fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77
  78 static struct kmem_cache *ocfs2_inode_cachep;
  79 struct kmem_cache *ocfs2_dquot_cachep;
  80 struct kmem_cache *ocfs2_qf_chunk_cachep;
  81
  82 /* OCFS2 needs to schedule several different types of work which
  83  * require cluster locking, disk I/O, recovery waits, etc. Since these
  84  * types of work tend to be heavy we avoid using the kernel events
  85  * workqueue and schedule on our own. */
  86 struct workqueue_struct *ocfs2_wq = NULL;
  87
  88 static struct dentry *ocfs2_debugfs_root;
  89
  90 MODULE_AUTHOR("Oracle");
  91 MODULE_LICENSE("GPL");
  92 MODULE_DESCRIPTION("OCFS2 cluster file system");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 147 {
 148         return OCFS2_I(inode)->i_dquot;
 149 }
 150
 151 static const struct super_operations ocfs2_sops = {
 152         .statfs         = ocfs2_statfs,
 153         .alloc_inode    = ocfs2_alloc_inode,
 154         .destroy_inode  = ocfs2_destroy_inode,
 155         .drop_inode     = ocfs2_drop_inode,
 156         .evict_inode    = ocfs2_evict_inode,
 157         .sync_fs        = ocfs2_sync_fs,
 158         .put_super      = ocfs2_put_super,
 159         .remount_fs     = ocfs2_remount,
 160         .show_options   = ocfs2_show_options,
 161         .quota_read     = ocfs2_quota_read,
 162         .quota_write    = ocfs2_quota_write,
 163         .get_dquots     = ocfs2_get_dquots,
 164 };
 165
 166 enum {
 167         Opt_barrier,
 168         Opt_err_panic,
 169         Opt_err_ro,
 170         Opt_intr,
 171         Opt_nointr,
 172         Opt_hb_none,
 173         Opt_hb_local,
 174         Opt_hb_global,
 175         Opt_data_ordered,
 176         Opt_data_writeback,
 177         Opt_atime_quantum,
 178         Opt_slot,
 179         Opt_commit,
 180         Opt_localalloc,
 181         Opt_localflocks,
 182         Opt_stack,
 183         Opt_user_xattr,
 184         Opt_nouser_xattr,
 185         Opt_inode64,
 186         Opt_acl,
 187         Opt_noacl,
 188         Opt_usrquota,
 189         Opt_grpquota,
 190         Opt_coherency_buffered,
 191         Opt_coherency_full,
 192         Opt_resv_level,
 193         Opt_dir_resv_level,
 194         Opt_journal_async_commit,
 195         Opt_err_cont,
 196         Opt_err,
 197 };
 198
 199 static const match_table_t tokens = {
 200         {Opt_barrier, "barrier=%u"},
 201         {Opt_err_panic, "errors=panic"},
 202         {Opt_err_ro, "errors=remount-ro"},
 203         {Opt_intr, "intr"},
 204         {Opt_nointr, "nointr"},
 205         {Opt_hb_none, OCFS2_HB_NONE},
 206         {Opt_hb_local, OCFS2_HB_LOCAL},
 207         {Opt_hb_global, OCFS2_HB_GLOBAL},
 208         {Opt_data_ordered, "data=ordered"},
 209         {Opt_data_writeback, "data=writeback"},
 210         {Opt_atime_quantum, "atime_quantum=%u"},
 211         {Opt_slot, "preferred_slot=%u"},
 212         {Opt_commit, "commit=%u"},
 213         {Opt_localalloc, "localalloc=%d"},
 214         {Opt_localflocks, "localflocks"},
 215         {Opt_stack, "cluster_stack=%s"},
 216         {Opt_user_xattr, "user_xattr"},
 217         {Opt_nouser_xattr, "nouser_xattr"},
 218         {Opt_inode64, "inode64"},
 219         {Opt_acl, "acl"},
 220         {Opt_noacl, "noacl"},
 221         {Opt_usrquota, "usrquota"},
 222         {Opt_grpquota, "grpquota"},
 223         {Opt_coherency_buffered, "coherency=buffered"},
 224         {Opt_coherency_full, "coherency=full"},
 225         {Opt_resv_level, "resv_level=%u"},
 226         {Opt_dir_resv_level, "dir_resv_level=%u"},
 227         {Opt_journal_async_commit, "journal_async_commit"},
 228         {Opt_err_cont, "errors=continue"},
 229         {Opt_err, NULL}
 230 };
 231
 232 #ifdef CONFIG_DEBUG_FS
 233 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 234 {
 235         struct ocfs2_cluster_connection *cconn = osb->cconn;
 236         struct ocfs2_recovery_map *rm = osb->recovery_map;
 237         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 238         int i, out = 0;
 239         unsigned long flags;
 240
 241         out += snprintf(buf + out, len - out,
 242                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 243                         "Device", osb->dev_str, osb->uuid_str,
 244                         osb->fs_generation, osb->vol_label);
 245
 246         out += snprintf(buf + out, len - out,
 247                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 248                         atomic_read(&osb->vol_state), osb->osb_flags);
 249
 250         out += snprintf(buf + out, len - out,
 251                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 252                         osb->sb->s_blocksize, osb->s_clustersize);
 253
 254         out += snprintf(buf + out, len - out,
 255                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 256                         "ROcompat: 0x%X\n",
 257                         "Features", osb->s_feature_compat,
 258                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 259
 260         out += snprintf(buf + out, len - out,
 261                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 262                         osb->s_mount_opt, osb->s_atime_quantum);
 263
 264         if (cconn) {
 265                 out += snprintf(buf + out, len - out,
 266                                 "%10s => Stack: %s  Name: %*s  "
 267                                 "Version: %d.%d\n", "Cluster",
 268                                 (*osb->osb_cluster_stack == '\0' ?
 269                                  "o2cb" : osb->osb_cluster_stack),
 270                                 cconn->cc_namelen, cconn->cc_name,
 271                                 cconn->cc_version.pv_major,
 272                                 cconn->cc_version.pv_minor);
 273         }
 274
 275         spin_lock_irqsave(&osb->dc_task_lock, flags);
 276         out += snprintf(buf + out, len - out,
 277                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 278                         "WorkSeq: %lu\n", "DownCnvt",
 279                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 280                         osb->blocked_lock_count, osb->dc_wake_sequence,
 281                         osb->dc_work_sequence);
 282         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 283
 284         spin_lock(&osb->osb_lock);
 285         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 286                         "Recovery",
 287                         (osb->recovery_thread_task ?
