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 #include <linux/signal.h>
  46
  47 #define CREATE_TRACE_POINTS
  48 #include "ocfs2_trace.h"
  49
  50 #include <cluster/masklog.h>
  51
  52 #include "ocfs2.h"
  53
  54 /* this should be the only file to include a version 1 header */
  55 #include "ocfs1_fs_compat.h"
  56
  57 #include "alloc.h"
  58 #include "aops.h"
  59 #include "blockcheck.h"
  60 #include "dlmglue.h"
  61 #include "export.h"
  62 #include "extent_map.h"
  63 #include "heartbeat.h"
  64 #include "inode.h"
  65 #include "journal.h"
  66 #include "localalloc.h"
  67 #include "namei.h"
  68 #include "slot_map.h"
  69 #include "super.h"
  70 #include "sysfile.h"
  71 #include "uptodate.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78 #include "filecheck.h"
  79
  80 static struct kmem_cache *ocfs2_inode_cachep;
  81 struct kmem_cache *ocfs2_dquot_cachep;
  82 struct kmem_cache *ocfs2_qf_chunk_cachep;
  83
  84 static struct dentry *ocfs2_debugfs_root;
  85
  86 MODULE_AUTHOR("Oracle");
  87 MODULE_LICENSE("GPL");
  88 MODULE_DESCRIPTION("OCFS2 cluster file system");
  89
  90 struct mount_options
  91 {
  92         unsigned long   commit_interval;
  93         unsigned long   mount_opt;
  94         unsigned int    atime_quantum;
  95         signed short    slot;
  96         int             localalloc_opt;
  97         unsigned int    resv_level;
  98         int             dir_resv_level;
  99         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 100 };
 101
 102 static int ocfs2_parse_options(struct super_block *sb, char *options,
 103                                struct mount_options *mopt,
 104                                int is_remount);
 105 static int ocfs2_check_set_options(struct super_block *sb,
 106                                    struct mount_options *options);
 107 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 108 static void ocfs2_put_super(struct super_block *sb);
 109 static int ocfs2_mount_volume(struct super_block *sb);
 110 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 111 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 112 static int ocfs2_initialize_mem_caches(void);
 113 static void ocfs2_free_mem_caches(void);
 114 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 115
 116 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 117
 118 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 119
 120 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 121 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 122 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 123 static int ocfs2_check_volume(struct ocfs2_super *osb);
 124 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 125                                struct buffer_head *bh,
 126                                u32 sectsize,
 127                                struct ocfs2_blockcheck_stats *stats);
 128 static int ocfs2_initialize_super(struct super_block *sb,
 129                                   struct buffer_head *bh,
 130                                   int sector_size,
 131                                   struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_get_sector(struct super_block *sb,
 133                             struct buffer_head **bh,
 134                             int block,
 135                             int sect_size);
 136 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 137 static void ocfs2_destroy_inode(struct inode *inode);
 138 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 139 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 140 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 141
 142 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 143 {
 144         return OCFS2_I(inode)->i_dquot;
 145 }
 146
 147 static const struct super_operations ocfs2_sops = {
 148         .statfs         = ocfs2_statfs,
 149         .alloc_inode    = ocfs2_alloc_inode,
 150         .destroy_inode  = ocfs2_destroy_inode,
 151         .drop_inode     = ocfs2_drop_inode,
 152         .evict_inode    = ocfs2_evict_inode,
 153         .sync_fs        = ocfs2_sync_fs,
 154         .put_super      = ocfs2_put_super,
 155         .remount_fs     = ocfs2_remount,
 156         .show_options   = ocfs2_show_options,
 157         .quota_read     = ocfs2_quota_read,
 158         .quota_write    = ocfs2_quota_write,
 159         .get_dquots     = ocfs2_get_dquots,
 160 };
 161
 162 enum {
 163         Opt_barrier,
 164         Opt_err_panic,
 165         Opt_err_ro,
 166         Opt_intr,
 167         Opt_nointr,
 168         Opt_hb_none,
 169         Opt_hb_local,
 170         Opt_hb_global,
 171         Opt_data_ordered,
 172         Opt_data_writeback,
 173         Opt_atime_quantum,
 174         Opt_slot,
 175         Opt_commit,
 176         Opt_localalloc,
 177         Opt_localflocks,
 178         Opt_stack,
 179         Opt_user_xattr,
 180         Opt_nouser_xattr,
 181         Opt_inode64,
 182         Opt_acl,
 183         Opt_noacl,
 184         Opt_usrquota,
 185         Opt_grpquota,
 186         Opt_coherency_buffered,
 187         Opt_coherency_full,
 188         Opt_resv_level,
 189         Opt_dir_resv_level,
 190         Opt_journal_async_commit,
 191         Opt_err_cont,
 192         Opt_err,
 193 };
 194
 195 static const match_table_t tokens = {
 196         {Opt_barrier, "barrier=%u"},
 197         {Opt_err_panic, "errors=panic"},
 198         {Opt_err_ro, "errors=remount-ro"},
 199         {Opt_intr, "intr"},
 200         {Opt_nointr, "nointr"},
 201         {Opt_hb_none, OCFS2_HB_NONE},
 202         {Opt_hb_local, OCFS2_HB_LOCAL},
 203         {Opt_hb_global, OCFS2_HB_GLOBAL},
 204         {Opt_data_ordered, "data=ordered"},
 205         {Opt_data_writeback, "data=writeback"},
 206         {Opt_atime_quantum, "atime_quantum=%u"},
 207         {Opt_slot, "preferred_slot=%u"},
 208         {Opt_commit, "commit=%u"},
 209         {Opt_localalloc, "localalloc=%d"},
 210         {Opt_localflocks, "localflocks"},
 211         {Opt_stack, "cluster_stack=%s"},
 212         {Opt_user_xattr, "user_xattr"},
 213         {Opt_nouser_xattr, "nouser_xattr"},
 214         {Opt_inode64, "inode64"},
 215         {Opt_acl, "acl"},
 216         {Opt_noacl, "noacl"},
 217         {Opt_usrquota, "usrquota"},
 218         {Opt_grpquota, "grpquota"},
 219         {Opt_coherency_buffered, "coherency=buffered"},
 220         {Opt_coherency_full, "coherency=full"},
 221         {Opt_resv_level, "resv_level=%u"},
 222         {Opt_dir_resv_level, "dir_resv_level=%u"},
 223         {Opt_journal_async_commit, "journal_async_commit"},
 224         {Opt_err_cont, "errors=continue"},
 225         {Opt_err, NULL}
 226 };
 227
 228 #ifdef CONFIG_DEBUG_FS
 229 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 230 {
 231         struct ocfs2_cluster_connection *cconn = osb->cconn;
 232         struct ocfs2_recovery_map *rm = osb->recovery_map;
 233         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 234         int i, out = 0;
 235         unsigned long flags;
 236
 237         out += snprintf(buf + out, len - out,
 238                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 239                         "Device", osb->dev_str, osb->uuid_str,
 240                         osb->fs_generation, osb->vol_label);
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 244                         atomic_read(&osb->vol_state), osb->osb_flags);
 245
 246         out += snprintf(buf + out, len - out,
 247                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 248                         osb->sb->s_blocksize, osb->s_clustersize);
 249
 250         out += snprintf(buf + out, len - out,
 251                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 252                         "ROcompat: 0x%X\n",
 253                         "Features", osb->s_feature_compat,
 254                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 255
 256         out += snprintf(buf + out, len - out,
 257                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 258                         osb->s_mount_opt, osb->s_atime_quantum);
 259
 260         if (cconn) {
 261                 out += snprintf(buf + out, len - out,
 262                                 "%10s => Stack: %s  Name: %*s  "
 263                                 "Version: %d.%d\n", "Cluster",
 264                                 (*osb->osb_cluster_stack == '\0' ?
 265                                  "o2cb" : osb->osb_cluster_stack),
 266                                 cconn->cc_namelen, cconn->cc_name,
 267                                 cconn->cc_version.pv_major,
 268                                 cconn->cc_version.pv_minor);
 269         }
 270
 271         spin_lock_irqsave(&osb->dc_task_lock, flags);
 272         out += snprintf(buf + out, len - out,
 273                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 274                         "WorkSeq: %lu\n", "DownCnvt",
 275                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 276                         osb->blocked_lock_count, osb->dc_wake_sequence,
 277                         osb->dc_work_sequence);
 278         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 279
 280         spin_lock(&osb->osb_lock);
 281         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 282                         "Recovery",
 283                         (osb->recovery_thread_task ?
