Pileus Git - ~andy/linux/blob - 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         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 635             !ocfs2_check_set_options(sb, &parsed_options)) {
 636                 ret = -EINVAL;
 637                 goto out;
 638         }
 639
 640         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 641                 OCFS2_MOUNT_HB_NONE;
 642         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 643                 ret = -EINVAL;
 644                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 645                 goto out;
 646         }
 647
 648         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 649             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 650                 ret = -EINVAL;
 651                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 652                 goto out;
 653         }
 654
 655         /* Probably don't want this on remount; it might
 656          * mess with other nodes */
 657         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 658             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 659                 ret = -EINVAL;
 660                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 661                 goto out;
 662         }
 663
 664         /* We're going to/from readonly mode. */
 665         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 666                 /* Disable quota accounting before remounting RO */
 667                 if (*flags & MS_RDONLY) {
 668                         ret = ocfs2_susp_quotas(osb, 0);
 669                         if (ret < 0)
 670                                 goto out;
 671                 }
 672                 /* Lock here so the check of HARD_RO and the potential
 673                  * setting of SOFT_RO is atomic. */
 674                 spin_lock(&osb->osb_lock);
 675                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 676                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 677                         ret = -EROFS;
 678                         goto unlock_osb;
 679                 }
 680
 681                 if (*flags & MS_RDONLY) {
 682                         sb->s_flags |= MS_RDONLY;
 683                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 684                 } else {
 685                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 686                                 mlog(ML_ERROR, "Cannot remount RDWR "
 687                                      "filesystem due to previous errors.\n");
 688                                 ret = -EROFS;
 689                                 goto unlock_osb;
 690                         }
 691                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 692                         if (incompat_features) {
 693                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 694                                      "of unsupported optional features "
 695                                      "(%x).\n", incompat_features);
 696                                 ret = -EINVAL;
 697                                 goto unlock_osb;
 698                         }
 699                         sb->s_flags &= ~MS_RDONLY;
 700                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 701                 }
 702                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 703 unlock_osb:
 704                 spin_unlock(&osb->osb_lock);
 705                 /* Enable quota accounting after remounting RW */
 706                 if (!ret && !(*flags & MS_RDONLY)) {
 707                         if (sb_any_quota_suspended(sb))
 708                                 ret = ocfs2_susp_quotas(osb, 1);
 709                         else
 710                                 ret = ocfs2_enable_quotas(osb);
 711                         if (ret < 0) {
 712                                 /* Return back changes... */
 713                                 spin_lock(&osb->osb_lock);
 714                                 sb->s_flags |= MS_RDONLY;
 715                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 716                                 spin_unlock(&osb->osb_lock);
 717                                 goto out;
 718                         }
 719                 }
 720         }
 721
 722         if (!ret) {
 723                 /* Only save off the new mount options in case of a successful
 724                  * remount. */
 725                 osb->s_mount_opt = parsed_options.mount_opt;
 726                 osb->s_atime_quantum = parsed_options.atime_quantum;
 727                 osb->preferred_slot = parsed_options.slot;
 728                 if (parsed_options.commit_interval)
 729                         osb->osb_commit_interval = parsed_options.commit_interval;
 730
 731                 if (!ocfs2_is_hard_readonly(osb))
 732                         ocfs2_set_journal_params(osb);
 733
 734                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 735                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 736                                                         MS_POSIXACL : 0);
 737         }
 738 out:
 739         return ret;
 740 }
 741
 742 static int ocfs2_sb_probe(struct super_block *sb,
 743                           struct buffer_head **bh,
 744                           int *sector_size,
 745                           struct ocfs2_blockcheck_stats *stats)
 746 {
 747         int status, tmpstat;
 748         struct ocfs1_vol_disk_hdr *hdr;
 749         struct ocfs2_dinode *di;
 750         int blksize;
 751
 752         *bh = NULL;
 753
 754         /* may be > 512 */
 755         *sector_size = bdev_logical_block_size(sb->s_bdev);
 756         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 757                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 758                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 759                 status = -EINVAL;
 760                 goto bail;
 761         }
 762
 763         /* Can this really happen? */
 764         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 765                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 766
 767         /* check block zero for old format */
 768         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 769         if (status < 0) {
 770                 mlog_errno(status);
 771                 goto bail;
 772         }
 773         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 774         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 775                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 776                      hdr->major_version, hdr->minor_version);
 777                 status = -EINVAL;
 778         }
 779         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 780                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 781                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 782                      hdr->signature);
 783                 status = -EINVAL;
 784         }
 785         brelse(*bh);
 786         *bh = NULL;
 787         if (status < 0) {
 788                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 789                      "upgraded before mounting with ocfs v2\n");
 790                 goto bail;
 791         }
 792
 793         /*
 794          * Now check at magic offset for 512, 1024, 2048, 4096
 795          * blocksizes.  4096 is the maximum blocksize because it is
 796          * the minimum clustersize.
 797          */
 798         status = -EINVAL;
 799         for (blksize = *sector_size;
 800              blksize <= OCFS2_MAX_BLOCKSIZE;
 801              blksize <<= 1) {
 802                 tmpstat = ocfs2_get_sector(sb, bh,
 803                                            OCFS2_SUPER_BLOCK_BLKNO,
 804                                            blksize);
 805                 if (tmpstat < 0) {
 806                         status = tmpstat;
 807                         mlog_errno(status);
 808                         break;
 809                 }
 810                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 811                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 812                 spin_lock_init(&stats->b_lock);
 813                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 814                 if (tmpstat < 0) {
 815                         brelse(*bh);
 816                         *bh = NULL;
 817                 }
 818                 if (tmpstat != -EAGAIN) {
 819                         status = tmpstat;
 820                         break;
 821                 }
 822         }
 823
 824 bail:
 825         return status;
 826 }
 827
 828 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 829 {
 830         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 831
 832         if (osb->s_mount_opt & hb_enabled) {
 833                 if (ocfs2_mount_local(osb)) {
 834                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 835                              "mounted device.\n");
 836                         return -EINVAL;
 837                 }
 838                 if (ocfs2_userspace_stack(osb)) {
 839                         mlog(ML_ERROR, "Userspace stack expected, but "
 840                              "o2cb heartbeat arguments passed to mount\n");
 841                         return -EINVAL;
 842                 }
 843                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 844                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 845                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 846                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 847                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 848                         return -EINVAL;
 849                 }
 850         }
 851
 852         if (!(osb->s_mount_opt & hb_enabled)) {
 853                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 854                     !ocfs2_userspace_stack(osb)) {
 855                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 856                              "a read-write clustered device.\n");
 857                         return -EINVAL;
 858                 }
 859         }
 860
 861         return 0;
 862 }
 863
 864 /*
 865  * If we're using a userspace stack, mount should have passed
 866  * a name that matches the disk.  If not, mount should not
 867  * have passed a stack.
