return 1;
1288 }
1289
1290 static int ocfs2_parse_options(struct super_block *sb,
1291                                char *options,
1292                                struct mount_options *mopt,
1293                                int is_remount)
1294 {
1295         int status, user_stack = 0;
1296         char *p;
1297         u32 tmp;
1298
1299         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1300
1301         mopt->commit_interval = 0;
1302         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1303         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1304         mopt->slot = OCFS2_INVALID_SLOT;
1305         mopt->localalloc_opt = -1;
1306         mopt->cluster_stack[0] = '\0';
1307         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1308         mopt->dir_resv_level = -1;
1309
1310         if (!options) {
1311                 status = 1;
1312                 goto bail;
1313         }
1314
1315         while ((p = strsep(&options, ",")) != NULL) {
1316                 int token, option;
1317                 substring_t args[MAX_OPT_ARGS];
1318
1319                 if (!*p)
1320                         continue;
1321
1322                 token = match_token(p, tokens, args);
1323                 switch (token) {
1324                 case Opt_hb_local:
1325                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1326                         break;
1327                 case Opt_hb_none:
1328                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1329                         break;
1330                 case Opt_hb_global:
1331                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1332                         break;
1333                 case Opt_barrier:
1334                         if (match_int(&args[0], &option)) {
1335                                 status = 0;
1336                                 goto bail;
1337                         }
1338                         if (option)
1339                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1340                         else
1341                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1342                         break;
1343                 case Opt_intr:
1344                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1345                         break;
1346                 case Opt_nointr:
1347                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1348                         break;
1349                 case Opt_err_panic:
1350                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1351                         break;
1352                 case Opt_err_ro:
1353                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1354                         break;
1355                 case Opt_data_ordered:
1356                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1357                         break;
1358                 case Opt_data_writeback:
1359                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1360                         break;
1361                 case Opt_user_xattr:
1362                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1363                         break;
1364                 case Opt_nouser_xattr:
1365                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1366                         break;
1367                 case Opt_atime_quantum:
1368                         if (match_int(&args[0], &option)) {
1369                                 status = 0;
1370                                 goto bail;
1371                         }
1372                         if (option >= 0)
1373                                 mopt->atime_quantum = option;
1374                         break;
1375                 case Opt_slot:
1376                         option = 0;
1377                         if (match_int(&args[0], &option)) {
1378                                 status = 0;
1379                                 goto bail;
1380                         }
1381                         if (option)
1382                                 mopt->slot = (s16)option;
1383                         break;
1384                 case Opt_commit:
1385                         option = 0;
1386                         if (match_int(&args[0], &option)) {
1387                                 status = 0;
1388                                 goto bail;
1389                         }
1390                         if (option < 0)
1391                                 return 0;
1392                         if (option == 0)
1393                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1394                         mopt->commit_interval = HZ * option;
1395                         break;
1396                 case Opt_localalloc:
1397                         option = 0;
1398                         if (match_int(&args[0], &option)) {
1399                                 status = 0;
1400                                 goto bail;
1401                         }
1402                         if (option >= 0)
1403                                 mopt->localalloc_opt = option;
1404                         break;
1405                 case Opt_localflocks:
1406                         /*
1407                          * Changing this during remount could race
1408                          * flock() requests, or "unbalance" existing
1409                          * ones (e.g., a lock is taken in one mode but
1410                          * dropped in the other). If users care enough
1411                          * to flip locking modes during remount, we
1412                          * could add a "local" flag to individual
1413                          * flock structures for proper tracking of
1414                          * state.
1415                          */
1416                         if (!is_remount)
1417                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1418                         break;
1419                 case Opt_stack:
1420                         /* Check both that the option we were passed
1421                          * is of the right length and that it is a proper
1422                          * string of the right length.
1423                          */
1424                         if (((args[0].to - args[0].from) !=
1425                              OCFS2_STACK_LABEL_LEN) ||
1426                             (strnlen(args[0].from,
1427                                      OCFS2_STACK_LABEL_LEN) !=
1428                              OCFS2_STACK_LABEL_LEN)) {
1429                                 mlog(ML_ERROR,
1430                                      "Invalid cluster_stack option\n");
1431                                 status = 0;
1432                                 goto bail;
1433                         }
1434                         memcpy(mopt->cluster_stack, args[0].from,
1435                                OCFS2_STACK_LABEL_LEN);
1436                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1437                         /*
1438                          * Open code the memcmp here as we don't have
1439                          * an osb to pass to
1440                          * ocfs2_userspace_stack().