 284                          task_pid_nr(osb->recovery_thread_task) : -1));
 285         if (rm->rm_used == 0)
 286                 out += snprintf(buf + out, len - out, " None\n");
 287         else {
 288                 for (i = 0; i < rm->rm_used; i++)
 289                         out += snprintf(buf + out, len - out, " %d",
 290                                         rm->rm_entries[i]);
 291                 out += snprintf(buf + out, len - out, "\n");
 292         }
 293         spin_unlock(&osb->osb_lock);
 294
 295         out += snprintf(buf + out, len - out,
 296                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 297                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 298                         osb->osb_commit_interval);
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 302                         "Journal", osb->journal->j_state,
 303                         osb->journal->j_trans_id,
 304                         atomic_read(&osb->journal->j_num_trans));
 305
 306         out += snprintf(buf + out, len - out,
 307                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 308                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 309                         "Stats",
 310                         atomic_read(&osb->alloc_stats.bitmap_data),
 311                         atomic_read(&osb->alloc_stats.local_data),
 312                         atomic_read(&osb->alloc_stats.bg_allocs),
 313                         atomic_read(&osb->alloc_stats.moves),
 314                         atomic_read(&osb->alloc_stats.bg_extends));
 315
 316         out += snprintf(buf + out, len - out,
 317                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 318                         "Default: %u bits\n",
 319                         "LocalAlloc", osb->local_alloc_state,
 320                         (unsigned long long)osb->la_last_gd,
 321                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 322
 323         spin_lock(&osb->osb_lock);
 324         out += snprintf(buf + out, len - out,
 325                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 326                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 327                         osb->s_inode_steal_slot,
 328                         atomic_read(&osb->s_num_inodes_stolen),
 329                         osb->s_meta_steal_slot,
 330                         atomic_read(&osb->s_num_meta_stolen));
 331         spin_unlock(&osb->osb_lock);
 332
 333         out += snprintf(buf + out, len - out, "OrphanScan => ");
 334         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 335                         os->os_count, os->os_seqno);
 336         out += snprintf(buf + out, len - out, " Last Scan: ");
 337         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 338                 out += snprintf(buf + out, len - out, "Disabled\n");
 339         else
 340                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 341                                 (unsigned long)(ktime_get_seconds() - os->os_scantime));
 342
 343         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 344                         "Slots", "Num", "RecoGen");
 345         for (i = 0; i < osb->max_slots; ++i) {
 346                 out += snprintf(buf + out, len - out,
 347                                 "%10s  %c %3d  %10d\n",
 348                                 " ",
 349                                 (i == osb->slot_num ? '*' : ' '),
 350                                 i, osb->slot_recovery_generations[i]);
 351         }
 352
 353         return out;
 354 }
 355
 356 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 357 {
 358         struct ocfs2_super *osb = inode->i_private;
 359         char *buf = NULL;
 360
 361         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 362         if (!buf)
 363                 goto bail;
 364
 365         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 366
 367         file->private_data = buf;
 368
 369         return 0;
 370 bail:
 371         return -ENOMEM;
 372 }
 373
 374 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 375 {
 376         kfree(file->private_data);
 377         return 0;
 378 }
 379
 380 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 381                                 size_t nbytes, loff_t *ppos)
 382 {
 383         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 384                                        i_size_read(file->f_mapping->host));
 385 }
 386 #else
 387 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 388 {
 389         return 0;
 390 }
 391 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 392 {
 393         return 0;
 394 }
 395 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 396                                 size_t nbytes, loff_t *ppos)
 397 {
 398         return 0;
 399 }
 400 #endif  /* CONFIG_DEBUG_FS */
 401
 402 static const struct file_operations ocfs2_osb_debug_fops = {
 403         .open =         ocfs2_osb_debug_open,
 404         .release =      ocfs2_debug_release,
 405         .read =         ocfs2_debug_read,
 406         .llseek =       generic_file_llseek,
 407 };
 408
 409 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 410 {
 411         int status;
 412         tid_t target;
 413         struct ocfs2_super *osb = OCFS2_SB(sb);
 414
 415         if (ocfs2_is_hard_readonly(osb))
 416                 return -EROFS;
 417
 418         if (wait) {
 419                 status = ocfs2_flush_truncate_log(osb);
 420                 if (status < 0)
 421                         mlog_errno(status);
 422         } else {
 423                 ocfs2_schedule_truncate_log_flush(osb, 0);
 424         }
 425
 426         if (jbd2_journal_start_commit(osb->journal->j_journal,
 427                                       &target)) {
 428                 if (wait)
 429                         jbd2_log_wait_commit(osb->journal->j_journal,
 430                                              target);
 431         }
 432         return 0;
 433 }
 434
 435 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 436 {
 437         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 438             && (ino == USER_QUOTA_SYSTEM_INODE
 439                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 440                 return 0;
 441         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 442             && (ino == GROUP_QUOTA_SYSTEM_INODE
 443                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 444                 return 0;
 445         return 1;
 446 }
 447
 448 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 449 {
 450         struct inode *new = NULL;
 451         int status = 0;
 452         int i;
 453
 454         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 455         if (IS_ERR(new)) {
 456                 status = PTR_ERR(new);
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460         osb->root_inode = new;
 461
 462         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 463         if (IS_ERR(new)) {
 464                 status = PTR_ERR(new);
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468         osb->sys_root_inode = new;
 469
 470         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 471              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 472                 if (!ocfs2_need_system_inode(osb, i))
 473                         continue;
 474                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 475                 if (!new) {
 476                         ocfs2_release_system_inodes(osb);
 477                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 478                         mlog_errno(status);
 479                         mlog(ML_ERROR, "Unable to load system inode %d, "
 480                              "possibly corrupt fs?", i);
 481                         goto bail;
 482                 }
 483                 // the array now has one ref, so drop this one
 484                 iput(new);
 485         }
 486
 487 bail:
 488         if (status)
 489                 mlog_errno(status);
 490         return status;
 491 }
 492
 493 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 494 {
 495         struct inode *new = NULL;
 496         int status = 0;
 497         int i;
 498
 499         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 500              i < NUM_SYSTEM_INODES;
 501              i++) {
 502                 if (!ocfs2_need_system_inode(osb, i))
 503                         continue;
 504                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 505                 if (!new) {
 506                         ocfs2_release_system_inodes(osb);
 507                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 508                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 509                              status, i, osb->slot_num);
 510                         goto bail;
 511                 }
 512                 /* the array now has one ref, so drop this one */
 513                 iput(new);
 514         }
 515
 516 bail:
 517         if (status)
 518                 mlog_errno(status);
 519         return status;
 520 }
 521
 522 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 523 {
 524         int i;
 525         struct inode *inode;
 526
 527         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 528                 inode = osb->global_system_inodes[i];
 529                 if (inode) {
 530                         iput(inode);
 531                         osb->global_system_inodes[i] = NULL;
 532                 }
 533         }
 534
 535         inode = osb->sys_root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->sys_root_inode = NULL;
 539         }
 540
 541         inode = osb->root_inode;
 542         if (inode) {
 543                 iput(inode);
 544                 osb->root_inode = NULL;
 545         }
 546
 547         if (!osb->local_system_inodes)
 548                 return;
 549
 550         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 551                 if (osb->local_system_inodes[i]) {
 552                         iput(osb->local_system_inodes[i]);
 553                         osb->local_system_inodes[i] = NULL;
 554                 }
 555         }
 556
 557         kfree(osb->local_system_inodes);
 558         osb->local_system_inodes = NULL;
 559 }
 560
 561 /* We're allocating fs objects, use GFP_NOFS */
 562 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 563 {
 564         struct ocfs2_inode_info *oi;
 565
 566         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 567         if (!oi)
 568                 return NULL;
 569
 570         oi->i_sync_tid = 0;
 571         oi->i_datasync_tid = 0;
 572         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 573
 574         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 575         return &oi->vfs_inode;
 576 }
 577
 578 static void ocfs2_i_callback(struct rcu_head *head)
 579 {
 580         struct inode *inode = container_of(head, struct inode, i_rcu);
 581         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 582 }
 583
 584 static void ocfs2_destroy_inode(struct inode *inode)
 585 {
 586         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 587 }
 588
 589 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 590                                                 unsigned int cbits)
 591 {
 592         unsigned int bytes = 1 << cbits;
 593         unsigned int trim = bytes;
 594         unsigned int bitshift = 32;
 595
 596         /*
 597          * i_size and all block offsets in ocfs2 are always 64 bits
 598          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 599          * 64 bit platforms, cluster size will be the limiting factor.
 600          */
 601
 602 #if BITS_PER_LONG == 32
 603 # if defined(CONFIG_LBDAF)
 604         BUILD_BUG_ON(sizeof(sector_t) != 8);
 605         /*
 606          * We might be limited by page cache size.
 607          */
 608         if (bytes > PAGE_SIZE) {
 609                 bytes = PAGE_SIZE;
 610                 trim = 1;
 611                 /*
 612                  * Shift by 31 here so that we don't get larger than
 613                  * MAX_LFS_FILESIZE
 614                  */
 615                 bitshift = 31;
 616         }
 617 # else
 618         /*
 619          * We are limited by the size of sector_t. Use block size, as
 620          * that's what we expose to the VFS.
 621          */
 622         bytes = 1 << bbits;
 623         trim = 1;
 624         bitshift = 31;
 625 # endif
 626 #endif
 627
 628         /*
 629          * Trim by a whole cluster when we can actually approach the
 630          * on-disk limits. Otherwise we can overflow i_clusters when
 631          * an extent start is at the max offset.