 288                          task_pid_nr(osb->recovery_thread_task) : -1));
 289         if (rm->rm_used == 0)
 290                 out += snprintf(buf + out, len - out, " None\n");
 291         else {
 292                 for (i = 0; i < rm->rm_used; i++)
 293                         out += snprintf(buf + out, len - out, " %d",
 294                                         rm->rm_entries[i]);
 295                 out += snprintf(buf + out, len - out, "\n");
 296         }
 297         spin_unlock(&osb->osb_lock);
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 301                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 302                         osb->osb_commit_interval);
 303
 304         out += snprintf(buf + out, len - out,
 305                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 306                         "Journal", osb->journal->j_state,
 307                         osb->journal->j_trans_id,
 308                         atomic_read(&osb->journal->j_num_trans));
 309
 310         out += snprintf(buf + out, len - out,
 311                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 312                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 313                         "Stats",
 314                         atomic_read(&osb->alloc_stats.bitmap_data),
 315                         atomic_read(&osb->alloc_stats.local_data),
 316                         atomic_read(&osb->alloc_stats.bg_allocs),
 317                         atomic_read(&osb->alloc_stats.moves),
 318                         atomic_read(&osb->alloc_stats.bg_extends));
 319
 320         out += snprintf(buf + out, len - out,
 321                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 322                         "Default: %u bits\n",
 323                         "LocalAlloc", osb->local_alloc_state,
 324                         (unsigned long long)osb->la_last_gd,
 325                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 326
 327         spin_lock(&osb->osb_lock);
 328         out += snprintf(buf + out, len - out,
 329                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 330                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 331                         osb->s_inode_steal_slot,
 332                         atomic_read(&osb->s_num_inodes_stolen),
 333                         osb->s_meta_steal_slot,
 334                         atomic_read(&osb->s_num_meta_stolen));
 335         spin_unlock(&osb->osb_lock);
 336
 337         out += snprintf(buf + out, len - out, "OrphanScan => ");
 338         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 339                         os->os_count, os->os_seqno);
 340         out += snprintf(buf + out, len - out, " Last Scan: ");
 341         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 342                 out += snprintf(buf + out, len - out, "Disabled\n");
 343         else
 344                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 345                                 (get_seconds() - os->os_scantime.tv_sec));
 346
 347         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 348                         "Slots", "Num", "RecoGen");
 349         for (i = 0; i < osb->max_slots; ++i) {
 350                 out += snprintf(buf + out, len - out,
 351                                 "%10s  %c %3d  %10d\n",
 352                                 " ",
 353                                 (i == osb->slot_num ? '*' : ' '),
 354                                 i, osb->slot_recovery_generations[i]);
 355         }
 356
 357         return out;
 358 }
 359
 360 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 361 {
 362         struct ocfs2_super *osb = inode->i_private;
 363         char *buf = NULL;
 364
 365         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 366         if (!buf)
 367                 goto bail;
 368
 369         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 370
 371         file->private_data = buf;
 372
 373         return 0;
 374 bail:
 375         return -ENOMEM;
 376 }
 377
 378 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 379 {
 380         kfree(file->private_data);
 381         return 0;
 382 }
 383
 384 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 385                                 size_t nbytes, loff_t *ppos)
 386 {
 387         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 388                                        i_size_read(file->f_mapping->host));
 389 }
 390 #else
 391 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 392 {
 393         return 0;
 394 }
 395 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 396 {
 397         return 0;
 398 }
 399 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 400                                 size_t nbytes, loff_t *ppos)
 401 {
 402         return 0;
 403 }
 404 #endif  /* CONFIG_DEBUG_FS */
 405
 406 static const struct file_operations ocfs2_osb_debug_fops = {
 407         .open =         ocfs2_osb_debug_open,
 408         .release =      ocfs2_debug_release,
 409         .read =         ocfs2_debug_read,
 410         .llseek =       generic_file_llseek,
 411 };
 412
 413 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 414 {
 415         int status;
 416         tid_t target;
 417         struct ocfs2_super *osb = OCFS2_SB(sb);
 418
 419         if (ocfs2_is_hard_readonly(osb))
 420                 return -EROFS;
 421
 422         if (wait) {
 423                 status = ocfs2_flush_truncate_log(osb);
 424                 if (status < 0)
 425                         mlog_errno(status);
 426         } else {
 427                 ocfs2_schedule_truncate_log_flush(osb, 0);
 428         }
 429
 430         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 431                                       &target)) {
 432                 if (wait)
 433                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 434                                              target);
 435         }
 436         return 0;
 437 }
 438
 439 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 440 {
 441         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 442             && (ino == USER_QUOTA_SYSTEM_INODE
 443                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 444                 return 0;
 445         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 446             && (ino == GROUP_QUOTA_SYSTEM_INODE
 447                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 448                 return 0;
 449         return 1;
 450 }
 451
 452 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 453 {
 454         struct inode *new = NULL;
 455         int status = 0;
 456         int i;
 457
 458         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 459         if (IS_ERR(new)) {
 460                 status = PTR_ERR(new);
 461                 mlog_errno(status);
 462                 goto bail;
 463         }
 464         osb->root_inode = new;
 465
 466         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 467         if (IS_ERR(new)) {
 468                 status = PTR_ERR(new);
 469                 mlog_errno(status);
 470                 goto bail;
 471         }
 472         osb->sys_root_inode = new;
 473
 474         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 475              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 476                 if (!ocfs2_need_system_inode(osb, i))
 477                         continue;
 478                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 479                 if (!new) {
 480                         ocfs2_release_system_inodes(osb);
 481                         status = -EINVAL;
 482                         mlog_errno(status);
 483                         /* FIXME: Should ERROR_RO_FS */
 484                         mlog(ML_ERROR, "Unable to load system inode %d, "
 485                              "possibly corrupt fs?", i);
 486                         goto bail;
 487                 }
 488                 // the array now has one ref, so drop this one
 489                 iput(new);
 490         }
 491
 492 bail:
 493         if (status)
 494                 mlog_errno(status);
 495         return status;
 496 }
 497
 498 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 499 {
 500         struct inode *new = NULL;
 501         int status = 0;
 502         int i;
 503
 504         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 505              i < NUM_SYSTEM_INODES;
 506              i++) {
 507                 if (!ocfs2_need_system_inode(osb, i))
 508                         continue;
 509                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 510                 if (!new) {
 511                         ocfs2_release_system_inodes(osb);
 512                         status = -EINVAL;
 513                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 514                              status, i, osb->slot_num);
 515                         goto bail;
 516                 }
 517                 /* the array now has one ref, so drop this one */
 518                 iput(new);
 519         }
 520
 521 bail:
 522         if (status)
 523                 mlog_errno(status);
 524         return status;
 525 }
 526
 527 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 528 {
 529         int i;
 530         struct inode *inode;
 531
 532         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 533                 inode = osb->global_system_inodes[i];