1441                          */
1442                         if (memcmp(mopt->cluster_stack,
1443                                    OCFS2_CLASSIC_CLUSTER_STACK,
1444                                    OCFS2_STACK_LABEL_LEN))
1445                                 user_stack = 1;
1446                         break;
1447                 case Opt_inode64:
1448                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1449                         break;
1450                 case Opt_usrquota:
1451                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1452                         break;
1453                 case Opt_grpquota:
1454                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1455                         break;
1456                 case Opt_coherency_buffered:
1457                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1458                         break;
1459                 case Opt_coherency_full:
1460                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1461                         break;
1462                 case Opt_acl:
1463                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1464                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1465                         break;
1466                 case Opt_noacl:
1467                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1468                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1469                         break;
1470                 case Opt_resv_level:
1471                         if (is_remount)
1472                                 break;
1473                         if (match_int(&args[0], &option)) {
1474                                 status = 0;
1475                                 goto bail;
1476                         }
1477                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1478                             option < OCFS2_MAX_RESV_LEVEL)
1479                                 mopt->resv_level = option;
1480                         break;
1481                 case Opt_dir_resv_level:
1482                         if (is_remount)
1483                                 break;
1484                         if (match_int(&args[0], &option)) {
1485                                 status = 0;
1486                                 goto bail;
1487                         }
1488                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1489                             option < OCFS2_MAX_RESV_LEVEL)
1490                                 mopt->dir_resv_level = option;
1491                         break;
1492                 default:
1493                         mlog(ML_ERROR,
1494                              "Unrecognized mount option \"%s\" "
1495                              "or missing value\n", p);
1496                         status = 0;
1497                         goto bail;
1498                 }
1499         }
1500
1501         if (user_stack == 0) {
1502                 /* Ensure only one heartbeat mode */
1503                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1504                                          OCFS2_MOUNT_HB_GLOBAL |
1505                                          OCFS2_MOUNT_HB_NONE);
1506                 if (hweight32(tmp) != 1) {
1507                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1508                         status = 0;
1509                         goto bail;
1510                 }
1511         }
1512
1513         status = 1;
1514
1515 bail:
1516         return status;
1517 }
1518
1519 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1520 {
1521         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1522         unsigned long opts = osb->s_mount_opt;
1523         unsigned int local_alloc_megs;
1524
1525         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1526                 seq_printf(s, ",_netdev");
1527                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1528                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1529                 else
1530                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1531         } else
1532                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1533
1534         if (opts & OCFS2_MOUNT_NOINTR)
1535                 seq_printf(s, ",nointr");
1536
1537         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1538                 seq_printf(s, ",data=writeback");
1539         else
1540                 seq_printf(s, ",data=ordered");
1541
1542         if (opts & OCFS2_MOUNT_BARRIER)
1543                 seq_printf(s, ",barrier=1");
1544
1545         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1546                 seq_printf(s, ",errors=panic");
1547         else
1548                 seq_printf(s, ",errors=remount-ro");
1549
1550         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1551                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1552
1553         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1554
1555         if (osb->osb_commit_interval)
1556                 seq_printf(s, ",commit=%u",
1557                            (unsigned) (osb->osb_commit_interval / HZ));
1558
1559         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1560         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1561                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1562
1563         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1564                 seq_printf(s, ",localflocks,");
1565
1566         if (osb->osb_cluster_stack[0])
1567                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1568                            osb->osb_cluster_stack);
1569         if (opts & OCFS2_MOUNT_USRQUOTA)
1570                 seq_printf(s, ",usrquota");
1571         if (opts & OCFS2_MOUNT_GRPQUOTA)
1572                 seq_printf(s, ",grpquota");
1573
1574         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1575                 seq_printf(s, ",coherency=buffered");
1576         else
1577                 seq_printf(s, ",coherency=full");
1578
1579         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1580                 seq_printf(s, ",nouser_xattr");
1581         else
1582                 seq_printf(s, ",user_xattr");
1583
1584         if (opts & OCFS2_MOUNT_INODE64)
1585                 seq_printf(s, ",inode64");
1586
1587         if (opts & OCFS2_MOUNT_POSIX_ACL)
1588                 seq_printf(s, ",acl");
1589         else
1590                 seq_printf(s, ",noacl");
1591
1592         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1593                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1594
1595         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1596                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1597
1598         return 0;
1599 }
1600
1601 static int __init ocfs2_init(void)
1602 {
1603         int status;
1604
1605         status = init_ocfs2_uptodate_cache();
1606         if (status < 0)
