fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root = NULL;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static const struct super_operations ocfs2_sops = {
 147         .statfs         = ocfs2_statfs,
 148         .alloc_inode    = ocfs2_alloc_inode,
 149         .destroy_inode  = ocfs2_destroy_inode,
 150         .drop_inode     = ocfs2_drop_inode,
 151         .evict_inode    = ocfs2_evict_inode,
 152         .sync_fs        = ocfs2_sync_fs,
 153         .put_super      = ocfs2_put_super,
 154         .remount_fs     = ocfs2_remount,
 155         .show_options   = ocfs2_show_options,
 156         .quota_read     = ocfs2_quota_read,
 157         .quota_write    = ocfs2_quota_write,
 158 };
 159
 160 enum {
 161         Opt_barrier,
 162         Opt_err_panic,
 163         Opt_err_ro,
 164         Opt_intr,
 165         Opt_nointr,
 166         Opt_hb_none,
 167         Opt_hb_local,
 168         Opt_hb_global,
 169         Opt_data_ordered,
 170         Opt_data_writeback,
 171         Opt_atime_quantum,
 172         Opt_slot,
 173         Opt_commit,
 174         Opt_localalloc,
 175         Opt_localflocks,
 176         Opt_stack,
 177         Opt_user_xattr,
 178         Opt_nouser_xattr,
 179         Opt_inode64,
 180         Opt_acl,
 181         Opt_noacl,
 182         Opt_usrquota,
 183         Opt_grpquota,
 184         Opt_coherency_buffered,
 185         Opt_coherency_full,
 186         Opt_resv_level,
 187         Opt_dir_resv_level,
 188         Opt_err,
 189 };
 190
 191 static const match_table_t tokens = {
 192         {Opt_barrier, "barrier=%u"},
 193         {Opt_err_panic, "errors=panic"},
 194         {Opt_err_ro, "errors=remount-ro"},
 195         {Opt_intr, "intr"},
 196         {Opt_nointr, "nointr"},
 197         {Opt_hb_none, OCFS2_HB_NONE},
 198         {Opt_hb_local, OCFS2_HB_LOCAL},
 199         {Opt_hb_global, OCFS2_HB_GLOBAL},
 200         {Opt_data_ordered, "data=ordered"},
 201         {Opt_data_writeback, "data=writeback"},
 202         {Opt_atime_quantum, "atime_quantum=%u"},
 203         {Opt_slot, "preferred_slot=%u"},
 204         {Opt_commit, "commit=%u"},
 205         {Opt_localalloc, "localalloc=%d"},
 206         {Opt_localflocks, "localflocks"},
 207         {Opt_stack, "cluster_stack=%s"},
 208         {Opt_user_xattr, "user_xattr"},
 209         {Opt_nouser_xattr, "nouser_xattr"},
 210         {Opt_inode64, "inode64"},
 211         {Opt_acl, "acl"},
 212         {Opt_noacl, "noacl"},
 213         {Opt_usrquota, "usrquota"},
 214         {Opt_grpquota, "grpquota"},
 215         {Opt_coherency_buffered, "coherency=buffered"},
 216         {Opt_coherency_full, "coherency=full"},
 217         {Opt_resv_level, "resv_level=%u"},
 218         {Opt_dir_resv_level, "dir_resv_level=%u"},
 219         {Opt_err, NULL}
 220 };
 221
 222 #ifdef CONFIG_DEBUG_FS
 223 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 224 {
 225         struct ocfs2_cluster_connection *cconn = osb->cconn;
 226         struct ocfs2_recovery_map *rm = osb->recovery_map;
 227         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 228         int i, out = 0;
 229
 230         out += snprintf(buf + out, len - out,
 231                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 232                         "Device", osb->dev_str, osb->uuid_str,
 233                         osb->fs_generation, osb->vol_label);
 234
 235         out += snprintf(buf + out, len - out,
 236                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 237                         atomic_read(&osb->vol_state), osb->osb_flags);
 238
 239         out += snprintf(buf + out, len - out,
 240                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 241                         osb->sb->s_blocksize, osb->s_clustersize);
 242
 243         out += snprintf(buf + out, len - out,
 244                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 245                         "ROcompat: 0x%X\n",
 246                         "Features", osb->s_feature_compat,
 247                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 251                         osb->s_mount_opt, osb->s_atime_quantum);
 252
 253         if (cconn) {
 254                 out += snprintf(buf + out, len - out,
 255                                 "%10s => Stack: %s  Name: %*s  "
 256                                 "Version: %d.%d\n", "Cluster",
 257                                 (*osb->osb_cluster_stack == '\0' ?
 258                                  "o2cb" : osb->osb_cluster_stack),
 259                                 cconn->cc_namelen, cconn->cc_name,
 260                                 cconn->cc_version.pv_major,
 261                                 cconn->cc_version.pv_minor);
 262         }
 263
 264         spin_lock(&osb->dc_task_lock);
 265         out += snprintf(buf + out, len - out,
 266                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 267                         "WorkSeq: %lu\n", "DownCnvt",
 268                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 269                         osb->blocked_lock_count, osb->dc_wake_sequence,
 270                         osb->dc_work_sequence);
 271         spin_unlock(&osb->dc_task_lock);
 272
 273         spin_lock(&osb->osb_lock);
 274         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 275                         "Recovery",
 276                         (osb->recovery_thread_task ?
 277                          task_pid_nr(osb->recovery_thread_task) : -1));
 278         if (rm->rm_used == 0)
 279                 out += snprintf(buf + out, len - out, " None\n");
 280         else {
 281                 for (i = 0; i < rm->rm_used; i++)
 282                         out += snprintf(buf + out, len - out, " %d",
 283                                         rm->rm_entries[i]);
 284                 out += snprintf(buf + out, len - out, "\n");
 285         }
 286         spin_unlock(&osb->osb_lock);
 287
 288         out += snprintf(buf + out, len - out,
 289                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 290                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 291                         osb->osb_commit_interval);
 292
 293         out += snprintf(buf + out, len - out,
 294                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 295                         "Journal", osb->journal->j_state,
 296                         osb->journal->j_trans_id,
 297                         atomic_read(&osb->journal->j_num_trans));
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 301                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 302                         "Stats",
 303                         atomic_read(&osb->alloc_stats.bitmap_data),
 304                         atomic_read(&osb->alloc_stats.local_data),
 305                         atomic_read(&osb->alloc_stats.bg_allocs),
 306                         atomic_read(&osb->alloc_stats.moves),
 307                         atomic_read(&osb->alloc_stats.bg_extends));
 308
 309         out += snprintf(buf + out, len - out,
 310                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 311                         "Default: %u bits\n",
 312                         "LocalAlloc", osb->local_alloc_state,
 313                         (unsigned long long)osb->la_last_gd,
 314                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 315
 316         spin_lock(&osb->osb_lock);
 317         out += snprintf(buf + out, len - out,
 318                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 319                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 320                         osb->s_inode_steal_slot,
 321                         atomic_read(&osb->s_num_inodes_stolen),
 322                         osb->s_meta_steal_slot,
 323                         atomic_read(&osb->s_num_meta_stolen));
 324         spin_unlock(&osb->osb_lock);
 325
 326         out += snprintf(buf + out, len - out, "OrphanScan => ");
 327         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 328                         os->os_count, os->os_seqno);
 329         out += snprintf(buf + out, len - out, " Last Scan: ");
 330         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 331                 out += snprintf(buf + out, len - out, "Disabled\n");
 332         else
 333                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 334                                 (get_seconds() - os->os_scantime.tv_sec));
 335
 336         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 337                         "Slots", "Num", "RecoGen");
 338         for (i = 0; i < osb->max_slots; ++i) {
 339                 out += snprintf(buf + out, len - out,
 340                                 "%10s  %c %3d  %10d\n",
 341                                 " ",
 342                                 (i == osb->slot_num ? '*' : ' '),
 343                                 i, osb->slot_recovery_generations[i]);
 344         }
 345
 346         return out;
 347 }
 348
 349 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 350 {
 351         struct ocfs2_super *osb = inode->i_private;
 352         char *buf = NULL;
 353
 354         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 355         if (!buf)
 356                 goto bail;
 357
 358         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 359
 360         file->private_data = buf;
 361
 362         return 0;
 363 bail:
 364         return -ENOMEM;
 365 }
 366
 367 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 368 {
 369         kfree(file->private_data);
 370         return 0;
 371 }
 372
 373 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 374                                 size_t nbytes, loff_t *ppos)
 375 {
 376         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 377                                        i_size_read(file->f_mapping->host));
 378 }
 379 #else
 380 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 381 {
 382         return 0;
 383 }
 384 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 385 {
 386         return 0;
 387 }
 388 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 389                                 size_t nbytes, loff_t *ppos)
 390 {
 391         return 0;
 392 }
 393 #endif  /* CONFIG_DEBUG_FS */
 394
 395 static const struct file_operations ocfs2_osb_debug_fops = {
 396         .open =         ocfs2_osb_debug_open,
 397         .release =      ocfs2_debug_release,
 398         .read =         ocfs2_debug_read,
 399         .llseek =       generic_file_llseek,
 400 };
 401
 402 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 403 {
 404         int status;
 405         tid_t target;
 406         struct ocfs2_super *osb = OCFS2_SB(sb);
 407
 408         if (ocfs2_is_hard_readonly(osb))
 409                 return -EROFS;
 410
 411         if (wait) {
 412                 status = ocfs2_flush_truncate_log(osb);
 413                 if (status < 0)
 414                         mlog_errno(status);
 415         } else {
 416                 ocfs2_schedule_truncate_log_flush(osb, 0);
 417         }
 418
 419         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 420                                       &target)) {
 421                 if (wait)
 422                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 423                                              target);
 424         }
 425         return 0;
 426 }
 427
 428 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 429 {
 430         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 431             && (ino == USER_QUOTA_SYSTEM_INODE
 432                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 433                 return 0;
 434         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 435             && (ino == GROUP_QUOTA_SYSTEM_INODE
 436                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 437                 return 0;
 438         return 1;
 439 }
 440
 441 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 442 {
 443         struct inode *new = NULL;
 444         int status = 0;
 445         int i;
 446
 447         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 448         if (IS_ERR(new)) {
 449                 status = PTR_ERR(new);
 450                 mlog_errno(status);
 451                 goto bail;
 452         }
 453         osb->root_inode = new;
 454
 455         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 456         if (IS_ERR(new)) {
 457                 status = PTR_ERR(new);
 458                 mlog_errno(status);
 459                 goto bail;
 460         }
 461         osb->sys_root_inode = new;
 462
 463         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 464              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 465                 if (!ocfs2_need_system_inode(osb, i))
 466                         continue;
 467                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 468                 if (!new) {
