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 "ver.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78
  79 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  80 struct kmem_cache *ocfs2_dquot_cachep;
  81 struct kmem_cache *ocfs2_qf_chunk_cachep;
  82
  83 /* OCFS2 needs to schedule several different types of work which
  84  * require cluster locking, disk I/O, recovery waits, etc. Since these
  85  * types of work tend to be heavy we avoid using the kernel events
  86  * workqueue and schedule on our own. */
  87 struct workqueue_struct *ocfs2_wq = NULL;
  88
  89 static struct dentry *ocfs2_debugfs_root = NULL;
  90
  91 MODULE_AUTHOR("Oracle");
  92 MODULE_LICENSE("GPL");
  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  Needs: %d\n", "Commit",
 290                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 291                         osb->osb_commit_interval,
 292                         atomic_read(&osb->needs_checkpoint));
 293
 294         out += snprintf(buf + out, len - out,
 295                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 296                         "Journal", osb->journal->j_state,
 297                         osb->journal->j_trans_id,
 298                         atomic_read(&osb->journal->j_num_trans));
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 302                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 303                         "Stats",
 304                         atomic_read(&osb->alloc_stats.bitmap_data),
 305                         atomic_read(&osb->alloc_stats.local_data),
 306                         atomic_read(&osb->alloc_stats.bg_allocs),
 307                         atomic_read(&osb->alloc_stats.moves),
 308                         atomic_read(&osb->alloc_stats.bg_extends));
 309
 310         out += snprintf(buf + out, len - out,
 311                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 312                         "Default: %u bits\n",
 313                         "LocalAlloc", osb->local_alloc_state,
 314                         (unsigned long long)osb->la_last_gd,
 315                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 316
 317         spin_lock(&osb->osb_lock);
 318         out += snprintf(buf + out, len - out,
 319                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 320                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 321                         osb->s_inode_steal_slot,
 322                         atomic_read(&osb->s_num_inodes_stolen),
 323                         osb->s_meta_steal_slot,
 324                         atomic_read(&osb->s_num_meta_stolen));
 325         spin_unlock(&osb->osb_lock);
 326
 327         out += snprintf(buf + out, len - out, "OrphanScan => ");
 328         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 329                         os->os_count, os->os_seqno);
 330         out += snprintf(buf + out, len - out, " Last Scan: ");
 331         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 332                 out += snprintf(buf + out, len - out, "Disabled\n");
 333         else
 334                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 335                                 (get_seconds() - os->os_scantime.tv_sec));
 336
 337         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 338                         "Slots", "Num", "RecoGen");
 339         for (i = 0; i < osb->max_slots; ++i) {
 340                 out += snprintf(buf + out, len - out,
 341                                 "%10s  %c %3d  %10d\n",
 342                                 " ",
 343                                 (i == osb->slot_num ? '*' : ' '),
 344                                 i, osb->slot_recovery_generations[i]);
 345         }
 346
 347         return out;
 348 }
 349
 350 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 351 {
 352         struct ocfs2_super *osb = inode->i_private;
 353         char *buf = NULL;
 354
 355         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 356         if (!buf)
 357                 goto bail;
 358
 359         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 360
 361         file->private_data = buf;
 362
 363         return 0;
 364 bail:
 365         return -ENOMEM;
 366 }
 367
 368 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 369 {
 370         kfree(file->private_data);
 371         return 0;
 372 }
 373
 374 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 375                                 size_t nbytes, loff_t *ppos)
 376 {
 377         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 378                                        i_size_read(file->f_mapping->host));
 379 }
 380 #else
 381 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 382 {
 383         return 0;
 384 }
 385 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 386 {
 387         return 0;
 388 }
 389 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 390                                 size_t nbytes, loff_t *ppos)
 391 {
 392         return 0;
 393 }
 394 #endif  /* CONFIG_DEBUG_FS */
 395
 396 static const struct file_operations ocfs2_osb_debug_fops = {
 397         .open =         ocfs2_osb_debug_open,
 398         .release =      ocfs2_debug_release,
 399         .read =         ocfs2_debug_read,
 400         .llseek =       generic_file_llseek,
 401 };
 402
 403 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 404 {
 405         int status;
 406         tid_t target;
 407         struct ocfs2_super *osb = OCFS2_SB(sb);
 408
 409         if (ocfs2_is_hard_readonly(osb))
 410                 return -EROFS;
 411
 412         if (wait) {
 413                 status = ocfs2_flush_truncate_log(osb);
 414                 if (status < 0)
 415                         mlog_errno(status);
 416         } else {
 417                 ocfs2_schedule_truncate_log_flush(osb, 0);
 418         }
 419
 420         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 421                                       &target)) {
 422                 if (wait)
 423                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 424                                              target);
 425         }
 426         return 0;
 427 }
 428
 429 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 430 {
 431         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 432             && (ino == USER_QUOTA_SYSTEM_INODE
 433                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 434                 return 0;
 435         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 436             && (ino == GROUP_QUOTA_SYSTEM_INODE
 437                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 438                 return 0;
 439         return 1;
 440 }
 441
 442 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 443 {
 444         struct inode *new = NULL;
 445         int status = 0;
 446         int i;
 447
 448         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 449         if (IS_ERR(new)) {
 450                 status = PTR_ERR(new);
 451                 mlog_errno(status);
 452                 goto bail;
 453         }
 454         osb->root_inode = new;
 455
 456         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 457         if (IS_ERR(new)) {
 458                 status = PTR_ERR(new);
 459                 mlog_errno(status);
 460                 goto bail;
 461         }
 462         osb->sys_root_inode = new;
 463
 464         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 465              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 466                 if (!ocfs2_need_system_inode(osb, i))
 467                         continue;
 468                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 469                 if (!new) {
 470                         ocfs2_release_system_inodes(osb);
 471                         status = -EINVAL;
 472                         mlog_errno(status);
 473                         /* FIXME: Should ERROR_RO_FS */
 474                         mlog(ML_ERROR, "Unable to load system inode %d, "
 475                              "possibly corrupt fs?", i);
 476                         goto bail;
 477                 }
 478                 // the array now has one ref, so drop this one
 479                 iput(new);
 480         }
 481
 482 bail:
 483         if (status)
 484                 mlog_errno(status);
 485         return status;
 486 }
 487
 488 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 489 {
 490         struct inode *new = NULL;
 491         int status = 0;
 492         int i;
 493
 494         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 495              i < NUM_SYSTEM_INODES;
 496              i++) {
 497                 if (!ocfs2_need_system_inode(osb, i))
 498                         continue;
 499                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 500                 if (!new) {
 501                         ocfs2_release_system_inodes(osb);
 502                         status = -EINVAL;
 503                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 504                              status, i, osb->slot_num);
 505                         goto bail;
 506                 }
 507                 /* the array now has one ref, so drop this one */
 508                 iput(new);
 509         }
 510
 511 bail:
 512         if (status)
 513                 mlog_errno(status);
 514         return status;
 515 }
 516
 517 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 518 {
 519         int i;
 520         struct inode *inode;
 521
 522         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 523                 inode = osb->global_system_inodes[i];
 524                 if (inode) {
 525                         iput(inode);
 526                         osb->global_system_inodes[i] = NULL;
 527                 }
 528         }
 529
 530         inode = osb->sys_root_inode;
 531         if (inode) {
 532                 iput(inode);
 533                 osb->sys_root_inode = NULL;
 534         }
 535
 536         inode = osb->root_inode;
 537         if (inode) {
 538                 iput(inode);
 539                 osb->root_inode = NULL;
 540         }
 541
 542         if (!osb->local_system_inodes)
 543                 return;
 544
 545         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 546                 if (osb->local_system_inodes[i]) {
 547                         iput(osb->local_system_inodes[i]);
 548                         osb->local_system_inodes[i] = NULL;
 549                 }
 550         }
 551
 552         kfree(osb->local_system_inodes);
 553         osb->local_system_inodes = NULL;
 554 }
 555
 556 /* We're allocating fs objects, use GFP_NOFS */
 557 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 558 {
 559         struct ocfs2_inode_info *oi;
 560
 561         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 562         if (!oi)
 563                 return NULL;
 564
 565         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 566         return &oi->vfs_inode;
 567 }
 568
 569 static void ocfs2_i_callback(struct rcu_head *head)
 570 {
 571         struct inode *inode = container_of(head, struct inode, i_rcu);
 572         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 573 }
 574
 575 static void ocfs2_destroy_inode(struct inode *inode)
 576 {
 577         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 578 }
 579
 580 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 581                                                 unsigned int cbits)
 582 {
 583         unsigned int bytes = 1 << cbits;
 584         unsigned int trim = bytes;
 585         unsigned int bitshift = 32;
 586
 587         /*
 588          * i_size and all block offsets in ocfs2 are always 64 bits
 589          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 590          * 64 bit platforms, cluster size will be the limiting factor.