1607                 goto out1;
1608
1609         status = ocfs2_initialize_mem_caches();
1610         if (status < 0)
1611                 goto out2;
1612
1613         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1614         if (!ocfs2_wq) {
1615                 status = -ENOMEM;
1616                 goto out3;
1617         }
1618
1619         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1620         if (!ocfs2_debugfs_root) {
1621                 status = -EFAULT;
1622                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1623         }
1624
1625         ocfs2_set_locking_protocol();
1626
1627         status = register_quota_format(&ocfs2_quota_format);
1628         if (status < 0)
1629                 goto out4;
1630         status = register_filesystem(&ocfs2_fs_type);
1631         if (!status)
1632                 return 0;
1633
1634         unregister_quota_format(&ocfs2_quota_format);
1635 out4:
1636         destroy_workqueue(ocfs2_wq);
1637         debugfs_remove(ocfs2_debugfs_root);
1638 out3:
1639         ocfs2_free_mem_caches();
1640 out2:
1641         exit_ocfs2_uptodate_cache();
1642 out1:
1643         mlog_errno(status);
1644         return status;
1645 }
1646
1647 static void __exit ocfs2_exit(void)
1648 {
1649         if (ocfs2_wq) {
1650                 flush_workqueue(ocfs2_wq);
1651                 destroy_workqueue(ocfs2_wq);
1652         }
1653
1654         unregister_quota_format(&ocfs2_quota_format);
1655
1656         debugfs_remove(ocfs2_debugfs_root);
1657
1658         ocfs2_free_mem_caches();
1659
1660         unregister_filesystem(&ocfs2_fs_type);
1661
1662         exit_ocfs2_uptodate_cache();
1663 }
1664
1665 static void ocfs2_put_super(struct super_block *sb)
1666 {
1667         trace_ocfs2_put_super(sb);
1668
1669         ocfs2_sync_blockdev(sb);
1670         ocfs2_dismount_volume(sb, 0);
1671 }
1672
1673 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1674 {
1675         struct ocfs2_super *osb;
1676         u32 numbits, freebits;
1677         int status;
1678         struct ocfs2_dinode *bm_lock;
1679         struct buffer_head *bh = NULL;
1680         struct inode *inode = NULL;
1681
1682         trace_ocfs2_statfs(dentry->d_sb, buf);
1683
1684         osb = OCFS2_SB(dentry->d_sb);
1685
1686         inode = ocfs2_get_system_file_inode(osb,
1687                                             GLOBAL_BITMAP_SYSTEM_INODE,
1688                                             OCFS2_INVALID_SLOT);
1689         if (!inode) {
1690                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1691                 status = -EIO;
1692                 goto bail;
1693         }
1694
1695         status = ocfs2_inode_lock(inode, &bh, 0);
1696         if (status < 0) {
1697                 mlog_errno(status);
1698                 goto bail;
1699         }
1700
1701         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1702
1703         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1704         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1705
1706         buf->f_type = OCFS2_SUPER_MAGIC;
1707         buf->f_bsize = dentry->d_sb->s_blocksize;
1708         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1709         buf->f_blocks = ((sector_t) numbits) *
1710                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1711         buf->f_bfree = ((sector_t) freebits) *
1712                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1713         buf->f_bavail = buf->f_bfree;
1714         buf->f_files = numbits;
1715         buf->f_ffree = freebits;
1716         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1717                                 & 0xFFFFFFFFUL;
1718         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1719                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1720
1721         brelse(bh);
1722
1723         ocfs2_inode_unlock(inode, 0);
1724         status = 0;
1725 bail:
1726         if (inode)
1727                 iput(inode);
1728
1729         if (status)
1730                 mlog_errno(status);
1731
1732         return status;
1733 }
1734
1735 static void ocfs2_inode_init_once(void *data)
1736 {
1737         struct ocfs2_inode_info *oi = data;
1738
1739         oi->ip_flags = 0;
1740         oi->ip_open_count = 0;
1741         spin_lock_init(&oi->ip_lock);
1742         ocfs2_extent_map_init(&oi->vfs_inode);
1743         INIT_LIST_HEAD(&oi->ip_io_markers);
1744         oi->ip_dir_start_lookup = 0;
1745         mutex_init(&oi->ip_unaligned_aio);
1746         init_rwsem(&oi->ip_alloc_sem);
1747         init_rwsem(&oi->ip_xattr_sem);
1748         mutex_init(&oi->ip_io_mutex);
1749
1750         oi->ip_blkno = 0ULL;
1751         oi->ip_clusters = 0;
1752
1753         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1754
1755         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1756         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1757         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1758
1759         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1760                                   &ocfs2_inode_caching_ops);
1761
1762         inode_init_once(&oi->vfs_inode);
1763 }
1764
1765 static int ocfs2_initialize_mem_caches(void)
1766 {
1767         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1768                                        sizeof(struct ocfs2_inode_info),
1769                                        0,
1770                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1771                                                 SLAB_MEM_SPREAD),
1772                                        ocfs2_inode_init_once);
1773         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1774                                         sizeof(struct ocfs2_dquot),
1775                                         0,
1776                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1777                                                 SLAB_MEM_SPREAD),
1778                                         NULL);
1779         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1780                                         sizeof(struct ocfs2_quota_chunk),
1781                                         0,
1782                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1783                                         NULL);
1784         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1785             !ocfs2_qf_chunk_cachep) {
1786                 if (ocfs2_inode_cachep)
1787                         kmem_cache_destroy(ocfs2_inode_cachep);
1788                 if (ocfs2_dquot_cachep)
1789                         kmem_cache_destroy(ocfs2_dquot_cachep);
1790                 if (ocfs2_qf_chunk_cachep)
1791                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1792                 return -ENOMEM;
1793         }
1794
1795         return 0;
1796 }
1797
1798 static void ocfs2_free_mem_caches(void)
1799 {
1800         /*
1801          * Make sure all delayed rcu free inodes are flushed before we
1802          * destroy cache.