 868  */
 869 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 870                                         struct mount_options *mopt)
 871 {
 872         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 873                 mlog(ML_ERROR,
 874                      "cluster stack passed to mount, but this filesystem "
 875                      "does not support it\n");
 876                 return -EINVAL;
 877         }
 878
 879         if (ocfs2_userspace_stack(osb) &&
 880             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 881                     OCFS2_STACK_LABEL_LEN)) {
 882                 mlog(ML_ERROR,
 883                      "cluster stack passed to mount (\"%s\") does not "
 884                      "match the filesystem (\"%s\")\n",
 885                      mopt->cluster_stack,
 886                      osb->osb_cluster_stack);
 887                 return -EINVAL;
 888         }
 889
 890         return 0;
 891 }
 892
 893 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 894 {
 895         int type;
 896         struct super_block *sb = osb->sb;
 897         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 898                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 899         int status = 0;
 900
 901         for (type = 0; type < MAXQUOTAS; type++) {
 902                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 903                         continue;
 904                 if (unsuspend)
 905                         status = dquot_resume(sb, type);
 906                 else {
 907                         struct ocfs2_mem_dqinfo *oinfo;
 908
 909                         /* Cancel periodic syncing before suspending */
 910                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 911                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 912                         status = dquot_suspend(sb, type);
 913                 }
 914                 if (status < 0)
 915                         break;
 916         }
 917         if (status < 0)
 918                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 919                      "remount (error = %d).\n", status);
 920         return status;
 921 }
 922
 923 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 924 {
 925         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 926         struct super_block *sb = osb->sb;
 927         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 928                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 929         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 930                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 931         int status;
 932         int type;
 933
 934         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 935         for (type = 0; type < MAXQUOTAS; type++) {
 936                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 937                         continue;
 938                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 939                                                         osb->slot_num);
 940                 if (!inode[type]) {
 941                         status = -ENOENT;
 942                         goto out_quota_off;
 943                 }
 944                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 945                                       DQUOT_USAGE_ENABLED);
 946                 if (status < 0)
 947                         goto out_quota_off;
 948         }
 949
 950         for (type = 0; type < MAXQUOTAS; type++)
 951                 iput(inode[type]);
 952         return 0;
 953 out_quota_off:
 954         ocfs2_disable_quotas(osb);
 955         for (type = 0; type < MAXQUOTAS; type++)
 956                 iput(inode[type]);
 957         mlog_errno(status);
 958         return status;
 959 }
 960
 961 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 962 {
 963         int type;
 964         struct inode *inode;
 965         struct super_block *sb = osb->sb;
 966         struct ocfs2_mem_dqinfo *oinfo;
 967
 968         /* We mostly ignore errors in this function because there's not much
 969          * we can do when we see them */
 970         for (type = 0; type < MAXQUOTAS; type++) {
 971                 if (!sb_has_quota_loaded(sb, type))
 972                         continue;
 973                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 974                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 975                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 976                 inode = igrab(sb->s_dquot.files[type]);
 977                 /* Turn off quotas. This will remove all dquot structures from
 978                  * memory and so they will be automatically synced to global
 979                  * quota files */
 980                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 981                                         DQUOT_LIMITS_ENABLED);
 982                 if (!inode)
 983                         continue;
 984                 iput(inode);
 985         }
 986 }
 987
 988 /* Handle quota on quotactl */
 989 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 990 {
 991         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 992                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 993
 994         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 995                 return -EINVAL;
 996
 997         return dquot_enable(sb_dqopt(sb)->files[type], type,
 998                             format_id, DQUOT_LIMITS_ENABLED);
 999 }
1000
1001 /* Handle quota off quotactl */
1002 static int ocfs2_quota_off(struct super_block *sb, int type)
1003 {
1004         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1005 }
1006
1007 static const struct quotactl_ops ocfs2_quotactl_ops = {
1008         .quota_on_meta  = ocfs2_quota_on,
1009         .quota_off      = ocfs2_quota_off,
1010         .quota_sync     = dquot_quota_sync,
1011         .get_info       = dquot_get_dqinfo,
1012         .set_info       = dquot_set_dqinfo,
1013         .get_dqblk      = dquot_get_dqblk,
1014         .set_dqblk      = dquot_set_dqblk,
1015 };
1016
1017 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1018 {
1019         struct dentry *root;
1020         int status, sector_size;
1021         struct mount_options parsed_options;
1022         struct inode *inode = NULL;
1023         struct ocfs2_super *osb = NULL;
1024         struct buffer_head *bh = NULL;
1025         char nodestr[12];
1026         struct ocfs2_blockcheck_stats stats;
1027
1028         trace_ocfs2_fill_super(sb, data, silent);
1029
1030         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1031                 status = -EINVAL;
1032                 goto read_super_error;
1033         }
1034
1035         /* probe for superblock */
1036         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1037         if (status < 0) {
1038                 mlog(ML_ERROR, "superblock probe failed!\n");
1039                 goto read_super_error;
1040         }
1041
1042         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1043         osb = OCFS2_SB(sb);
1044         if (status < 0) {
1045                 mlog_errno(status);
1046                 goto read_super_error;
1047         }
1048         brelse(bh);
1049         bh = NULL;
1050
1051         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1052                 status = -EINVAL;
1053                 goto read_super_error;
1054         }
1055         osb->s_mount_opt = parsed_options.mount_opt;
1056         osb->s_atime_quantum = parsed_options.atime_quantum;
1057         osb->preferred_slot = parsed_options.slot;
1058         osb->osb_commit_interval = parsed_options.commit_interval;
1059
1060         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1061         osb->osb_resv_level = parsed_options.resv_level;
1062         osb->osb_dir_resv_level = parsed_options.resv_level;
1063         if (parsed_options.dir_resv_level == -1)
1064                 osb->osb_dir_resv_level = parsed_options.resv_level;
1065         else
1066                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1067
1068         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1069         if (status)
1070                 goto read_super_error;
1071
1072         sb->s_magic = OCFS2_SUPER_MAGIC;
1073
1074         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1075                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1076
1077         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1078          * heartbeat=none */
1079         if (bdev_read_only(sb->s_bdev)) {
1080                 if (!(sb->s_flags & MS_RDONLY)) {
1081                         status = -EACCES;
1082                         mlog(ML_ERROR, "Readonly device detected but readonly "
1083                              "mount was not specified.\n");
1084                         goto read_super_error;
1085                 }
1086
1087                 /* You should not be able to start a local heartbeat
1088                  * on a readonly device. */
1089                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1090                         status = -EROFS;
1091                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1092                              "device.\n");
1093                         goto read_super_error;
1094                 }
1095
1096                 status = ocfs2_check_journals_nolocks(osb);
1097                 if (status < 0) {
1098                         if (status == -EROFS)
1099                                 mlog(ML_ERROR, "Recovery required on readonly "
1100                                      "file system, but write access is "
1101                                      "unavailable.\n");
1102                         else
1103                                 mlog_errno(status);
1104                         goto read_super_error;
1105                 }
1106
1107                 ocfs2_set_ro_flag(osb, 1);
1108
1109                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1110                        "Cluster services will not be used for this mount. "
1111                        "Recovery will be skipped.\n", osb->dev_str);
1112         }
1113
1114         if (!ocfs2_is_hard_readonly(osb)) {
1115                 if (sb->s_flags & MS_RDONLY)
1116                         ocfs2_set_ro_flag(osb, 0);
1117         }
1118
1119         status = ocfs2_verify_heartbeat(osb);
1120         if (status < 0) {
1121                 mlog_errno(status);
1122                 goto read_super_error;
1123         }
1124
1125         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1126                                                  ocfs2_debugfs_root);
1127         if (!osb->osb_debug_root) {
1128                 status = -EINVAL;
1129                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1130                 goto read_super_error;
1131         }
1132
1133         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1134                                             osb->osb_debug_root,
1135                                             osb,
1136                                             &ocfs2_osb_debug_fops);
1137         if (!osb->osb_ctxt) {
1138                 status = -EINVAL;
1139                 mlog_errno(status);
1140                 goto read_super_error;
1141         }
1142
1143         if (ocfs2_meta_ecc(osb)) {
1144                 status = ocfs2_blockcheck_stats_debugfs_install(
1145                                                 &osb->osb_ecc_stats,
1146                                                 osb->osb_debug_root);
1147                 if (status) {
1148                         mlog(ML_ERROR,
1149                              "Unable to create blockcheck statistics "