 632          */
 633         return (((unsigned long long)bytes) << bitshift) - trim;
 634 }
 635
 636 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 637 {
 638         int incompat_features;
 639         int ret = 0;
 640         struct mount_options parsed_options;
 641         struct ocfs2_super *osb = OCFS2_SB(sb);
 642         u32 tmp;
 643
 644         sync_filesystem(sb);
 645
 646         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 647             !ocfs2_check_set_options(sb, &parsed_options)) {
 648                 ret = -EINVAL;
 649                 goto out;
 650         }
 651
 652         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 653                 OCFS2_MOUNT_HB_NONE;
 654         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 657                 goto out;
 658         }
 659
 660         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 661             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 662                 ret = -EINVAL;
 663                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 664                 goto out;
 665         }
 666
 667         /* Probably don't want this on remount; it might
 668          * mess with other nodes */
 669         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 670             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 671                 ret = -EINVAL;
 672                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 673                 goto out;
 674         }
 675
 676         /* We're going to/from readonly mode. */
 677         if ((bool)(*flags & SB_RDONLY) != sb_rdonly(sb)) {
 678                 /* Disable quota accounting before remounting RO */
 679                 if (*flags & SB_RDONLY) {
 680                         ret = ocfs2_susp_quotas(osb, 0);
 681                         if (ret < 0)
 682                                 goto out;
 683                 }
 684                 /* Lock here so the check of HARD_RO and the potential
 685                  * setting of SOFT_RO is atomic. */
 686                 spin_lock(&osb->osb_lock);
 687                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 688                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 689                         ret = -EROFS;
 690                         goto unlock_osb;
 691                 }
 692
 693                 if (*flags & SB_RDONLY) {
 694                         sb->s_flags |= SB_RDONLY;
 695                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 696                 } else {
 697                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 698                                 mlog(ML_ERROR, "Cannot remount RDWR "
 699                                      "filesystem due to previous errors.\n");
 700                                 ret = -EROFS;
 701                                 goto unlock_osb;
 702                         }
 703                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 704                         if (incompat_features) {
 705                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 706                                      "of unsupported optional features "
 707                                      "(%x).\n", incompat_features);
 708                                 ret = -EINVAL;
 709                                 goto unlock_osb;
 710                         }
 711                         sb->s_flags &= ~SB_RDONLY;
 712                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 713                 }
 714                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 715 unlock_osb:
 716                 spin_unlock(&osb->osb_lock);
 717                 /* Enable quota accounting after remounting RW */
 718                 if (!ret && !(*flags & SB_RDONLY)) {
 719                         if (sb_any_quota_suspended(sb))
 720                                 ret = ocfs2_susp_quotas(osb, 1);
 721                         else
 722                                 ret = ocfs2_enable_quotas(osb);
 723                         if (ret < 0) {
 724                                 /* Return back changes... */
 725                                 spin_lock(&osb->osb_lock);
 726                                 sb->s_flags |= SB_RDONLY;
 727                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 728                                 spin_unlock(&osb->osb_lock);
 729                                 goto out;
 730                         }
 731                 }
 732         }
 733
 734         if (!ret) {
 735                 /* Only save off the new mount options in case of a successful
 736                  * remount. */
 737                 osb->s_mount_opt = parsed_options.mount_opt;
 738                 osb->s_atime_quantum = parsed_options.atime_quantum;
 739                 osb->preferred_slot = parsed_options.slot;
 740                 if (parsed_options.commit_interval)
 741                         osb->osb_commit_interval = parsed_options.commit_interval;
 742
 743                 if (!ocfs2_is_hard_readonly(osb))
 744                         ocfs2_set_journal_params(osb);
 745
 746                 sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
 747                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 748                                                         SB_POSIXACL : 0);
 749         }
 750 out:
 751         return ret;
 752 }
 753
 754 static int ocfs2_sb_probe(struct super_block *sb,
 755                           struct buffer_head **bh,
 756                           int *sector_size,
 757                           struct ocfs2_blockcheck_stats *stats)
 758 {
 759         int status, tmpstat;
 760         struct ocfs1_vol_disk_hdr *hdr;
 761         struct ocfs2_dinode *di;
 762         int blksize;
 763
 764         *bh = NULL;
 765
 766         /* may be > 512 */
 767         *sector_size = bdev_logical_block_size(sb->s_bdev);
 768         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 769                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 770                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 771                 status = -EINVAL;
 772                 goto bail;
 773         }
 774
 775         /* Can this really happen? */
 776         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 777                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 778
 779         /* check block zero for old format */
 780         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 781         if (status < 0) {
 782                 mlog_errno(status);
 783                 goto bail;
 784         }
 785         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 786         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 787                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 788                      hdr->major_version, hdr->minor_version);
 789                 status = -EINVAL;
 790         }
 791         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 792                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 793                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 794                      hdr->signature);
 795                 status = -EINVAL;
 796         }
 797         brelse(*bh);
 798         *bh = NULL;
 799         if (status < 0) {
 800                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 801                      "upgraded before mounting with ocfs v2\n");
 802                 goto bail;
 803         }
 804
 805         /*
 806          * Now check at magic offset for 512, 1024, 2048, 4096
 807          * blocksizes.  4096 is the maximum blocksize because it is
 808          * the minimum clustersize.
 809          */
 810         status = -EINVAL;
 811         for (blksize = *sector_size;
 812              blksize <= OCFS2_MAX_BLOCKSIZE;
 813              blksize <<= 1) {
 814                 tmpstat = ocfs2_get_sector(sb, bh,
 815                                            OCFS2_SUPER_BLOCK_BLKNO,
 816                                            blksize);
 817                 if (tmpstat < 0) {
 818                         status = tmpstat;
 819                         mlog_errno(status);
 820                         break;
 821                 }
 822                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 823                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 824                 spin_lock_init(&stats->b_lock);
 825                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 826                 if (tmpstat < 0) {
 827                         brelse(*bh);
 828                         *bh = NULL;
 829                 }
 830                 if (tmpstat != -EAGAIN) {
 831                         status = tmpstat;
 832                         break;
 833                 }
 834         }
 835
 836 bail:
 837         return status;
 838 }
 839
 840 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 841 {
 842         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 843
 844         if (osb->s_mount_opt & hb_enabled) {
 845                 if (ocfs2_mount_local(osb)) {
 846                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 847                              "mounted device.\n");
 848                         return -EINVAL;
 849                 }
 850                 if (ocfs2_userspace_stack(osb)) {
 851                         mlog(ML_ERROR, "Userspace stack expected, but "
 852                              "o2cb heartbeat arguments passed to mount\n");
 853                         return -EINVAL;
 854                 }
 855                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 856                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 857                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 858                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 859                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 860                         return -EINVAL;
 861                 }
 862         }
 863
 864         if (!(osb->s_mount_opt & hb_enabled)) {
 865                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 866                     !ocfs2_userspace_stack(osb)) {
 867                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 868                              "a read-write clustered device.\n");
 869                         return -EINVAL;
 870                 }
 871         }
 872
 873         return 0;
 874 }
 875
 876 /*
 877  * If we're using a userspace stack, mount should have passed
 878  * a name that matches the disk.  If not, mount should not
 879  * have passed a stack.
 880  */
 881 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 882                                         struct mount_options *mopt)
 883 {
 884         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 885                 mlog(ML_ERROR,
 886                      "cluster stack passed to mount, but this filesystem "
 887                      "does not support it\n");
 888                 return -EINVAL;
 889         }
 890
 891         if (ocfs2_userspace_stack(osb) &&
 892             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 893                     OCFS2_STACK_LABEL_LEN)) {
 894                 mlog(ML_ERROR,
 895                      "cluster stack passed to mount (\"%s\") does not "
 896                      "match the filesystem (\"%s\")\n",
 897                      mopt->cluster_stack,
 898                      osb->osb_cluster_stack);
 899                 return -EINVAL;
 900         }
 901
 902         return 0;
 903 }
 904
 905 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 906 {
 907         int type;
 908         struct super_block *sb = osb->sb;
 909         unsigned int feature[OCFS2_MAXQUOTAS] = {
 910                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 911                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 912         int status = 0;
 913
 914         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 915                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 916                         continue;
 917                 if (unsuspend)
 918                         status = dquot_resume(sb, type);
 919                 else {
 920                         struct ocfs2_mem_dqinfo *oinfo;
 921
 922                         /* Cancel periodic syncing before suspending */
 923                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 924                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 925                         status = dquot_suspend(sb, type);
 926                 }
 927                 if (status < 0)
 928                         break;
 929         }
 930         if (status < 0)
 931                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 932                      "remount (error = %d).\n", status);
 933         return status;
 934 }
 935
 936 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 937 {
 938         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 939         struct super_block *sb = osb->sb;
 940         unsigned int feature[OCFS2_MAXQUOTAS] = {
 941                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 942                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 943         unsigned int ino[OCFS2_MAXQUOTAS] = {
 944                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 945                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 946         int status;
 947         int type;
 948
 949         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 950         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 951                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 952                         continue;
 953                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 954                                                         osb->slot_num);
 955                 if (!inode[type]) {
 956                         status = -ENOENT;
 957                         goto out_quota_off;
 958                 }
 959                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 960                                       DQUOT_USAGE_ENABLED);
 961                 if (status < 0)
 962                         goto out_quota_off;
 963         }
 964
 965         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 966                 iput(inode[type]);
 967         return 0;
 968 out_quota_off:
 969         ocfs2_disable_quotas(osb);
 970         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 971                 iput(inode[type]);
 972         mlog_errno(status);
 973         return status;
 974 }
 975
 976 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 977 {
 978         int type;
 979         struct inode *inode;
 980         struct super_block *sb = osb->sb;
 981         struct ocfs2_mem_dqinfo *oinfo;
 982
 983         /* We mostly ignore errors in this function because there's not much
 984          * we can do when we see them */
 985         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 986                 if (!sb_has_quota_loaded(sb, type))
 987                         continue;
 988                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 989                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 990                 inode = igrab(sb->s_dquot.files[type]);
 991                 /* Turn off quotas. This will remove all dquot structures from
 992                  * memory and so they will be automatically synced to global
 993                  * quota files */
 994                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 995                                         DQUOT_LIMITS_ENABLED);
 996                 if (!inode)
 997                         continue;
 998                 iput(inode);
 999         }
1000 }
1001
1002 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1003 {
1004         struct dentry *root;
1005         int status, sector_size;
1006         struct mount_options parsed_options;
1007         struct inode *inode = NULL;
1008         struct ocfs2_super *osb = NULL;
1009         struct buffer_head *bh = NULL;
1010         char nodestr[12];
1011         struct ocfs2_blockcheck_stats stats;
1012
1013         trace_ocfs2_fill_super(sb, data, silent);
1014
1015         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1016                 status = -EINVAL;
1017                 goto read_super_error;
1018         }