 534                 if (inode) {
 535                         iput(inode);
 536                         osb->global_system_inodes[i] = NULL;
 537                 }
 538         }
 539
 540         inode = osb->sys_root_inode;
 541         if (inode) {
 542                 iput(inode);
 543                 osb->sys_root_inode = NULL;
 544         }
 545
 546         inode = osb->root_inode;
 547         if (inode) {
 548                 iput(inode);
 549                 osb->root_inode = NULL;
 550         }
 551
 552         if (!osb->local_system_inodes)
 553                 return;
 554
 555         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 556                 if (osb->local_system_inodes[i]) {
 557                         iput(osb->local_system_inodes[i]);
 558                         osb->local_system_inodes[i] = NULL;
 559                 }
 560         }
 561
 562         kfree(osb->local_system_inodes);
 563         osb->local_system_inodes = NULL;
 564 }
 565
 566 /* We're allocating fs objects, use GFP_NOFS */
 567 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 568 {
 569         struct ocfs2_inode_info *oi;
 570
 571         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 572         if (!oi)
 573                 return NULL;
 574
 575         oi->i_sync_tid = 0;
 576         oi->i_datasync_tid = 0;
 577         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 578
 579         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 580         return &oi->vfs_inode;
 581 }
 582
 583 static void ocfs2_i_callback(struct rcu_head *head)
 584 {
 585         struct inode *inode = container_of(head, struct inode, i_rcu);
 586         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 587 }
 588
 589 static void ocfs2_destroy_inode(struct inode *inode)
 590 {
 591         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 592 }
 593
 594 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 595                                                 unsigned int cbits)
 596 {
 597         unsigned int bytes = 1 << cbits;
 598         unsigned int trim = bytes;
 599         unsigned int bitshift = 32;
 600
 601         /*
 602          * i_size and all block offsets in ocfs2 are always 64 bits
 603          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 604          * 64 bit platforms, cluster size will be the limiting factor.
 605          */
 606
 607 #if BITS_PER_LONG == 32
 608 # if defined(CONFIG_LBDAF)
 609         BUILD_BUG_ON(sizeof(sector_t) != 8);
 610         /*
 611          * We might be limited by page cache size.
 612          */
 613         if (bytes > PAGE_CACHE_SIZE) {
 614                 bytes = PAGE_CACHE_SIZE;
 615                 trim = 1;
 616                 /*
 617                  * Shift by 31 here so that we don't get larger than
 618                  * MAX_LFS_FILESIZE
 619                  */
 620                 bitshift = 31;
 621         }
 622 # else
 623         /*
 624          * We are limited by the size of sector_t. Use block size, as
 625          * that's what we expose to the VFS.
 626          */
 627         bytes = 1 << bbits;
 628         trim = 1;
 629         bitshift = 31;
 630 # endif
 631 #endif
 632
 633         /*
 634          * Trim by a whole cluster when we can actually approach the
 635          * on-disk limits. Otherwise we can overflow i_clusters when
 636          * an extent start is at the max offset.
 637          */
 638         return (((unsigned long long)bytes) << bitshift) - trim;
 639 }
 640
 641 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 642 {
 643         int incompat_features;
 644         int ret = 0;
 645         struct mount_options parsed_options;
 646         struct ocfs2_super *osb = OCFS2_SB(sb);
 647         u32 tmp;
 648
 649         sync_filesystem(sb);
 650
 651         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 652             !ocfs2_check_set_options(sb, &parsed_options)) {
 653                 ret = -EINVAL;
 654                 goto out;
 655         }
 656
 657         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 658                 OCFS2_MOUNT_HB_NONE;
 659         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 660                 ret = -EINVAL;
 661                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 662                 goto out;
 663         }
 664
 665         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 666             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 667                 ret = -EINVAL;
 668                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 669                 goto out;
 670         }
 671
 672         /* Probably don't want this on remount; it might
 673          * mess with other nodes */
 674         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 675             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 676                 ret = -EINVAL;
 677                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 678                 goto out;
 679         }
 680
 681         /* We're going to/from readonly mode. */
 682         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 683                 /* Disable quota accounting before remounting RO */
 684                 if (*flags & MS_RDONLY) {
 685                         ret = ocfs2_susp_quotas(osb, 0);
 686                         if (ret < 0)
 687                                 goto out;
 688                 }
 689                 /* Lock here so the check of HARD_RO and the potential
 690                  * setting of SOFT_RO is atomic. */
 691                 spin_lock(&osb->osb_lock);
 692                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 693                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 694                         ret = -EROFS;
 695                         goto unlock_osb;
 696                 }
 697
 698                 if (*flags & MS_RDONLY) {
 699                         sb->s_flags |= MS_RDONLY;
 700                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 701                 } else {
 702                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 703                                 mlog(ML_ERROR, "Cannot remount RDWR "
 704                                      "filesystem due to previous errors.\n");
 705                                 ret = -EROFS;
 706                                 goto unlock_osb;
 707                         }
 708                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 709                         if (incompat_features) {
 710                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 711                                      "of unsupported optional features "
 712                                      "(%x).\n", incompat_features);
 713                                 ret = -EINVAL;
 714                                 goto unlock_osb;
 715                         }
 716                         sb->s_flags &= ~MS_RDONLY;
 717                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 718                 }
 719                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 720 unlock_osb:
 721                 spin_unlock(&osb->osb_lock);
 722                 /* Enable quota accounting after remounting RW */
 723                 if (!ret && !(*flags & MS_RDONLY)) {
 724                         if (sb_any_quota_suspended(sb))
 725                                 ret = ocfs2_susp_quotas(osb, 1);
 726                         else
 727                                 ret = ocfs2_enable_quotas(osb);
 728                         if (ret < 0) {
 729                                 /* Return back changes... */
 730                                 spin_lock(&osb->osb_lock);
 731                                 sb->s_flags |= MS_RDONLY;
 732                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 733                                 spin_unlock(&osb->osb_lock);
 734                                 goto out;
 735                         }
 736                 }
 737         }
 738
 739         if (!ret) {
 740                 /* Only save off the new mount options in case of a successful
 741                  * remount. */
 742                 osb->s_mount_opt = parsed_options.mount_opt;
 743                 osb->s_atime_quantum = parsed_options.atime_quantum;
 744                 osb->preferred_slot = parsed_options.slot;
 745                 if (parsed_options.commit_interval)
 746                         osb->osb_commit_interval = parsed_options.commit_interval;
 747
 748                 if (!ocfs2_is_hard_readonly(osb))
 749                         ocfs2_set_journal_params(osb);
 750
 751                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 752                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 753                                                         MS_POSIXACL : 0);
 754         }
 755 out:
 756         return ret;
 757 }
 758
 759 static int ocfs2_sb_probe(struct super_block *sb,
 760                           struct buffer_head **bh,
 761                           int *sector_size,
 762                           struct ocfs2_blockcheck_stats *stats)
 763 {
 764         int status, tmpstat;
 765         struct ocfs1_vol_disk_hdr *hdr;
 766         struct ocfs2_dinode *di;
 767         int blksize;
 768
 769         *bh = NULL;
 770
 771         /* may be > 512 */
 772         *sector_size = bdev_logical_block_size(sb->s_bdev);