1150                              "files\n");
1151                         goto read_super_error;
1152                 }
1153         }
1154
1155         status = ocfs2_mount_volume(sb);
1156         if (status < 0)
1157                 goto read_super_error;
1158
1159         if (osb->root_inode)
1160                 inode = igrab(osb->root_inode);
1161
1162         if (!inode) {
1163                 status = -EIO;
1164                 mlog_errno(status);
1165                 goto read_super_error;
1166         }
1167
1168         root = d_make_root(inode);
1169         if (!root) {
1170                 status = -ENOMEM;
1171                 mlog_errno(status);
1172                 goto read_super_error;
1173         }
1174
1175         sb->s_root = root;
1176
1177         ocfs2_complete_mount_recovery(osb);
1178
1179         if (ocfs2_mount_local(osb))
1180                 snprintf(nodestr, sizeof(nodestr), "local");
1181         else
1182                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1183
1184         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1185                "with %s data mode.\n",
1186                osb->dev_str, nodestr, osb->slot_num,
1187                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1188                "ordered");
1189
1190         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1191         wake_up(&osb->osb_mount_event);
1192
1193         /* Now we can initialize quotas because we can afford to wait
1194          * for cluster locks recovery now. That also means that truncation
1195          * log recovery can happen but that waits for proper quota setup */
1196         if (!(sb->s_flags & MS_RDONLY)) {
1197                 status = ocfs2_enable_quotas(osb);
1198                 if (status < 0) {
1199                         /* We have to err-out specially here because
1200                          * s_root is already set */
1201                         mlog_errno(status);
1202                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1203                         wake_up(&osb->osb_mount_event);
1204                         return status;
1205                 }
1206         }
1207
1208         ocfs2_complete_quota_recovery(osb);
1209
1210         /* Now we wake up again for processes waiting for quotas */
1211         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1212         wake_up(&osb->osb_mount_event);
1213
1214         /* Start this when the mount is almost sure of being successful */
1215         ocfs2_orphan_scan_start(osb);
1216
1217         return status;
1218
1219 read_super_error:
1220         brelse(bh);
1221
1222         if (osb) {
1223                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1224                 wake_up(&osb->osb_mount_event);
1225                 ocfs2_dismount_volume(sb, 1);
1226         }
1227
1228         if (status)
1229                 mlog_errno(status);
1230         return status;
1231 }
1232
1233 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1234                         int flags,
1235                         const char *dev_name,
1236                         void *data)
1237 {
1238         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1239 }
1240
1241 static void ocfs2_kill_sb(struct super_block *sb)
1242 {
1243         struct ocfs2_super *osb = OCFS2_SB(sb);
1244
1245         /* Failed mount? */
1246         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1247                 goto out;
1248
1249         /* Prevent further queueing of inode drop events */
1250         spin_lock(&dentry_list_lock);
1251         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1252         spin_unlock(&dentry_list_lock);
1253         /* Wait for work to finish and/or remove it */
1254         cancel_work_sync(&osb->dentry_lock_work);
1255 out:
1256         kill_block_super(sb);
1257 }
1258
1259 static struct file_system_type ocfs2_fs_type = {
1260         .owner          = THIS_MODULE,
1261         .name           = "ocfs2",
1262         .mount          = ocfs2_mount,
1263         .kill_sb        = ocfs2_kill_sb,
1264
1265         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1266         .next           = NULL
1267 };
1268 MODULE_ALIAS_FS("ocfs2");
1269
1270 static int ocfs2_check_set_options(struct super_block *sb,
1271                                    struct mount_options *options)
1272 {
1273         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1274             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1275                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1276                 mlog(ML_ERROR, "User quotas were requested, but this "
1277                      "filesystem does not have the feature enabled.\n");
1278                 return 0;
1279         }
1280         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1281             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1282                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1283                 mlog(ML_ERROR, "Group quotas were requested, but this "
1284                      "filesystem does not have the feature enabled.\n");
1285                 return 0;
1286         }
1287         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1288             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1289                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1290                      "feature is not enabled\n");
1291                 return 0;
1292         }
1293         /* No ACL setting specified? Use XATTR feature... */
1294         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1295                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1296                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1297                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1298                 else
1299                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1300         }
1301         return 1;
1302 }
1303
1304 static int ocfs2_parse_options(struct super_block *sb,
1305                                char *options,
1306                                struct mount_options *mopt,
1307                                int is_remount)
1308 {
1309         int status, user_stack = 0;
1310         char *p;
1311         u32 tmp;
1312
1313         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1314
1315         mopt->commit_interval = 0;
1316         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1317         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1318         mopt->slot = OCFS2_INVALID_SLOT;
1319         mopt->localalloc_opt = -1;
1320         mopt->cluster_stack[0] = '\0';
1321         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1322         mopt->dir_resv_level = -1;
1323
1324         if (!options) {
1325                 status = 1;
1326                 goto bail;
1327         }
1328
1329         while ((p = strsep(&options, ",")) != NULL) {
1330                 int token, option;
1331                 substring_t args[MAX_OPT_ARGS];
1332
1333                 if (!*p)
1334                         continue;
1335
1336                 token = match_token(p, tokens, args);
1337                 switch (token) {
1338                 case Opt_hb_local:
1339                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1340                         break;
1341                 case Opt_hb_none:
1342                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1343                         break;
1344                 case Opt_hb_global:
1345                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1346                         break;
1347                 case Opt_barrier:
1348                         if (match_int(&args[0], &option)) {
1349                                 status = 0;
1350                                 goto bail;
1351                         }
1352                         if (option)
1353                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1354                         else
1355                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1356                         break;
1357                 case Opt_intr:
1358                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1359                         break;
1360                 case Opt_nointr:
1361                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1362                         break;
1363                 case Opt_err_panic:
1364                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1365                         break;
1366                 case Opt_err_ro:
1367                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1368                         break;
1369                 case Opt_data_ordered:
1370                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1371                         break;
1372                 case Opt_data_writeback:
1373                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1374                         break;
1375                 case Opt_user_xattr:
1376                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1377                         break;
1378                 case Opt_nouser_xattr:
1379                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1380                         break;
1381                 case Opt_atime_quantum:
1382                         if (match_int(&args[0], &option)) {
1383                                 status = 0;
1384                                 goto bail;
1385                         }
1386                         if (option >= 0)
1387                                 mopt->atime_quantum = option;
1388                         break;
1389                 case Opt_slot:
1390                         option = 0;
1391                         if (match_int(&args[0], &option)) {
1392                                 status = 0;
1393                                 goto bail;
1394                         }
1395                         if (option)
1396                                 mopt->slot = (s16)option;
1397                         break;
1398                 case Opt_commit:
1399                         option = 0;
1400                         if (match_int(&args[0], &option)) {
1401                                 status = 0;
1402                                 goto bail;
1403                         }
1404                         if (option < 0)
1405                                 return 0;
1406                         if (option == 0)
1407                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1408                         mopt->commit_interval = HZ * option;
1409                         break;
1410                 case Opt_localalloc:
1411                         option = 0;
1412                         if (match_int(&args[0], &option)) {
1413                                 status = 0;
1414                                 goto bail;
1415                         }
1416                         if (option >= 0)
1417                                 mopt->localalloc_opt = option;
1418                         break;
1419                 case Opt_localflocks:
1420                         /*
1421                          * Changing this during remount could race
1422                          * flock() requests, or "unbalance" existing
1423                          * ones (e.g., a lock is taken in one mode but
1424                          * dropped in the other). If users care enough
1425                          * to flip locking modes during remount, we
1426                          * could add a "local" flag to individual
1427                          * flock structures for proper tracking of
1428                          * state.