1803          */
1804         rcu_barrier();
1805         if (ocfs2_inode_cachep)
1806                 kmem_cache_destroy(ocfs2_inode_cachep);
1807         ocfs2_inode_cachep = NULL;
1808
1809         if (ocfs2_dquot_cachep)
1810                 kmem_cache_destroy(ocfs2_dquot_cachep);
1811         ocfs2_dquot_cachep = NULL;
1812
1813         if (ocfs2_qf_chunk_cachep)
1814                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1815         ocfs2_qf_chunk_cachep = NULL;
1816 }
1817
1818 static int ocfs2_get_sector(struct super_block *sb,
1819                             struct buffer_head **bh,
1820                             int block,
1821                             int sect_size)
1822 {
1823         if (!sb_set_blocksize(sb, sect_size)) {
1824                 mlog(ML_ERROR, "unable to set blocksize\n");
1825                 return -EIO;
1826         }
1827
1828         *bh = sb_getblk(sb, block);
1829         if (!*bh) {
1830                 mlog_errno(-ENOMEM);
1831                 return -ENOMEM;
1832         }
1833         lock_buffer(*bh);
1834         if (!buffer_dirty(*bh))
1835                 clear_buffer_uptodate(*bh);
1836         unlock_buffer(*bh);
1837         ll_rw_block(READ, 1, bh);
1838         wait_on_buffer(*bh);
1839         if (!buffer_uptodate(*bh)) {
1840                 mlog_errno(-EIO);
1841                 brelse(*bh);
1842                 *bh = NULL;
1843                 return -EIO;
1844         }
1845
1846         return 0;
1847 }
1848
1849 static int ocfs2_mount_volume(struct super_block *sb)
1850 {
1851         int status = 0;
1852         int unlock_super = 0;
1853         struct ocfs2_super *osb = OCFS2_SB(sb);
1854
1855         if (ocfs2_is_hard_readonly(osb))
1856                 goto leave;
1857
1858         status = ocfs2_dlm_init(osb);
1859         if (status < 0) {
1860                 mlog_errno(status);
1861                 goto leave;
1862         }
1863
1864         status = ocfs2_super_lock(osb, 1);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 goto leave;
1868         }
1869         unlock_super = 1;
1870
1871         /* This will load up the node map and add ourselves to it. */
1872         status = ocfs2_find_slot(osb);
1873         if (status < 0) {
1874                 mlog_errno(status);
1875                 goto leave;
1876         }
1877
1878         /* load all node-local system inodes */
1879         status = ocfs2_init_local_system_inodes(osb);
1880         if (status < 0) {
1881                 mlog_errno(status);
1882                 goto leave;
1883         }
1884
1885         status = ocfs2_check_volume(osb);
1886         if (status < 0) {
1887                 mlog_errno(status);
1888                 goto leave;
1889         }
1890
1891         status = ocfs2_truncate_log_init(osb);
1892         if (status < 0)
1893                 mlog_errno(status);
1894
1895 leave:
1896         if (unlock_super)
1897                 ocfs2_super_unlock(osb, 1);
1898
1899         return status;
1900 }
1901
1902 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1903 {
1904         int tmp, hangup_needed = 0;
1905         struct ocfs2_super *osb = NULL;
1906         char nodestr[12];
1907
1908         trace_ocfs2_dismount_volume(sb);
1909
1910         BUG_ON(!sb);
1911         osb = OCFS2_SB(sb);
1912         BUG_ON(!osb);
1913
1914         debugfs_remove(osb->osb_ctxt);
1915
1916         /* Orphan scan should be stopped as early as possible */
1917         ocfs2_orphan_scan_stop(osb);
1918
1919         ocfs2_disable_quotas(osb);
1920
1921         /* All dquots should be freed by now */
1922         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1923         /* Wait for worker to be done with the work structure in osb */
1924         cancel_work_sync(&osb->dquot_drop_work);
1925
1926         ocfs2_shutdown_local_alloc(osb);
1927
1928         ocfs2_truncate_log_shutdown(osb);
1929
1930         /* This will disable recovery and flush any recovery work. */
1931         ocfs2_recovery_exit(osb);
1932
1933         ocfs2_journal_shutdown(osb);
1934
1935         ocfs2_sync_blockdev(sb);
1936
1937         ocfs2_purge_refcount_trees(osb);
1938
1939         /* No cluster connection means we've failed during mount, so skip
1940          * all the steps which depended on that to complete. */
1941         if (osb->cconn) {
1942                 tmp = ocfs2_super_lock(osb, 1);
1943                 if (tmp < 0) {
1944                         mlog_errno(tmp);
1945                         return;
1946                 }
1947         }
1948
1949         if (osb->slot_num != OCFS2_INVALID_SLOT)
1950                 ocfs2_put_slot(osb);
1951
1952         if (osb->cconn)
1953                 ocfs2_super_unlock(osb, 1);
1954
1955         ocfs2_release_system_inodes(osb);
1956
1957         /*
1958          * If we're dismounting due to mount error, mount.ocfs2 will clean
1959          * up heartbeat.  If we're a local mount, there is no heartbeat.
1960          * If we failed before we got a uuid_str yet, we can't stop
1961          * heartbeat.  Otherwise, do it.