1019
1020         /* probe for superblock */
1021         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1022         if (status < 0) {
1023                 mlog(ML_ERROR, "superblock probe failed!\n");
1024                 goto read_super_error;
1025         }
1026
1027         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1028         osb = OCFS2_SB(sb);
1029         if (status < 0) {
1030                 mlog_errno(status);
1031                 goto read_super_error;
1032         }
1033         brelse(bh);
1034         bh = NULL;
1035
1036         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1037                 status = -EINVAL;
1038                 goto read_super_error;
1039         }
1040         osb->s_mount_opt = parsed_options.mount_opt;
1041         osb->s_atime_quantum = parsed_options.atime_quantum;
1042         osb->preferred_slot = parsed_options.slot;
1043         osb->osb_commit_interval = parsed_options.commit_interval;
1044
1045         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1046         osb->osb_resv_level = parsed_options.resv_level;
1047         osb->osb_dir_resv_level = parsed_options.resv_level;
1048         if (parsed_options.dir_resv_level == -1)
1049                 osb->osb_dir_resv_level = parsed_options.resv_level;
1050         else
1051                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1052
1053         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1054         if (status)
1055                 goto read_super_error;
1056
1057         sb->s_magic = OCFS2_SUPER_MAGIC;
1058
1059         sb->s_flags = (sb->s_flags & ~(SB_POSIXACL | SB_NOSEC)) |
1060                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? SB_POSIXACL : 0);
1061
1062         /* Hard readonly mode only if: bdev_read_only, SB_RDONLY,
1063          * heartbeat=none */
1064         if (bdev_read_only(sb->s_bdev)) {
1065                 if (!sb_rdonly(sb)) {
1066                         status = -EACCES;
1067                         mlog(ML_ERROR, "Readonly device detected but readonly "
1068                              "mount was not specified.\n");
1069                         goto read_super_error;
1070                 }
1071
1072                 /* You should not be able to start a local heartbeat
1073                  * on a readonly device. */
1074                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1075                         status = -EROFS;
1076                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1077                              "device.\n");
1078                         goto read_super_error;
1079                 }
1080
1081                 status = ocfs2_check_journals_nolocks(osb);
1082                 if (status < 0) {
1083                         if (status == -EROFS)
1084                                 mlog(ML_ERROR, "Recovery required on readonly "
1085                                      "file system, but write access is "
1086                                      "unavailable.\n");
1087                         else
1088                                 mlog_errno(status);
1089                         goto read_super_error;
1090                 }
1091
1092                 ocfs2_set_ro_flag(osb, 1);
1093
1094                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1095                        "Cluster services will not be used for this mount. "
1096                        "Recovery will be skipped.\n", osb->dev_str);
1097         }
1098
1099         if (!ocfs2_is_hard_readonly(osb)) {
1100                 if (sb_rdonly(sb))
1101                         ocfs2_set_ro_flag(osb, 0);
1102         }
1103
1104         status = ocfs2_verify_heartbeat(osb);
1105         if (status < 0) {
1106                 mlog_errno(status);
1107                 goto read_super_error;
1108         }
1109
1110         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1111                                                  ocfs2_debugfs_root);
1112         if (!osb->osb_debug_root) {
1113                 status = -EINVAL;
1114                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1115                 goto read_super_error;
1116         }
1117
1118         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1119                                             osb->osb_debug_root,
1120                                             osb,
1121                                             &ocfs2_osb_debug_fops);
1122         if (!osb->osb_ctxt) {
1123                 status = -EINVAL;
1124                 mlog_errno(status);
1125                 goto read_super_error;
1126         }
1127
1128         if (ocfs2_meta_ecc(osb)) {
1129                 status = ocfs2_blockcheck_stats_debugfs_install(
1130                                                 &osb->osb_ecc_stats,
1131                                                 osb->osb_debug_root);
1132                 if (status) {
1133                         mlog(ML_ERROR,
1134                              "Unable to create blockcheck statistics "
1135                              "files\n");
1136                         goto read_super_error;
1137                 }
1138         }
1139
1140         status = ocfs2_mount_volume(sb);
1141         if (status < 0)
1142                 goto read_super_error;
1143
1144         if (osb->root_inode)
1145                 inode = igrab(osb->root_inode);
1146
1147         if (!inode) {
1148                 status = -EIO;
1149                 mlog_errno(status);
1150                 goto read_super_error;
1151         }
1152
1153         root = d_make_root(inode);
1154         if (!root) {
1155                 status = -ENOMEM;
1156                 mlog_errno(status);
1157                 goto read_super_error;
1158         }
1159
1160         sb->s_root = root;
1161
1162         ocfs2_complete_mount_recovery(osb);
1163
1164         osb->osb_dev_kset = kset_create_and_add(sb->s_id, NULL,
1165                                                 &ocfs2_kset->kobj);
1166         if (!osb->osb_dev_kset) {
1167                 status = -ENOMEM;
1168                 mlog(ML_ERROR, "Unable to create device kset %s.\n", sb->s_id);
1169                 goto read_super_error;
1170         }
1171
1172         /* Create filecheck sysfs related directories/files at
1173          * /sys/fs/ocfs2/<devname>/filecheck */
1174         if (ocfs2_filecheck_create_sysfs(osb)) {
1175                 status = -ENOMEM;
1176                 mlog(ML_ERROR, "Unable to create filecheck sysfs directory at "
1177                         "/sys/fs/ocfs2/%s/filecheck.\n", sb->s_id);
1178                 goto read_super_error;
1179         }
1180
1181         if (ocfs2_mount_local(osb))
1182                 snprintf(nodestr, sizeof(nodestr), "local");
1183         else
1184                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1185
1186         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1187                "with %s data mode.\n",
1188                osb->dev_str, nodestr, osb->slot_num,
1189                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1190                "ordered");
1191
1192         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1193         wake_up(&osb->osb_mount_event);
1194
1195         /* Now we can initialize quotas because we can afford to wait
1196          * for cluster locks recovery now. That also means that truncation
1197          * log recovery can happen but that waits for proper quota setup */
1198         if (!sb_rdonly(sb)) {
1199                 status = ocfs2_enable_quotas(osb);
1200                 if (status < 0) {
1201                         /* We have to err-out specially here because
1202                          * s_root is already set */
1203                         mlog_errno(status);
1204                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1205                         wake_up(&osb->osb_mount_event);
1206                         return status;
1207                 }
1208         }
1209
1210         ocfs2_complete_quota_recovery(osb);
1211
1212         /* Now we wake up again for processes waiting for quotas */
1213         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1214         wake_up(&osb->osb_mount_event);
1215
1216         /* Start this when the mount is almost sure of being successful */
1217         ocfs2_orphan_scan_start(osb);
1218
1219         return status;
1220
1221 read_super_error:
1222         brelse(bh);
1223
1224         if (status)
1225                 mlog_errno(status);
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         return status;
1234 }
1235
1236 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1237                         int flags,
1238                         const char *dev_name,
1239                         void *data)
1240 {
1241         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1242 }
1243
1244 static struct file_system_type ocfs2_fs_type = {
1245         .owner          = THIS_MODULE,
1246         .name           = "ocfs2",
1247         .mount          = ocfs2_mount,
1248         .kill_sb        = kill_block_super,
1249         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1250         .next           = NULL
1251 };
1252 MODULE_ALIAS_FS("ocfs2");
1253
1254 static int ocfs2_check_set_options(struct super_block *sb,
1255                                    struct mount_options *options)
1256 {
1257         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1258             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1259                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1260                 mlog(ML_ERROR, "User quotas were requested, but this "
1261                      "filesystem does not have the feature enabled.\n");
1262                 return 0;
1263         }
1264         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1265             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1266                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1267                 mlog(ML_ERROR, "Group quotas were requested, but this "
1268                      "filesystem does not have the feature enabled.\n");
1269                 return 0;
1270         }
1271         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1272             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1273                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1274                      "feature is not enabled\n");
1275                 return 0;
1276         }
1277         /* No ACL setting specified? Use XATTR feature... */
1278         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1279                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1280                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1281                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1282                 else
1283                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1284         }
1285         return 1;
1286 }
1287
1288 static int ocfs2_parse_options(struct super_block *sb,
1289                                char *options,
1290                                struct mount_options *mopt,
1291                                int is_remount)
1292 {
1293         int status, user_stack = 0;
1294         char *p;
1295         u32 tmp;
1296         int token, option;
1297         substring_t args[MAX_OPT_ARGS];
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                 if (!*p)
1317                         continue;
1318
1319                 token = match_token(p, tokens, args);
1320                 switch (token) {
1321                 case Opt_hb_local:
1322                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1323                         break;
1324                 case Opt_hb_none:
1325                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1326                         break;
1327                 case Opt_hb_global:
1328                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1329                         break;
1330                 case Opt_barrier:
1331                         if (match_int(&args[0], &option)) {
1332                                 status = 0;
1333                                 goto bail;
1334                         }
1335                         if (option)
1336                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1337                         else
1338                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1339                         break;
1340                 case Opt_intr:
1341                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1342                         break;
1343                 case Opt_nointr:
1344                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1345                         break;
1346                 case Opt_err_panic:
1347                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1348                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1349                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1350                         break;
1351                 case Opt_err_ro:
1352                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1353                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1354                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1355                         break;
1356                 case Opt_err_cont:
1357                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1358                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1359                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1360                         break;
1361                 case Opt_data_ordered:
1362                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1363                         break;
1364                 case Opt_data_writeback:
1365                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1366                         break;
1367                 case Opt_user_xattr:
1368                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1369                         break;
1370                 case Opt_nouser_xattr:
1371                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1372                         break;
1373                 case Opt_atime_quantum:
1374                         if (match_int(&args[0], &option)) {
1375                                 status = 0;
1376                                 goto bail;
1377                         }
1378                         if (option >= 0)
1379                                 mopt->atime_quantum = option;
1380                         break;
1381                 case Opt_slot:
1382                         if (match_int(&args[0], &option)) {
1383                                 status = 0;
1384                                 goto bail;
1385                         }
1386                         if (option)
1387                                 mopt->slot = (s16)option;
1388                         break;
1389                 case Opt_commit:
1390                         if (match_int(&args[0], &option)) {
1391                                 status = 0;
1392                                 goto bail;
1393                         }
1394                         if (option < 0)
1395                                 return 0;
1396                         if (option == 0)
1397                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1398                         mopt->commit_interval = HZ * option;
1399                         break;
1400                 case Opt_localalloc:
1401                         if (match_int(&args[0], &option)) {
1402                                 status = 0;
1403                                 goto bail;
1404                         }
1405                         if (option >= 0)
1406                                 mopt->localalloc_opt = option;
1407                         break;
1408                 case Opt_localflocks:
1409                         /*
1410                          * Changing this during remount could race
1411                          * flock() requests, or "unbalance" existing
1412                          * ones (e.g., a lock is taken in one mode but
1413                          * dropped in the other). If users care enough
1414                          * to flip locking modes during remount, we
1415                          * could add a "local" flag to individual
1416                          * flock structures for proper tracking of
1417                          * state.