1429                          */
1430                         if (!is_remount)
1431                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1432                         break;
1433                 case Opt_stack:
1434                         /* Check both that the option we were passed
1435                          * is of the right length and that it is a proper
1436                          * string of the right length.
1437                          */
1438                         if (((args[0].to - args[0].from) !=
1439                              OCFS2_STACK_LABEL_LEN) ||
1440                             (strnlen(args[0].from,
1441                                      OCFS2_STACK_LABEL_LEN) !=
1442                              OCFS2_STACK_LABEL_LEN)) {
1443                                 mlog(ML_ERROR,
1444                                      "Invalid cluster_stack option\n");
1445                                 status = 0;
1446                                 goto bail;
1447                         }
1448                         memcpy(mopt->cluster_stack, args[0].from,
1449                                OCFS2_STACK_LABEL_LEN);
1450                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1451                         /*
1452                          * Open code the memcmp here as we don't have
1453                          * an osb to pass to
1454                          * ocfs2_userspace_stack().
1455                          */
1456                         if (memcmp(mopt->cluster_stack,
1457                                    OCFS2_CLASSIC_CLUSTER_STACK,
1458                                    OCFS2_STACK_LABEL_LEN))
1459                                 user_stack = 1;
1460                         break;
1461                 case Opt_inode64:
1462                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1463                         break;
1464                 case Opt_usrquota:
1465                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1466                         break;
1467                 case Opt_grpquota:
1468                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1469                         break;
1470                 case Opt_coherency_buffered:
1471                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1472                         break;
1473                 case Opt_coherency_full:
1474                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1475                         break;
1476                 case Opt_acl:
1477                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1478                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1479                         break;
1480                 case Opt_noacl:
1481                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1482                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1483                         break;
1484                 case Opt_resv_level:
1485                         if (is_remount)
1486                                 break;
1487                         if (match_int(&args[0], &option)) {
1488                                 status = 0;
1489                                 goto bail;
1490                         }
1491                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1492                             option < OCFS2_MAX_RESV_LEVEL)
1493                                 mopt->resv_level = option;
1494                         break;
1495                 case Opt_dir_resv_level:
1496                         if (is_remount)
1497                                 break;
1498                         if (match_int(&args[0], &option)) {
1499                                 status = 0;
1500                                 goto bail;
1501                         }
1502                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1503                             option < OCFS2_MAX_RESV_LEVEL)
1504                                 mopt->dir_resv_level = option;
1505                         break;
1506                 default:
1507                         mlog(ML_ERROR,
1508                              "Unrecognized mount option \"%s\" "
1509                              "or missing value\n", p);
1510                         status = 0;
1511                         goto bail;
1512                 }
1513         }
1514
1515         if (user_stack == 0) {
1516                 /* Ensure only one heartbeat mode */
1517                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1518                                          OCFS2_MOUNT_HB_GLOBAL |
1519                                          OCFS2_MOUNT_HB_NONE);
1520                 if (hweight32(tmp) != 1) {
1521                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1522                         status = 0;
1523                         goto bail;
1524                 }
1525         }
1526
1527         status = 1;
1528
1529 bail:
1530         return status;
1531 }
1532
1533 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1534 {
1535         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1536         unsigned long opts = osb->s_mount_opt;
1537         unsigned int local_alloc_megs;
1538
1539         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1540                 seq_printf(s, ",_netdev");
1541                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1542                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1543                 else
1544                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1545         } else
1546                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1547
1548         if (opts & OCFS2_MOUNT_NOINTR)
1549                 seq_printf(s, ",nointr");
1550
1551         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1552                 seq_printf(s, ",data=writeback");
1553         else
1554                 seq_printf(s, ",data=ordered");
1555
1556         if (opts & OCFS2_MOUNT_BARRIER)
1557                 seq_printf(s, ",barrier=1");
1558
1559         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1560                 seq_printf(s, ",errors=panic");
1561         else
1562                 seq_printf(s, ",errors=remount-ro");
1563
1564         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1565                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1566
1567         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1568
1569         if (osb->osb_commit_interval)
1570                 seq_printf(s, ",commit=%u",
1571                            (unsigned) (osb->osb_commit_interval / HZ));
1572
1573         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1574         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1575                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1576
1577         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1578                 seq_printf(s, ",localflocks,");
1579
1580         if (osb->osb_cluster_stack[0])
1581                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1582                            osb->osb_cluster_stack);
1583         if (opts & OCFS2_MOUNT_USRQUOTA)
1584                 seq_printf(s, ",usrquota");
1585         if (opts & OCFS2_MOUNT_GRPQUOTA)
1586                 seq_printf(s, ",grpquota");
1587
1588         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1589                 seq_printf(s, ",coherency=buffered");
1590         else
1591                 seq_printf(s, ",coherency=full");
1592
1593         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1594                 seq_printf(s, ",nouser_xattr");
1595         else
1596                 seq_printf(s, ",user_xattr");
1597
1598         if (opts & OCFS2_MOUNT_INODE64)
1599                 seq_printf(s, ",inode64");
1600
1601         if (opts & OCFS2_MOUNT_POSIX_ACL)
1602                 seq_printf(s, ",acl");
1603         else
1604                 seq_printf(s, ",noacl");
1605
1606         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1607                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1608
1609         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1610                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1611
1612         return 0;
1613 }
1614
1615 wait_queue_head_t ocfs2__ioend_wq[OCFS2_IOEND_WQ_HASH_SZ];
1616
1617 static int __init ocfs2_init(void)
1618 {
1619         int status, i;
1620
1621         for (i = 0; i < OCFS2_IOEND_WQ_HASH_SZ; i++)
1622                 init_waitqueue_head(&ocfs2__ioend_wq[i]);
1623
1624         status = init_ocfs2_uptodate_cache();
1625         if (status < 0)
1626                 goto out1;
1627
1628         status = ocfs2_initialize_mem_caches();
1629         if (status < 0)
1630                 goto out2;
1631
1632         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1633         if (!ocfs2_wq) {
1634                 status = -ENOMEM;
1635                 goto out3;
1636         }
1637
1638         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1639         if (!ocfs2_debugfs_root) {
1640                 status = -EFAULT;
1641                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1642         }
1643
1644         ocfs2_set_locking_protocol();
1645
1646         status = register_quota_format(&ocfs2_quota_format);
1647         if (status < 0)
1648                 goto out4;
1649         status = register_filesystem(&ocfs2_fs_type);
1650         if (!status)
1651                 return 0;
1652
1653         unregister_quota_format(&ocfs2_quota_format);
1654 out4:
1655         destroy_workqueue(ocfs2_wq);
1656         debugfs_remove(ocfs2_debugfs_root);