 591          */
 592
 593 #if BITS_PER_LONG == 32
 594 # if defined(CONFIG_LBDAF)
 595         BUILD_BUG_ON(sizeof(sector_t) != 8);
 596         /*
 597          * We might be limited by page cache size.
 598          */
 599         if (bytes > PAGE_CACHE_SIZE) {
 600                 bytes = PAGE_CACHE_SIZE;
 601                 trim = 1;
 602                 /*
 603                  * Shift by 31 here so that we don't get larger than
 604                  * MAX_LFS_FILESIZE
 605                  */
 606                 bitshift = 31;
 607         }
 608 # else
 609         /*
 610          * We are limited by the size of sector_t. Use block size, as
 611          * that's what we expose to the VFS.
 612          */
 613         bytes = 1 << bbits;
 614         trim = 1;
 615         bitshift = 31;
 616 # endif
 617 #endif
 618
 619         /*
 620          * Trim by a whole cluster when we can actually approach the
 621          * on-disk limits. Otherwise we can overflow i_clusters when
 622          * an extent start is at the max offset.
 623          */
 624         return (((unsigned long long)bytes) << bitshift) - trim;
 625 }
 626
 627 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 628 {
 629         int incompat_features;
 630         int ret = 0;
 631         struct mount_options parsed_options;
 632         struct ocfs2_super *osb = OCFS2_SB(sb);
 633         u32 tmp;
 634
 635         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 636             !ocfs2_check_set_options(sb, &parsed_options)) {
 637                 ret = -EINVAL;
 638                 goto out;
 639         }
 640
 641         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 642                 OCFS2_MOUNT_HB_NONE;
 643         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 644                 ret = -EINVAL;
 645                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 646                 goto out;
 647         }
 648
 649         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 650             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 651                 ret = -EINVAL;
 652                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 653                 goto out;
 654         }
 655
 656         /* Probably don't want this on remount; it might
 657          * mess with other nodes */
 658         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 659             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 660                 ret = -EINVAL;
 661                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 662                 goto out;
 663         }
 664
 665         /* We're going to/from readonly mode. */
 666         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 667                 /* Disable quota accounting before remounting RO */
 668                 if (*flags & MS_RDONLY) {
 669                         ret = ocfs2_susp_quotas(osb, 0);
 670                         if (ret < 0)
 671                                 goto out;
 672                 }
 673                 /* Lock here so the check of HARD_RO and the potential
 674                  * setting of SOFT_RO is atomic. */
 675                 spin_lock(&osb->osb_lock);
 676                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 677                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 678                         ret = -EROFS;
 679                         goto unlock_osb;
 680                 }
 681
 682                 if (*flags & MS_RDONLY) {
 683                         sb->s_flags |= MS_RDONLY;
 684                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 685                 } else {
 686                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 687                                 mlog(ML_ERROR, "Cannot remount RDWR "
 688                                      "filesystem due to previous errors.\n");
 689                                 ret = -EROFS;
 690                                 goto unlock_osb;
 691                         }
 692                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 693                         if (incompat_features) {
 694                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 695                                      "of unsupported optional features "
 696                                      "(%x).\n", incompat_features);
 697                                 ret = -EINVAL;
 698                                 goto unlock_osb;
 699                         }
 700                         sb->s_flags &= ~MS_RDONLY;
 701                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 702                 }
 703                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 704 unlock_osb:
 705                 spin_unlock(&osb->osb_lock);
 706                 /* Enable quota accounting after remounting RW */
 707                 if (!ret && !(*flags & MS_RDONLY)) {
 708                         if (sb_any_quota_suspended(sb))
 709                                 ret = ocfs2_susp_quotas(osb, 1);
 710                         else
 711                                 ret = ocfs2_enable_quotas(osb);
 712                         if (ret < 0) {
 713                                 /* Return back changes... */
 714                                 spin_lock(&osb->osb_lock);
 715                                 sb->s_flags |= MS_RDONLY;
 716                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 717                                 spin_unlock(&osb->osb_lock);
 718                                 goto out;
 719                         }
 720                 }
 721         }
 722
 723         if (!ret) {
 724                 /* Only save off the new mount options in case of a successful
 725                  * remount. */
 726                 osb->s_mount_opt = parsed_options.mount_opt;
 727                 osb->s_atime_quantum = parsed_options.atime_quantum;
 728                 osb->preferred_slot = parsed_options.slot;
 729                 if (parsed_options.commit_interval)
 730                         osb->osb_commit_interval = parsed_options.commit_interval;
 731
 732                 if (!ocfs2_is_hard_readonly(osb))
 733                         ocfs2_set_journal_params(osb);
 734
 735                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 736                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 737                                                         MS_POSIXACL : 0);
 738         }
 739 out:
 740         return ret;
 741 }
 742
 743 static int ocfs2_sb_probe(struct super_block *sb,
 744                           struct buffer_head **bh,
 745                           int *sector_size,
 746                           struct ocfs2_blockcheck_stats *stats)
 747 {
 748         int status, tmpstat;
 749         struct ocfs1_vol_disk_hdr *hdr;
 750         struct ocfs2_dinode *di;
 751         int blksize;
 752
 753         *bh = NULL;
 754
 755         /* may be > 512 */
 756         *sector_size = bdev_logical_block_size(sb->s_bdev);
 757         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 758                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 759                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 760                 status = -EINVAL;
 761                 goto bail;
 762         }
 763
 764         /* Can this really happen? */
 765         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 766                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 767
 768         /* check block zero for old format */
 769         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 770         if (status < 0) {
 771                 mlog_errno(status);
 772                 goto bail;
 773         }
 774         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 775         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 776                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 777                      hdr->major_version, hdr->minor_version);
 778                 status = -EINVAL;
 779         }
 780         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 781                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 782                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 783                      hdr->signature);
 784                 status = -EINVAL;
 785         }
 786         brelse(*bh);
 787         *bh = NULL;
 788         if (status < 0) {
 789                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 790                      "upgraded before mounting with ocfs v2\n");
 791                 goto bail;
 792         }
 793
 794         /*
 795          * Now check at magic offset for 512, 1024, 2048, 4096
 796          * blocksizes.  4096 is the maximum blocksize because it is
 797          * the minimum clustersize.