 469                         ocfs2_release_system_inodes(osb);
 470                         status = -EINVAL;
 471                         mlog_errno(status);
 472                         /* FIXME: Should ERROR_RO_FS */
 473                         mlog(ML_ERROR, "Unable to load system inode %d, "
 474                              "possibly corrupt fs?", i);
 475                         goto bail;
 476                 }
 477                 // the array now has one ref, so drop this one
 478                 iput(new);
 479         }
 480
 481 bail:
 482         if (status)
 483                 mlog_errno(status);
 484         return status;
 485 }
 486
 487 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 488 {
 489         struct inode *new = NULL;
 490         int status = 0;
 491         int i;
 492
 493         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 494              i < NUM_SYSTEM_INODES;
 495              i++) {
 496                 if (!ocfs2_need_system_inode(osb, i))
 497                         continue;
 498                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 499                 if (!new) {
 500                         ocfs2_release_system_inodes(osb);
 501                         status = -EINVAL;
 502                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 503                              status, i, osb->slot_num);
 504                         goto bail;
 505                 }
 506                 /* the array now has one ref, so drop this one */
 507                 iput(new);
 508         }
 509
 510 bail:
 511         if (status)
 512                 mlog_errno(status);
 513         return status;
 514 }
 515
 516 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 517 {
 518         int i;
 519         struct inode *inode;
 520
 521         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 522                 inode = osb->global_system_inodes[i];
 523                 if (inode) {
 524                         iput(inode);
 525                         osb->global_system_inodes[i] = NULL;
 526                 }
 527         }
 528
 529         inode = osb->sys_root_inode;
 530         if (inode) {
 531                 iput(inode);
 532                 osb->sys_root_inode = NULL;
 533         }
 534
 535         inode = osb->root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->root_inode = NULL;
 539         }
 540
 541         if (!osb->local_system_inodes)
 542                 return;
 543
 544         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 545                 if (osb->local_system_inodes[i]) {
 546                         iput(osb->local_system_inodes[i]);
 547                         osb->local_system_inodes[i] = NULL;
 548                 }
 549         }
 550
 551         kfree(osb->local_system_inodes);
 552         osb->local_system_inodes = NULL;
 553 }
 554
 555 /* We're allocating fs objects, use GFP_NOFS */
 556 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 557 {
 558         struct ocfs2_inode_info *oi;
 559
 560         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 561         if (!oi)
 562                 return NULL;
 563
 564         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 565         return &oi->vfs_inode;
 566 }
 567
 568 static void ocfs2_i_callback(struct rcu_head *head)
 569 {
 570         struct inode *inode = container_of(head, struct inode, i_rcu);
 571         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 572 }
 573
 574 static void ocfs2_destroy_inode(struct inode *inode)
 575 {
 576         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 577 }
 578
 579 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 580                                                 unsigned int cbits)
 581 {
 582         unsigned int bytes = 1 << cbits;
 583         unsigned int trim = bytes;
 584         unsigned int bitshift = 32;
 585
 586         /*
 587          * i_size and all block offsets in ocfs2 are always 64 bits
 588          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 589          * 64 bit platforms, cluster size will be the limiting factor.
 590          */
 591
 592 #if BITS_PER_LONG == 32
 593 # if defined(CONFIG_LBDAF)
 594         BUILD_BUG_ON(sizeof(sector_t) != 8);
 595         /*
 596          * We might be limited by page cache size.
 597          */
 598         if (bytes > PAGE_CACHE_SIZE) {
 599                 bytes = PAGE_CACHE_SIZE;
 600                 trim = 1;
 601                 /*
 602                  * Shift by 31 here so that we don't get larger than
 603                  * MAX_LFS_FILESIZE
 604                  */
 605                 bitshift = 31;
 606         }
 607 # else
 608         /*
 609          * We are limited by the size of sector_t. Use block size, as
 610          * that's what we expose to the VFS.
 611          */
 612         bytes = 1 << bbits;
 613         trim = 1;
 614         bitshift = 31;
 615 # endif
 616 #endif
 617
 618         /*
 619          * Trim by a whole cluster when we can actually approach the
 620          * on-disk limits. Otherwise we can overflow i_clusters when
 621          * an extent start is at the max offset.
 622          */
 623         return (((unsigned long long)bytes) << bitshift) - trim;
 624 }
 625
 626 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 627 {
 628         int incompat_features;
 629         int ret = 0;
 630         struct mount_options parsed_options;
 631         struct ocfs2_super *osb = OCFS2_SB(sb);
 632         u32 tmp;
 633
 634         sync_filesystem(sb);
 635
 636         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 637             !ocfs2_check_set_options(sb, &parsed_options)) {
 638                 ret = -EINVAL;
 639                 goto out;
 640         }
 641
 642         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 643                 OCFS2_MOUNT_HB_NONE;
 644         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 645                 ret = -EINVAL;
 646                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 647                 goto out;
 648         }
 649
 650         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 651             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 652                 ret = -EINVAL;
 653                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 654                 goto out;
 655         }
 656
 657         /* Probably don't want this on remount; it might
 658          * mess with other nodes */
 659         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 660             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 661                 ret = -EINVAL;
 662                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 663                 goto out;
 664         }
 665
 666         /* We're going to/from readonly mode. */
 667         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 668                 /* Disable quota accounting before remounting RO */
 669                 if (*flags & MS_RDONLY) {
 670                         ret = ocfs2_susp_quotas(osb, 0);
 671                         if (ret < 0)
 672                                 goto out;
 673                 }
 674                 /* Lock here so the check of HARD_RO and the potential
 675                  * setting of SOFT_RO is atomic. */
 676                 spin_lock(&osb->osb_lock);
 677                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 678                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 679                         ret = -EROFS;
 680                         goto unlock_osb;
 681                 }
 682
 683                 if (*flags & MS_RDONLY) {
 684                         sb->s_flags |= MS_RDONLY;
 685                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 686                 } else {
 687                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 688                                 mlog(ML_ERROR, "Cannot remount RDWR "
 689                                      "filesystem due to previous errors.\n");
 690                                 ret = -EROFS;
 691                                 goto unlock_osb;
 692                         }
 693                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 694                         if (incompat_features) {
 695                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 696                                      "of unsupported optional features "
 697                                      "(%x).\n", incompat_features);
 698                                 ret = -EINVAL;
 699                                 goto unlock_osb;
 700                         }
 701                         sb->s_flags &= ~MS_RDONLY;
 702                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 703                 }
 704                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 705 unlock_osb:
 706                 spin_unlock(&osb->osb_lock);
 707                 /* Enable quota accounting after remounting RW */
 708                 if (!ret && !(*flags & MS_RDONLY)) {
 709                         if (sb_any_quota_suspended(sb))
 710                                 ret = ocfs2_susp_quotas(osb, 1);
 711                         else
 712                                 ret = ocfs2_enable_quotas(osb);
 713                         if (ret < 0) {
 714                                 /* Return back changes... */
 715                                 spin_lock(&osb->osb_lock);
 716                                 sb->s_flags |= MS_RDONLY;
 717                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 718                                 spin_unlock(&osb->osb_lock);
 719                                 goto out;
 720                         }
 721                 }
 722         }
 723
 724         if (!ret) {
 725                 /* Only save off the new mount options in case of a successful
 726                  * remount. */
 727                 osb->s_mount_opt = parsed_options.mount_opt;
 728                 osb->s_atime_quantum = parsed_options.atime_quantum;
 729                 osb->preferred_slot = parsed_options.slot;
 730                 if (parsed_options.commit_interval)
 731                         osb->osb_commit_interval = parsed_options.commit_interval;
 732
 733                 if (!ocfs2_is_hard_readonly(osb))
 734                         ocfs2_set_journal_params(osb);
 735
 736                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 737                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 738                                                         MS_POSIXACL : 0);
 739         }
 740 out:
 741         return ret;
 742 }
 743
 744 static int ocfs2_sb_probe(struct super_block *sb,
 745                           struct buffer_head **bh,
 746                           int *sector_size,
 747                           struct ocfs2_blockcheck_stats *stats)
 748 {
 749         int status, tmpstat;
 750         struct ocfs1_vol_disk_hdr *hdr;
 751         struct ocfs2_dinode *di;
 752         int blksize;
 753
 754         *bh = NULL;
 755
 756         /* may be > 512 */
 757         *sector_size = bdev_logical_block_size(sb->s_bdev);
 758         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 759                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 760                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 761                 status = -EINVAL;
 762                 goto bail;
 763         }
 764
 765         /* Can this really happen? */
 766         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 767                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 768
 769         /* check block zero for old format */
 770         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 771         if (status < 0) {
 772                 mlog_errno(status);
 773                 goto bail;
 774         }
 775         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 776         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 777                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 778                      hdr->major_version, hdr->minor_version);
 779                 status = -EINVAL;
 780         }
 781         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 782                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 783                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 784                      hdr->signature);
 785                 status = -EINVAL;
 786         }
 787         brelse(*bh);
 788         *bh = NULL;
 789         if (status < 0) {
 790                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 791                      "upgraded before mounting with ocfs v2\n");
 792                 goto bail;
 793         }
 794
 795         /*
 796          * Now check at magic offset for 512, 1024, 2048, 4096
 797          * blocksizes.  4096 is the maximum blocksize because it is
 798          * the minimum clustersize.