1657 out3:
1658         ocfs2_free_mem_caches();
1659 out2:
1660         exit_ocfs2_uptodate_cache();
1661 out1:
1662         mlog_errno(status);
1663         return status;
1664 }
1665
1666 static void __exit ocfs2_exit(void)
1667 {
1668         if (ocfs2_wq) {
1669                 flush_workqueue(ocfs2_wq);
1670                 destroy_workqueue(ocfs2_wq);
1671         }
1672
1673         unregister_quota_format(&ocfs2_quota_format);
1674
1675         debugfs_remove(ocfs2_debugfs_root);
1676
1677         ocfs2_free_mem_caches();
1678
1679         unregister_filesystem(&ocfs2_fs_type);
1680
1681         exit_ocfs2_uptodate_cache();
1682 }
1683
1684 static void ocfs2_put_super(struct super_block *sb)
1685 {
1686         trace_ocfs2_put_super(sb);
1687
1688         ocfs2_sync_blockdev(sb);
1689         ocfs2_dismount_volume(sb, 0);
1690 }
1691
1692 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1693 {
1694         struct ocfs2_super *osb;
1695         u32 numbits, freebits;
1696         int status;
1697         struct ocfs2_dinode *bm_lock;
1698         struct buffer_head *bh = NULL;
1699         struct inode *inode = NULL;
1700
1701         trace_ocfs2_statfs(dentry->d_sb, buf);
1702
1703         osb = OCFS2_SB(dentry->d_sb);
1704
1705         inode = ocfs2_get_system_file_inode(osb,
1706                                             GLOBAL_BITMAP_SYSTEM_INODE,
1707                                             OCFS2_INVALID_SLOT);
1708         if (!inode) {
1709                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1710                 status = -EIO;
1711                 goto bail;
1712         }
1713
1714         status = ocfs2_inode_lock(inode, &bh, 0);
1715         if (status < 0) {
1716                 mlog_errno(status);
1717                 goto bail;
1718         }
1719
1720         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1721
1722         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1723         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1724
1725         buf->f_type = OCFS2_SUPER_MAGIC;
1726         buf->f_bsize = dentry->d_sb->s_blocksize;
1727         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1728         buf->f_blocks = ((sector_t) numbits) *
1729                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1730         buf->f_bfree = ((sector_t) freebits) *
1731                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1732         buf->f_bavail = buf->f_bfree;
1733         buf->f_files = numbits;
1734         buf->f_ffree = freebits;
1735         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1736                                 & 0xFFFFFFFFUL;
1737         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1738                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1739
1740         brelse(bh);
1741
1742         ocfs2_inode_unlock(inode, 0);
1743         status = 0;
1744 bail:
1745         if (inode)
1746                 iput(inode);
1747
1748         if (status)
1749                 mlog_errno(status);
1750
1751         return status;
1752 }
1753
1754 static void ocfs2_inode_init_once(void *data)
1755 {
1756         struct ocfs2_inode_info *oi = data;
1757
1758         oi->ip_flags = 0;
1759         oi->ip_open_count = 0;
1760         spin_lock_init(&oi->ip_lock);
1761         ocfs2_extent_map_init(&oi->vfs_inode);
1762         INIT_LIST_HEAD(&oi->ip_io_markers);
1763         oi->ip_dir_start_lookup = 0;
1764         atomic_set(&oi->ip_unaligned_aio, 0);
1765         init_rwsem(&oi->ip_alloc_sem);
1766         init_rwsem(&oi->ip_xattr_sem);
1767         mutex_init(&oi->ip_io_mutex);
1768
1769         oi->ip_blkno = 0ULL;
1770         oi->ip_clusters = 0;
1771
1772         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1773
1774         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1775         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1776         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1777
1778         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1779                                   &ocfs2_inode_caching_ops);
1780
1781         inode_init_once(&oi->vfs_inode);
1782 }
1783
1784 static int ocfs2_initialize_mem_caches(void)
1785 {
1786         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1787                                        sizeof(struct ocfs2_inode_info),
1788                                        0,
1789                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1790                                                 SLAB_MEM_SPREAD),
1791                                        ocfs2_inode_init_once);
1792         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1793                                         sizeof(struct ocfs2_dquot),
1794                                         0,
1795                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1796                                                 SLAB_MEM_SPREAD),
1797                                         NULL);
1798         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1799                                         sizeof(struct ocfs2_quota_chunk),
1800                                         0,
1801                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1802                                         NULL);
1803         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1804             !ocfs2_qf_chunk_cachep) {
1805                 if (ocfs2_inode_cachep)
1806                         kmem_cache_destroy(ocfs2_inode_cachep);
1807                 if (ocfs2_dquot_cachep)
1808                         kmem_cache_destroy(ocfs2_dquot_cachep);
1809                 if (ocfs2_qf_chunk_cachep)
1810                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1811                 return -ENOMEM;
1812         }
1813
1814         return 0;
1815 }
1816
1817 static void ocfs2_free_mem_caches(void)
1818 {
1819         /*
1820          * Make sure all delayed rcu free inodes are flushed before we
1821          * destroy cache.
1822          */
1823         rcu_barrier();
1824         if (ocfs2_inode_cachep)
1825                 kmem_cache_destroy(ocfs2_inode_cachep);
1826         ocfs2_inode_cachep = NULL;
1827
1828         if (ocfs2_dquot_cachep)
1829                 kmem_cache_destroy(ocfs2_dquot_cachep);
1830         ocfs2_dquot_cachep = NULL;
1831
1832         if (ocfs2_qf_chunk_cachep)
1833                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1834         ocfs2_qf_chunk_cachep = NULL;
1835 }
1836
1837 static int ocfs2_get_sector(struct super_block *sb,
1838                             struct buffer_head **bh,
1839                             int block,
1840                             int sect_size)
1841 {
1842         if (!sb_set_blocksize(sb, sect_size)) {
1843                 mlog(ML_ERROR, "unable to set blocksize\n");
1844                 return -EIO;
1845         }
1846
1847         *bh = sb_getblk(sb, block);
1848         if (!*bh) {
1849                 mlog_errno(-ENOMEM);
1850                 return -ENOMEM;
1851         }
1852         lock_buffer(*bh);
1853         if (!buffer_dirty(*bh))
1854                 clear_buffer_uptodate(*bh);
1855         unlock_buffer(*bh);
1856         ll_rw_block(READ, 1, bh);
1857         wait_on_buffer(*bh);
1858         if (!buffer_uptodate(*bh)) {
1859                 mlog_errno(-EIO);
1860                 brelse(*bh);
1861                 *bh = NULL;
1862                 return -EIO;
1863         }
1864
1865         return 0;
1866 }
1867
1868 static int ocfs2_mount_volume(struct super_block *sb)
1869 {
1870         int status = 0;
1871         int unlock_super = 0;
1872         struct ocfs2_super *osb = OCFS2_SB(sb);
1873
1874         if (ocfs2_is_hard_readonly(osb))
1875                 goto leave;
1876
1877         status = ocfs2_dlm_init(osb);
1878         if (status < 0) {
1879                 mlog_errno(status);
1880                 goto leave;
1881         }
1882
1883         status = ocfs2_super_lock(osb, 1);
1884         if (status < 0) {
1885                 mlog_errno(status);
1886                 goto leave;
1887         }
1888         unlock_super = 1;
1889
1890         /* This will load up the node map and add ourselves to it. */
1891         status = ocfs2_find_slot(osb);
1892         if (status < 0) {
1893                 mlog_errno(status);
1894                 goto leave;
1895         }
1896
1897         /* load all node-local system inodes */
1898         status = ocfs2_init_local_system_inodes(osb);
1899         if (status < 0) {
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         status = ocfs2_check_volume(osb);
1905         if (status < 0) {
1906                 mlog_errno(status);
1907                 goto leave;
1908         }
1909
1910         status = ocfs2_truncate_log_init(osb);
1911         if (status < 0)
1912                 mlog_errno(status);
1913
1914 leave:
1915         if (unlock_super)
1916                 ocfs2_super_unlock(osb, 1);
1917
1918         return status;
1919 }
1920
1921 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1922 {
1923         int tmp, hangup_needed = 0;
1924         struct ocfs2_super *osb = NULL;
1925         char nodestr[12];
1926
1927         trace_ocfs2_dismount_volume(sb);
1928
1929         BUG_ON(!sb);
1930         osb = OCFS2_SB(sb);
1931         BUG_ON(!osb);
1932
1933         debugfs_remove(osb->osb_ctxt);
1934
1935         /*
1936          * Flush inode dropping work queue so that deletes are
1937          * performed while the filesystem is still working
1938          */
1939         ocfs2_drop_all_dl_inodes(osb);
1940
1941         /* Orphan scan should be stopped as early as possible */