 798          */
 799         status = -EINVAL;
 800         for (blksize = *sector_size;
 801              blksize <= OCFS2_MAX_BLOCKSIZE;
 802              blksize <<= 1) {
 803                 tmpstat = ocfs2_get_sector(sb, bh,
 804                                            OCFS2_SUPER_BLOCK_BLKNO,
 805                                            blksize);
 806                 if (tmpstat < 0) {
 807                         status = tmpstat;
 808                         mlog_errno(status);
 809                         break;
 810                 }
 811                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 812                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 813                 spin_lock_init(&stats->b_lock);
 814                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 815                 if (tmpstat < 0) {
 816                         brelse(*bh);
 817                         *bh = NULL;
 818                 }
 819                 if (tmpstat != -EAGAIN) {
 820                         status = tmpstat;
 821                         break;
 822                 }
 823         }
 824
 825 bail:
 826         return status;
 827 }
 828
 829 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 830 {
 831         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 832
 833         if (osb->s_mount_opt & hb_enabled) {
 834                 if (ocfs2_mount_local(osb)) {
 835                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 836                              "mounted device.\n");
 837                         return -EINVAL;
 838                 }
 839                 if (ocfs2_userspace_stack(osb)) {
 840                         mlog(ML_ERROR, "Userspace stack expected, but "
 841                              "o2cb heartbeat arguments passed to mount\n");
 842                         return -EINVAL;
 843                 }
 844                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 845                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 846                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 847                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 848                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 849                         return -EINVAL;
 850                 }
 851         }
 852
 853         if (!(osb->s_mount_opt & hb_enabled)) {
 854                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 855                     !ocfs2_userspace_stack(osb)) {
 856                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 857                              "a read-write clustered device.\n");
 858                         return -EINVAL;
 859                 }
 860         }
 861
 862         return 0;
 863 }
 864
 865 /*
 866  * If we're using a userspace stack, mount should have passed
 867  * a name that matches the disk.  If not, mount should not
 868  * have passed a stack.
 869  */
 870 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 871                                         struct mount_options *mopt)
 872 {
 873         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 874                 mlog(ML_ERROR,
 875                      "cluster stack passed to mount, but this filesystem "
 876                      "does not support it\n");
 877                 return -EINVAL;
 878         }
 879
 880         if (ocfs2_userspace_stack(osb) &&
 881             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 882                     OCFS2_STACK_LABEL_LEN)) {
 883                 mlog(ML_ERROR,
 884                      "cluster stack passed to mount (\"%s\") does not "
 885                      "match the filesystem (\"%s\")\n",
 886                      mopt->cluster_stack,
 887                      osb->osb_cluster_stack);
 888                 return -EINVAL;
 889         }
 890
 891         return 0;
 892 }
 893
 894 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 895 {
 896         int type;
 897         struct super_block *sb = osb->sb;
 898         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 899                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 900         int status = 0;
 901
 902         for (type = 0; type < MAXQUOTAS; type++) {
 903                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 904                         continue;
 905                 if (unsuspend)
 906                         status = dquot_resume(sb, type);
 907                 else {
 908                         struct ocfs2_mem_dqinfo *oinfo;
 909
 910                         /* Cancel periodic syncing before suspending */
 911                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 912                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 913                         status = dquot_suspend(sb, type);
 914                 }
 915                 if (status < 0)
 916                         break;
 917         }
 918         if (status < 0)
 919                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 920                      "remount (error = %d).\n", status);
 921         return status;
 922 }
 923
 924 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 925 {
 926         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 927         struct super_block *sb = osb->sb;
 928         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 929                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 930         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 931                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 932         int status;
 933         int type;
 934
 935         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 936         for (type = 0; type < MAXQUOTAS; type++) {
 937                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 938                         continue;
 939                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 940                                                         osb->slot_num);
 941                 if (!inode[type]) {
 942                         status = -ENOENT;
 943                         goto out_quota_off;
 944                 }
 945                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 946                                       DQUOT_USAGE_ENABLED);
 947                 if (status < 0)
 948                         goto out_quota_off;
 949         }
 950
 951         for (type = 0; type < MAXQUOTAS; type++)
 952                 iput(inode[type]);
 953         return 0;
 954 out_quota_off:
 955         ocfs2_disable_quotas(osb);
 956         for (type = 0; type < MAXQUOTAS; type++)
 957                 iput(inode[type]);
 958         mlog_errno(status);
 959         return status;
 960 }
 961
 962 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 963 {
 964         int type;
 965         struct inode *inode;
 966         struct super_block *sb = osb->sb;
 967         struct ocfs2_mem_dqinfo *oinfo;
 968
 969         /* We mostly ignore errors in this function because there's not much
 970          * we can do when we see them */
 971         for (type = 0; type < MAXQUOTAS; type++) {
 972                 if (!sb_has_quota_loaded(sb, type))
 973                         continue;
 974                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 975                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 976                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 977                 inode = igrab(sb->s_dquot.files[type]);
 978                 /* Turn off quotas. This will remove all dquot structures from
 979                  * memory and so they will be automatically synced to global
 980                  * quota files */
 981                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 982                                         DQUOT_LIMITS_ENABLED);
 983                 if (!inode)
 984                         continue;
 985                 iput(inode);
 986         }
 987 }
 988
 989 /* Handle quota on quotactl */
 990 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 991 {
 992         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 993                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 994
 995         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 996                 return -EINVAL;
 997
 998         return dquot_enable(sb_dqopt(sb)->files[type], type,
 999                             format_id, DQUOT_LIMITS_ENABLED);
1000 }
1001
1002 /* Handle quota off quotactl */
1003 static int ocfs2_quota_off(struct super_block *sb, int type)
1004 {
1005         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1006 }
1007
1008 static const struct quotactl_ops ocfs2_quotactl_ops = {
1009         .quota_on_meta  = ocfs2_quota_on,
1010         .quota_off      = ocfs2_quota_off,
1011         .quota_sync     = dquot_quota_sync,
1012         .get_info       = dquot_get_dqinfo,
1013         .set_info       = dquot_set_dqinfo,
1014         .get_dqblk      = dquot_get_dqblk,
1015         .set_dqblk      = dquot_set_dqblk,
1016 };
1017
1018 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1019 {
1020         struct dentry *root;
1021         int status, sector_size;
1022         struct mount_options parsed_options;
1023         struct inode *inode = NULL;
1024         struct ocfs2_super *osb = NULL;
1025         struct buffer_head *bh = NULL;
1026         char nodestr[8];
1027         struct ocfs2_blockcheck_stats stats;
1028
1029         trace_ocfs2_fill_super(sb, data, silent);
1030
1031         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1032                 status = -EINVAL;
1033                 goto read_super_error;
1034         }
1035
1036         /* probe for superblock */
1037         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1038         if (status < 0) {