 799          */
 800         status = -EINVAL;
 801         for (blksize = *sector_size;
 802              blksize <= OCFS2_MAX_BLOCKSIZE;
 803              blksize <<= 1) {
 804                 tmpstat = ocfs2_get_sector(sb, bh,
 805                                            OCFS2_SUPER_BLOCK_BLKNO,
 806                                            blksize);
 807                 if (tmpstat < 0) {
 808                         status = tmpstat;
 809                         mlog_errno(status);
 810                         break;
 811                 }
 812                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 813                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 814                 spin_lock_init(&stats->b_lock);
 815                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 816                 if (tmpstat < 0) {
 817                         brelse(*bh);
 818                         *bh = NULL;
 819                 }
 820                 if (tmpstat != -EAGAIN) {
 821                         status = tmpstat;
 822                         break;
 823                 }
 824         }
 825
 826 bail:
 827         return status;
 828 }
 829
 830 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 831 {
 832         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 833
 834         if (osb->s_mount_opt & hb_enabled) {
 835                 if (ocfs2_mount_local(osb)) {
 836                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 837                              "mounted device.\n");
 838                         return -EINVAL;
 839                 }
 840                 if (ocfs2_userspace_stack(osb)) {
 841                         mlog(ML_ERROR, "Userspace stack expected, but "
 842                              "o2cb heartbeat arguments passed to mount\n");
 843                         return -EINVAL;
 844                 }
 845                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 846                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 847                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 848                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 849                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 850                         return -EINVAL;
 851                 }
 852         }
 853
 854         if (!(osb->s_mount_opt & hb_enabled)) {
 855                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 856                     !ocfs2_userspace_stack(osb)) {
 857                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 858                              "a read-write clustered device.\n");
 859                         return -EINVAL;
 860                 }
 861         }
 862
 863         return 0;
 864 }
 865
 866 /*
 867  * If we're using a userspace stack, mount should have passed
 868  * a name that matches the disk.  If not, mount should not
 869  * have passed a stack.
 870  */
 871 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 872                                         struct mount_options *mopt)
 873 {
 874         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 875                 mlog(ML_ERROR,
 876                      "cluster stack passed to mount, but this filesystem "
 877                      "does not support it\n");
 878                 return -EINVAL;
 879         }
 880
 881         if (ocfs2_userspace_stack(osb) &&
 882             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 883                     OCFS2_STACK_LABEL_LEN)) {
 884                 mlog(ML_ERROR,
 885                      "cluster stack passed to mount (\"%s\") does not "
 886                      "match the filesystem (\"%s\")\n",
 887                      mopt->cluster_stack,
 888                      osb->osb_cluster_stack);
 889                 return -EINVAL;
 890         }
 891
 892         return 0;
 893 }
 894
 895 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 896 {
 897         int type;
 898         struct super_block *sb = osb->sb;
 899         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 900                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 901         int status = 0;
 902
 903         for (type = 0; type < MAXQUOTAS; type++) {
 904                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 905                         continue;
 906                 if (unsuspend)
 907                         status = dquot_resume(sb, type);
 908                 else {
 909                         struct ocfs2_mem_dqinfo *oinfo;
 910
 911                         /* Cancel periodic syncing before suspending */
 912                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 913                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 914                         status = dquot_suspend(sb, type);
 915                 }
 916                 if (status < 0)
 917                         break;
 918         }
 919         if (status < 0)
 920                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 921                      "remount (error = %d).\n", status);
 922         return status;
 923 }
 924
 925 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 926 {
 927         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 928         struct super_block *sb = osb->sb;
 929         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 930                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 931         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 932                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 933         int status;
 934         int type;
 935
 936         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 937         for (type = 0; type < MAXQUOTAS; type++) {
 938                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 939                         continue;
 940                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 941                                                         osb->slot_num);
 942                 if (!inode[type]) {
 943                         status = -ENOENT;
 944                         goto out_quota_off;
 945                 }
 946                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 947                                       DQUOT_USAGE_ENABLED);
 948                 if (status < 0)
 949                         goto out_quota_off;
 950         }
 951
 952         for (type = 0; type < MAXQUOTAS; type++)
 953                 iput(inode[type]);
 954         return 0;
 955 out_quota_off:
 956         ocfs2_disable_quotas(osb);
 957         for (type = 0; type < MAXQUOTAS; type++)
 958                 iput(inode[type]);
 959         mlog_errno(status);
 960         return status;
 961 }
 962
 963 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 964 {
 965         int type;
 966         struct inode *inode;
 967         struct super_block *sb = osb->sb;
 968         struct ocfs2_mem_dqinfo *oinfo;
 969
 970         /* We mostly ignore errors in this function because there's not much
 971          * we can do when we see them */
 972         for (type = 0; type < MAXQUOTAS; type++) {
 973                 if (!sb_has_quota_loaded(sb, type))
 974                         continue;
 975                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 976                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 977                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 978                 inode = igrab(sb->s_dquot.files[type]);
 979                 /* Turn off quotas. This will remove all dquot structures from
 980                  * memory and so they will be automatically synced to global
 981                  * quota files */
 982                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 983                                         DQUOT_LIMITS_ENABLED);
 984                 if (!inode)
 985                         continue;
 986                 iput(inode);
 987         }
 988 }
 989
 990 /* Handle quota on quotactl */
 991 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 992 {
 993         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 994                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 995
 996         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 997                 return -EINVAL;
 998
 999         return dquot_enable(sb_dqopt(sb)->files[type], type,
1000                             format_id, DQUOT_LIMITS_ENABLED);
1001 }
1002
1003 /* Handle quota off quotactl */
1004 static int ocfs2_quota_off(struct super_block *sb, int type)
1005 {
1006         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1007 }
1008
1009 static const struct quotactl_ops ocfs2_quotactl_ops = {
1010         .quota_on_meta  = ocfs2_quota_on,
1011         .quota_off      = ocfs2_quota_off,
1012         .quota_sync     = dquot_quota_sync,
1013         .get_info       = dquot_get_dqinfo,
1014         .set_info       = dquot_set_dqinfo,
1015         .get_dqblk      = dquot_get_dqblk,
1016         .set_dqblk      = dquot_set_dqblk,
1017 };
1018
1019 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1020 {
1021         struct dentry *root;
1022         int status, sector_size;
1023         struct mount_options parsed_options;
1024         struct inode *inode = NULL;
1025         struct ocfs2_super *osb = NULL;
1026         struct buffer_head *bh = NULL;
1027         char nodestr[12];
1028         struct ocfs2_blockcheck_stats stats;
1029
1030         trace_ocfs2_fill_super(sb, data, silent);
1031
1032         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1033                 status = -EINVAL;
1034                 goto read_super_error;
1035         }
1036
1037         /* probe for superblock */
1038         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1039         if (status < 0) {
1040                 mlog(ML_ERROR, "superblock probe failed!\n");
1041                 goto read_super_error;
1042         }
1043
1044         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1045         osb = OCFS2_SB(sb);
1046         if (status < 0) {
1047                 mlog_errno(status);
1048                 goto read_super_error;
1049         }
1050         brelse(bh);
1051         bh = NULL;
1052
1053         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1054                 status = -EINVAL;
1055                 goto read_super_error;
1056         }
1057         osb->s_mount_opt = parsed_options.mount_opt;
1058         osb->s_atime_quantum = parsed_options.atime_quantum;
1059         osb->preferred_slot = parsed_options.slot;
1060         osb->osb_commit_interval = parsed_options.commit_interval;
1061
1062         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1063         osb->osb_resv_level = parsed_options.resv_level;
1064         osb->osb_dir_resv_level = parsed_options.resv_level;
1065         if (parsed_options.dir_resv_level == -1)
1066                 osb->osb_dir_resv_level = parsed_options.resv_level;
1067         else
1068                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1069
1070         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1071         if (status)
1072                 goto read_super_error;
1073
1074         sb->s_magic = OCFS2_SUPER_MAGIC;
1075
1076         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1077                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1078
1079         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1080          * heartbeat=none */
1081         if (bdev_read_only(sb->s_bdev)) {
1082                 if (!(sb->s_flags & MS_RDONLY)) {
1083                         status = -EACCES;
1084                         mlog(ML_ERROR, "Readonly device detected but readonly "
1085                              "mount was not specified.\n");
1086                         goto read_super_error;
1087                 }
1088
1089                 /* You should not be able to start a local heartbeat
1090                  * on a readonly device. */
1091                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1092                         status = -EROFS;
1093                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1094                              "device.\n");
1095                         goto read_super_error;
1096                 }
1097
1098                 status = ocfs2_check_journals_nolocks(osb);
1099                 if (status < 0) {
1100                         if (status == -EROFS)
1101                                 mlog(ML_ERROR, "Recovery required on readonly "