1942         ocfs2_orphan_scan_stop(osb);
1943
1944         ocfs2_disable_quotas(osb);
1945
1946         ocfs2_shutdown_local_alloc(osb);
1947
1948         ocfs2_truncate_log_shutdown(osb);
1949
1950         /* This will disable recovery and flush any recovery work. */
1951         ocfs2_recovery_exit(osb);
1952
1953         ocfs2_journal_shutdown(osb);
1954
1955         ocfs2_sync_blockdev(sb);
1956
1957         ocfs2_purge_refcount_trees(osb);
1958
1959         /* No cluster connection means we've failed during mount, so skip
1960          * all the steps which depended on that to complete. */
1961         if (osb->cconn) {
1962                 tmp = ocfs2_super_lock(osb, 1);
1963                 if (tmp < 0) {
1964                         mlog_errno(tmp);
1965                         return;
1966                 }
1967         }
1968
1969         if (osb->slot_num != OCFS2_INVALID_SLOT)
1970                 ocfs2_put_slot(osb);
1971
1972         if (osb->cconn)
1973                 ocfs2_super_unlock(osb, 1);
1974
1975         ocfs2_release_system_inodes(osb);
1976
1977         /*
1978          * If we're dismounting due to mount error, mount.ocfs2 will clean
1979          * up heartbeat.  If we're a local mount, there is no heartbeat.
1980          * If we failed before we got a uuid_str yet, we can't stop
1981          * heartbeat.  Otherwise, do it.
1982          */
1983         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1984             !ocfs2_is_hard_readonly(osb))
1985                 hangup_needed = 1;
1986
1987         if (osb->cconn)
1988                 ocfs2_dlm_shutdown(osb, hangup_needed);
1989
1990         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1991         debugfs_remove(osb->osb_debug_root);
1992
1993         if (hangup_needed)
1994                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1995
1996         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1997
1998         if (ocfs2_mount_local(osb))
1999                 snprintf(nodestr, sizeof(nodestr), "local");
2000         else
2001                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2002
2003         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2004                osb->dev_str, nodestr);
2005
2006         ocfs2_delete_osb(osb);
2007         kfree(osb);
2008         sb->s_dev = 0;
2009         sb->s_fs_info = NULL;
2010 }
2011
2012 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2013                                 unsigned uuid_bytes)
2014 {
2015         int i, ret;
2016         char *ptr;
2017
2018         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2019
2020         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2021         if (osb->uuid_str == NULL)
2022                 return -ENOMEM;
2023
2024         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2025                 /* print with null */
2026                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2027                 if (ret != 2) /* drop super cleans up */
2028                         return -EINVAL;
2029                 /* then only advance past the last char */
2030                 ptr += 2;
2031         }
2032
2033         return 0;
2034 }
2035
2036 /* Make sure entire volume is addressable by our journal.  Requires
2037    osb_clusters_at_boot to be valid and for the journal to have been
2038    initialized by ocfs2_journal_init(). */
2039 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2040 {
2041         int status = 0;
2042         u64 max_block =
2043                 ocfs2_clusters_to_blocks(osb->sb,
2044                                          osb->osb_clusters_at_boot) - 1;
2045
2046         /* 32-bit block number is always OK. */
2047         if (max_block <= (u32)~0ULL)
2048                 goto out;
2049
2050         /* Volume is "huge", so see if our journal is new enough to
2051            support it. */
2052         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2053                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2054               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2055                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2056                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2057                      "Enable the 'block64' journal option with tunefs.ocfs2");
2058                 status = -EFBIG;
2059                 goto out;
2060         }
2061
2062  out:
2063         return status;
2064 }
2065
2066 static int ocfs2_initialize_super(struct super_block *sb,
2067                                   struct buffer_head *bh,
2068                                   int sector_size,
2069                                   struct ocfs2_blockcheck_stats *stats)
2070 {
2071         int status;
2072         int i, cbits, bbits;
2073         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2074         struct inode *inode = NULL;
2075         struct ocfs2_journal *journal;
2076         __le32 uuid_net_key;
2077         struct ocfs2_super *osb;
2078         u64 total_blocks;
2079
2080         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2081         if (!osb) {
2082                 status = -ENOMEM;
2083                 mlog_errno(status);
2084                 goto bail;
2085         }
2086
2087         sb->s_fs_info = osb;
2088         sb->s_op = &ocfs2_sops;
2089         sb->s_d_op = &ocfs2_dentry_ops;
2090         sb->s_export_op = &ocfs2_export_ops;
2091         sb->s_qcop = &ocfs2_quotactl_ops;
2092         sb->dq_op = &ocfs2_quota_operations;
2093         sb->s_xattr = ocfs2_xattr_handlers;
2094         sb->s_time_gran = 1;
2095         sb->s_flags |= MS_NOATIME;
2096         /* this is needed to support O_LARGEFILE */
2097         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2098         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2099         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2100
2101         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2102
2103         for (i = 0; i < 3; i++)
2104                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2105         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2106
2107         osb->sb = sb;
2108         /* Save off for ocfs2_rw_direct */
2109         osb->s_sectsize_bits = blksize_bits(sector_size);
2110         BUG_ON(!osb->s_sectsize_bits);
2111
2112         spin_lock_init(&osb->dc_task_lock);
2113         init_waitqueue_head(&osb->dc_event);
2114         osb->dc_work_sequence = 0;
2115         osb->dc_wake_sequence = 0;
2116         INIT_LIST_HEAD(&osb->blocked_lock_list);
2117         osb->blocked_lock_count = 0;
2118         spin_lock_init(&osb->osb_lock);
2119         spin_lock_init(&osb->osb_xattr_lock);
2120         ocfs2_init_steal_slots(osb);
2121
2122         atomic_set(&osb->alloc_stats.moves, 0);
2123         atomic_set(&osb->alloc_stats.local_data, 0);
2124         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2125         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2126         atomic_set(&osb->alloc_stats.bg_extends, 0);
2127
2128         /* Copy the blockcheck stats from the superblock probe */
2129         osb->osb_ecc_stats = *stats;
2130
2131         ocfs2_init_node_maps(osb);
2132
2133         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2134                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2135
2136         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2137         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2138                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2139                      osb->max_slots);
2140                 status = -EINVAL;
2141                 goto bail;
2142         }
2143
2144         ocfs2_orphan_scan_init(osb);
2145
2146         status = ocfs2_recovery_init(osb);
2147         if (status) {
2148                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2149                 mlog_errno(status);
2150                 goto bail;
2151         }
2152
2153         init_waitqueue_head(&osb->checkpoint_event);
2154
2155         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2156
2157         osb->slot_num = OCFS2_INVALID_SLOT;
2158
2159         osb->s_xattr_inline_size = le16_to_cpu(
2160                                         di->id2.i_super.s_xattr_inline_size);
2161
2162         osb->local_alloc_state = OCFS2_LA_UNUSED;
2163         osb->local_alloc_bh = NULL;
2164         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2165
2166         init_waitqueue_head(&osb->osb_mount_event);
2167
2168         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2169         if (status) {
2170                 mlog_errno(status);
2171                 goto bail;
2172         }
2173
2174         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2175         if (!osb->vol_label) {
2176                 mlog(ML_ERROR, "unable to alloc vol label\n");
2177                 status = -ENOMEM;
2178                 goto bail;
2179         }
2180
2181         osb->slot_recovery_generations =
2182                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2183                         GFP_KERNEL);
2184         if (!osb->slot_recovery_generations) {
2185                 status = -ENOMEM;
2186                 mlog_errno(status);
2187                 goto bail;
2188         }
2189
2190         init_waitqueue_head(&osb->osb_wipe_event);
2191         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2192                                         sizeof(*osb->osb_orphan_wipes),
2193                                         GFP_KERNEL);
2194         if (!osb->osb_orphan_wipes) {
2195                 status = -ENOMEM;