 773         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 774                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 775                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 776                 status = -EINVAL;
 777                 goto bail;
 778         }
 779
 780         /* Can this really happen? */
 781         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 782                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 783
 784         /* check block zero for old format */
 785         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 786         if (status < 0) {
 787                 mlog_errno(status);
 788                 goto bail;
 789         }
 790         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 791         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 792                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 793                      hdr->major_version, hdr->minor_version);
 794                 status = -EINVAL;
 795         }
 796         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 797                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 798                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 799                      hdr->signature);
 800                 status = -EINVAL;
 801         }
 802         brelse(*bh);
 803         *bh = NULL;
 804         if (status < 0) {
 805                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 806                      "upgraded before mounting with ocfs v2\n");
 807                 goto bail;
 808         }
 809
 810         /*
 811          * Now check at magic offset for 512, 1024, 2048, 4096
 812          * blocksizes.  4096 is the maximum blocksize because it is
 813          * the minimum clustersize.
 814          */
 815         status = -EINVAL;
 816         for (blksize = *sector_size;
 817              blksize <= OCFS2_MAX_BLOCKSIZE;
 818              blksize <<= 1) {
 819                 tmpstat = ocfs2_get_sector(sb, bh,
 820                                            OCFS2_SUPER_BLOCK_BLKNO,
 821                                            blksize);
 822                 if (tmpstat < 0) {
 823                         status = tmpstat;
 824                         mlog_errno(status);
 825                         break;
 826                 }
 827                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 828                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 829                 spin_lock_init(&stats->b_lock);
 830                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 831                 if (tmpstat < 0) {
 832                         brelse(*bh);
 833                         *bh = NULL;
 834                 }
 835                 if (tmpstat != -EAGAIN) {
 836                         status = tmpstat;
 837                         break;
 838                 }
 839         }
 840
 841 bail:
 842         return status;
 843 }
 844
 845 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 846 {
 847         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 848
 849         if (osb->s_mount_opt & hb_enabled) {
 850                 if (ocfs2_mount_local(osb)) {
 851                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 852                              "mounted device.\n");
 853                         return -EINVAL;
 854                 }
 855                 if (ocfs2_userspace_stack(osb)) {
 856                         mlog(ML_ERROR, "Userspace stack expected, but "
 857                              "o2cb heartbeat arguments passed to mount\n");
 858                         return -EINVAL;
 859                 }
 860                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 861                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 862                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 863                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 864                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 865                         return -EINVAL;
 866                 }
 867         }
 868
 869         if (!(osb->s_mount_opt & hb_enabled)) {
 870                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 871                     !ocfs2_userspace_stack(osb)) {
 872                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 873                              "a read-write clustered device.\n");
 874                         return -EINVAL;
 875                 }
 876         }
 877
 878         return 0;
 879 }
 880
 881 /*
 882  * If we're using a userspace stack, mount should have passed
 883  * a name that matches the disk.  If not, mount should not
 884  * have passed a stack.
 885  */
 886 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 887                                         struct mount_options *mopt)
 888 {
 889         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 890                 mlog(ML_ERROR,
 891                      "cluster stack passed to mount, but this filesystem "
 892                      "does not support it\n");
 893                 return -EINVAL;
 894         }
 895
 896         if (ocfs2_userspace_stack(osb) &&
 897             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 898                     OCFS2_STACK_LABEL_LEN)) {
 899                 mlog(ML_ERROR,
 900                      "cluster stack passed to mount (\"%s\") does not "
 901                      "match the filesystem (\"%s\")\n",
 902                      mopt->cluster_stack,
 903                      osb->osb_cluster_stack);
 904                 return -EINVAL;
 905         }
 906
 907         return 0;
 908 }
 909
 910 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 911 {
 912         int type;
 913         struct super_block *sb = osb->sb;
 914         unsigned int feature[OCFS2_MAXQUOTAS] = {
 915                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 916                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 917         int status = 0;
 918
 919         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 920                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 921                         continue;
 922                 if (unsuspend)
 923                         status = dquot_resume(sb, type);
 924                 else {
 925                         struct ocfs2_mem_dqinfo *oinfo;
 926
 927                         /* Cancel periodic syncing before suspending */
 928                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 929                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 930                         status = dquot_suspend(sb, type);
 931                 }
 932                 if (status < 0)
 933                         break;
 934         }
 935         if (status < 0)
 936                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 937                      "remount (error = %d).\n", status);
 938         return status;
 939 }
 940
 941 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 942 {
 943         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 944         struct super_block *sb = osb->sb;
 945         unsigned int feature[OCFS2_MAXQUOTAS] = {
 946                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 947                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 948         unsigned int ino[OCFS2_MAXQUOTAS] = {
 949                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 950                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 951         int status;
 952         int type;
 953
 954         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 955         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 956                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 957                         continue;
 958                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 959                                                         osb->slot_num);
 960                 if (!inode[type]) {
 961                         status = -ENOENT;
 962                         goto out_quota_off;
 963                 }
 964                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 965                                       DQUOT_USAGE_ENABLED);
 966                 if (status < 0)
 967                         goto out_quota_off;
 968         }
 969
 970         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 971                 iput(inode[type]);
 972         return 0;
 973 out_quota_off:
 974         ocfs2_disable_quotas(osb);
 975         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 976                 iput(inode[type]);
 977         mlog_errno(status);
 978         return status;
 979 }
 980
 981 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 982 {
 983         int type;
 984         struct inode *inode;
 985         struct super_block *sb = osb->sb;
 986         struct ocfs2_mem_dqinfo *oinfo;
 987
 988         /* We mostly ignore errors in this function because there's not much
 989          * we can do when we see them */
 990         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 991                 if (!sb_has_quota_loaded(sb, type))
 992                         continue;
 993                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 994                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 995                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 996                 inode = igrab(sb->s_dquot.files[type]);