1039                 mlog(ML_ERROR, "superblock probe failed!\n");
1040                 goto read_super_error;
1041         }
1042
1043         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1044         osb = OCFS2_SB(sb);
1045         if (status < 0) {
1046                 mlog_errno(status);
1047                 goto read_super_error;
1048         }
1049         brelse(bh);
1050         bh = NULL;
1051
1052         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1053                 status = -EINVAL;
1054                 goto read_super_error;
1055         }
1056         osb->s_mount_opt = parsed_options.mount_opt;
1057         osb->s_atime_quantum = parsed_options.atime_quantum;
1058         osb->preferred_slot = parsed_options.slot;
1059         osb->osb_commit_interval = parsed_options.commit_interval;
1060
1061         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1062         osb->osb_resv_level = parsed_options.resv_level;
1063         osb->osb_dir_resv_level = parsed_options.resv_level;
1064         if (parsed_options.dir_resv_level == -1)
1065                 osb->osb_dir_resv_level = parsed_options.resv_level;
1066         else
1067                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1068
1069         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1070         if (status)
1071                 goto read_super_error;
1072
1073         sb->s_magic = OCFS2_SUPER_MAGIC;
1074
1075         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1076                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1077
1078         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1079          * heartbeat=none */
1080         if (bdev_read_only(sb->s_bdev)) {
1081                 if (!(sb->s_flags & MS_RDONLY)) {
1082                         status = -EACCES;
1083                         mlog(ML_ERROR, "Readonly device detected but readonly "
1084                              "mount was not specified.\n");
1085                         goto read_super_error;
1086                 }
1087
1088                 /* You should not be able to start a local heartbeat
1089                  * on a readonly device. */
1090                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1091                         status = -EROFS;
1092                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1093                              "device.\n");
1094                         goto read_super_error;
1095                 }
1096
1097                 status = ocfs2_check_journals_nolocks(osb);
1098                 if (status < 0) {
1099                         if (status == -EROFS)
1100                                 mlog(ML_ERROR, "Recovery required on readonly "
1101                                      "file system, but write access is "
1102                                      "unavailable.\n");
1103                         else
1104                                 mlog_errno(status);
1105                         goto read_super_error;
1106                 }
1107
1108                 ocfs2_set_ro_flag(osb, 1);
1109
1110                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1111                        "Cluster services will not be used for this mount. "
1112                        "Recovery will be skipped.\n", osb->dev_str);
1113         }
1114
1115         if (!ocfs2_is_hard_readonly(osb)) {
1116                 if (sb->s_flags & MS_RDONLY)
1117                         ocfs2_set_ro_flag(osb, 0);
1118         }
1119
1120         status = ocfs2_verify_heartbeat(osb);
1121         if (status < 0) {
1122                 mlog_errno(status);
1123                 goto read_super_error;
1124         }
1125
1126         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1127                                                  ocfs2_debugfs_root);
1128         if (!osb->osb_debug_root) {
1129                 status = -EINVAL;
1130                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1131                 goto read_super_error;
1132         }
1133
1134         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1135                                             osb->osb_debug_root,
1136                                             osb,
1137                                             &ocfs2_osb_debug_fops);
1138         if (!osb->osb_ctxt) {
1139                 status = -EINVAL;
1140                 mlog_errno(status);
1141                 goto read_super_error;
1142         }
1143
1144         if (ocfs2_meta_ecc(osb)) {
1145                 status = ocfs2_blockcheck_stats_debugfs_install(
1146                                                 &osb->osb_ecc_stats,
1147                                                 osb->osb_debug_root);
1148                 if (status) {
1149                         mlog(ML_ERROR,
1150                              "Unable to create blockcheck statistics "
1151                              "files\n");
1152                         goto read_super_error;
1153                 }
1154         }
1155
1156         status = ocfs2_mount_volume(sb);
1157         if (status < 0)
1158                 goto read_super_error;
1159
1160         if (osb->root_inode)
1161                 inode = igrab(osb->root_inode);
1162
1163         if (!inode) {
1164                 status = -EIO;
1165                 mlog_errno(status);
1166                 goto read_super_error;
1167         }
1168
1169         root = d_make_root(inode);
1170         if (!root) {
1171                 status = -ENOMEM;
1172                 mlog_errno(status);
1173                 goto read_super_error;
1174         }
1175
1176         sb->s_root = root;
1177
1178         ocfs2_complete_mount_recovery(osb);
1179
1180         if (ocfs2_mount_local(osb))
1181                 snprintf(nodestr, sizeof(nodestr), "local");
1182         else
1183                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1184
1185         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1186                "with %s data mode.\n",
1187                osb->dev_str, nodestr, osb->slot_num,
1188                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1189                "ordered");
1190
1191         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1192         wake_up(&osb->osb_mount_event);
1193
1194         /* Now we can initialize quotas because we can afford to wait
1195          * for cluster locks recovery now. That also means that truncation
1196          * log recovery can happen but that waits for proper quota setup */
1197         if (!(sb->s_flags & MS_RDONLY)) {
1198                 status = ocfs2_enable_quotas(osb);
1199                 if (status < 0) {
1200                         /* We have to err-out specially here because
1201                          * s_root is already set */
1202                         mlog_errno(status);
1203                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1204                         wake_up(&osb->osb_mount_event);
1205                         return status;
1206                 }
1207         }
1208
1209         ocfs2_complete_quota_recovery(osb);
1210
1211         /* Now we wake up again for processes waiting for quotas */
1212         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1213         wake_up(&osb->osb_mount_event);
1214
1215         /* Start this when the mount is almost sure of being successful */
1216         ocfs2_orphan_scan_start(osb);
1217
1218         return status;
1219
1220 read_super_error:
1221         brelse(bh);
1222
1223         if (osb) {
1224                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1225                 wake_up(&osb->osb_mount_event);
1226                 ocfs2_dismount_volume(sb, 1);
1227         }
1228
1229         if (status)
1230                 mlog_errno(status);
1231         return status;
1232 }
1233
1234 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1235                         int flags,
1236                         const char *dev_name,
1237                         void *data)
1238 {
1239         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1240 }
1241
1242 static void ocfs2_kill_sb(struct super_block *sb)
1243 {
1244         struct ocfs2_super *osb = OCFS2_SB(sb);
1245
1246         /* Failed mount? */
1247         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1248                 goto out;
1249
1250         /* Prevent further queueing of inode drop events */
1251         spin_lock(&dentry_list_lock);
1252         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1253         spin_unlock(&dentry_list_lock);
1254         /* Wait for work to finish and/or remove it */
1255         cancel_work_sync(&osb->dentry_lock_work);
1256 out:
1257         kill_block_super(sb);
1258 }
1259
1260 static struct file_system_type ocfs2_fs_type = {
1261         .owner          = THIS_MODULE,
1262         .name           = "ocfs2",
1263         .mount          = ocfs2_mount,
1264         .kill_sb        = ocfs2_kill_sb,
1265
1266         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1267         .next           = NULL
1268 };
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         ocfs2_print_version();
1622
1623         for (i = 0; i < OCFS2_IOEND_WQ_HASH_SZ; i++)
1624                 init_waitqueue_head(&ocfs2__ioend_wq[i]);
1625
1626         status = init_ocfs2_uptodate_cache();
1627         if (status < 0)
1628                 goto out1;
1629
1630         status = ocfs2_initialize_mem_caches();
1631         if (status < 0)
1632                 goto out2;
1633
1634         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1635         if (!ocfs2_wq) {
1636                 status = -ENOMEM;
1637                 goto out3;
1638         }
1639
1640         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1641         if (!ocfs2_debugfs_root) {