1102                                      "file system, but write access is "
1103                                      "unavailable.\n");
1104                         else
1105                                 mlog_errno(status);
1106                         goto read_super_error;
1107                 }
1108
1109                 ocfs2_set_ro_flag(osb, 1);
1110
1111                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1112                        "Cluster services will not be used for this mount. "
1113                        "Recovery will be skipped.\n", osb->dev_str);
1114         }
1115
1116         if (!ocfs2_is_hard_readonly(osb)) {
1117                 if (sb->s_flags & MS_RDONLY)
1118                         ocfs2_set_ro_flag(osb, 0);
1119         }
1120
1121         status = ocfs2_verify_heartbeat(osb);
1122         if (status < 0) {
1123                 mlog_errno(status);
1124                 goto read_super_error;
1125         }
1126
1127         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1128                                                  ocfs2_debugfs_root);
1129         if (!osb->osb_debug_root) {
1130                 status = -EINVAL;
1131                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1132                 goto read_super_error;
1133         }
1134
1135         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1136                                             osb->osb_debug_root,
1137                                             osb,
1138                                             &ocfs2_osb_debug_fops);
1139         if (!osb->osb_ctxt) {
1140                 status = -EINVAL;
1141                 mlog_errno(status);
1142                 goto read_super_error;
1143         }
1144
1145         if (ocfs2_meta_ecc(osb)) {
1146                 status = ocfs2_blockcheck_stats_debugfs_install(
1147                                                 &osb->osb_ecc_stats,
1148                                                 osb->osb_debug_root);
1149                 if (status) {
1150                         mlog(ML_ERROR,
1151                              "Unable to create blockcheck statistics "
1152                              "files\n");
1153                         goto read_super_error;
1154                 }
1155         }
1156
1157         status = ocfs2_mount_volume(sb);
1158         if (status < 0)
1159                 goto read_super_error;
1160
1161         if (osb->root_inode)
1162                 inode = igrab(osb->root_inode);
1163
1164         if (!inode) {
1165                 status = -EIO;
1166                 mlog_errno(status);
1167                 goto read_super_error;
1168         }
1169
1170         root = d_make_root(inode);
1171         if (!root) {
1172                 status = -ENOMEM;
1173                 mlog_errno(status);
1174                 goto read_super_error;
1175         }
1176
1177         sb->s_root = root;
1178
1179         ocfs2_complete_mount_recovery(osb);
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->s_flags & MS_RDONLY)) {
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 (osb) {
1225                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1226                 wake_up(&osb->osb_mount_event);
1227                 ocfs2_dismount_volume(sb, 1);
1228         }
1229
1230         if (status)
1231                 mlog_errno(status);
1232         return status;
1233 }
1234
1235 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1236                         int flags,
1237                         const char *dev_name,
1238                         void *data)
1239 {
1240         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1241 }
1242
1243 static void ocfs2_kill_sb(struct super_block *sb)
1244 {
1245         struct ocfs2_super *osb = OCFS2_SB(sb);
1246
1247         /* Failed mount? */
1248         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1249                 goto out;
1250
1251         /* Prevent further queueing of inode drop events */
1252         spin_lock(&dentry_list_lock);
1253         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1254         spin_unlock(&dentry_list_lock);
1255         /* Wait for work to finish and/or remove it */
1256         cancel_work_sync(&osb->dentry_lock_work);
1257 out:
1258         kill_block_super(sb);
1259 }
1260
1261 static struct file_system_type ocfs2_fs_type = {
1262         .owner          = THIS_MODULE,
1263         .name           = "ocfs2",
1264         .mount          = ocfs2_mount,
1265         .kill_sb        = ocfs2_kill_sb,
1266
1267         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1268         .next           = NULL
1269 };
1270 MODULE_ALIAS_FS("ocfs2");
1271
1272 static int ocfs2_check_set_options(struct super_block *sb,
1273                                    struct mount_options *options)
1274 {
1275         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1276             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1277                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1278                 mlog(ML_ERROR, "User quotas were requested, but this "
1279                      "filesystem does not have the feature enabled.\n");
1280                 return 0;
1281         }
1282         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1283             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1284                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1285                 mlog(ML_ERROR, "Group quotas were requested, but this "
1286                      "filesystem does not have the feature enabled.\n");
1287                 return 0;
1288         }
1289         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1290             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1291                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1292                      "feature is not enabled\n");
1293                 return 0;
1294         }
1295         /* No ACL setting specified? Use XATTR feature... */
1296         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1297                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1298                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1299                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1300                 else
1301                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1302         }
1303         return 1;
1304 }
1305
1306 static int ocfs2_parse_options(struct super_block *sb,
1307                                char *options,
1308                                struct mount_options *mopt,
1309                                int is_remount)
1310 {
1311         int status, user_stack = 0;
1312         char *p;
1313         u32 tmp;
1314
1315         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1316
1317         mopt->commit_interval = 0;
1318         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1319         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1320         mopt->slot = OCFS2_INVALID_SLOT;
1321         mopt->localalloc_opt = -1;
1322         mopt->cluster_stack[0] = '\0';
1323         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1324         mopt->dir_resv_level = -1;
1325
1326         if (!options) {
1327                 status = 1;
1328                 goto bail;
1329         }
1330
1331         while ((p = strsep(&options, ",")) != NULL) {
1332                 int token, option;
1333                 substring_t args[MAX_OPT_ARGS];
1334
1335                 if (!*p)
1336                         continue;
1337
1338                 token = match_token(p, tokens, args);
1339                 switch (token) {
1340                 case Opt_hb_local:
1341                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1342                         break;
1343                 case Opt_hb_none:
1344                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1345                         break;
1346                 case Opt_hb_global:
1347                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1348                         break;
1349                 case Opt_barrier:
1350                         if (match_int(&args[0], &option)) {
1351                                 status = 0;
1352                                 goto bail;
1353                         }
1354                         if (option)
1355                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1356                         else
1357                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1358                         break;
1359                 case Opt_intr:
1360                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1361                         break;
1362                 case Opt_nointr:
1363                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1364                         break;
1365                 case Opt_err_panic:
1366                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1367                         break;
1368                 case Opt_err_ro:
1369                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1370                         break;
1371                 case Opt_data_ordered:
1372                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1373                         break;
1374                 case Opt_data_writeback:
1375                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1376                         break;
1377                 case Opt_user_xattr:
1378                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1379                         break;
1380                 case Opt_nouser_xattr:
1381                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1382                         break;
1383                 case Opt_atime_quantum:
1384                         if (match_int(&args[0], &option)) {
1385                                 status = 0;
1386                                 goto bail;
1387                         }
1388                         if (option >= 0)
1389                                 mopt->atime_quantum = option;
1390                         break;
1391                 case Opt_slot:
1392                         option = 0;
1393                         if (match_int(&args[0], &option)) {
1394                                 status = 0;
1395                                 goto bail;
1396                         }
1397                         if (option)
1398                                 mopt->slot = (s16)option;
1399                         break;
1400                 case Opt_commit:
1401                         option = 0;
1402                         if (match_int(&args[0], &option)) {
1403                                 status = 0;
1404                                 goto bail;
1405                         }
1406                         if (option < 0)
1407                                 return 0;
1408                         if (option == 0)
1409                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1410                         mopt->commit_interval = HZ * option;
1411                         break;
1412                 case Opt_localalloc:
1413                         option = 0;
1414                         if (match_int(&args[0], &option)) {
1415                                 status = 0;
1416                                 goto bail;
1417                         }
1418                         if (option >= 0)
1419                                 mopt->localalloc_opt = option;
1420                         break;
1421                 case Opt_localflocks:
1422                         /*
1423                          * Changing this during remount could race
1424                          * flock() requests, or "unbalance" existing
1425                          * ones (e.g., a lock is taken in one mode but
1426                          * dropped in the other). If users care enough
1427                          * to flip locking modes during remount, we
1428                          * could add a "local" flag to individual
1429                          * flock structures for proper tracking of
1430                          * state.
1431                          */
1432                         if (!is_remount)
1433                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1434                         break;
1435                 case Opt_stack:
1436                         /* Check both that the option we were passed
1437                          * is of the right length and that it is a proper
1438                          * string of the right length.