 997                 /* Turn off quotas. This will remove all dquot structures from
 998                  * memory and so they will be automatically synced to global
 999                  * quota files */
1000                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
1001                                         DQUOT_LIMITS_ENABLED);
1002                 if (!inode)
1003                         continue;
1004                 iput(inode);
1005         }
1006 }
1007
1008 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1009 {
1010         struct dentry *root;
1011         int status, sector_size;
1012         struct mount_options parsed_options;
1013         struct inode *inode = NULL;
1014         struct ocfs2_super *osb = NULL;
1015         struct buffer_head *bh = NULL;
1016         char nodestr[12];
1017         struct ocfs2_blockcheck_stats stats;
1018
1019         trace_ocfs2_fill_super(sb, data, silent);
1020
1021         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1022                 status = -EINVAL;
1023                 goto read_super_error;
1024         }
1025
1026         /* probe for superblock */
1027         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1028         if (status < 0) {
1029                 mlog(ML_ERROR, "superblock probe failed!\n");
1030                 goto read_super_error;
1031         }
1032
1033         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1034         osb = OCFS2_SB(sb);
1035         if (status < 0) {
1036                 mlog_errno(status);
1037                 goto read_super_error;
1038         }
1039         brelse(bh);
1040         bh = NULL;
1041
1042         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1043                 status = -EINVAL;
1044                 goto read_super_error;
1045         }
1046         osb->s_mount_opt = parsed_options.mount_opt;
1047         osb->s_atime_quantum = parsed_options.atime_quantum;
1048         osb->preferred_slot = parsed_options.slot;
1049         osb->osb_commit_interval = parsed_options.commit_interval;
1050
1051         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1052         osb->osb_resv_level = parsed_options.resv_level;
1053         osb->osb_dir_resv_level = parsed_options.resv_level;
1054         if (parsed_options.dir_resv_level == -1)
1055                 osb->osb_dir_resv_level = parsed_options.resv_level;
1056         else
1057                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1058
1059         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1060         if (status)
1061                 goto read_super_error;
1062
1063         sb->s_magic = OCFS2_SUPER_MAGIC;
1064
1065         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1066                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1067
1068         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1069          * heartbeat=none */
1070         if (bdev_read_only(sb->s_bdev)) {
1071                 if (!(sb->s_flags & MS_RDONLY)) {
1072                         status = -EACCES;
1073                         mlog(ML_ERROR, "Readonly device detected but readonly "
1074                              "mount was not specified.\n");
1075                         goto read_super_error;
1076                 }
1077
1078                 /* You should not be able to start a local heartbeat
1079                  * on a readonly device. */
1080                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1081                         status = -EROFS;
1082                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1083                              "device.\n");
1084                         goto read_super_error;
1085                 }
1086
1087                 status = ocfs2_check_journals_nolocks(osb);
1088                 if (status < 0) {
1089                         if (status == -EROFS)
1090                                 mlog(ML_ERROR, "Recovery required on readonly "
1091                                      "file system, but write access is "
1092                                      "unavailable.\n");
1093                         else
1094                                 mlog_errno(status);
1095                         goto read_super_error;
1096                 }
1097
1098                 ocfs2_set_ro_flag(osb, 1);
1099
1100                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1101                        "Cluster services will not be used for this mount. "
1102                        "Recovery will be skipped.\n", osb->dev_str);
1103         }
1104
1105         if (!ocfs2_is_hard_readonly(osb)) {
1106                 if (sb->s_flags & MS_RDONLY)
1107                         ocfs2_set_ro_flag(osb, 0);
1108         }
1109
1110         status = ocfs2_verify_heartbeat(osb);
1111         if (status < 0) {
1112                 mlog_errno(status);
1113                 goto read_super_error;
1114         }
1115
1116         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1117                                                  ocfs2_debugfs_root);
1118         if (!osb->osb_debug_root) {
1119                 status = -EINVAL;
1120                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1121                 goto read_super_error;
1122         }
1123
1124         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1125                                             osb->osb_debug_root,
1126                                             osb,
1127                                             &ocfs2_osb_debug_fops);
1128         if (!osb->osb_ctxt) {
1129                 status = -EINVAL;
1130                 mlog_errno(status);
1131                 goto read_super_error;
1132         }
1133
1134         if (ocfs2_meta_ecc(osb)) {
1135                 status = ocfs2_blockcheck_stats_debugfs_install(
1136                                                 &osb->osb_ecc_stats,
1137                                                 osb->osb_debug_root);
1138                 if (status) {
1139                         mlog(ML_ERROR,
1140                              "Unable to create blockcheck statistics "
1141                              "files\n");
1142                         goto read_super_error;
1143                 }
1144         }
1145
1146         status = ocfs2_mount_volume(sb);
1147         if (status < 0)
1148                 goto read_super_error;
1149
1150         if (osb->root_inode)
1151                 inode = igrab(osb->root_inode);
1152
1153         if (!inode) {
1154                 status = -EIO;
1155                 mlog_errno(status);
1156                 goto read_super_error;
1157         }
1158
1159         root = d_make_root(inode);
1160         if (!root) {
1161                 status = -ENOMEM;
1162                 mlog_errno(status);
1163                 goto read_super_error;
1164         }
1165
1166         sb->s_root = root;
1167
1168         ocfs2_complete_mount_recovery(osb);
1169
1170         if (ocfs2_mount_local(osb))
1171                 snprintf(nodestr, sizeof(nodestr), "local");
1172         else
1173                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1174
1175         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1176                "with %s data mode.\n",
1177                osb->dev_str, nodestr, osb->slot_num,
1178                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1179                "ordered");
1180
1181         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1182         wake_up(&osb->osb_mount_event);
1183
1184         /* Now we can initialize quotas because we can afford to wait
1185          * for cluster locks recovery now. That also means that truncation
1186          * log recovery can happen but that waits for proper quota setup */
1187         if (!(sb->s_flags & MS_RDONLY)) {
1188                 status = ocfs2_enable_quotas(osb);
1189                 if (status < 0) {
1190                         /* We have to err-out specially here because
1191                          * s_root is already set */
1192                         mlog_errno(status);
1193                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1194                         wake_up(&osb->osb_mount_event);
1195                         return status;
1196                 }
1197         }
1198
1199         ocfs2_complete_quota_recovery(osb);
1200
1201         /* Now we wake up again for processes waiting for quotas */
1202         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1203         wake_up(&osb->osb_mount_event);
1204
1205         /* Start this when the mount is almost sure of being successful */
1206         ocfs2_orphan_scan_start(osb);
1207
1208         return status;
1209
1210 read_super_error:
1211         brelse(bh);
1212
1213         if (osb) {
1214                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1215                 wake_up(&osb->osb_mount_event);
1216                 ocfs2_dismount_volume(sb, 1);
1217         }
1218
1219         if (status)
1220                 mlog_errno(status);
1221         return status;
1222 }
1223
1224 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1225                         int flags,
1226                         const char *dev_name,
1227                         void *data)
1228 {
1229         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1230 }
1231
1232 static struct file_system_type ocfs2_fs_type = {
1233         .owner          = THIS_MODULE,
1234         .name           = "ocfs2",
1235         .mount          = ocfs2_mount,
1236         .kill_sb        = kill_block_super,