1642                 status = -EFAULT;
1643                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1644         }
1645
1646         ocfs2_set_locking_protocol();
1647
1648         status = register_quota_format(&ocfs2_quota_format);
1649         if (status < 0)
1650                 goto out4;
1651         status = register_filesystem(&ocfs2_fs_type);
1652         if (!status)
1653                 return 0;
1654
1655         unregister_quota_format(&ocfs2_quota_format);
1656 out4:
1657         destroy_workqueue(ocfs2_wq);
1658         debugfs_remove(ocfs2_debugfs_root);
1659 out3:
1660         ocfs2_free_mem_caches();
1661 out2:
1662         exit_ocfs2_uptodate_cache();
1663 out1:
1664         mlog_errno(status);
1665         return status;
1666 }
1667
1668 static void __exit ocfs2_exit(void)
1669 {
1670         if (ocfs2_wq) {
1671                 flush_workqueue(ocfs2_wq);
1672                 destroy_workqueue(ocfs2_wq);
1673         }
1674
1675         unregister_quota_format(&ocfs2_quota_format);
1676
1677         debugfs_remove(ocfs2_debugfs_root);
1678
1679         ocfs2_free_mem_caches();
1680
1681         unregister_filesystem(&ocfs2_fs_type);
1682
1683         exit_ocfs2_uptodate_cache();
1684 }
1685
1686 static void ocfs2_put_super(struct super_block *sb)
1687 {
1688         trace_ocfs2_put_super(sb);
1689
1690         ocfs2_sync_blockdev(sb);
1691         ocfs2_dismount_volume(sb, 0);
1692 }
1693
1694 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1695 {
1696         struct ocfs2_super *osb;
1697         u32 numbits, freebits;
1698         int status;
1699         struct ocfs2_dinode *bm_lock;
1700         struct buffer_head *bh = NULL;
1701         struct inode *inode = NULL;
1702
1703         trace_ocfs2_statfs(dentry->d_sb, buf);
1704
1705         osb = OCFS2_SB(dentry->d_sb);
1706
1707         inode = ocfs2_get_system_file_inode(osb,
1708                                             GLOBAL_BITMAP_SYSTEM_INODE,
1709                                             OCFS2_INVALID_SLOT);
1710         if (!inode) {
1711                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1712                 status = -EIO;
1713                 goto bail;
1714         }
1715
1716         status = ocfs2_inode_lock(inode, &bh, 0);
1717         if (status < 0) {
1718                 mlog_errno(status);
1719                 goto bail;
1720         }
1721
1722         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1723
1724         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1725         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1726
1727         buf->f_type = OCFS2_SUPER_MAGIC;
1728         buf->f_bsize = dentry->d_sb->s_blocksize;
1729         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1730         buf->f_blocks = ((sector_t) numbits) *
1731                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1732         buf->f_bfree = ((sector_t) freebits) *
1733                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1734         buf->f_bavail = buf->f_bfree;
1735         buf->f_files = numbits;
1736         buf->f_ffree = freebits;
1737         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1738                                 & 0xFFFFFFFFUL;
1739         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1740                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1741
1742         brelse(bh);
1743
1744         ocfs2_inode_unlock(inode, 0);
1745         status = 0;
1746 bail:
1747         if (inode)
1748                 iput(inode);
1749
1750         if (status)
1751                 mlog_errno(status);
1752
1753         return status;
1754 }
1755
1756 static void ocfs2_inode_init_once(void *data)
1757 {
1758         struct ocfs2_inode_info *oi = data;
1759
1760         oi->ip_flags = 0;
1761         oi->ip_open_count = 0;
1762         spin_lock_init(&oi->ip_lock);
1763         ocfs2_extent_map_init(&oi->vfs_inode);
1764         INIT_LIST_HEAD(&oi->ip_io_markers);
1765         oi->ip_dir_start_lookup = 0;
1766         atomic_set(&oi->ip_unaligned_aio, 0);
1767         init_rwsem(&oi->ip_alloc_sem);
1768         init_rwsem(&oi->ip_xattr_sem);
1769         mutex_init(&oi->ip_io_mutex);
1770
1771         oi->ip_blkno = 0ULL;
1772         oi->ip_clusters = 0;
1773
1774         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1775
1776         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1777         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1778         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1779
1780         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1781                                   &ocfs2_inode_caching_ops);
1782
1783         inode_init_once(&oi->vfs_inode);
1784 }
1785
1786 static int ocfs2_initialize_mem_caches(void)
1787 {
1788         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1789                                        sizeof(struct ocfs2_inode_info),
1790                                        0,
1791                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1792                                                 SLAB_MEM_SPREAD),
1793                                        ocfs2_inode_init_once);
1794         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1795                                         sizeof(struct ocfs2_dquot),
1796                                         0,
1797                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1798                                                 SLAB_MEM_SPREAD),
1799                                         NULL);
1800         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1801                                         sizeof(struct ocfs2_quota_chunk),
1802                                         0,
1803                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1804                                         NULL);
1805         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1806             !ocfs2_qf_chunk_cachep) {
1807                 if (ocfs2_inode_cachep)
1808                         kmem_cache_destroy(ocfs2_inode_cachep);
1809                 if (ocfs2_dquot_cachep)
1810                         kmem_cache_destroy(ocfs2_dquot_cachep);
1811                 if (ocfs2_qf_chunk_cachep)
1812                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1813                 return -ENOMEM;
1814         }
1815
1816         return 0;
1817 }
1818
1819 static void ocfs2_free_mem_caches(void)
1820 {
1821         /*
1822          * Make sure all delayed rcu free inodes are flushed before we
1823          * destroy cache.
1824          */
1825         rcu_barrier();
1826         if (ocfs2_inode_cachep)
1827                 kmem_cache_destroy(ocfs2_inode_cachep);
1828         ocfs2_inode_cachep = NULL;
1829
1830         if (ocfs2_dquot_cachep)
1831                 kmem_cache_destroy(ocfs2_dquot_cachep);
1832         ocfs2_dquot_cachep = NULL;
1833
1834         if (ocfs2_qf_chunk_cachep)
1835                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1836         ocfs2_qf_chunk_cachep = NULL;
1837 }
1838
1839 static int ocfs2_get_sector(struct super_block *sb,
1840                             struct buffer_head **bh,
1841                             int block,
1842                             int sect_size)
1843 {
1844         if (!sb_set_blocksize(sb, sect_size)) {
1845                 mlog(ML_ERROR, "unable to set blocksize\n");
1846                 return -EIO;
1847         }
1848
1849         *bh = sb_getblk(sb, block);
1850         if (!*bh) {
1851                 mlog_errno(-EIO);
1852                 return -EIO;
1853         }
1854         lock_buffer(*bh);
1855         if (!buffer_dirty(*bh))
1856                 clear_buffer_uptodate(*bh);
1857         unlock_buffer(*bh);
1858         ll_rw_block(READ, 1, bh);
1859         wait_on_buffer(*bh);
1860         if (!buffer_uptodate(*bh)) {
1861                 mlog_errno(-EIO);
1862                 brelse(*bh);
1863                 *bh = NULL;
1864                 return -EIO;
1865         }
1866
1867         return 0;
1868 }
1869
1870 static int ocfs2_mount_volume(struct super_block *sb)
1871 {
1872         int status = 0;
1873         int unlock_super = 0;
1874         struct ocfs2_super *osb = OCFS2_SB(sb);
1875
1876         if (ocfs2_is_hard_readonly(osb))
1877                 goto leave;
1878
1879         status = ocfs2_dlm_init(osb);
1880         if (status < 0) {
1881                 mlog_errno(status);
1882                 goto leave;
1883         }
1884
1885         status = ocfs2_super_lock(osb, 1);
1886         if (status < 0) {
1887                 mlog_errno(status);
1888                 goto leave;
1889         }
1890         unlock_super = 1;
1891
1892         /* This will load up the node map and add ourselves to it. */
1893         status = ocfs2_find_slot(osb);
1894         if (status < 0) {
1895                 mlog_errno(status);
1896                 goto leave;
1897         }
1898
1899         /* load all node-local system inodes */
1900         status = ocfs2_init_local_system_inodes(osb);
1901         if (status < 0) {
1902                 mlog_errno(status);
1903                 goto leave;
1904         }
1905
1906         status = ocfs2_check_volume(osb);
1907         if (status < 0) {
1908                 mlog_errno(status);
1909                 goto leave;
1910         }
1911
1912         status = ocfs2_truncate_log_init(osb);
1913         if (status < 0)
1914                 mlog_errno(status);
1915
1916 leave:
1917         if (unlock_super)
1918                 ocfs2_super_unlock(osb, 1);
1919
1920         return status;
1921 }
1922
1923 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1924 {
1925         int tmp, hangup_needed = 0;
1926         struct ocfs2_super *osb = NULL;
1927         char nodestr[8];