1962          */
1963         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1964             !ocfs2_is_hard_readonly(osb))
1965                 hangup_needed = 1;
1966
1967         if (osb->cconn)
1968                 ocfs2_dlm_shutdown(osb, hangup_needed);
1969
1970         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1971         debugfs_remove(osb->osb_debug_root);
1972
1973         if (hangup_needed)
1974                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1975
1976         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1977
1978         if (ocfs2_mount_local(osb))
1979                 snprintf(nodestr, sizeof(nodestr), "local");
1980         else
1981                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1982
1983         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1984                osb->dev_str, nodestr);
1985
1986         ocfs2_delete_osb(osb);
1987         kfree(osb);
1988         sb->s_dev = 0;
1989         sb->s_fs_info = NULL;
1990 }
1991
1992 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1993                                 unsigned uuid_bytes)
1994 {
1995         int i, ret;
1996         char *ptr;
1997
1998         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1999
2000         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2001         if (osb->uuid_str == NULL)
2002                 return -ENOMEM;
2003
2004         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2005                 /* print with null */
2006                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2007                 if (ret != 2) /* drop super cleans up */
2008                         return -EINVAL;
2009                 /* then only advance past the last char */
2010                 ptr += 2;
2011         }
2012
2013         return 0;
2014 }
2015
2016 /* Make sure entire volume is addressable by our journal.  Requires
2017    osb_clusters_at_boot to be valid and for the journal to have been
2018    initialized by ocfs2_journal_init(). */
2019 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2020 {
2021         int status = 0;
2022         u64 max_block =
2023                 ocfs2_clusters_to_blocks(osb->sb,
2024                                          osb->osb_clusters_at_boot) - 1;
2025
2026         /* 32-bit block number is always OK. */
2027         if (max_block <= (u32)~0ULL)
2028                 goto out;
2029
2030         /* Volume is "huge", so see if our journal is new enough to
2031            support it. */
2032         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2033                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2034               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2035                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2036                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2037                      "Enable the 'block64' journal option with tunefs.ocfs2");
2038                 status = -EFBIG;
2039                 goto out;
2040         }
2041
2042  out:
2043         return status;
2044 }
2045
2046 static int ocfs2_initialize_super(struct super_block *sb,
2047                                   struct buffer_head *bh,
2048                                   int sector_size,
2049                                   struct ocfs2_blockcheck_stats *stats)
2050 {
2051         int status;
2052         int i, cbits, bbits;
2053         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2054         struct inode *inode = NULL;
2055         struct ocfs2_journal *journal;
2056         struct ocfs2_super *osb;
2057         u64 total_blocks;
2058
2059         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2060         if (!osb) {
2061                 status = -ENOMEM;
2062                 mlog_errno(status);
2063                 goto bail;
2064         }
2065
2066         sb->s_fs_info = osb;
2067         sb->s_op = &ocfs2_sops;
2068         sb->s_d_op = &ocfs2_dentry_ops;
2069         sb->s_export_op = &ocfs2_export_ops;
2070         sb->s_qcop = &ocfs2_quotactl_ops;
2071         sb->dq_op = &ocfs2_quota_operations;
2072         sb->s_xattr = ocfs2_xattr_handlers;
2073         sb->s_time_gran = 1;
2074         sb->s_flags |= MS_NOATIME;
2075         /* this is needed to support O_LARGEFILE */
2076         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2077         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2078         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2079
2080         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2081
2082         for (i = 0; i < 3; i++)
2083                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2084         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2085
2086         osb->sb = sb;
2087         /* Save off for ocfs2_rw_direct */
2088         osb->s_sectsize_bits = blksize_bits(sector_size);
2089         BUG_ON(!osb->s_sectsize_bits);
2090
2091         spin_lock_init(&osb->dc_task_lock);
2092         init_waitqueue_head(&osb->dc_event);
2093         osb->dc_work_sequence = 0;
2094         osb->dc_wake_sequence = 0;
2095         INIT_LIST_HEAD(&osb->blocked_lock_list);
2096         osb->blocked_lock_count = 0;
2097         spin_lock_init(&osb->osb_lock);
2098         spin_lock_init(&osb->osb_xattr_lock);
2099         ocfs2_init_steal_slots(osb);
2100
2101         mutex_init(&osb->system_file_mutex);
2102
2103         atomic_set(&osb->alloc_stats.moves, 0);
2104         atomic_set(&osb->alloc_stats.local_data, 0);
2105         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2106         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2107         atomic_set(&osb->alloc_stats.bg_extends, 0);
2108
2109         /* Copy the blockcheck stats from the superblock probe */
2110         osb->osb_ecc_stats = *stats;
2111
2112         ocfs2_init_node_maps(osb);
2113
2114         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2115                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2116
2117         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2118         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2119                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2120                      osb->max_slots);
2121                 status = -EINVAL;
2122                 goto bail;
2123         }
2124
2125         ocfs2_orphan_scan_init(osb);
2126
2127         status = ocfs2_recovery_init(osb);
2128         if (status) {
2129                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2130                 mlog_errno(status);
2131                 goto bail;
2132         }
2133
2134         init_waitqueue_head(&osb->checkpoint_event);
2135
2136         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2137
2138         osb->slot_num = OCFS2_INVALID_SLOT;
2139
2140         osb->s_xattr_inline_size = le16_to_cpu(
2141                                         di->id2.i_super.s_xattr_inline_size);
2142
2143         osb->local_alloc_state = OCFS2_LA_UNUSED;
2144         osb->local_alloc_bh = NULL;
2145         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2146
2147         init_waitqueue_head(&osb->osb_mount_event);
2148
2149         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2150         if (status) {
2151                 mlog_errno(status);
2152                 goto bail;
2153         }
2154
2155         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2156         if (!osb->vol_label) {
2157                 mlog(ML_ERROR, "unable to alloc vol label\n");
2158                 status = -ENOMEM;
2159                 goto bail;
2160         }
2161
2162         osb->slot_recovery_generations =
2163                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2164                         GFP_KERNEL);