1439                          */
1440                         if (((args[0].to - args[0].from) !=
1441                              OCFS2_STACK_LABEL_LEN) ||
1442                             (strnlen(args[0].from,
1443                                      OCFS2_STACK_LABEL_LEN) !=
1444                              OCFS2_STACK_LABEL_LEN)) {
1445                                 mlog(ML_ERROR,
1446                                      "Invalid cluster_stack option\n");
1447                                 status = 0;
1448                                 goto bail;
1449                         }
1450                         memcpy(mopt->cluster_stack, args[0].from,
1451                                OCFS2_STACK_LABEL_LEN);
1452                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1453                         /*
1454                          * Open code the memcmp here as we don't have
1455                          * an osb to pass to
1456                          * ocfs2_userspace_stack().
1457                          */
1458                         if (memcmp(mopt->cluster_stack,
1459                                    OCFS2_CLASSIC_CLUSTER_STACK,
1460                                    OCFS2_STACK_LABEL_LEN))
1461                                 user_stack = 1;
1462                         break;
1463                 case Opt_inode64:
1464                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1465                         break;
1466                 case Opt_usrquota:
1467                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1468                         break;
1469                 case Opt_grpquota:
1470                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1471                         break;
1472                 case Opt_coherency_buffered:
1473                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1474                         break;
1475                 case Opt_coherency_full:
1476                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1477                         break;
1478                 case Opt_acl:
1479                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1480                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1481                         break;
1482                 case Opt_noacl:
1483                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1484                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1485                         break;
1486                 case Opt_resv_level:
1487                         if (is_remount)
1488                                 break;
1489                         if (match_int(&args[0], &option)) {
1490                                 status = 0;
1491                                 goto bail;
1492                         }
1493                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1494                             option < OCFS2_MAX_RESV_LEVEL)
1495                                 mopt->resv_level = option;
1496                         break;
1497                 case Opt_dir_resv_level:
1498                         if (is_remount)
1499                                 break;
1500                         if (match_int(&args[0], &option)) {
1501                                 status = 0;
1502                                 goto bail;
1503                         }
1504                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1505                             option < OCFS2_MAX_RESV_LEVEL)
1506                                 mopt->dir_resv_level = option;
1507                         break;
1508                 default:
1509                         mlog(ML_ERROR,
1510                              "Unrecognized mount option \"%s\" "
1511                              "or missing value\n", p);
1512                         status = 0;
1513                         goto bail;
1514                 }
1515         }
1516
1517         if (user_stack == 0) {
1518                 /* Ensure only one heartbeat mode */
1519                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1520                                          OCFS2_MOUNT_HB_GLOBAL |
1521                                          OCFS2_MOUNT_HB_NONE);
1522                 if (hweight32(tmp) != 1) {
1523                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1524                         status = 0;
1525                         goto bail;
1526                 }
1527         }
1528
1529         status = 1;
1530
1531 bail:
1532         return status;
1533 }
1534
1535 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1536 {
1537         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1538         unsigned long opts = osb->s_mount_opt;
1539         unsigned int local_alloc_megs;
1540
1541         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1542                 seq_printf(s, ",_netdev");
1543                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1544                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1545                 else
1546                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1547         } else
1548                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1549
1550         if (opts & OCFS2_MOUNT_NOINTR)
1551                 seq_printf(s, ",nointr");
1552
1553         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1554                 seq_printf(s, ",data=writeback");
1555         else
1556                 seq_printf(s, ",data=ordered");
1557
1558         if (opts & OCFS2_MOUNT_BARRIER)
1559                 seq_printf(s, ",barrier=1");
1560
1561         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1562                 seq_printf(s, ",errors=panic");
1563         else
1564                 seq_printf(s, ",errors=remount-ro");
1565
1566         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1567                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1568
1569         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1570
1571         if (osb->osb_commit_interval)
1572                 seq_printf(s, ",commit=%u",
1573                            (unsigned) (osb->osb_commit_interval / HZ));
1574
1575         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1576         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1577                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1578
1579         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1580                 seq_printf(s, ",localflocks,");
1581
1582         if (osb->osb_cluster_stack[0])
1583                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1584                            osb->osb_cluster_stack);
1585         if (opts & OCFS2_MOUNT_USRQUOTA)
1586                 seq_printf(s, ",usrquota");
1587         if (opts & OCFS2_MOUNT_GRPQUOTA)
1588                 seq_printf(s, ",grpquota");
1589
1590         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1591                 seq_printf(s, ",coherency=buffered");
1592         else
1593                 seq_printf(s, ",coherency=full");
1594
1595         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1596                 seq_printf(s, ",nouser_xattr");
1597         else
1598                 seq_printf(s, ",user_xattr");
1599
1600         if (opts & OCFS2_MOUNT_INODE64)
1601                 seq_printf(s, ",inode64");
1602
1603         if (opts & OCFS2_MOUNT_POSIX_ACL)
1604                 seq_printf(s, ",acl");
1605         else
1606                 seq_printf(s, ",noacl");
1607
1608         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1609                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1610
1611         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1612                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1613
1614         return 0;
1615 }
1616
1617 wait_queue_head_t ocfs2__ioend_wq[OCFS2_IOEND_WQ_HASH_SZ];
1618
1619 static int __init ocfs2_init(void)
1620 {
1621         int status, i;
1622
1623         for (i = 0; i < OCFS2_IOEND_WQ_HASH_SZ; i++)
1624                 init_waitqueue_head(&ocfs2__ioend_wq[i]);
1625
1626         status = init_ocfs2_uptodate_cache();
1627         if (status < 0)
1628                 goto out1;
1629
1630         status = ocfs2_initialize_mem_caches();
1631         if (status < 0)
1632                 goto out2;
1633
1634         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1635         if (!ocfs2_wq) {
1636                 status = -ENOMEM;
1637                 goto out3;
1638         }
1639
1640         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1641         if (!ocfs2_debugfs_root) {
1642                 status = -EFAULT;
1643                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1644         }
1645
1646         ocfs2_set_locking_protocol();
1647
1648         status = register_quota_format(&ocfs2_quota_format);
1649         if (status < 0)
1650                 goto out4;
1651         status = register_filesystem(&ocfs2_fs_type);
1652         if (!status)
1653                 return 0;
1654
1655         unregister_quota_format(&ocfs2_quota_format);
1656 out4:
1657         destroy_workqueue(ocfs2_wq);
1658         debugfs_remove(ocfs2_debugfs_root);
1659 out3:
1660         ocfs2_free_mem_caches();
1661 out2:
1662         exit_ocfs2_uptodate_cache();
1663 out1:
1664         mlog_errno(status);
1665         return status;
1666 }
1667
1668 static void __exit ocfs2_exit(void)
1669 {
1670         if (ocfs2_wq) {
1671                 flush_workqueue(ocfs2_wq);
1672                 destroy_workqueue(ocfs2_wq);
1673         }
1674
1675         unregister_quota_format(&ocfs2_quota_format);
1676
1677         debugfs_remove(ocfs2_debugfs_root);
1678
1679         ocfs2_free_mem_caches();
1680
1681         unregister_filesystem(&ocfs2_fs_type);
1682
1683         exit_ocfs2_uptodate_cache();
1684 }
1685
1686 static void ocfs2_put_super(struct super_block *sb)
1687 {
1688         trace_ocfs2_put_super(sb);
1689
1690         ocfs2_sync_blockdev(sb);
1691         ocfs2_dismount_volume(sb, 0);
1692 }
1693
1694 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1695 {
1696         struct ocfs2_super *osb;
1697         u32 numbits, freebits;
1698         int status;
1699         struct ocfs2_dinode *bm_lock;
1700         struct buffer_head *bh = NULL;
1701         struct inode *inode = NULL;
1702
1703         trace_ocfs2_statfs(dentry->d_sb, buf);
1704
1705         osb = OCFS2_SB(dentry->d_sb);
1706
1707         inode = ocfs2_get_system_file_inode(osb,
1708                                             GLOBAL_BITMAP_SYSTEM_INODE,
1709                                             OCFS2_INVALID_SLOT);
1710         if (!inode) {
1711                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1712                 status = -EIO;
1713                 goto bail;
1714         }
1715
1716         status = ocfs2_inode_lock(inode, &bh, 0);
1717         if (status < 0) {
1718                 mlog_errno(status);
1719                 goto bail;
1720         }
1721
1722         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1723
1724         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1725         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1726
1727         buf->f_type = OCFS2_SUPER_MAGIC;
1728         buf->f_bsize = dentry->d_sb->s_blocksize;
1729         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1730         buf->f_blocks = ((sector_t) numbits) *
1731                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1732         buf->f_bfree = ((sector_t) freebits) *
1733                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1734         buf->f_bavail = buf->f_bfree;
1735         buf->f_files = numbits;
1736         buf->f_ffree = freebits;
1737         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1738                                 & 0xFFFFFFFFUL;
1739         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1740                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1741
1742         brelse(bh);
1743
1744         ocfs2_inode_unlock(inode, 0);
1745         status = 0;
1746 bail:
1747         if (inode)
1748                 iput(inode);
1749
1750         if (status)
1751                 mlog_errno(status);
1752
1753         return status;
1754 }
1755
1756 static void ocfs2_inode_init_once(void *data)
1757 {
1758         struct ocfs2_inode_info *oi = data;
1759
1760         oi->ip_flags = 0;
1761         oi->ip_open_count = 0;
1762         spin_lock_init(&oi->ip_lock);
1763         ocfs2_extent_map_init(&oi->vfs_inode);
1764         INIT_LIST_HEAD(&oi->ip_io_markers);
1765         oi->ip_dir_start_lookup = 0;
1766         atomic_set(&oi->ip_unaligned_aio, 0);
1767         init_rwsem(&oi->ip_alloc_sem);
1768         init_rwsem(&oi->ip_xattr_sem);
1769         mutex_init(&oi->ip_io_mutex);
1770
1771         oi->ip_blkno = 0ULL;
1772         oi->ip_clusters = 0;
1773
1774         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1775
1776         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1777         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1778         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1779
1780         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1781                                   &ocfs2_inode_caching_ops);
1782
1783         inode_init_once(&oi->vfs_inode);
1784 }
1785
1786 static int ocfs2_initialize_mem_caches(void)
1787 {
1788         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1789                                        sizeof(struct ocfs2_inode_info),
1790                                        0,
1791                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1792                                                 SLAB_MEM_SPREAD),
1793                                        ocfs2_inode_init_once);
1794         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1795                                         sizeof(struct ocfs2_dquot),
1796                                         0,
1797                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1798                                                 SLAB_MEM_SPREAD),
1799                                         NULL);
1800         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1801                                         sizeof(struct ocfs2_quota_chunk),
1802                                         0,
1803                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1804                                         NULL);
1805         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1806             !ocfs2_qf_chunk_cachep) {
1807                 if (ocfs2_inode_cachep)
1808                         kmem_cache_destroy(ocfs2_inode_cachep);
1809                 if (ocfs2_dquot_cachep)
1810                         kmem_cache_destroy(ocfs2_dquot_cachep);
1811                 if (ocfs2_qf_chunk_cachep)
1812                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1813                 return -ENOMEM;
1814         }
1815
1816         return 0;
1817 }
1818
1819 static void ocfs2_free_mem_caches(void)
1820 {
1821         /*
1822          * Make sure all delayed rcu free inodes are flushed before we
1823          * destroy cache.