1928
1929         trace_ocfs2_dismount_volume(sb);
1930
1931         BUG_ON(!sb);
1932         osb = OCFS2_SB(sb);
1933         BUG_ON(!osb);
1934
1935         debugfs_remove(osb->osb_ctxt);
1936
1937         /*
1938          * Flush inode dropping work queue so that deletes are
1939          * performed while the filesystem is still working
1940          */
1941         ocfs2_drop_all_dl_inodes(osb);
1942
1943         /* Orphan scan should be stopped as early as possible */
1944         ocfs2_orphan_scan_stop(osb);
1945
1946         ocfs2_disable_quotas(osb);
1947
1948         ocfs2_shutdown_local_alloc(osb);
1949
1950         ocfs2_truncate_log_shutdown(osb);
1951
1952         /* This will disable recovery and flush any recovery work. */
1953         ocfs2_recovery_exit(osb);
1954
1955         ocfs2_journal_shutdown(osb);
1956
1957         ocfs2_sync_blockdev(sb);
1958
1959         ocfs2_purge_refcount_trees(osb);
1960
1961         /* No cluster connection means we've failed during mount, so skip
1962          * all the steps which depended on that to complete. */
1963         if (osb->cconn) {
1964                 tmp = ocfs2_super_lock(osb, 1);
1965                 if (tmp < 0) {
1966                         mlog_errno(tmp);
1967                         return;
1968                 }
1969         }
1970
1971         if (osb->slot_num != OCFS2_INVALID_SLOT)
1972                 ocfs2_put_slot(osb);
1973
1974         if (osb->cconn)
1975                 ocfs2_super_unlock(osb, 1);
1976
1977         ocfs2_release_system_inodes(osb);
1978
1979         /*
1980          * If we're dismounting due to mount error, mount.ocfs2 will clean
1981          * up heartbeat.  If we're a local mount, there is no heartbeat.
1982          * If we failed before we got a uuid_str yet, we can't stop
1983          * heartbeat.  Otherwise, do it.
1984          */
1985         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1986             !ocfs2_is_hard_readonly(osb))
1987                 hangup_needed = 1;
1988
1989         if (osb->cconn)
1990                 ocfs2_dlm_shutdown(osb, hangup_needed);
1991
1992         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1993         debugfs_remove(osb->osb_debug_root);
1994
1995         if (hangup_needed)
1996                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1997
1998         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1999
2000         if (ocfs2_mount_local(osb))
2001                 snprintf(nodestr, sizeof(nodestr), "local");
2002         else
2003                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2004
2005         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2006                osb->dev_str, nodestr);
2007
2008         ocfs2_delete_osb(osb);
2009         kfree(osb);
2010         sb->s_dev = 0;
2011         sb->s_fs_info = NULL;
2012 }
2013
2014 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2015                                 unsigned uuid_bytes)
2016 {
2017         int i, ret;
2018         char *ptr;
2019
2020         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2021
2022         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2023         if (osb->uuid_str == NULL)
2024                 return -ENOMEM;
2025
2026         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2027                 /* print with null */
2028                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2029                 if (ret != 2) /* drop super cleans up */
2030                         return -EINVAL;
2031                 /* then only advance past the last char */
2032                 ptr += 2;
2033         }
2034
2035         return 0;
2036 }
2037
2038 /* Make sure entire volume is addressable by our journal.  Requires
2039    osb_clusters_at_boot to be valid and for the journal to have been
2040    initialized by ocfs2_journal_init(). */
2041 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2042 {
2043         int status = 0;
2044         u64 max_block =
2045                 ocfs2_clusters_to_blocks(osb->sb,
2046                                          osb->osb_clusters_at_boot) - 1;
2047
2048         /* 32-bit block number is always OK. */
2049         if (max_block <= (u32)~0ULL)
2050                 goto out;
2051
2052         /* Volume is "huge", so see if our journal is new enough to
2053            support it. */
2054         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2055                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2056               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2057                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2058                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2059                      "Enable the 'block64' journal option with tunefs.ocfs2");
2060                 status = -EFBIG;
2061                 goto out;
2062         }
2063
2064  out:
2065         return status;
2066 }
2067
2068 static int ocfs2_initialize_super(struct super_block *sb,
2069                                   struct buffer_head *bh,
2070                                   int sector_size,
2071                                   struct ocfs2_blockcheck_stats *stats)
2072 {
2073         int status;
2074         int i, cbits, bbits;
2075         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2076         struct inode *inode = NULL;
2077         struct ocfs2_journal *journal;
2078         __le32 uuid_net_key;
2079         struct ocfs2_super *osb;
2080         u64 total_blocks;
2081
2082         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2083         if (!osb) {
2084                 status = -ENOMEM;
2085                 mlog_errno(status);
2086                 goto bail;
2087         }
2088
2089         sb->s_fs_info = osb;
2090         sb->s_op = &ocfs2_sops;
2091         sb->s_d_op = &ocfs2_dentry_ops;
2092         sb->s_export_op = &ocfs2_export_ops;
2093         sb->s_qcop = &ocfs2_quotactl_ops;
2094         sb->dq_op = &ocfs2_quota_operations;
2095         sb->s_xattr = ocfs2_xattr_handlers;
2096         sb->s_time_gran = 1;
2097         sb->s_flags |= MS_NOATIME;
2098         /* this is needed to support O_LARGEFILE */
2099         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2100         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2101         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2102
2103         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2104
2105         for (i = 0; i < 3; i++)
2106                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2107         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2108
2109         osb->sb = sb;
2110         /* Save off for ocfs2_rw_direct */
2111         osb->s_sectsize_bits = blksize_bits(sector_size);
2112         BUG_ON(!osb->s_sectsize_bits);
2113
2114         spin_lock_init(&osb->dc_task_lock);
2115         init_waitqueue_head(&osb->dc_event);
2116         osb->dc_work_sequence = 0;
2117         osb->dc_wake_sequence = 0;
2118         INIT_LIST_HEAD(&osb->blocked_lock_list);
2119         osb->blocked_lock_count = 0;
2120         spin_lock_init(&osb->osb_lock);
2121         spin_lock_init(&osb->osb_xattr_lock);
2122         ocfs2_init_steal_slots(osb);
2123
2124         atomic_set(&osb->alloc_stats.moves, 0);
2125         atomic_set(&osb->alloc_stats.local_data, 0);
2126         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2127         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2128         atomic_set(&osb->alloc_stats.bg_extends, 0);
2129
2130         /* Copy the blockcheck stats from the superblock probe */
2131         osb->osb_ecc_stats = *stats;
2132
2133         ocfs2_init_node_maps(osb);
2134
2135         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2136                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2137
2138         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2139         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2140                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2141                      osb->max_slots);
2142                 status = -EINVAL;
2143                 goto bail;
2144         }
2145
2146         ocfs2_orphan_scan_init(osb);
2147
2148         status = ocfs2_recovery_init(osb);
2149         if (status) {
2150                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2151                 mlog_errno(status);
2152                 goto bail;
2153         }
2154
2155         init_waitqueue_head(&osb->checkpoint_event);
2156         atomic_set(&osb->needs_checkpoint, 0);
2157
2158         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2159
2160         osb->slot_num = OCFS2_INVALID_SLOT;
2161
2162         osb->s_xattr_inline_size = le16_to_cpu(
2163                                         di->id2.i_super.s_xattr_inline_size);
2164
2165         osb->local_alloc_state = OCFS2_LA_UNUSED;
2166         osb->local_alloc_bh = NULL;
2167         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2168
2169         init_waitqueue_head(&osb->osb_mount_event);
2170
2171         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2172         if (status) {
2173                 mlog_errno(status);
2174                 goto bail;
2175         }
2176
2177         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2178         if (!osb->vol_label) {
2179                 mlog(ML_ERROR, "unable to alloc vol label\n");
2180                 status = -ENOMEM;
2181                 goto bail;
2182         }
2183
2184         osb->slot_recovery_generations =
2185                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2186                         GFP_KERNEL);
2187         if (!osb->slot_recovery_generations) {
2188                 status = -ENOMEM;
2189                 mlog_errno(status);
2190                 goto bail;
2191         }