1237         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1238         .next           = NULL
1239 };
1240 MODULE_ALIAS_FS("ocfs2");
1241
1242 static int ocfs2_check_set_options(struct super_block *sb,
1243                                    struct mount_options *options)
1244 {
1245         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1246             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1247                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1248                 mlog(ML_ERROR, "User quotas were requested, but this "
1249                      "filesystem does not have the feature enabled.\n");
1250                 return 0;
1251         }
1252         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1253             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1254                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1255                 mlog(ML_ERROR, "Group quotas were requested, but this "
1256                      "filesystem does not have the feature enabled.\n");
1257                 return 0;
1258         }
1259         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1260             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1261                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1262                      "feature is not enabled\n");
1263                 return 0;
1264         }
1265         /* No ACL setting specified? Use XATTR feature... */
1266         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1267                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1268                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1269                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1270                 else
1271                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1272         }
1273         return 1;
1274 }
1275
1276 static int ocfs2_parse_options(struct super_block *sb,
1277                                char *options,
1278                                struct mount_options *mopt,
1279                                int is_remount)
1280 {
1281         int status, user_stack = 0;
1282         char *p;
1283         u32 tmp;
1284         int token, option;
1285         substring_t args[MAX_OPT_ARGS];
1286
1287         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1288
1289         mopt->commit_interval = 0;
1290         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1291         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1292         mopt->slot = OCFS2_INVALID_SLOT;
1293         mopt->localalloc_opt = -1;
1294         mopt->cluster_stack[0] = '\0';
1295         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1296         mopt->dir_resv_level = -1;
1297
1298         if (!options) {
1299                 status = 1;
1300                 goto bail;
1301         }
1302
1303         while ((p = strsep(&options, ",")) != NULL) {
1304                 if (!*p)
1305                         continue;
1306
1307                 token = match_token(p, tokens, args);
1308                 switch (token) {
1309                 case Opt_hb_local:
1310                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1311                         break;
1312                 case Opt_hb_none:
1313                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1314                         break;
1315                 case Opt_hb_global:
1316                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1317                         break;
1318                 case Opt_barrier:
1319                         if (match_int(&args[0], &option)) {
1320                                 status = 0;
1321                                 goto bail;
1322                         }
1323                         if (option)
1324                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1325                         else
1326                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1327                         break;
1328                 case Opt_intr:
1329                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1330                         break;
1331                 case Opt_nointr:
1332                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1333                         break;
1334                 case Opt_err_panic:
1335                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1336                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1337                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1338                         break;
1339                 case Opt_err_ro:
1340                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1341                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1342                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1343                         break;
1344                 case Opt_err_cont:
1345                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1346                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1347                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1348                         break;
1349                 case Opt_data_ordered:
1350                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1351                         break;
1352                 case Opt_data_writeback:
1353                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1354                         break;
1355                 case Opt_user_xattr:
1356                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1357                         break;
1358                 case Opt_nouser_xattr:
1359                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1360                         break;
1361                 case Opt_atime_quantum:
1362                         if (match_int(&args[0], &option)) {
1363                                 status = 0;
1364                                 goto bail;
1365                         }
1366                         if (option >= 0)
1367                                 mopt->atime_quantum = option;
1368                         break;
1369                 case Opt_slot:
1370                         if (match_int(&args[0], &option)) {
1371                                 status = 0;
1372                                 goto bail;
1373                         }
1374                         if (option)
1375                                 mopt->slot = (s16)option;
1376                         break;
1377                 case Opt_commit:
1378                         if (match_int(&args[0], &option)) {
1379                                 status = 0;
1380                                 goto bail;
1381                         }
1382                         if (option < 0)
1383                                 return 0;
1384                         if (option == 0)
1385                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1386                         mopt->commit_interval = HZ * option;
1387                         break;
1388                 case Opt_localalloc:
1389                         if (match_int(&args[0], &option)) {
1390                                 status = 0;
1391                                 goto bail;
1392                         }
1393                         if (option >= 0)
1394                                 mopt->localalloc_opt = option;
1395                         break;
1396                 case Opt_localflocks:
1397                         /*
1398                          * Changing this during remount could race
1399                          * flock() requests, or "unbalance" existing
1400                          * ones (e.g., a lock is taken in one mode but
1401                          * dropped in the other). If users care enough
1402                          * to flip locking modes during remount, we
1403                          * could add a "local" flag to individual
1404                          * flock structures for proper tracking of
1405                          * state.
1406                          */
1407                         if (!is_remount)
1408                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1409                         break;
1410                 case Opt_stack:
1411                         /* Check both that the option we were passed
1412                          * is of the right length and that it is a proper
1413                          * string of the right length.
1414                          */
1415                         if (((args[0].to - args[0].from) !=
1416                              OCFS2_STACK_LABEL_LEN) ||
1417                             (strnlen(args[0].from,
1418                                      OCFS2_STACK_LABEL_LEN) !=
1419                              OCFS2_STACK_LABEL_LEN)) {
1420                                 mlog(ML_ERROR,
1421                                      "Invalid cluster_stack option\n");
1422                                 status = 0;
1423                                 goto bail;
1424                         }
1425                         memcpy(mopt->cluster_stack, args[0].from,
1426                                OCFS2_STACK_LABEL_LEN);
1427                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1428                         /*
1429                          * Open code the memcmp here as we don't have
1430                          * an osb to pass to
1431                          * ocfs2_userspace_stack().