2165         if (!osb->slot_recovery_generations) {
2166                 status = -ENOMEM;
2167                 mlog_errno(status);
2168                 goto bail;
2169         }
2170
2171         init_waitqueue_head(&osb->osb_wipe_event);
2172         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2173                                         sizeof(*osb->osb_orphan_wipes),
2174                                         GFP_KERNEL);
2175         if (!osb->osb_orphan_wipes) {
2176                 status = -ENOMEM;
2177                 mlog_errno(status);
2178                 goto bail;
2179         }
2180
2181         osb->osb_rf_lock_tree = RB_ROOT;
2182
2183         osb->s_feature_compat =
2184                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2185         osb->s_feature_ro_compat =
2186                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2187         osb->s_feature_incompat =
2188                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2189
2190         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2191                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2192                      "optional features (%x).\n", i);
2193                 status = -EINVAL;
2194                 goto bail;
2195         }
2196         if (!(osb->sb->s_flags & MS_RDONLY) &&
2197             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2198                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2199                      "unsupported optional features (%x).\n", i);
2200                 status = -EINVAL;
2201                 goto bail;
2202         }
2203
2204         if (ocfs2_clusterinfo_valid(osb)) {
2205                 osb->osb_stackflags =
2206                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2207                 strlcpy(osb->osb_cluster_stack,
2208                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2209                        OCFS2_STACK_LABEL_LEN + 1);
2210                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2211                         mlog(ML_ERROR,
2212                              "couldn't mount because of an invalid "
2213                              "cluster stack label (%s) \n",
2214                              osb->osb_cluster_stack);
2215                         status = -EINVAL;
2216                         goto bail;
2217                 }
2218                 strlcpy(osb->osb_cluster_name,
2219                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2220                         OCFS2_CLUSTER_NAME_LEN + 1);
2221         } else {
2222                 /* The empty string is identical with classic tools that
2223                  * don't know about s_cluster_info. */
2224                 osb->osb_cluster_stack[0] = '\0';
2225         }
2226
2227         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2228
2229         /* FIXME
2230          * This should be done in ocfs2_journal_init(), but unknown
2231          * ordering issues will cause the filesystem to crash.
2232          * If anyone wants to figure out what part of the code
2233          * refers to osb->journal before ocfs2_journal_init() is run,
2234          * be my guest.
2235          */
2236         /* initialize our journal structure */
2237
2238         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2239         if (!journal) {
2240                 mlog(ML_ERROR, "unable to alloc journal\n");
2241                 status = -ENOMEM;
2242                 goto bail;
2243         }
2244         osb->journal = journal;
2245         journal->j_osb = osb;
2246
2247         atomic_set(&journal->j_num_trans, 0);
2248         init_rwsem(&journal->j_trans_barrier);
2249         init_waitqueue_head(&journal->j_checkpointed);
2250         spin_lock_init(&journal->j_lock);
2251         journal->j_trans_id = (unsigned long) 1;
2252         INIT_LIST_HEAD(&journal->j_la_cleanups);
2253         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2254         journal->j_state = OCFS2_JOURNAL_FREE;
2255
2256         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2257         init_llist_head(&osb->dquot_drop_list);
2258
2259         /* get some pseudo constants for clustersize bits */
2260         osb->s_clustersize_bits =
2261                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2262         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2263
2264         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2265             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2266                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2267                      osb->s_clustersize);
2268                 status = -EINVAL;
2269                 goto bail;
2270         }
2271
2272         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2273                                                 le32_to_cpu(di->i_clusters));
2274
2275         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2276                                            total_blocks);
2277         if (status) {
2278                 mlog(ML_ERROR, "Volume too large "
2279                      "to mount safely on this system");
2280                 status = -EFBIG;
2281                 goto bail;
2282         }
2283
2284         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2285                                  sizeof(di->id2.i_super.s_uuid))) {
2286                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2287                 status = -ENOMEM;
2288                 goto bail;
2289         }
2290
2291         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2292                 OCFS2_MAX_VOL_LABEL_LEN);
2293         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2294         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2295         osb->first_cluster_group_blkno =
2296                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2297         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2298         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2299         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2300                                      (unsigned long long)osb->root_blkno,
2301                                      (unsigned long long)osb->system_dir_blkno,
2302                                      osb->s_clustersize_bits);
2303
2304         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2305         if (!osb->osb_dlm_debug) {
2306                 status = -ENOMEM;
2307                 mlog_errno(status);
2308                 goto bail;
2309         }
2310
2311         atomic_set(&osb->vol_state, VOLUME_INIT);
2312
2313         /* load root, system_dir, and all global system inodes */
2314         status = ocfs2_init_global_system_inodes(osb);
2315         if (status < 0) {
2316                 mlog_errno(status);
2317                 goto bail;
2318         }
2319
2320         /*
2321          * global bitmap
2322          */
2323         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2324                                             OCFS2_INVALID_SLOT);
2325         if (!inode) {
2326                 status = -EINVAL;
2327                 mlog_errno(status);
2328                 goto bail;
2329         }
2330
2331         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2332         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2333         iput(inode);
2334
2335         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2336                                  osb->s_feature_incompat) * 8;
2337
2338         status = ocfs2_init_slot_info(osb);
2339         if (status < 0) {
2340                 mlog_errno(status);
2341                 goto bail;