2192
2193         init_waitqueue_head(&osb->osb_wipe_event);
2194         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2195                                         sizeof(*osb->osb_orphan_wipes),
2196                                         GFP_KERNEL);
2197         if (!osb->osb_orphan_wipes) {
2198                 status = -ENOMEM;
2199                 mlog_errno(status);
2200                 goto bail;
2201         }
2202
2203         osb->osb_rf_lock_tree = RB_ROOT;
2204
2205         osb->s_feature_compat =
2206                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2207         osb->s_feature_ro_compat =
2208                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2209         osb->s_feature_incompat =
2210                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2211
2212         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2213                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2214                      "optional features (%x).\n", i);
2215                 status = -EINVAL;
2216                 goto bail;
2217         }
2218         if (!(osb->sb->s_flags & MS_RDONLY) &&
2219             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2220                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2221                      "unsupported optional features (%x).\n", i);
2222                 status = -EINVAL;
2223                 goto bail;
2224         }
2225
2226         if (ocfs2_clusterinfo_valid(osb)) {
2227                 osb->osb_stackflags =
2228                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2229                 memcpy(osb->osb_cluster_stack,
2230                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2231                        OCFS2_STACK_LABEL_LEN);
2232                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
2233                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2234                         mlog(ML_ERROR,
2235                              "couldn't mount because of an invalid "
2236                              "cluster stack label (%s) \n",
2237                              osb->osb_cluster_stack);
2238                         status = -EINVAL;
2239                         goto bail;
2240                 }
2241         } else {
2242                 /* The empty string is identical with classic tools that
2243                  * don't know about s_cluster_info. */
2244                 osb->osb_cluster_stack[0] = '\0';
2245         }
2246
2247         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2248
2249         /* FIXME
2250          * This should be done in ocfs2_journal_init(), but unknown
2251          * ordering issues will cause the filesystem to crash.
2252          * If anyone wants to figure out what part of the code
2253          * refers to osb->journal before ocfs2_journal_init() is run,
2254          * be my guest.
2255          */
2256         /* initialize our journal structure */
2257
2258         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2259         if (!journal) {
2260                 mlog(ML_ERROR, "unable to alloc journal\n");
2261                 status = -ENOMEM;
2262                 goto bail;
2263         }
2264         osb->journal = journal;
2265         journal->j_osb = osb;
2266
2267         atomic_set(&journal->j_num_trans, 0);
2268         init_rwsem(&journal->j_trans_barrier);
2269         init_waitqueue_head(&journal->j_checkpointed);
2270         spin_lock_init(&journal->j_lock);
2271         journal->j_trans_id = (unsigned long) 1;
2272         INIT_LIST_HEAD(&journal->j_la_cleanups);
2273         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2274         journal->j_state = OCFS2_JOURNAL_FREE;
2275
2276         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2277         osb->dentry_lock_list = NULL;
2278
2279         /* get some pseudo constants for clustersize bits */
2280         osb->s_clustersize_bits =
2281                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2282         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2283
2284         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2285             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2286                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2287                      osb->s_clustersize);
2288                 status = -EINVAL;
2289                 goto bail;
2290         }
2291
2292         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2293                                                 le32_to_cpu(di->i_clusters));
2294
2295         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2296                                            total_blocks);
2297         if (status) {
2298                 mlog(ML_ERROR, "Volume too large "
2299                      "to mount safely on this system");
2300                 status = -EFBIG;
2301                 goto bail;
2302         }
2303
2304         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2305                                  sizeof(di->id2.i_super.s_uuid))) {
2306                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2307                 status = -ENOMEM;
2308                 goto bail;
2309         }
2310
2311         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2312
2313         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2314         osb->vol_label[63] = '\0';
2315         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2316         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2317         osb->first_cluster_group_blkno =
2318                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2319         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2320         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2321         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2322                                      (unsigned long long)osb->root_blkno,
2323                                      (unsigned long long)osb->system_dir_blkno,
2324                                      osb->s_clustersize_bits);
2325
2326         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2327         if (!osb->osb_dlm_debug) {
2328                 status = -ENOMEM;
2329                 mlog_errno(status);
2330                 goto bail;
2331         }
2332
2333         atomic_set(&osb->vol_state, VOLUME_INIT);
2334
2335         /* load root, system_dir, and all global system inodes */
2336         status = ocfs2_init_global_system_inodes(osb);
2337         if (status < 0) {
2338                 mlog_errno(status);
2339                 goto bail;
2340         }
2341
2342         /*
2343          * global bitmap
2344          */
2345         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2346                                             OCFS2_INVALID_SLOT);
2347         if (!inode) {
2348                 status = -EINVAL;
2349                 mlog_errno(status);
2350                 goto bail;
2351         }
2352
2353         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2354         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2355         iput(inode);
2356
2357         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2358                                  osb->s_feature_incompat) * 8;
2359
2360         status = ocfs2_init_slot_info(osb);
2361         if (status < 0) {
2362                 mlog_errno(status);
2363                 goto bail;
2364         }
2365         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2366
2367 bail:
2368         return status;
2369 }
2370
2371 /*
2372  * will return: -EAGAIN if it is ok to keep searching for superblocks
2373  *              -EINVAL if there is a bad superblock
2374  *              0 on success
2375  */
2376 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2377                                struct buffer_head *bh,
2378                                u32 blksz,
2379                                struct ocfs2_blockcheck_stats *stats)
2380 {
2381         int status = -EAGAIN;
2382
2383         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2384                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2385                 /* We have to do a raw check of the feature here */
2386                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2387                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2388                         status = ocfs2_block_check_validate(bh->b_data,
2389                                                             bh->b_size,
2390                                                             &di->i_check,
2391                                                             stats);
2392                         if (status)
2393                                 goto out;
2394                 }
2395                 status = -EINVAL;
2396                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2397                         mlog(ML_ERROR, "found superblock with incorrect block "
2398                              "size: found %u, should be %u\n",
2399                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2400                                blksz);
2401                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2402                            OCFS2_MAJOR_REV_LEVEL ||
2403                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2404                            OCFS2_MINOR_REV_LEVEL) {
2405                         mlog(ML_ERROR, "found superblock with bad version: "
2406                              "found %u.%u, should be %u.%u\n",
2407                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2408                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2409                              OCFS2_MAJOR_REV_LEVEL,