1824          */
1825         rcu_barrier();
1826         if (ocfs2_inode_cachep)
1827                 kmem_cache_destroy(ocfs2_inode_cachep);
1828         ocfs2_inode_cachep = NULL;
1829
1830         if (ocfs2_dquot_cachep)
1831                 kmem_cache_destroy(ocfs2_dquot_cachep);
1832         ocfs2_dquot_cachep = NULL;
1833
1834         if (ocfs2_qf_chunk_cachep)
1835                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1836         ocfs2_qf_chunk_cachep = NULL;
1837 }
1838
1839 static int ocfs2_get_sector(struct super_block *sb,
1840                             struct buffer_head **bh,
1841                             int block,
1842                             int sect_size)
1843 {
1844         if (!sb_set_blocksize(sb, sect_size)) {
1845                 mlog(ML_ERROR, "unable to set blocksize\n");
1846                 return -EIO;
1847         }
1848
1849         *bh = sb_getblk(sb, block);
1850         if (!*bh) {
1851                 mlog_errno(-ENOMEM);
1852                 return -ENOMEM;
1853         }
1854         lock_buffer(*bh);
1855         if (!buffer_dirty(*bh))
1856                 clear_buffer_uptodate(*bh);
1857         unlock_buffer(*bh);
1858         ll_rw_block(READ, 1, bh);
1859         wait_on_buffer(*bh);
1860         if (!buffer_uptodate(*bh)) {
1861                 mlog_errno(-EIO);
1862                 brelse(*bh);
1863                 *bh = NULL;
1864                 return -EIO;
1865         }
1866
1867         return 0;
1868 }
1869
1870 static int ocfs2_mount_volume(struct super_block *sb)
1871 {
1872         int status = 0;
1873         int unlock_super = 0;
1874         struct ocfs2_super *osb = OCFS2_SB(sb);
1875
1876         if (ocfs2_is_hard_readonly(osb))
1877                 goto leave;
1878
1879         status = ocfs2_dlm_init(osb);
1880         if (status < 0) {
1881                 mlog_errno(status);
1882                 goto leave;
1883         }
1884
1885         status = ocfs2_super_lock(osb, 1);
1886         if (status < 0) {
1887                 mlog_errno(status);
1888                 goto leave;
1889         }
1890         unlock_super = 1;
1891
1892         /* This will load up the node map and add ourselves to it. */
1893         status = ocfs2_find_slot(osb);
1894         if (status < 0) {
1895                 mlog_errno(status);
1896                 goto leave;
1897         }
1898
1899         /* load all node-local system inodes */
1900         status = ocfs2_init_local_system_inodes(osb);
1901         if (status < 0) {
1902                 mlog_errno(status);
1903                 goto leave;
1904         }
1905
1906         status = ocfs2_check_volume(osb);
1907         if (status < 0) {
1908                 mlog_errno(status);
1909                 goto leave;
1910         }
1911
1912         status = ocfs2_truncate_log_init(osb);
1913         if (status < 0)
1914                 mlog_errno(status);
1915
1916 leave:
1917         if (unlock_super)
1918                 ocfs2_super_unlock(osb, 1);
1919
1920         return status;
1921 }
1922
1923 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1924 {
1925         int tmp, hangup_needed = 0;
1926         struct ocfs2_super *osb = NULL;
1927         char nodestr[12];
1928
1929         trace_ocfs2_dismount_volume(sb);
1930
1931         BUG_ON(!sb);
1932         osb = OCFS2_SB(sb);
1933         BUG_ON(!osb);
1934
1935         debugfs_remove(osb->osb_ctxt);
1936
1937         /*
1938          * Flush inode dropping work queue so that deletes are
1939          * performed while the filesystem is still working
1940          */
1941         ocfs2_drop_all_dl_inodes(osb);
1942
1943         /* Orphan scan should be stopped as early as possible */
1944         ocfs2_orphan_scan_stop(osb);
1945
1946         ocfs2_disable_quotas(osb);
1947
1948         ocfs2_shutdown_local_alloc(osb);
1949
1950         /* This will disable recovery and flush any recovery work. */
1951         ocfs2_recovery_exit(osb);
1952
1953         /*
1954          * During dismount, when it recovers another node it will call
1955          * ocfs2_recover_orphans and queue delayed work osb_truncate_log_wq.
1956          */
1957         ocfs2_truncate_log_shutdown(osb);
1958
1959         ocfs2_journal_shutdown(osb);
1960
1961         ocfs2_sync_blockdev(sb);
1962
1963         ocfs2_purge_refcount_trees(osb);
1964
1965         /* No cluster connection means we've failed during mount, so skip
1966          * all the steps which depended on that to complete. */
1967         if (osb->cconn) {
1968                 tmp = ocfs2_super_lock(osb, 1);
1969                 if (tmp < 0) {
1970                         mlog_errno(tmp);
1971                         return;
1972                 }
1973         }
1974
1975         if (osb->slot_num != OCFS2_INVALID_SLOT)
1976                 ocfs2_put_slot(osb);
1977
1978         if (osb->cconn)
1979                 ocfs2_super_unlock(osb, 1);
1980
1981         ocfs2_release_system_inodes(osb);
1982
1983         /*
1984          * If we're dismounting due to mount error, mount.ocfs2 will clean
1985          * up heartbeat.  If we're a local mount, there is no heartbeat.
1986          * If we failed before we got a uuid_str yet, we can't stop
1987          * heartbeat.  Otherwise, do it.