1418                          */
1419                         if (!is_remount)
1420                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1421                         break;
1422                 case Opt_stack:
1423                         /* Check both that the option we were passed
1424                          * is of the right length and that it is a proper
1425                          * string of the right length.
1426                          */
1427                         if (((args[0].to - args[0].from) !=
1428                              OCFS2_STACK_LABEL_LEN) ||
1429                             (strnlen(args[0].from,
1430                                      OCFS2_STACK_LABEL_LEN) !=
1431                              OCFS2_STACK_LABEL_LEN)) {
1432                                 mlog(ML_ERROR,
1433                                      "Invalid cluster_stack option\n");
1434                                 status = 0;
1435                                 goto bail;
1436                         }
1437                         memcpy(mopt->cluster_stack, args[0].from,
1438                                OCFS2_STACK_LABEL_LEN);
1439                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1440                         /*
1441                          * Open code the memcmp here as we don't have
1442                          * an osb to pass to
1443                          * ocfs2_userspace_stack().
1444                          */
1445                         if (memcmp(mopt->cluster_stack,
1446                                    OCFS2_CLASSIC_CLUSTER_STACK,
1447                                    OCFS2_STACK_LABEL_LEN))
1448                                 user_stack = 1;
1449                         break;
1450                 case Opt_inode64:
1451                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1452                         break;
1453                 case Opt_usrquota:
1454                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1455                         break;
1456                 case Opt_grpquota:
1457                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1458                         break;
1459                 case Opt_coherency_buffered:
1460                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1461                         break;
1462                 case Opt_coherency_full:
1463                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1464                         break;
1465                 case Opt_acl:
1466                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1467                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1468                         break;
1469                 case Opt_noacl:
1470                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1471                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1472                         break;
1473                 case Opt_resv_level:
1474                         if (is_remount)
1475                                 break;
1476                         if (match_int(&args[0], &option)) {
1477                                 status = 0;
1478                                 goto bail;
1479                         }
1480                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1481                             option < OCFS2_MAX_RESV_LEVEL)
1482                                 mopt->resv_level = option;
1483                         break;
1484                 case Opt_dir_resv_level:
1485                         if (is_remount)
1486                                 break;
1487                         if (match_int(&args[0], &option)) {
1488                                 status = 0;
1489                                 goto bail;
1490                         }
1491                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1492                             option < OCFS2_MAX_RESV_LEVEL)
1493                                 mopt->dir_resv_level = option;
1494                         break;
1495                 case Opt_journal_async_commit:
1496                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1497                         break;
1498                 default:
1499                         mlog(ML_ERROR,
1500                              "Unrecognized mount option \"%s\" "
1501                              "or missing value\n", p);
1502                         status = 0;
1503                         goto bail;
1504                 }
1505         }
1506
1507         if (user_stack == 0) {
1508                 /* Ensure only one heartbeat mode */
1509                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1510                                          OCFS2_MOUNT_HB_GLOBAL |
1511                                          OCFS2_MOUNT_HB_NONE);
1512                 if (hweight32(tmp) != 1) {
1513                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1514                         status = 0;
1515                         goto bail;
1516                 }
1517         }
1518
1519         status = 1;
1520
1521 bail:
1522         return status;
1523 }
1524
1525 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1526 {
1527         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1528         unsigned long opts = osb->s_mount_opt;
1529         unsigned int local_alloc_megs;
1530
1531         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1532                 seq_printf(s, ",_netdev");
1533                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1534                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1535                 else
1536                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1537         } else
1538                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1539
1540         if (opts & OCFS2_MOUNT_NOINTR)
1541                 seq_printf(s, ",nointr");
1542
1543         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1544                 seq_printf(s, ",data=writeback");
1545         else
1546                 seq_printf(s, ",data=ordered");
1547
1548         if (opts & OCFS2_MOUNT_BARRIER)
1549                 seq_printf(s, ",barrier=1");
1550
1551         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1552                 seq_printf(s, ",errors=panic");
1553         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1554                 seq_printf(s, ",errors=continue");
1555         else
1556                 seq_printf(s, ",errors=remount-ro");
1557
1558         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1559                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1560
1561         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1562
1563         if (osb->osb_commit_interval)
1564                 seq_printf(s, ",commit=%u",
1565                            (unsigned) (osb->osb_commit_interval / HZ));
1566
1567         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1568         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1569                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1570
1571         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1572                 seq_printf(s, ",localflocks,");
1573
1574         if (osb->osb_cluster_stack[0])
1575                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1576                                   OCFS2_STACK_LABEL_LEN);
1577         if (opts & OCFS2_MOUNT_USRQUOTA)
1578                 seq_printf(s, ",usrquota");
1579         if (opts & OCFS2_MOUNT_GRPQUOTA)
1580                 seq_printf(s, ",grpquota");
1581
1582         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1583                 seq_printf(s, ",coherency=buffered");
1584         else
1585                 seq_printf(s, ",coherency=full");
1586
1587         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1588                 seq_printf(s, ",nouser_xattr");
1589         else
1590                 seq_printf(s, ",user_xattr");
1591
1592         if (opts & OCFS2_MOUNT_INODE64)
1593                 seq_printf(s, ",inode64");
1594
1595         if (opts & OCFS2_MOUNT_POSIX_ACL)
1596                 seq_printf(s, ",acl");
1597         else
1598                 seq_printf(s, ",noacl");
1599
1600         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1601                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1602
1603         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1604                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1605
1606         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1607                 seq_printf(s, ",journal_async_commit");
1608
1609         return 0;
1610 }
1611
1612 static int __init ocfs2_init(void)
1613 {
1614         int status;
1615
1616         status = init_ocfs2_uptodate_cache();
1617         if (status < 0)
1618                 goto out1;
1619
1620         status = ocfs2_initialize_mem_caches();
1621         if (status < 0)
1622                 goto out2;
1623
1624         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1625         if (!ocfs2_debugfs_root) {
1626                 status = -ENOMEM;
1627                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1628                 goto out3;
1629         }
1630
1631         ocfs2_set_locking_protocol();
1632
1633         status = register_quota_format(&ocfs2_quota_format);
1634         if (status < 0)
1635                 goto out3;
1636         status = register_filesystem(&ocfs2_fs_type);
1637         if (!status)
1638                 return 0;
1639
1640         unregister_quota_format(&ocfs2_quota_format);
1641 out3:
1642         debugfs_remove(ocfs2_debugfs_root);
1643         ocfs2_free_mem_caches();
1644 out2:
1645         exit_ocfs2_uptodate_cache();
1646 out1:
1647         mlog_errno(status);
1648         return status;
1649 }
1650
1651 static void __exit ocfs2_exit(void)
1652 {
1653         unregister_quota_format(&ocfs2_quota_format);
1654
1655         debugfs_remove(ocfs2_debugfs_root);
1656
1657         ocfs2_free_mem_caches();
1658
1659         unregister_filesystem(&ocfs2_fs_type);
1660
1661         exit_ocfs2_uptodate_cache();
1662 }
1663
1664 static void ocfs2_put_super(struct super_block *sb)
1665 {
1666         trace_ocfs2_put_super(sb);
1667
1668         ocfs2_sync_blockdev(sb);
1669         ocfs2_dismount_volume(sb, 0);
1670 }
1671
1672 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1673 {
1674         struct ocfs2_super *osb;
1675         u32 numbits, freebits;
1676         int status;
1677         struct ocfs2_dinode *bm_lock;
1678         struct buffer_head *bh = NULL;
1679         struct inode *inode = NULL;
1680
1681         trace_ocfs2_statfs(dentry->d_sb, buf);
1682
1683         osb = OCFS2_SB(dentry->d_sb);
1684
1685         inode = ocfs2_get_system_file_inode(osb,
1686                                             GLOBAL_BITMAP_SYSTEM_INODE,
1687                                             OCFS2_INVALID_SLOT);
1688         if (!inode) {
1689                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1690                 status = -EIO;
1691                 goto bail;
1692         }
1693
1694         status = ocfs2_inode_lock(inode, &bh, 0);
1695         if (status < 0) {
1696                 mlog_errno(status);
1697                 goto bail;
1698         }
1699
1700         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1701
1702         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1703         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1704
1705         buf->f_type = OCFS2_SUPER_MAGIC;
1706         buf->f_bsize = dentry->d_sb->s_blocksize;
1707         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1708         buf->f_blocks = ((sector_t) numbits) *
1709                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1710         buf->f_bfree = ((sector_t) freebits) *
1711                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1712         buf->f_bavail = buf->f_bfree;
1713         buf->f_files = numbits;
1714         buf->f_ffree = freebits;
1715         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1716                                 & 0xFFFFFFFFUL;
1717         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1718                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1719
1720         brelse(bh);
1721
1722         ocfs2_inode_unlock(inode, 0);
1723         status = 0;
1724 bail:
1725         iput(inode);
1726
1727         if (status)
1728                 mlog_errno(status);
1729
1730         return status;
1731 }
1732
1733 static void ocfs2_inode_init_once(void *data)
1734 {
1735         struct ocfs2_inode_info *oi = data;
1736
1737         oi->ip_flags = 0;
1738         oi->ip_open_count = 0;
1739         spin_lock_init(&oi->ip_lock);
1740         ocfs2_extent_map_init(&oi->vfs_inode);
1741         INIT_LIST_HEAD(&oi->ip_io_markers);
1742         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1743         oi->ip_dir_start_lookup = 0;
1744         init_rwsem(&oi->ip_alloc_sem);
1745         init_rwsem(&oi->ip_xattr_sem);
1746         mutex_init(&oi->ip_io_mutex);
1747
1748         oi->ip_blkno = 0ULL;
1749         oi->ip_clusters = 0;
1750
1751         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1752
1753         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1754         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1755         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1756
1757         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1758                                   &ocfs2_inode_caching_ops);
1759
1760         inode_init_once(&oi->vfs_inode);
1761 }
1762
1763 static int ocfs2_initialize_mem_caches(void)
1764 {
1765         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1766                                        sizeof(struct ocfs2_inode_info),
1767                                        0,
1768                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1769                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1770                                        ocfs2_inode_init_once);
1771         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1772                                         sizeof(struct ocfs2_dquot),
1773                                         0,
1774                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1775                                                 SLAB_MEM_SPREAD),
1776                                         NULL);
1777         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1778                                         sizeof(struct ocfs2_quota_chunk),
1779                                         0,
1780                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1781                                         NULL);
1782         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1783             !ocfs2_qf_chunk_cachep) {
1784                 kmem_cache_destroy(ocfs2_inode_cachep);
1785                 kmem_cache_destroy(ocfs2_dquot_cachep);
1786                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1787                 return -ENOMEM;
1788         }
1789
1790         return 0;
1791 }
1792
1793 static void ocfs2_free_mem_caches(void)
1794 {
1795         /*
1796          * Make sure all delayed rcu free inodes are flushed before we
1797          * destroy cache.