1432                          */
1433                         if (memcmp(mopt->cluster_stack,
1434                                    OCFS2_CLASSIC_CLUSTER_STACK,
1435                                    OCFS2_STACK_LABEL_LEN))
1436                                 user_stack = 1;
1437                         break;
1438                 case Opt_inode64:
1439                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1440                         break;
1441                 case Opt_usrquota:
1442                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1443                         break;
1444                 case Opt_grpquota:
1445                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1446                         break;
1447                 case Opt_coherency_buffered:
1448                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1449                         break;
1450                 case Opt_coherency_full:
1451                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1452                         break;
1453                 case Opt_acl:
1454                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1455                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1456                         break;
1457                 case Opt_noacl:
1458                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1459                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1460                         break;
1461                 case Opt_resv_level:
1462                         if (is_remount)
1463                                 break;
1464                         if (match_int(&args[0], &option)) {
1465                                 status = 0;
1466                                 goto bail;
1467                         }
1468                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1469                             option < OCFS2_MAX_RESV_LEVEL)
1470                                 mopt->resv_level = option;
1471                         break;
1472                 case Opt_dir_resv_level:
1473                         if (is_remount)
1474                                 break;
1475                         if (match_int(&args[0], &option)) {
1476                                 status = 0;
1477                                 goto bail;
1478                         }
1479                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1480                             option < OCFS2_MAX_RESV_LEVEL)
1481                                 mopt->dir_resv_level = option;
1482                         break;
1483                 case Opt_journal_async_commit:
1484                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1485                         break;
1486                 default:
1487                         mlog(ML_ERROR,
1488                              "Unrecognized mount option \"%s\" "
1489                              "or missing value\n", p);
1490                         status = 0;
1491                         goto bail;
1492                 }
1493         }
1494
1495         if (user_stack == 0) {
1496                 /* Ensure only one heartbeat mode */
1497                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1498                                          OCFS2_MOUNT_HB_GLOBAL |
1499                                          OCFS2_MOUNT_HB_NONE);
1500                 if (hweight32(tmp) != 1) {
1501                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1502                         status = 0;
1503                         goto bail;
1504                 }
1505         }
1506
1507         status = 1;
1508
1509 bail:
1510         return status;
1511 }
1512
1513 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1514 {
1515         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1516         unsigned long opts = osb->s_mount_opt;
1517         unsigned int local_alloc_megs;
1518
1519         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1520                 seq_printf(s, ",_netdev");
1521                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1522                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1523                 else
1524                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1525         } else
1526                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1527
1528         if (opts & OCFS2_MOUNT_NOINTR)
1529                 seq_printf(s, ",nointr");
1530
1531         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1532                 seq_printf(s, ",data=writeback");
1533         else
1534                 seq_printf(s, ",data=ordered");
1535
1536         if (opts & OCFS2_MOUNT_BARRIER)
1537                 seq_printf(s, ",barrier=1");
1538
1539         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1540                 seq_printf(s, ",errors=panic");
1541         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1542                 seq_printf(s, ",errors=continue");
1543         else
1544                 seq_printf(s, ",errors=remount-ro");
1545
1546         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1547                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1548
1549         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1550
1551         if (osb->osb_commit_interval)
1552                 seq_printf(s, ",commit=%u",
1553                            (unsigned) (osb->osb_commit_interval / HZ));
1554
1555         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1556         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1557                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1558
1559         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1560                 seq_printf(s, ",localflocks,");
1561
1562         if (osb->osb_cluster_stack[0])
1563                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1564                                   OCFS2_STACK_LABEL_LEN);
1565         if (opts & OCFS2_MOUNT_USRQUOTA)
1566                 seq_printf(s, ",usrquota");
1567         if (opts & OCFS2_MOUNT_GRPQUOTA)
1568                 seq_printf(s, ",grpquota");
1569
1570         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1571                 seq_printf(s, ",coherency=buffered");
1572         else
1573                 seq_printf(s, ",coherency=full");
1574
1575         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1576                 seq_printf(s, ",nouser_xattr");
1577         else
1578                 seq_printf(s, ",user_xattr");
1579
1580         if (opts & OCFS2_MOUNT_INODE64)
1581                 seq_printf(s, ",inode64");
1582
1583         if (opts & OCFS2_MOUNT_POSIX_ACL)
1584                 seq_printf(s, ",acl");
1585         else
1586                 seq_printf(s, ",noacl");
1587
1588         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1589                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1590
1591         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1592                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1593
1594         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1595                 seq_printf(s, ",journal_async_commit");
1596
1597         return 0;
1598 }
1599
1600 static int __init ocfs2_init(void)
1601 {
1602         int status;
1603
1604         status = init_ocfs2_uptodate_cache();
1605         if (status < 0)
1606                 goto out1;
1607
1608         status = ocfs2_initialize_mem_caches();
1609         if (status < 0)
1610                 goto out2;
1611
1612         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1613         if (!ocfs2_wq) {
1614                 status = -ENOMEM;
1615                 goto out3;
1616         }
1617
1618         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1619         if (!ocfs2_debugfs_root) {
1620                 status = -ENOMEM;
1621                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1622                 goto out4;
1623         }
1624
1625         ocfs2_set_locking_protocol();
1626
1627         status = register_quota_format(&ocfs2_quota_format);
1628         if (status < 0)
1629                 goto out4;
1630         status = register_filesystem(&ocfs2_fs_type);
1631         if (!status)
1632                 return 0;
1633
1634         unregister_quota_format(&ocfs2_quota_format);
1635 out4:
1636         destroy_workqueue(ocfs2_wq);