1988          */
1989         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1990             !ocfs2_is_hard_readonly(osb))
1991                 hangup_needed = 1;
1992
1993         if (osb->cconn)
1994                 ocfs2_dlm_shutdown(osb, hangup_needed);
1995
1996         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1997         debugfs_remove(osb->osb_debug_root);
1998
1999         if (hangup_needed)
2000                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
2001
2002         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
2003
2004         if (ocfs2_mount_local(osb))
2005                 snprintf(nodestr, sizeof(nodestr), "local");
2006         else
2007                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2008
2009         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2010                osb->dev_str, nodestr);
2011
2012         ocfs2_delete_osb(osb);
2013         kfree(osb);
2014         sb->s_dev = 0;
2015         sb->s_fs_info = NULL;
2016 }
2017
2018 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2019                                 unsigned uuid_bytes)
2020 {
2021         int i, ret;
2022         char *ptr;
2023
2024         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2025
2026         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2027         if (osb->uuid_str == NULL)
2028                 return -ENOMEM;
2029
2030         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2031                 /* print with null */
2032                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2033                 if (ret != 2) /* drop super cleans up */
2034                         return -EINVAL;
2035                 /* then only advance past the last char */
2036                 ptr += 2;
2037         }
2038
2039         return 0;
2040 }
2041
2042 /* Make sure entire volume is addressable by our journal.  Requires
2043    osb_clusters_at_boot to be valid and for the journal to have been
2044    initialized by ocfs2_journal_init(). */
2045 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2046 {
2047         int status = 0;
2048         u64 max_block =
2049                 ocfs2_clusters_to_blocks(osb->sb,
2050                                          osb->osb_clusters_at_boot) - 1;
2051
2052         /* 32-bit block number is always OK. */
2053         if (max_block <= (u32)~0ULL)
2054                 goto out;
2055
2056         /* Volume is "huge", so see if our journal is new enough to
2057            support it. */
2058         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2059                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2060               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2061                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2062                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2063                      "Enable the 'block64' journal option with tunefs.ocfs2");
2064                 status = -EFBIG;
2065                 goto out;
2066         }
2067
2068  out:
2069         return status;
2070 }
2071
2072 static int ocfs2_initialize_super(struct super_block *sb,
2073                                   struct buffer_head *bh,
2074                                   int sector_size,
2075                                   struct ocfs2_blockcheck_stats *stats)
2076 {
2077         int status;
2078         int i, cbits, bbits;
2079         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2080         struct inode *inode = NULL;
2081         struct ocfs2_journal *journal;
2082         __le32 uuid_net_key;
2083         struct ocfs2_super *osb;
2084         u64 total_blocks;
2085
2086         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2087         if (!osb) {
2088                 status = -ENOMEM;
2089                 mlog_errno(status);
2090                 goto bail;
2091         }
2092
2093         sb->s_fs_info = osb;
2094         sb->s_op = &ocfs2_sops;
2095         sb->s_d_op = &ocfs2_dentry_ops;
2096         sb->s_export_op = &ocfs2_export_ops;
2097         sb->s_qcop = &ocfs2_quotactl_ops;
2098         sb->dq_op = &ocfs2_quota_operations;
2099         sb->s_xattr = ocfs2_xattr_handlers;
2100         sb->s_time_gran = 1;
2101         sb->s_flags |= MS_NOATIME;
2102         /* this is needed to support O_LARGEFILE */
2103         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2104         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2105         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2106
2107         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2108
2109         for (i = 0; i < 3; i++)
2110                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2111         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2112
2113         osb->sb = sb;
2114         /* Save off for ocfs2_rw_direct */
2115         osb->s_sectsize_bits = blksize_bits(sector_size);
2116         BUG_ON(!osb->s_sectsize_bits);
2117
2118         spin_lock_init(&osb->dc_task_lock);
2119         init_waitqueue_head(&osb->dc_event);
2120         osb->dc_work_sequence = 0;
2121         osb->dc_wake_sequence = 0;
2122         INIT_LIST_HEAD(&osb->blocked_lock_list);
2123         osb->blocked_lock_count = 0;
2124         spin_lock_init(&osb->osb_lock);
2125         spin_lock_init(&osb->osb_xattr_lock);
2126         ocfs2_init_steal_slots(osb);
2127
2128         atomic_set(&osb->alloc_stats.moves, 0);
2129         atomic_set(&osb->alloc_stats.local_data, 0);
2130         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2131         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2132         atomic_set(&osb->alloc_stats.bg_extends, 0);
2133
2134         /* Copy the blockcheck stats from the superblock probe */
2135         osb->osb_ecc_stats = *stats;
2136
2137         ocfs2_init_node_maps(osb);
2138
2139         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2140                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2141
2142         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2143         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2144                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2145                      osb->max_slots);
2146                 status = -EINVAL;
2147                 goto bail;
2148         }
2149
2150         ocfs2_orphan_scan_init(osb);
2151
2152         status = ocfs2_recovery_init(osb);
2153         if (status) {
2154                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2155                 mlog_errno(status);
2156                 goto bail;
2157         }
2158
2159         init_waitqueue_head(&osb->checkpoint_event);
2160
2161         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2162
2163         osb->slot_num = OCFS2_INVALID_SLOT;
2164
2165         osb->s_xattr_inline_size = le16_to_cpu(
2166                                         di->id2.i_super.s_xattr_inline_size);
2167
2168         osb->local_alloc_state = OCFS2_LA_UNUSED;
2169         osb->local_alloc_bh = NULL;
2170         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2171
2172         init_waitqueue_head(&osb->osb_mount_event);
2173
2174         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2175         if (status) {
2176                 mlog_errno(status);
2177                 goto bail;
2178         }
2179
2180         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2181         if (!osb->vol_label) {
2182                 mlog(ML_ERROR, "unable to alloc vol label\n");
2183                 status = -ENOMEM;
2184                 goto bail;
2185         }
2186
2187         osb->slot_recovery_generations =
2188                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2189                         GFP_KERNEL);
2190         if (!osb->slot_recovery_generations) {
2191                 status = -ENOMEM;
2192                 mlog_errno(status);
2193                 goto bail;
2194         }
2195
2196         init_waitqueue_head(&osb->osb_wipe_event);
2197         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2198                                         sizeof(*osb->osb_orphan_wipes),
2199                                         GFP_KERNEL);
2200         if (!osb->osb_orphan_wipes) {
2201                 status = -ENOMEM;
2202                 mlog_errno(status);
2203                 goto bail;
2204         }
2205
2206         osb->osb_rf_lock_tree = RB_ROOT;
2207
2208         osb->s_feature_compat =
2209                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2210         osb->s_feature_ro_compat =
2211                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2212         osb->s_feature_incompat =
2213                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2214
2215         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2216                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2217                      "optional features (%x).\n", i);
2218                 status = -EINVAL;
2219                 goto bail;
2220         }
2221         if (!(osb->sb->s_flags & MS_RDONLY) &&
2222             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2223                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2224                      "unsupported optional features (%x).\n", i);
2225                 status = -EINVAL;
2226                 goto bail;
2227         }
2228
2229         if (ocfs2_clusterinfo_valid(osb)) {
2230                 osb->osb_stackflags =
2231                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2232                 strlcpy(osb->osb_cluster_stack,
2233                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2234                        OCFS2_STACK_LABEL_LEN + 1);
2235                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2236                         mlog(ML_ERROR,
2237                              "couldn't mount because of an invalid "
2238                              "cluster stack label (%s) \n",
2239                              osb->osb_cluster_stack);
2240                         status = -EINVAL;
2241                         goto bail;
2242                 }
2243                 strlcpy(osb->osb_cluster_name,
2244                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2245                         OCFS2_CLUSTER_NAME_LEN + 1);
2246         } else {
2247                 /* The empty string is identical with classic tools that
2248                  * don't know about s_cluster_info. */
2249                 osb->osb_cluster_stack[0] = '\0';
2250         }
2251
2252         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2253
2254         /* FIXME
2255          * This should be done in ocfs2_journal_init(), but unknown
2256          * ordering issues will cause the filesystem to crash.
2257          * If anyone wants to figure out what part of the code
2258          * refers to osb->journal before ocfs2_journal_init() is run,
2259          * be my guest.