1798          */
1799         rcu_barrier();
1800         kmem_cache_destroy(ocfs2_inode_cachep);
1801         ocfs2_inode_cachep = NULL;
1802
1803         kmem_cache_destroy(ocfs2_dquot_cachep);
1804         ocfs2_dquot_cachep = NULL;
1805
1806         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1807         ocfs2_qf_chunk_cachep = NULL;
1808 }
1809
1810 static int ocfs2_get_sector(struct super_block *sb,
1811                             struct buffer_head **bh,
1812                             int block,
1813                             int sect_size)
1814 {
1815         if (!sb_set_blocksize(sb, sect_size)) {
1816                 mlog(ML_ERROR, "unable to set blocksize\n");
1817                 return -EIO;
1818         }
1819
1820         *bh = sb_getblk(sb, block);
1821         if (!*bh) {
1822                 mlog_errno(-ENOMEM);
1823                 return -ENOMEM;
1824         }
1825         lock_buffer(*bh);
1826         if (!buffer_dirty(*bh))
1827                 clear_buffer_uptodate(*bh);
1828         unlock_buffer(*bh);
1829         ll_rw_block(REQ_OP_READ, 0, 1, bh);
1830         wait_on_buffer(*bh);
1831         if (!buffer_uptodate(*bh)) {
1832                 mlog_errno(-EIO);
1833                 brelse(*bh);
1834                 *bh = NULL;
1835                 return -EIO;
1836         }
1837
1838         return 0;
1839 }
1840
1841 static int ocfs2_mount_volume(struct super_block *sb)
1842 {
1843         int status = 0;
1844         int unlock_super = 0;
1845         struct ocfs2_super *osb = OCFS2_SB(sb);
1846
1847         if (ocfs2_is_hard_readonly(osb))
1848                 goto leave;
1849
1850         status = ocfs2_dlm_init(osb);
1851         if (status < 0) {
1852                 mlog_errno(status);
1853                 if (status == -EBADR && ocfs2_userspace_stack(osb))
1854                         mlog(ML_ERROR, "couldn't mount because cluster name on"
1855                         " disk does not match the running cluster name.\n");
1856                 goto leave;
1857         }
1858
1859         status = ocfs2_super_lock(osb, 1);
1860         if (status < 0) {
1861                 mlog_errno(status);
1862                 goto leave;
1863         }
1864         unlock_super = 1;
1865
1866         /* This will load up the node map and add ourselves to it. */
1867         status = ocfs2_find_slot(osb);
1868         if (status < 0) {
1869                 mlog_errno(status);
1870                 goto leave;
1871         }
1872
1873         /* load all node-local system inodes */
1874         status = ocfs2_init_local_system_inodes(osb);
1875         if (status < 0) {
1876                 mlog_errno(status);
1877                 goto leave;
1878         }
1879
1880         status = ocfs2_check_volume(osb);
1881         if (status < 0) {
1882                 mlog_errno(status);
1883                 goto leave;
1884         }
1885
1886         status = ocfs2_truncate_log_init(osb);
1887         if (status < 0)
1888                 mlog_errno(status);
1889
1890 leave:
1891         if (unlock_super)
1892                 ocfs2_super_unlock(osb, 1);
1893
1894         return status;
1895 }
1896
1897 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1898 {
1899         int tmp, hangup_needed = 0;
1900         struct ocfs2_super *osb = NULL;
1901         char nodestr[12];
1902
1903         trace_ocfs2_dismount_volume(sb);
1904
1905         BUG_ON(!sb);
1906         osb = OCFS2_SB(sb);
1907         BUG_ON(!osb);
1908
1909         /* Remove file check sysfs related directores/files,
1910          * and wait for the pending file check operations */
1911         ocfs2_filecheck_remove_sysfs(osb);
1912
1913         kset_unregister(osb->osb_dev_kset);
1914
1915         debugfs_remove(osb->osb_ctxt);
1916
1917         /* Orphan scan should be stopped as early as possible */
1918         ocfs2_orphan_scan_stop(osb);
1919
1920         ocfs2_disable_quotas(osb);
1921
1922         /* All dquots should be freed by now */
1923         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1924         /* Wait for worker to be done with the work structure in osb */
1925         cancel_work_sync(&osb->dquot_drop_work);
1926
1927         ocfs2_shutdown_local_alloc(osb);
1928
1929         ocfs2_truncate_log_shutdown(osb);
1930
1931         /* This will disable recovery and flush any recovery work. */
1932         ocfs2_recovery_exit(osb);
1933
1934         ocfs2_journal_shutdown(osb);
1935
1936         ocfs2_sync_blockdev(sb);
1937
1938         ocfs2_purge_refcount_trees(osb);
1939
1940         /* No cluster connection means we've failed during mount, so skip
1941          * all the steps which depended on that to complete. */
1942         if (osb->cconn) {
1943                 tmp = ocfs2_super_lock(osb, 1);
1944                 if (tmp < 0) {
1945                         mlog_errno(tmp);
1946                         return;
1947                 }
1948         }
1949
1950         if (osb->slot_num != OCFS2_INVALID_SLOT)
1951                 ocfs2_put_slot(osb);
1952
1953         if (osb->cconn)
1954                 ocfs2_super_unlock(osb, 1);
1955
1956         ocfs2_release_system_inodes(osb);
1957
1958         /*
1959          * If we're dismounting due to mount error, mount.ocfs2 will clean
1960          * up heartbeat.  If we're a local mount, there is no heartbeat.
1961          * If we failed before we got a uuid_str yet, we can't stop
1962          * heartbeat.  Otherwise, do it.