2410                              OCFS2_MINOR_REV_LEVEL);
2411                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2412                         mlog(ML_ERROR, "bad block number on superblock: "
2413                              "found %llu, should be %llu\n",
2414                              (unsigned long long)le64_to_cpu(di->i_blkno),
2415                              (unsigned long long)bh->b_blocknr);
2416                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2417                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2418                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2419                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2420                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2421                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2422                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2423                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2424                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2425                         mlog(ML_ERROR,
2426                              "Superblock slots found greater than file system "
2427                              "maximum: found %u, max %u\n",
2428                              le16_to_cpu(di->id2.i_super.s_max_slots),
2429                              OCFS2_MAX_SLOTS);
2430                 } else {
2431                         /* found it! */
2432                         status = 0;
2433                 }
2434         }
2435
2436 out:
2437         if (status && status != -EAGAIN)
2438                 mlog_errno(status);
2439         return status;
2440 }
2441
2442 static int ocfs2_check_volume(struct ocfs2_super *osb)
2443 {
2444         int status;
2445         int dirty;
2446         int local;
2447         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2448                                                   * recover
2449                                                   * ourselves. */
2450
2451         /* Init our journal object. */
2452         status = ocfs2_journal_init(osb->journal, &dirty);
2453         if (status < 0) {
2454                 mlog(ML_ERROR, "Could not initialize journal!\n");
2455                 goto finally;
2456         }
2457
2458         /* Now that journal has been initialized, check to make sure
2459            entire volume is addressable. */
2460         status = ocfs2_journal_addressable(osb);
2461         if (status)
2462                 goto finally;
2463
2464         /* If the journal was unmounted cleanly then we don't want to
2465          * recover anything. Otherwise, journal_load will do that
2466          * dirty work for us :) */
2467         if (!dirty) {
2468                 status = ocfs2_journal_wipe(osb->journal, 0);
2469                 if (status < 0) {
2470                         mlog_errno(status);
2471                         goto finally;
2472                 }
2473         } else {
2474                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2475                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2476         }
2477
2478         local = ocfs2_mount_local(osb);
2479
2480         /* will play back anything left in the journal. */
2481         status = ocfs2_journal_load(osb->journal, local, dirty);
2482         if (status < 0) {
2483                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2484                 goto finally;
2485         }
2486
2487         if (dirty) {
2488                 /* recover my local alloc if we didn't unmount cleanly. */
2489                 status = ocfs2_begin_local_alloc_recovery(osb,
2490                                                           osb->slot_num,
2491                                                           &local_alloc);
2492                 if (status < 0) {
2493                         mlog_errno(status);
2494                         goto finally;
2495                 }
2496                 /* we complete the recovery process after we've marked
2497                  * ourselves as mounted. */
2498         }
2499
2500         status = ocfs2_load_local_alloc(osb);
2501         if (status < 0) {
2502                 mlog_errno(status);
2503                 goto finally;
2504         }
2505
2506         if (dirty) {
2507                 /* Recovery will be completed after we've mounted the
2508                  * rest of the volume. */
2509                 osb->dirty = 1;
2510                 osb->local_alloc_copy = local_alloc;
2511                 local_alloc = NULL;
2512         }
2513
2514         /* go through each journal, trylock it and if you get the
2515          * lock, and it's marked as dirty, set the bit in the recover
2516          * map and launch a recovery thread for it. */
2517         status = ocfs2_mark_dead_nodes(osb);
2518         if (status < 0) {
2519                 mlog_errno(status);
2520                 goto finally;
2521         }
2522
2523         status = ocfs2_compute_replay_slots(osb);
2524         if (status < 0)
2525                 mlog_errno(status);
2526
2527 finally:
2528         if (local_alloc)
2529                 kfree(local_alloc);
2530
2531         if (status)
2532                 mlog_errno(status);
2533         return status;
2534 }
2535
2536 /*
2537  * The routine gets called from dismount or close whenever a dismount on
2538  * volume is requested and the osb open count becomes 1.
2539  * It will remove the osb from the global list and also free up all the
2540  * initialized resources and fileobject.
2541  */
2542 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2543 {
2544         /* This function assumes that the caller has the main osb resource */
2545
2546         ocfs2_free_slot_info(osb);
2547
2548         kfree(osb->osb_orphan_wipes);
2549         kfree(osb->slot_recovery_generations);
2550         /* FIXME
2551          * This belongs in journal shutdown, but because we have to
2552          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2553          * we free it here.
2554          */
2555         kfree(osb->journal);
2556         if (osb->local_alloc_copy)
2557                 kfree(osb->local_alloc_copy);
2558         kfree(osb->uuid_str);
2559         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2560         memset(osb, 0, sizeof(struct ocfs2_super));
2561 }
2562
2563 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2564  * panic(). We do not support continue-on-error operation. */
2565 static void ocfs2_handle_error(struct super_block *sb)
2566 {
2567         struct ocfs2_super *osb = OCFS2_SB(sb);
2568
2569         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2570                 panic("OCFS2: (device %s): panic forced after error\n",
2571                       sb->s_id);
2572
2573         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2574
2575         if (sb->s_flags & MS_RDONLY &&
2576             (ocfs2_is_soft_readonly(osb) ||
2577              ocfs2_is_hard_readonly(osb)))
2578                 return;
2579
2580         printk(KERN_CRIT "File system is now read-only due to the potential "
2581                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2582                "system is unmounted.\n");
2583         sb->s_flags |= MS_RDONLY;
2584         ocfs2_set_ro_flag(osb, 0);
2585 }
2586
2587 static char error_buf[1024];
2588
2589 void __ocfs2_error(struct super_block *sb,
2590                    const char *function,
2591                    const char *fmt, ...)
2592 {
2593         va_list args;
2594
2595         va_start(args, fmt);
2596         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2597         va_end(args);
2598
2599         /* Not using mlog here because we want to show the actual
2600          * function the error came from. */
2601         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2602                sb->s_id, function, error_buf);
2603
2604         ocfs2_handle_error(sb);
2605 }
2606
2607 /* Handle critical errors. This is intentionally more drastic than
2608  * ocfs2_handle_error, so we only use for things like journal errors,
2609  * etc. */
2610 void __ocfs2_abort(struct super_block* sb,
2611                    const char *function,
2612                    const char *fmt, ...)
2613 {
2614         va_list args;
2615
2616         va_start(args, fmt);
2617         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2618         va_end(args);
2619
2620         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2621                sb->s_id, function, error_buf);
2622
2623         /* We don't have the cluster support yet to go straight to
2624          * hard readonly in here. Until then, we want to keep
2625          * ocfs2_abort() so that we can at least mark critical
2626          * errors.
2627          *
2628          * TODO: This should abort the journal and alert other nodes
2629          * that our slot needs recovery. */
2630
2631         /* Force a panic(). This stinks, but it's better than letting
2632          * things continue without having a proper hard readonly
2633          * here. */
2634         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2635                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2636         ocfs2_handle_error(sb);
2637 }
2638
2639 /*
2640  * Void signal blockers, because in-kernel sigprocmask() only fails
2641  * when SIG_* is wrong.
2642  */
2643 void ocfs2_block_signals(sigset_t *oldset)
2644 {
2645         int rc;
2646         sigset_t blocked;
2647
2648         sigfillset(&blocked);
2649         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2650         BUG_ON(rc);
2651 }
2652
2653 void ocfs2_unblock_signals(sigset_t *oldset)
2654 {
2655         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2656         BUG_ON(rc);
2657 }
2658
2659 module_init(ocfs2_init);
2660 module_exit(ocfs2_exit);