2260          */
2261         /* initialize our journal structure */
2262
2263         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2264         if (!journal) {
2265                 mlog(ML_ERROR, "unable to alloc journal\n");
2266                 status = -ENOMEM;
2267                 goto bail;
2268         }
2269         osb->journal = journal;
2270         journal->j_osb = osb;
2271
2272         atomic_set(&journal->j_num_trans, 0);
2273         init_rwsem(&journal->j_trans_barrier);
2274         init_waitqueue_head(&journal->j_checkpointed);
2275         spin_lock_init(&journal->j_lock);
2276         journal->j_trans_id = (unsigned long) 1;
2277         INIT_LIST_HEAD(&journal->j_la_cleanups);
2278         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2279         journal->j_state = OCFS2_JOURNAL_FREE;
2280
2281         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2282         osb->dentry_lock_list = NULL;
2283
2284         /* get some pseudo constants for clustersize bits */
2285         osb->s_clustersize_bits =
2286                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2287         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2288
2289         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2290             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2291                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2292                      osb->s_clustersize);
2293                 status = -EINVAL;
2294                 goto bail;
2295         }
2296
2297         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2298                                                 le32_to_cpu(di->i_clusters));
2299
2300         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2301                                            total_blocks);
2302         if (status) {
2303                 mlog(ML_ERROR, "Volume too large "
2304                      "to mount safely on this system");
2305                 status = -EFBIG;
2306                 goto bail;
2307         }
2308
2309         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2310                                  sizeof(di->id2.i_super.s_uuid))) {
2311                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2312                 status = -ENOMEM;
2313                 goto bail;
2314         }
2315
2316         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2317
2318         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2319         osb->vol_label[63] = '\0';
2320         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2321         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2322         osb->first_cluster_group_blkno =
2323                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2324         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2325         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2326         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2327                                      (unsigned long long)osb->root_blkno,
2328                                      (unsigned long long)osb->system_dir_blkno,
2329                                      osb->s_clustersize_bits);
2330
2331         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2332         if (!osb->osb_dlm_debug) {
2333                 status = -ENOMEM;
2334                 mlog_errno(status);
2335                 goto bail;
2336         }
2337
2338         atomic_set(&osb->vol_state, VOLUME_INIT);
2339
2340         /* load root, system_dir, and all global system inodes */
2341         status = ocfs2_init_global_system_inodes(osb);
2342         if (status < 0) {
2343                 mlog_errno(status);
2344                 goto bail;
2345         }
2346
2347         /*
2348          * global bitmap
2349          */
2350         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2351                                             OCFS2_INVALID_SLOT);
2352         if (!inode) {
2353                 status = -EINVAL;
2354                 mlog_errno(status);
2355                 goto bail;
2356         }
2357
2358         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2359         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2360         iput(inode);
2361
2362         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2363                                  osb->s_feature_incompat) * 8;
2364
2365         status = ocfs2_init_slot_info(osb);
2366         if (status < 0) {
2367                 mlog_errno(status);
2368                 goto bail;
2369         }
2370         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2371
2372 bail:
2373         return status;
2374 }
2375
2376 /*
2377  * will return: -EAGAIN if it is ok to keep searching for superblocks
2378  *              -EINVAL if there is a bad superblock
2379  *              0 on success
2380  */
2381 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2382                                struct buffer_head *bh,
2383                                u32 blksz,
2384                                struct ocfs2_blockcheck_stats *stats)
2385 {
2386         int status = -EAGAIN;
2387
2388         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2389                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2390                 /* We have to do a raw check of the feature here */
2391                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2392                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2393                         status = ocfs2_block_check_validate(bh->b_data,
2394                                                             bh->b_size,
2395                                                             &di->i_check,
2396                                                             stats);
2397                         if (status)
2398                                 goto out;
2399                 }
2400                 status = -EINVAL;
2401                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2402                         mlog(ML_ERROR, "found superblock with incorrect block "
2403                              "size: found %u, should be %u\n",
2404                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2405                                blksz);
2406                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2407                            OCFS2_MAJOR_REV_LEVEL ||
2408                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2409                            OCFS2_MINOR_REV_LEVEL) {
2410                         mlog(ML_ERROR, "found superblock with bad version: "
2411                              "found %u.%u, should be %u.%u\n",
2412                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2413                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2414                              OCFS2_MAJOR_REV_LEVEL,
2415                              OCFS2_MINOR_REV_LEVEL);
2416                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2417                         mlog(ML_ERROR, "bad block number on superblock: "
2418                              "found %llu, should be %llu\n",
2419                              (unsigned long long)le64_to_cpu(di->i_blkno),
2420                              (unsigned long long)bh->b_blocknr);
2421                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2422                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2423                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2424                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2425                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2426                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2427                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2428                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2429                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2430                         mlog(ML_ERROR,
2431                              "Superblock slots found greater than file system "
2432                              "maximum: found %u, max %u\n",
2433                              le16_to_cpu(di->id2.i_super.s_max_slots),
2434                              OCFS2_MAX_SLOTS);
2435                 } else {
2436                         /* found it! */
2437                         status = 0;
2438                 }
2439         }
2440
2441 out:
2442         if (status && status != -EAGAIN)
2443                 mlog_errno(status);
2444         return status;
2445 }
2446
2447 static int ocfs2_check_volume(struct ocfs2_super *osb)
2448 {
2449         int status;
2450         int dirty;
2451         int local;
2452         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2453                                                   * recover
2454                                                   * ourselves. */
2455
2456         /* Init our journal object. */
2457         status = ocfs2_journal_init(osb->journal, &dirty);
2458         if (status < 0) {
2459                 mlog(ML_ERROR, "Could not initialize journal!\n");
2460                 goto finally;
2461         }
2462
2463         /* Now that journal has been initialized, check to make sure
2464            entire volume is addressable. */
2465         status = ocfs2_journal_addressable(osb);
2466         if (status)
2467                 goto finally;
2468
2469         /* If the journal was unmounted cleanly then we don't want to
2470          * recover anything. Otherwise, journal_load will do that
2471          * dirty work for us :) */
2472         if (!dirty) {
2473                 status = ocfs2_journal_wipe(osb->journal, 0);
2474                 if (status < 0) {
2475                         mlog_errno(status);
2476                         goto finally;
2477                 }
2478         } else {
2479                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2480                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2481         }
2482
2483         local = ocfs2_mount_local(osb);
2484
2485         /* will play back anything left in the journal. */
2486         status = ocfs2_journal_load(osb->journal, local, dirty);
2487         if (status < 0) {
2488                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2489                 goto finally;
2490         }
2491
2492         if (dirty) {
2493                 /* recover my local alloc if we didn't unmount cleanly. */
2494                 status = ocfs2_begin_local_alloc_recovery(osb,
2495                                                           osb->slot_num,
2496                                                           &local_alloc);
2497                 if (status < 0) {
2498                         mlog_errno(status);
2499                         goto finally;
2500                 }
2501                 /* we complete the recovery process after we've marked
2502                  * ourselves as mounted. */
2503         }
2504
2505         status = ocfs2_load_local_alloc(osb);
2506         if (status < 0) {
2507                 mlog_errno(status);
2508                 goto finally;
2509         }
2510
2511         if (dirty) {
2512                 /* Recovery will be completed after we've mounted the
2513                  * rest of the volume. */
2514                 osb->dirty = 1;
2515                 osb->local_alloc_copy = local_alloc;
2516                 local_alloc = NULL;
2517         }
2518
2519         /* go through each journal, trylock it and if you get the
2520          * lock, and it's marked as dirty, set the bit in the recover
2521          * map and launch a recovery thread for it. */
2522         status = ocfs2_mark_dead_nodes(osb);
2523         if (status < 0) {
2524                 mlog_errno(status);
2525                 goto finally;
2526         }
2527
2528         status = ocfs2_compute_replay_slots(osb);
2529         if (status < 0)
2530                 mlog_errno(status);
2531
2532 finally:
2533         kfree(local_alloc);
2534
2535         if (status)
2536                 mlog_errno(status);
2537         return status;
2538 }
2539
2540 /*
2541  * The routine gets called from dismount or close whenever a dismount on
2542  * volume is requested and the osb open count becomes 1.
2543  * It will remove the osb from the global list and also free up all the
2544  * initialized resources and fileobject.
2545  */
2546 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2547 {
2548         /* This function assumes that the caller has the main osb resource */
2549
2550         ocfs2_free_slot_info(osb);
2551
2552         kfree(osb->osb_orphan_wipes);
2553         kfree(osb->slot_recovery_generations);
2554         /* FIXME
2555          * This belongs in journal shutdown, but because we have to
2556          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2557          * we free it here.
2558          */
2559         kfree(osb->journal);
2560         kfree(osb->local_alloc_copy);
2561         kfree(osb->uuid_str);
2562         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2563         memset(osb, 0, sizeof(struct ocfs2_super));
2564 }
2565
2566 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2567  * panic(). We do not support continue-on-error operation. */
2568 static void ocfs2_handle_error(struct super_block *sb)
2569 {
2570         struct ocfs2_super *osb = OCFS2_SB(sb);
2571
2572         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2573                 panic("OCFS2: (device %s): panic forced after error\n",
2574                       sb->s_id);
2575
2576         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2577
2578         if (sb->s_flags & MS_RDONLY &&
2579             (ocfs2_is_soft_readonly(osb) ||
2580              ocfs2_is_hard_readonly(osb)))
2581                 return;
2582
2583         printk(KERN_CRIT "File system is now read-only due to the potential "
2584                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2585                "system is unmounted.\n");
2586         sb->s_flags |= MS_RDONLY;
2587         ocfs2_set_ro_flag(osb, 0);
2588 }
2589
2590 static char error_buf[1024];
2591
2592 void __ocfs2_error(struct super_block *sb,
2593                    const char *function,
2594                    const char *fmt, ...)
2595 {
2596         va_list args;
2597
2598         va_start(args, fmt);
2599         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2600         va_end(args);
2601
2602         /* Not using mlog here because we want to show the actual
2603          * function the error came from. */
2604         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2605                sb->s_id, function, error_buf);
2606
2607         ocfs2_handle_error(sb);
2608 }
2609
2610 /* Handle critical errors. This is intentionally more drastic than
2611  * ocfs2_handle_error, so we only use for things like journal errors,
2612  * etc. */
2613 void __ocfs2_abort(struct super_block* sb,
2614                    const char *function,
2615                    const char *fmt, ...)
2616 {
2617         va_list args;
2618
2619         va_start(args, fmt);
2620         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2621         va_end(args);
2622
2623         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2624                sb->s_id, function, error_buf);
2625
2626         /* We don't have the cluster support yet to go straight to
2627          * hard readonly in here. Until then, we want to keep
2628          * ocfs2_abort() so that we can at least mark critical
2629          * errors.
2630          *
2631          * TODO: This should abort the journal and alert other nodes
2632          * that our slot needs recovery. */
2633
2634         /* Force a panic(). This stinks, but it's better than letting
2635          * things continue without having a proper hard readonly
2636          * here. */
2637         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2638                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2639         ocfs2_handle_error(sb);
2640 }
2641
2642 /*
2643  * Void signal blockers, because in-kernel sigprocmask() only fails
2644  * when SIG_* is wrong.
2645  */
2646 void ocfs2_block_signals(sigset_t *oldset)
2647 {
2648         int rc;
2649         sigset_t blocked;
2650
2651         sigfillset(&blocked);
2652         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2653         BUG_ON(rc);
2654 }
2655
2656 void ocfs2_unblock_signals(sigset_t *oldset)
2657 {
2658         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2659         BUG_ON(rc);
2660 }
2661
2662 module_init(ocfs2_init);
2663 module_exit(ocfs2_exit);