1963          */
1964         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1965             !ocfs2_is_hard_readonly(osb))
1966                 hangup_needed = 1;
1967
1968         if (osb->cconn)
1969                 ocfs2_dlm_shutdown(osb, hangup_needed);
1970
1971         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1972         debugfs_remove(osb->osb_debug_root);
1973
1974         if (hangup_needed)
1975                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1976
1977         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1978
1979         if (ocfs2_mount_local(osb))
1980                 snprintf(nodestr, sizeof(nodestr), "local");
1981         else
1982                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1983
1984         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1985                osb->dev_str, nodestr);
1986
1987         ocfs2_delete_osb(osb);
1988         kfree(osb);
1989         sb->s_dev = 0;
1990         sb->s_fs_info = NULL;
1991 }
1992
1993 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1994                                 unsigned uuid_bytes)
1995 {
1996         int i, ret;
1997         char *ptr;
1998
1999         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2000
2001         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2002         if (osb->uuid_str == NULL)
2003                 return -ENOMEM;
2004
2005         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2006                 /* print with null */
2007                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2008                 if (ret != 2) /* drop super cleans up */
2009                         return -EINVAL;
2010                 /* then only advance past the last char */
2011                 ptr += 2;
2012         }
2013
2014         return 0;
2015 }
2016
2017 /* Make sure entire volume is addressable by our journal.  Requires
2018    osb_clusters_at_boot to be valid and for the journal to have been
2019    initialized by ocfs2_journal_init(). */
2020 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2021 {
2022         int status = 0;
2023         u64 max_block =
2024                 ocfs2_clusters_to_blocks(osb->sb,
2025                                          osb->osb_clusters_at_boot) - 1;
2026
2027         /* 32-bit block number is always OK. */
2028         if (max_block <= (u32)~0ULL)
2029                 goto out;
2030
2031         /* Volume is "huge", so see if our journal is new enough to
2032            support it. */
2033         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2034                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2035               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2036                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2037                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2038                      "Enable the 'block64' journal option with tunefs.ocfs2");
2039                 status = -EFBIG;
2040                 goto out;
2041         }
2042
2043  out:
2044         return status;
2045 }
2046
2047 static int ocfs2_initialize_super(struct super_block *sb,
2048                                   struct buffer_head *bh,
2049                                   int sector_size,
2050                                   struct ocfs2_blockcheck_stats *stats)
2051 {
2052         int status;
2053         int i, cbits, bbits;
2054         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2055         struct inode *inode = NULL;
2056         struct ocfs2_journal *journal;
2057         struct ocfs2_super *osb;
2058         u64 total_blocks;
2059
2060         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2061         if (!osb) {
2062                 status = -ENOMEM;
2063                 mlog_errno(status);
2064                 goto bail;
2065         }
2066
2067         sb->s_fs_info = osb;
2068         sb->s_op = &ocfs2_sops;
2069         sb->s_d_op = &ocfs2_dentry_ops;
2070         sb->s_export_op = &ocfs2_export_ops;
2071         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2072         sb->dq_op = &ocfs2_quota_operations;
2073         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2074         sb->s_xattr = ocfs2_xattr_handlers;
2075         sb->s_time_gran = 1;
2076         sb->s_flags |= SB_NOATIME;
2077         /* this is needed to support O_LARGEFILE */
2078         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2079         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2080         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2081         memcpy(&sb->s_uuid, di->id2.i_super.s_uuid,
2082                sizeof(di->id2.i_super.s_uuid));
2083
2084         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2085
2086         for (i = 0; i < 3; i++)
2087                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2088         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2089
2090         osb->sb = sb;
2091         osb->s_sectsize_bits = blksize_bits(sector_size);
2092         BUG_ON(!osb->s_sectsize_bits);
2093
2094         spin_lock_init(&osb->dc_task_lock);
2095         init_waitqueue_head(&osb->dc_event);
2096         osb->dc_work_sequence = 0;
2097         osb->dc_wake_sequence = 0;
2098         INIT_LIST_HEAD(&osb->blocked_lock_list);
2099         osb->blocked_lock_count = 0;
2100         spin_lock_init(&osb->osb_lock);
2101         spin_lock_init(&osb->osb_xattr_lock);
2102         ocfs2_init_steal_slots(osb);
2103
2104         mutex_init(&osb->system_file_mutex);
2105
2106         atomic_set(&osb->alloc_stats.moves, 0);
2107         atomic_set(&osb->alloc_stats.local_data, 0);
2108         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2109         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2110         atomic_set(&osb->alloc_stats.bg_extends, 0);
2111
2112         /* Copy the blockcheck stats from the superblock probe */
2113         osb->osb_ecc_stats = *stats;
2114
2115         ocfs2_init_node_maps(osb);
2116
2117         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2118                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2119
2120         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2121         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2122                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2123                      osb->max_slots);
2124                 status = -EINVAL;
2125                 goto bail;
2126         }
2127
2128         ocfs2_orphan_scan_init(osb);
2129
2130         status = ocfs2_recovery_init(osb);
2131         if (status) {
2132                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2133                 mlog_errno(status);
2134                 goto bail;
2135         }
2136
2137         init_waitqueue_head(&osb->checkpoint_event);
2138
2139         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2140
2141         osb->slot_num = OCFS2_INVALID_SLOT;
2142
2143         osb->s_xattr_inline_size = le16_to_cpu(
2144                                         di->id2.i_super.s_xattr_inline_size);
2145
2146         osb->local_alloc_state = OCFS2_LA_UNUSED;
2147         osb->local_alloc_bh = NULL;
2148         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2149
2150         init_waitqueue_head(&osb->osb_mount_event);
2151
2152         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2153         if (status) {
2154                 mlog_errno(status);
2155                 goto bail;
2156         }
2157
2158         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2159         if (!osb->vol_label) {
2160                 mlog(ML_ERROR, "unable to alloc vol label\n");
2161                 status = -ENOMEM;
2162                 goto bail;
2163         }
2164
2165         osb->slot_recovery_generations =
2166                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2167                         GFP_KERNEL);
2168         if (!osb->slot_recovery_generations) {
2169                 status = -ENOMEM;
2170                 mlog_errno(status);
2171                 goto bail;
2172         }
2173
2174         init_waitqueue_head(&osb->osb_wipe_event);
2175         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2176                                         sizeof(*osb->osb_orphan_wipes),
2177                                         GFP_KERNEL);
2178         if (!osb->osb_orphan_wipes) {
2179                 status = -ENOMEM;
2180                 mlog_errno(status);
2181                 goto bail;
2182         }
2183
2184         osb->osb_rf_lock_tree = RB_ROOT;
2185
2186         osb->s_feature_compat =
2187                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2188         osb->s_feature_ro_compat =
2189                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2190         osb->s_feature_incompat =
2191                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2192
2193         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2194                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2195                      "optional features (%x).\n", i);
2196                 status = -EINVAL;
2197                 goto bail;
2198         }
2199         if (!sb_rdonly(osb->sb) && (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2200                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2201                      "unsupported optional features (%x).\n", i);
2202                 status = -EINVAL;
2203                 goto bail;
2204         }
2205
2206         if (ocfs2_clusterinfo_valid(osb)) {
2207                 osb->osb_stackflags =
2208                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2209                 strlcpy(osb->osb_cluster_stack,
2210                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2211                        OCFS2_STACK_LABEL_LEN + 1);
2212                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2213                         mlog(ML_ERROR,
2214                              "couldn't mount because of an invalid "
2215                              "cluster stack label (%s) \n",
2216                              osb->osb_cluster_stack);
2217                         status = -EINVAL;
2218                         goto bail;
2219                 }
2220                 strlcpy(osb->osb_cluster_name,
2221                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2222                         OCFS2_CLUSTER_NAME_LEN + 1);
2223         } else {
2224                 /* The empty string is identical with classic tools that
2225                  * don't know about s_cluster_info. */
2226                 osb->osb_cluster_stack[0] = '\0';
2227         }
2228
2229         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2230
2231         /* FIXME
2232          * This should be done in ocfs2_journal_init(), but unknown
2233          * ordering issues will cause the filesystem to crash.
2234          * If anyone wants to figure out what part of the code
2235          * refers to osb->journal before ocfs2_journal_init() is run,
2236          * be my guest.
2237          */
2238         /* initialize our journal structure */
2239
2240         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2241         if (!journal) {
2242                 mlog(ML_ERROR, "unable to alloc journal\n");
2243                 status = -ENOMEM;
2244                 goto bail;
2245         }
2246         osb->journal = journal;
2247         journal->j_osb = osb;
2248
2249         atomic_set(&journal->j_num_trans, 0);
2250         init_rwsem(&journal->j_trans_barrier);
2251         init_waitqueue_head(&journal->j_checkpointed);
2252         spin_lock_init(&journal->j_lock);
2253         journal->j_trans_id = (unsigned long) 1;
2254         INIT_LIST_HEAD(&journal->j_la_cleanups);
2255         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2256         journal->j_state = OCFS2_JOURNAL_FREE;
2257
2258         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2259         init_llist_head(&osb->dquot_drop_list);
2260
2261         /* get some pseudo constants for clustersize bits */
2262         osb->s_clustersize_bits =
2263                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2264         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2265
2266         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2267             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2268                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2269                      osb->s_clustersize);
2270                 status = -EINVAL;
2271                 goto bail;
2272         }
2273
2274         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2275                                                 le32_to_cpu(di->i_clusters));
2276
2277         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2278                                            total_blocks);
2279         if (status) {
2280                 mlog(ML_ERROR, "Volume too large "
2281                      "to mount safely on this system");
2282                 status = -EFBIG;
2283                 goto bail;
2284         }
2285
2286         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2287                                  sizeof(di->id2.i_super.s_uuid))) {
2288                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2289                 status = -ENOMEM;
2290                 goto bail;
2291         }
2292
2293         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2294                 OCFS2_MAX_VOL_LABEL_LEN);
2295         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2296         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2297         osb->first_cluster_group_blkno =
2298                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2299         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2300         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2301         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2302                                      (unsigned long long)osb->root_blkno,
2303                                      (unsigned long long)osb->system_dir_blkno,
2304                                      osb->s_clustersize_bits);
2305
2306         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2307         if (!osb->osb_dlm_debug) {
2308                 status = -ENOMEM;
2309                 mlog_errno(status);
2310                 goto bail;
2311         }
2312