2196                 mlog_errno(status);
2197                 goto bail;
2198         }
2199
2200         osb->osb_rf_lock_tree = RB_ROOT;
2201
2202         osb->s_feature_compat =
2203                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2204         osb->s_feature_ro_compat =
2205                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2206         osb->s_feature_incompat =
2207                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2208
2209         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2210                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2211                      "optional features (%x).\n", i);
2212                 status = -EINVAL;
2213                 goto bail;
2214         }
2215         if (!(osb->sb->s_flags & MS_RDONLY) &&
2216             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2217                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2218                      "unsupported optional features (%x).\n", i);
2219                 status = -EINVAL;
2220                 goto bail;
2221         }
2222
2223         if (ocfs2_clusterinfo_valid(osb)) {
2224                 osb->osb_stackflags =
2225                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2226                 strlcpy(osb->osb_cluster_stack,
2227                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2228                        OCFS2_STACK_LABEL_LEN + 1);
2229                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2230                         mlog(ML_ERROR,
2231                              "couldn't mount because of an invalid "
2232                              "cluster stack label (%s) \n",
2233                              osb->osb_cluster_stack);
2234                         status = -EINVAL;
2235                         goto bail;
2236                 }
2237                 strlcpy(osb->osb_cluster_name,
2238                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2239                         OCFS2_CLUSTER_NAME_LEN + 1);
2240         } else {
2241                 /* The empty string is identical with classic tools that
2242                  * don't know about s_cluster_info. */
2243                 osb->osb_cluster_stack[0] = '\0';
2244         }
2245
2246         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2247
2248         /* FIXME
2249          * This should be done in ocfs2_journal_init(), but unknown
2250          * ordering issues will cause the filesystem to crash.
2251          * If anyone wants to figure out what part of the code
2252          * refers to osb->journal before ocfs2_journal_init() is run,
2253          * be my guest.
2254          */
2255         /* initialize our journal structure */
2256
2257         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2258         if (!journal) {
2259                 mlog(ML_ERROR, "unable to alloc journal\n");
2260                 status = -ENOMEM;
2261                 goto bail;
2262         }
2263         osb->journal = journal;
2264         journal->j_osb = osb;
2265
2266         atomic_set(&journal->j_num_trans, 0);
2267         init_rwsem(&journal->j_trans_barrier);
2268         init_waitqueue_head(&journal->j_checkpointed);
2269         spin_lock_init(&journal->j_lock);
2270         journal->j_trans_id = (unsigned long) 1;
2271         INIT_LIST_HEAD(&journal->j_la_cleanups);
2272         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2273         journal->j_state = OCFS2_JOURNAL_FREE;
2274
2275         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2276         osb->dentry_lock_list = NULL;
2277
2278         /* get some pseudo constants for clustersize bits */
2279         osb->s_clustersize_bits =
2280                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2281         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2282
2283         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2284             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2285                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2286                      osb->s_clustersize);
2287                 status = -EINVAL;
2288                 goto bail;
2289         }
2290
2291         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2292                                                 le32_to_cpu(di->i_clusters));
2293
2294         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2295                                            total_blocks);
2296         if (status) {
2297                 mlog(ML_ERROR, "Volume too large "
2298                      "to mount safely on this system");
2299                 status = -EFBIG;
2300                 goto bail;
2301         }
2302
2303         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2304                                  sizeof(di->id2.i_super.s_uuid))) {
2305                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2306                 status = -ENOMEM;
2307                 goto bail;
2308         }
2309
2310         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2311
2312         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2313         osb->vol_label[63] = '\0';
2314         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2315         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2316         osb->first_cluster_group_blkno =
2317                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2318         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2319         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2320         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2321                                      (unsigned long long)osb->root_blkno,
2322                                      (unsigned long long)osb->system_dir_blkno,
2323                                      osb->s_clustersize_bits);
2324
2325         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2326         if (!osb->osb_dlm_debug) {
2327                 status = -ENOMEM;
2328                 mlog_errno(status);
2329                 goto bail;
2330         }
2331
2332         atomic_set(&osb->vol_state, VOLUME_INIT);
2333
2334         /* load root, system_dir, and all global system inodes */
2335         status = ocfs2_init_global_system_inodes(osb);
2336         if (status < 0) {
2337                 mlog_errno(status);
2338                 goto bail;
2339         }
2340
2341         /*
2342          * global bitmap
2343          */
2344         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2345                                             OCFS2_INVALID_SLOT);
2346         if (!inode) {
2347                 status = -EINVAL;
2348                 mlog_errno(status);
2349                 goto bail;
2350         }
2351
2352         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2353         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2354         iput(inode);
2355
2356         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2357                                  osb->s_feature_incompat) * 8;
2358
2359         status = ocfs2_init_slot_info(osb);
2360         if (status < 0) {
2361                 mlog_errno(status);
2362                 goto bail;
2363         }
2364         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2365
2366 bail:
2367         return status;
2368 }
2369
2370 /*
2371  * will return: -EAGAIN if it is ok to keep searching for superblocks
2372  *              -EINVAL if there is a bad superblock
2373  *              0 on success
2374  */
2375 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2376                                struct buffer_head *bh,
2377                                u32 blksz,
2378                                struct ocfs2_blockcheck_stats *stats)
2379 {
2380         int status = -EAGAIN;
2381
2382         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2383                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2384                 /* We have to do a raw check of the feature here */
2385                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2386                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2387                         status = ocfs2_block_check_validate(bh->b_data,
2388                                                             bh->b_size,
2389                                                             &di->i_check,
2390                                                             stats);
2391                         if (status)
2392                                 goto out;
2393                 }
2394                 status = -EINVAL;
2395                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2396                         mlog(ML_ERROR, "found superblock with incorrect block "
2397                              "size: found %u, should be %u\n",
2398                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2399                                blksz);
2400                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2401                            OCFS2_MAJOR_REV_LEVEL ||
2402                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2403                            OCFS2_MINOR_REV_LEVEL) {
2404                         mlog(ML_ERROR, "found superblock with bad version: "
2405                              "found %u.%u, should be %u.%u\n",
2406                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2407                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2408                              OCFS2_MAJOR_REV_LEVEL,
2409                              OCFS2_MINOR_REV_LEVEL);