2342         }
2343         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2344
2345 bail:
2346         return status;
2347 }
2348
2349 /*
2350  * will return: -EAGAIN if it is ok to keep searching for superblocks
2351  *              -EINVAL if there is a bad superblock
2352  *              0 on success
2353  */
2354 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2355                                struct buffer_head *bh,
2356                                u32 blksz,
2357                                struct ocfs2_blockcheck_stats *stats)
2358 {
2359         int status = -EAGAIN;
2360
2361         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2362                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2363                 /* We have to do a raw check of the feature here */
2364                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2365                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2366                         status = ocfs2_block_check_validate(bh->b_data,
2367                                                             bh->b_size,
2368                                                             &di->i_check,
2369                                                             stats);
2370                         if (status)
2371                                 goto out;
2372                 }
2373                 status = -EINVAL;
2374                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2375                         mlog(ML_ERROR, "found superblock with incorrect block "
2376                              "size: found %u, should be %u\n",
2377                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2378                                blksz);
2379                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2380                            OCFS2_MAJOR_REV_LEVEL ||
2381                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2382                            OCFS2_MINOR_REV_LEVEL) {
2383                         mlog(ML_ERROR, "found superblock with bad version: "
2384                              "found %u.%u, should be %u.%u\n",
2385                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2386                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2387                              OCFS2_MAJOR_REV_LEVEL,
2388                              OCFS2_MINOR_REV_LEVEL);
2389                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2390                         mlog(ML_ERROR, "bad block number on superblock: "
2391                              "found %llu, should be %llu\n",
2392                              (unsigned long long)le64_to_cpu(di->i_blkno),
2393                              (unsigned long long)bh->b_blocknr);
2394                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2395                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2396                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2397                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2398                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2399                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2400                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2401                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2402                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2403                         mlog(ML_ERROR,
2404                              "Superblock slots found greater than file system "
2405                              "maximum: found %u, max %u\n",
2406                              le16_to_cpu(di->id2.i_super.s_max_slots),
2407                              OCFS2_MAX_SLOTS);
2408                 } else {
2409                         /* found it! */
2410                         status = 0;
2411                 }
2412         }
2413
2414 out:
2415         if (status && status != -EAGAIN)
2416                 mlog_errno(status);
2417         return status;
2418 }
2419
2420 static int ocfs2_check_volume(struct ocfs2_super *osb)
2421 {
2422         int status;
2423         int dirty;
2424         int local;
2425         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2426                                                   * recover
2427                                                   * ourselves. */
2428
2429         /* Init our journal object. */
2430         status = ocfs2_journal_init(osb->journal, &dirty);
2431         if (status < 0) {
2432                 mlog(ML_ERROR, "Could not initialize journal!\n");
2433                 goto finally;
2434         }
2435
2436         /* Now that journal has been initialized, check to make sure
2437            entire volume is addressable. */
2438         status = ocfs2_journal_addressable(osb);
2439         if (status)
2440                 goto finally;
2441
2442         /* If the journal was unmounted cleanly then we don't want to
2443          * recover anything. Otherwise, journal_load will do that
2444          * dirty work for us :) */
2445         if (!dirty) {
2446                 status = ocfs2_journal_wipe(osb->journal, 0);
2447                 if (status < 0) {
2448                         mlog_errno(status);
2449                         goto finally;
2450                 }
2451         } else {
2452                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2453                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2454         }
2455
2456         local = ocfs2_mount_local(osb);
2457
2458         /* will play back anything left in the journal. */
2459         status = ocfs2_journal_load(osb->journal, local, dirty);
2460         if (status < 0) {
2461                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2462                 goto finally;
2463         }
2464
2465         if (dirty) {
2466                 /* recover my local alloc if we didn't unmount cleanly. */
2467                 status = ocfs2_begin_local_alloc_recovery(osb,
2468                                                           osb->slot_num,
2469                                                           &local_alloc);
2470                 if (status < 0) {
2471                         mlog_errno(status);
2472                         goto finally;
2473                 }
2474                 /* we complete the recovery process after we've marked
2475                  * ourselves as mounted. */
2476         }
2477
2478         status = ocfs2_load_local_alloc(osb);
2479         if (status < 0) {
2480                 mlog_errno(status);
2481                 goto finally;
2482         }
2483
2484         if (dirty) {
2485                 /* Recovery will be completed after we've mounted the
2486                  * rest of the volume. */
2487                 osb->dirty = 1;
2488                 osb->local_alloc_copy = local_alloc;
2489                 local_alloc = NULL;
2490         }
2491
2492         /* go through each journal, trylock it and if you get the
2493          * lock, and it's marked as dirty, set the bit in the recover
2494          * map and launch a recovery thread for it. */
2495         status = ocfs2_mark_dead_nodes(osb);
2496         if (status < 0) {
2497                 mlog_errno(status);
2498                 goto finally;
2499         }
2500
2501         status = ocfs2_compute_replay_slots(osb);
2502         if (status < 0)
2503                 mlog_errno(status);
2504
2505 finally:
2506         kfree(local_alloc);
2507
2508         if (status)
2509                 mlog_errno(status);
2510         return status;
2511 }
2512
2513 /*
2514  * The routine gets called from dismount or close whenever a dismount on
2515  * volume is requested and the osb open count becomes 1.