2313         atomic_set(&osb->vol_state, VOLUME_INIT);
2314
2315         /* load root, system_dir, and all global system inodes */
2316         status = ocfs2_init_global_system_inodes(osb);
2317         if (status < 0) {
2318                 mlog_errno(status);
2319                 goto bail;
2320         }
2321
2322         /*
2323          * global bitmap
2324          */
2325         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2326                                             OCFS2_INVALID_SLOT);
2327         if (!inode) {
2328                 status = -EINVAL;
2329                 mlog_errno(status);
2330                 goto bail;
2331         }
2332
2333         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2334         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2335         iput(inode);
2336
2337         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2338                                  osb->s_feature_incompat) * 8;
2339
2340         status = ocfs2_init_slot_info(osb);
2341         if (status < 0) {
2342                 mlog_errno(status);
2343                 goto bail;
2344         }
2345         cleancache_init_shared_fs(sb);
2346
2347         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2348         if (!osb->ocfs2_wq) {
2349                 status = -ENOMEM;
2350                 mlog_errno(status);
2351         }
2352
2353 bail:
2354         return status;
2355 }
2356
2357 /*
2358  * will return: -EAGAIN if it is ok to keep searching for superblocks
2359  *              -EINVAL if there is a bad superblock
2360  *              0 on success
2361  */
2362 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2363                                struct buffer_head *bh,
2364                                u32 blksz,
2365                                struct ocfs2_blockcheck_stats *stats)
2366 {
2367         int status = -EAGAIN;
2368
2369         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2370                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2371                 /* We have to do a raw check of the feature here */
2372                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2373                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2374                         status = ocfs2_block_check_validate(bh->b_data,
2375                                                             bh->b_size,
2376                                                             &di->i_check,
2377                                                             stats);
2378                         if (status)
2379                                 goto out;
2380                 }
2381                 status = -EINVAL;
2382                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2383                         mlog(ML_ERROR, "found superblock with incorrect block "
2384                              "size: found %u, should be %u\n",
2385                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2386                                blksz);
2387                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2388                            OCFS2_MAJOR_REV_LEVEL ||
2389                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2390                            OCFS2_MINOR_REV_LEVEL) {
2391                         mlog(ML_ERROR, "found superblock with bad version: "
2392                              "found %u.%u, should be %u.%u\n",
2393                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2394                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2395                              OCFS2_MAJOR_REV_LEVEL,
2396                              OCFS2_MINOR_REV_LEVEL);
2397                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2398                         mlog(ML_ERROR, "bad block number on superblock: "
2399                              "found %llu, should be %llu\n",
2400                              (unsigned long long)le64_to_cpu(di->i_blkno),
2401                              (unsigned long long)bh->b_blocknr);
2402                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2403                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2404                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2405                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2406                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2407                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2408                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2409                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2410                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2411                         mlog(ML_ERROR,
2412                              "Superblock slots found greater than file system "
2413                              "maximum: found %u, max %u\n",
2414                              le16_to_cpu(di->id2.i_super.s_max_slots),
2415                              OCFS2_MAX_SLOTS);
2416                 } else {
2417                         /* found it! */
2418                         status = 0;
2419                 }
2420         }
2421
2422 out:
2423         if (status && status != -EAGAIN)
2424                 mlog_errno(status);
2425         return status;
2426 }
2427
2428 static int ocfs2_check_volume(struct ocfs2_super *osb)
2429 {
2430         int status;
2431         int dirty;
2432         int local;
2433         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2434                                                   * recover
2435                                                   * ourselves. */
2436
2437         /* Init our journal object. */
2438         status = ocfs2_journal_init(osb->journal, &dirty);
2439         if (status < 0) {
2440                 mlog(ML_ERROR, "Could not initialize journal!\n");
2441                 goto finally;
2442         }
2443
2444         /* Now that journal has been initialized, check to make sure
2445            entire volume is addressable. */
2446         status = ocfs2_journal_addressable(osb);
2447         if (status)
2448                 goto finally;
2449
2450         /* If the journal was unmounted cleanly then we don't want to
2451          * recover anything. Otherwise, journal_load will do that
2452          * dirty work for us :) */
2453         if (!dirty) {
2454                 status = ocfs2_journal_wipe(osb->journal, 0);
2455                 if (status < 0) {
2456                         mlog_errno(status);
2457                         goto finally;
2458                 }
2459         } else {
2460                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2461                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2462         }
2463
2464         local = ocfs2_mount_local(osb);
2465
2466         /* will play back anything left in the journal. */
2467         status = ocfs2_journal_load(osb->journal, local, dirty);
2468         if (status < 0) {
2469                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2470                 goto finally;
2471         }
2472
2473         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2474                 jbd2_journal_set_features(osb->journal->j_journal,
2475                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2476                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2477         else
2478                 jbd2_journal_clear_features(osb->journal->j_journal,
2479                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2480                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2481
2482         if (dirty) {
2483                 /* recover my local alloc if we didn't unmount cleanly. */
2484                 status = ocfs2_begin_local_alloc_recovery(osb,
2485                                                           osb->slot_num,
2486                                                           &local_alloc);
2487                 if (status < 0) {
2488                         mlog_errno(status);
2489                         goto finally;
2490                 }
2491                 /* we complete the recovery process after we've marked
2492                  * ourselves as mounted. */
2493         }
2494
2495         status = ocfs2_load_local_alloc(osb);
2496         if (status < 0) {
2497                 mlog_errno(status);
2498                 goto finally;
2499         }
2500
2501         if (dirty) {
2502                 /* Recovery will be completed after we've mounted the
2503                  * rest of the volume. */
2504                 osb->local_alloc_copy = local_alloc;
2505                 local_alloc = NULL;
2506         }
2507
2508         /* go through each journal, trylock it and if you get the
2509          * lock, and it's marked as dirty, set the bit in the recover
2510          * map and launch a recovery thread for it. */
2511         status = ocfs2_mark_dead_nodes(osb);
2512         if (status < 0) {
2513                 mlog_errno(status);
2514                 goto finally;
2515         }
2516
2517         status = ocfs2_compute_replay_slots(osb);
2518         if (status < 0)
2519                 mlog_errno(status);
2520
2521 finally:
2522         kfree(local_alloc);
2523
2524         if (status)
2525                 mlog_errno(status);
2526         return status;
2527 }
2528
2529 /*
2530  * The routine gets called from dismount or close whenever a dismount on
2531  * volume is requested and the osb open count becomes 1.
2532  * It will remove the osb from the global list and also free up all the
2533  * initialized resources and fileobject.
2534  */
2535 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2536 {
2537         /* This function assumes that the caller has the main osb resource */
2538
2539         /* ocfs2_initializer_super have already created this workqueue */
2540         if (osb->ocfs2_wq)
2541                 destroy_workqueue(osb->ocfs2_wq);
2542
2543         ocfs2_free_slot_info(osb);
2544
2545         kfree(osb->osb_orphan_wipes);
2546         kfree(osb->slot_recovery_generations);
2547         /* FIXME
2548          * This belongs in journal shutdown, but because we have to
2549          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2550          * we free it here.
2551          */
2552         kfree(osb->journal);
2553         kfree(osb->local_alloc_copy);
2554         kfree(osb->uuid_str);
2555         kfree(osb->vol_label);
2556         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2557         memset(osb, 0, sizeof(struct ocfs2_super));
2558 }
2559
2560 /* Depending on the mount option passed, perform one of the following:
2561  * Put OCFS2 into a readonly state (default)
2562  * Return EIO so that only the process errs
2563  * Fix the error as if fsck.ocfs2 -y
2564  * panic
2565  */
2566 static int ocfs2_handle_error(struct super_block *sb)
2567 {
2568         struct ocfs2_super *osb = OCFS2_SB(sb);
2569         int rv = 0;
2570
2571         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2572         pr_crit("On-disk corruption discovered. "
2573                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2574
2575         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2576                 panic("OCFS2: (device %s): panic forced after error\n",
2577                       sb->s_id);
2578         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2579                 pr_crit("OCFS2: Returning error to the calling process.\n");
2580                 rv = -EIO;
2581         } else { /* default option */
2582                 rv = -EROFS;
2583                 if (sb_rdonly(sb) && (ocfs2_is_soft_readonly(osb) || ocfs2_is_hard_readonly(osb)))
2584                         return rv;
2585
2586                 pr_crit("OCFS2: File system is now read-only.\n");
2587                 sb->s_flags |= SB_RDONLY;
2588                 ocfs2_set_ro_flag(osb, 0);
2589         }
2590
2591         return rv;
2592 }
2593
2594 int __ocfs2_error(struct super_block *sb, const char *function,
2595                   const char *fmt, ...)
2596 {
2597         struct va_format vaf;
2598         va_list args;
2599
2600         va_start(args, fmt);
2601         vaf.fmt = fmt;
2602         vaf.va = &args;
2603
2604         /* Not using mlog here because we want to show the actual
2605          * function the error came from. */
2606         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2607                sb->s_id, function, &vaf);
2608
2609         va_end(args);
2610
2611         return ocfs2_handle_error(sb);
2612 }
2613
2614 /* Handle critical errors. This is intentionally more drastic than
2615  * ocfs2_handle_error, so we only use for things like journal errors,
2616  * etc. */
2617 void __ocfs2_abort(struct super_block *sb, const char *function,
2618                    const char *fmt, ...)
2619 {
2620         struct va_format vaf;
2621         va_list args;
2622
2623         va_start(args, fmt);
2624
2625         vaf.fmt = fmt;
2626         vaf.va = &args;
2627
2628         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2629                sb->s_id, function, &vaf);
2630
2631         va_end(args);
2632
2633         /* We don't have the cluster support yet to go straight to
2634          * hard readonly in here. Until then, we want to keep
2635          * ocfs2_abort() so that we can at least mark critical
2636          * errors.
2637          *
2638          * TODO: This should abort the journal and alert other nodes
2639          * that our slot needs recovery. */
2640
2641         /* Force a panic(). This stinks, but it's better than letting
2642          * things continue without having a proper hard readonly
2643          * here. */
2644         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2645                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2646         ocfs2_handle_error(sb);
2647 }
2648
2649 /*
2650  * Void signal blockers, because in-kernel sigprocmask() only fails
2651  * when SIG_* is wrong.
2652  */
2653 void ocfs2_block_signals(sigset_t *oldset)
2654 {
2655         int rc;
2656         sigset_t blocked;
2657
2658         sigfillset(&blocked);
2659         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2660         BUG_ON(rc);
2661 }
2662
2663 void ocfs2_unblock_signals(sigset_t *oldset)
2664 {
2665         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2666         BUG_ON(rc);
2667 }
2668
2669 module_init(ocfs2_init);
2670 module_exit(ocfs2_exit);