2410                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2411                         mlog(ML_ERROR, "bad block number on superblock: "
2412                              "found %llu, should be %llu\n",
2413                              (unsigned long long)le64_to_cpu(di->i_blkno),
2414                              (unsigned long long)bh->b_blocknr);
2415                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2416                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2417                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2418                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2419                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2420                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2421                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2422                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2423                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2424                         mlog(ML_ERROR,
2425                              "Superblock slots found greater than file system "
2426                              "maximum: found %u, max %u\n",
2427                              le16_to_cpu(di->id2.i_super.s_max_slots),
2428                              OCFS2_MAX_SLOTS);
2429                 } else {
2430                         /* found it! */
2431                         status = 0;
2432                 }
2433         }
2434
2435 out:
2436         if (status && status != -EAGAIN)
2437                 mlog_errno(status);
2438         return status;
2439 }
2440
2441 static int ocfs2_check_volume(struct ocfs2_super *osb)
2442 {
2443         int status;
2444         int dirty;
2445         int local;
2446         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2447                                                   * recover
2448                                                   * ourselves. */
2449
2450         /* Init our journal object. */
2451         status = ocfs2_journal_init(osb->journal, &dirty);
2452         if (status < 0) {
2453                 mlog(ML_ERROR, "Could not initialize journal!\n");
2454                 goto finally;
2455         }
2456
2457         /* Now that journal has been initialized, check to make sure
2458            entire volume is addressable. */
2459         status = ocfs2_journal_addressable(osb);
2460         if (status)
2461                 goto finally;
2462
2463         /* If the journal was unmounted cleanly then we don't want to
2464          * recover anything. Otherwise, journal_load will do that
2465          * dirty work for us :) */
2466         if (!dirty) {
2467                 status = ocfs2_journal_wipe(osb->journal, 0);
2468                 if (status < 0) {
2469                         mlog_errno(status);
2470                         goto finally;
2471                 }
2472         } else {
2473                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2474                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2475         }
2476
2477         local = ocfs2_mount_local(osb);
2478
2479         /* will play back anything left in the journal. */
2480         status = ocfs2_journal_load(osb->journal, local, dirty);
2481         if (status < 0) {
2482                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2483                 goto finally;
2484         }
2485
2486         if (dirty) {
2487                 /* recover my local alloc if we didn't unmount cleanly. */
2488                 status = ocfs2_begin_local_alloc_recovery(osb,
2489                                                           osb->slot_num,
2490                                                           &local_alloc);
2491                 if (status < 0) {
2492                         mlog_errno(status);
2493                         goto finally;
2494                 }
2495                 /* we complete the recovery process after we've marked
2496                  * ourselves as mounted. */
2497         }
2498
2499         status = ocfs2_load_local_alloc(osb);
2500         if (status < 0) {
2501                 mlog_errno(status);
2502                 goto finally;
2503         }
2504
2505         if (dirty) {
2506                 /* Recovery will be completed after we've mounted the
2507                  * rest of the volume. */
2508                 osb->dirty = 1;
2509                 osb->local_alloc_copy = local_alloc;
2510                 local_alloc = NULL;
2511         }
2512
2513         /* go through each journal, trylock it and if you get the
2514          * lock, and it's marked as dirty, set the bit in the recover
2515          * map and launch a recovery thread for it. */
2516         status = ocfs2_mark_dead_nodes(osb);
2517         if (status < 0) {
2518                 mlog_errno(status);
2519                 goto finally;
2520         }
2521
2522         status = ocfs2_compute_replay_slots(osb);
2523         if (status < 0)
2524                 mlog_errno(status);
2525
2526 finally:
2527         kfree(local_alloc);
2528
2529         if (status)
2530                 mlog_errno(status);
2531         return status;
2532 }
2533
2534 /*
2535  * The routine gets called from dismount or close whenever a dismount on
2536  * volume is requested and the osb open count becomes 1.
2537  * It will remove the osb from the global list and also free up all the
2538  * initialized resources and fileobject.
2539  */
2540 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2541 {
2542         /* This function assumes that the caller has the main osb resource */
2543
2544         ocfs2_free_slot_info(osb);
2545
2546         kfree(osb->osb_orphan_wipes);
2547         kfree(osb->slot_recovery_generations);
2548         /* FIXME
2549          * This belongs in journal shutdown, but because we have to
2550          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2551          * we free it here.
2552          */
2553         kfree(osb->journal);
2554         kfree(osb->local_alloc_copy);
2555         kfree(osb->uuid_str);
2556         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2557         memset(osb, 0, sizeof(struct ocfs2_super));
2558 }
2559
2560 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2561  * panic(). We do not support continue-on-error operation. */
2562 static void ocfs2_handle_error(struct super_block *sb)
2563 {
2564         struct ocfs2_super *osb = OCFS2_SB(sb);
2565
2566         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2567                 panic("OCFS2: (device %s): panic forced after error\n",
2568                       sb->s_id);
2569
2570         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2571
2572         if (sb->s_flags & MS_RDONLY &&
2573             (ocfs2_is_soft_readonly(osb) ||
2574              ocfs2_is_hard_readonly(osb)))
2575                 return;
2576
2577         printk(KERN_CRIT "File system is now read-only due to the potential "
2578                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2579                "system is unmounted.\n");
2580         sb->s_flags |= MS_RDONLY;
2581         ocfs2_set_ro_flag(osb, 0);
2582 }
2583
2584 static char error_buf[1024];
2585
2586 void __ocfs2_error(struct super_block *sb,
2587                    const char *function,
2588                    const char *fmt, ...)
2589 {
2590         va_list args;
2591
2592         va_start(args, fmt);
2593         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2594         va_end(args);
2595
2596         /* Not using mlog here because we want to show the actual
2597          * function the error came from. */
2598         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2599                sb->s_id, function, error_buf);
2600
2601         ocfs2_handle_error(sb);
2602 }
2603
2604 /* Handle critical errors. This is intentionally more drastic than
2605  * ocfs2_handle_error, so we only use for things like journal errors,
2606  * etc. */
2607 void __ocfs2_abort(struct super_block* sb,
2608                    const char *function,
2609                    const char *fmt, ...)
2610 {
2611         va_list args;
2612
2613         va_start(args, fmt);
2614         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2615         va_end(args);
2616
2617         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2618                sb->s_id, function, error_buf);
2619
2620         /* We don't have the cluster support yet to go straight to
2621          * hard readonly in here. Until then, we want to keep
2622          * ocfs2_abort() so that we can at least mark critical
2623          * errors.
2624          *
2625          * TODO: This should abort the journal and alert other nodes
2626          * that our slot needs recovery. */
2627
2628         /* Force a panic(). This stinks, but it's better than letting
2629          * things continue without having a proper hard readonly
2630          * here. */
2631         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2632                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2633         ocfs2_handle_error(sb);
2634 }
2635
2636 /*
2637  * Void signal blockers, because in-kernel sigprocmask() only fails
2638  * when SIG_* is wrong.
2639  */
2640 void ocfs2_block_signals(sigset_t *oldset)
2641 {
2642         int rc;
2643         sigset_t blocked;
2644
2645         sigfillset(&blocked);
2646         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2647         BUG_ON(rc);
2648 }
2649
2650 void ocfs2_unblock_signals(sigset_t *oldset)
2651 {
2652         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2653         BUG_ON(rc);
2654 }
2655
2656 module_init(ocfs2_init);
2657 module_exit(ocfs2_exit);