2516  * It will remove the osb from the global list and also free up all the
2517  * initialized resources and fileobject.
2518  */
2519 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2520 {
2521         /* This function assumes that the caller has the main osb resource */
2522
2523         ocfs2_free_slot_info(osb);
2524
2525         kfree(osb->osb_orphan_wipes);
2526         kfree(osb->slot_recovery_generations);
2527         /* FIXME
2528          * This belongs in journal shutdown, but because we have to
2529          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2530          * we free it here.
2531          */
2532         kfree(osb->journal);
2533         kfree(osb->local_alloc_copy);
2534         kfree(osb->uuid_str);
2535         kfree(osb->vol_label);
2536         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2537         memset(osb, 0, sizeof(struct ocfs2_super));
2538 }
2539
2540 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2541  * panic(). We do not support continue-on-error operation. */
2542 static void ocfs2_handle_error(struct super_block *sb)
2543 {
2544         struct ocfs2_super *osb = OCFS2_SB(sb);
2545
2546         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2547                 panic("OCFS2: (device %s): panic forced after error\n",
2548                       sb->s_id);
2549
2550         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2551
2552         if (sb->s_flags & MS_RDONLY &&
2553             (ocfs2_is_soft_readonly(osb) ||
2554              ocfs2_is_hard_readonly(osb)))
2555                 return;
2556
2557         printk(KERN_CRIT "File system is now read-only due to the potential "
2558                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2559                "system is unmounted.\n");
2560         sb->s_flags |= MS_RDONLY;
2561         ocfs2_set_ro_flag(osb, 0);
2562 }
2563
2564 static char error_buf[1024];
2565
2566 void __ocfs2_error(struct super_block *sb,
2567                    const char *function,
2568                    const char *fmt, ...)
2569 {
2570         va_list args;
2571
2572         va_start(args, fmt);
2573         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2574         va_end(args);
2575
2576         /* Not using mlog here because we want to show the actual
2577          * function the error came from. */
2578         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2579                sb->s_id, function, error_buf);
2580
2581         ocfs2_handle_error(sb);
2582 }
2583
2584 /* Handle critical errors. This is intentionally more drastic than
2585  * ocfs2_handle_error, so we only use for things like journal errors,
2586  * etc. */
2587 void __ocfs2_abort(struct super_block* sb,
2588                    const char *function,
2589                    const char *fmt, ...)
2590 {
2591         va_list args;
2592
2593         va_start(args, fmt);
2594         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2595         va_end(args);
2596
2597         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2598                sb->s_id, function, error_buf);
2599
2600         /* We don't have the cluster support yet to go straight to
2601          * hard readonly in here. Until then, we want to keep
2602          * ocfs2_abort() so that we can at least mark critical
2603          * errors.
2604          *
2605          * TODO: This should abort the journal and alert other nodes
2606          * that our slot needs recovery. */
2607
2608         /* Force a panic(). This stinks, but it's better than letting
2609          * things continue without having a proper hard readonly
2610          * here. */
2611         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2612                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2613         ocfs2_handle_error(sb);
2614 }
2615
2616 /*
2617  * Void signal blockers, because in-kernel sigprocmask() only fails
2618  * when SIG_* is wrong.
2619  */
2620 void ocfs2_block_signals(sigset_t *oldset)
2621 {
2622         int rc;
2623         sigset_t blocked;
2624
2625         sigfillset(&blocked);
2626         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2627         BUG_ON(rc);
2628 }
2629
2630 void ocfs2_unblock_signals(sigset_t *oldset)
2631 {
2632         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2633         BUG_ON(rc);
2634 }
2635
2636 module_init(ocfs2_init);
2637 module_exit(ocfs2_exit);