1637         debugfs_remove(ocfs2_debugfs_root);
1638 out3:
1639         ocfs2_free_mem_caches();
1640 out2:
1641         exit_ocfs2_uptodate_cache();
1642 out1:
1643         mlog_errno(status);
1644         return status;
1645 }
1646
1647 static void __exit ocfs2_exit(void)
1648 {
1649         if (ocfs2_wq) {
1650                 flush_workqueue(ocfs2_wq);
1651                 destroy_workqueue(ocfs2_wq);
1652         }
1653
1654         unregister_quota_format(&ocfs2_quota_format);
1655
1656         debugfs_remove(ocfs2_debugfs_root);
1657
1658         ocfs2_free_mem_caches();
1659
1660         unregister_filesystem(&ocfs2_fs_type);
1661
1662         exit_ocfs2_uptodate_cache();
1663 }
1664
1665 static void ocfs2_put_super(struct super_block *sb)
1666 {
1667         trace_ocfs2_put_super(sb);
1668
1669         ocfs2_sync_blockdev(sb);
1670         ocfs2_dismount_volume(sb, 0);
1671 }
1672
1673 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1674 {
1675         struct ocfs2_super *osb;
1676         u32 numbits, freebits;
1677         int status;
1678         struct ocfs2_dinode *bm_lock;
1679         struct buffer_head *bh = NULL;
1680         struct inode *inode = NULL;
1681
1682         trace_ocfs2_statfs(dentry->d_sb, buf);
1683
1684         osb = OCFS2_SB(dentry->d_sb);
1685
1686         inode = ocfs2_get_system_file_inode(osb,
1687                                             GLOBAL_BITMAP_SYSTEM_INODE,
1688                                             OCFS2_INVALID_SLOT);
1689         if (!inode) {
1690                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1691                 status = -EIO;
1692                 goto bail;
1693         }
1694
1695         status = ocfs2_inode_lock(inode, &bh, 0);
1696         if (status < 0) {
1697                 mlog_errno(status);
1698                 goto bail;
1699         }
1700
1701         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1702
1703         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1704         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1705
1706         buf->f_type = OCFS2_SUPER_MAGIC;
1707         buf->f_bsize = dentry->d_sb->s_blocksize;
1708         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1709         buf->f_blocks = ((sector_t) numbits) *
1710                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1711         buf->f_bfree = ((sector_t) freebits) *
1712                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1713         buf->f_bavail = buf->f_bfree;
1714         buf->f_files = numbits;
1715         buf->f_ffree = freebits;
1716         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1717                                 & 0xFFFFFFFFUL;
1718         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1719                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1720
1721         brelse(bh);
1722
1723         ocfs2_inode_unlock(inode, 0);
1724         status = 0;
1725 bail:
1726         iput(inode);
1727
1728         if (status)
1729                 mlog_errno(status);
1730
1731         return status;
1732 }
1733
1734 static void ocfs2_inode_init_once(void *data)
1735 {
1736         struct ocfs2_inode_info *oi = data;
1737
1738         oi->ip_flags = 0;
1739         oi->ip_open_count = 0;
1740         spin_lock_init(&oi->ip_lock);
1741         ocfs2_extent_map_init(&oi->vfs_inode);
1742         INIT_LIST_HEAD(&oi->ip_io_markers);
1743         oi->ip_dir_start_lookup = 0;
1744         mutex_init(&oi->ip_unaligned_aio);
1745         init_rwsem(&oi->ip_alloc_sem);
1746         init_rwsem(&oi->ip_xattr_sem);
1747         mutex_init(&oi->ip_io_mutex);
1748
1749         oi->ip_blkno = 0ULL;
1750         oi->ip_clusters = 0;
1751
1752         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1753
1754         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1755         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1756         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1757
1758         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1759                                   &ocfs2_inode_caching_ops);
1760
1761         inode_init_once(&oi->vfs_inode);
1762 }
1763
1764 static int ocfs2_initialize_mem_caches(void)
1765 {
1766         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1767                                        sizeof(struct ocfs2_inode_info),
1768                                        0,
1769                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1770                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1771                                        ocfs2_inode_init_once);
1772         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1773                                         sizeof(struct ocfs2_dquot),
1774                                         0,
1775                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1776                                                 SLAB_MEM_SPREAD),
1777                                         NULL);
1778         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1779                                         sizeof(struct ocfs2_quota_chunk),
1780                                         0,
1781                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1782                                         NULL);
1783         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1784             !ocfs2_qf_chunk_cachep) {
1785                 if (ocfs2_inode_cachep)
1786                         kmem_cache_destroy(ocfs2_inode_cachep);
1787                 if (ocfs2_dquot_cachep)
1788                         kmem_cache_destroy(ocfs2_dquot_cachep);
1789                 if (ocfs2_qf_chunk_cachep)
1790                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1791                 return -ENOMEM;
1792         }
1793
1794         return 0;
1795 }
1796
1797 static void ocfs2_free_mem_caches(void)
1798 {
1799         /*
1800          * Make sure all delayed rcu free inodes are flushed before we
1801          * destroy cache.
1802          */
1803         rcu_barrier();
1804         if (ocfs2_inode_cachep)
1805                 kmem_cache_destroy(ocfs2_inode_cachep);
1806         ocfs2_inode_cachep = NULL;
1807
1808         if (ocfs2_dquot_cachep)
1809                 kmem_cache_destroy(ocfs2_dquot_cachep);
1810         ocfs2_dquot_cachep = NULL;
1811
1812         if (ocfs2_qf_chunk_cachep)
1813                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1814         ocfs2_qf_chunk_cachep = NULL;
1815 }
1816
1817 static int ocfs2_get_sector(struct super_block *sb,
1818                             struct buffer_head **bh,
1819                             int block,
1820                             int sect_size)
1821 {
1822         if (!sb_set_blocksize(sb, sect_size)) {
1823                 mlog(ML_ERROR, "unable to set blocksize\n");
1824                 return -EIO;
1825         }
1826
1827         *bh = sb_getblk(sb, block);
1828         if (!*bh) {
1829                 mlog_errno(-ENOMEM);
1830                 return -ENOMEM;
1831         }
1832         lock_buffer(*bh);
1833         if (!buffer_dirty(*bh))
1834                 clear_buffer_uptodate(*bh);
1835         unlock_buffer(*bh);
1836         ll_rw_block(READ, 1, bh);
1837         wait_on_buffer(*bh);
1838         if (!buffer_uptodate(*bh)) {
1839                 mlog_errno(-EIO);
1840                 brelse(*bh);
1841                 *bh = NULL;
1842                 return -EIO;
1843         }
1844
1845         return 0;
1846 }
1847
1848 static int ocfs2_mount_volume(struct super_block *sb)
1849 {
1850         int status = 0;
1851         int unlock_super = 0;
1852         struct ocfs2_super *osb = OCFS2_SB(sb);
1853
1854         if (ocfs2_is_hard_readonly(osb))
1855                 goto leave;
1856
1857         status = ocfs2_dlm_init(osb);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto leave;
1861         }
1862
1863         status = ocfs2_super_lock(osb, 1);
1864         if (status < 0) {
1865                 mlog_errno(status);
1866                 goto leave;
1867         }
1868         unlock_super = 1;
1869
1870         /* This will load up the node map and add ourselves to it. */
1871         status = ocfs2_find_slot(osb);
1872         if (status < 0) {
1873                 mlog_errno(status);
1874                 goto leave;
1875         }
1876
1877         /* load all node-local system inodes */
1878         status = ocfs2_init_local_system_inodes(osb);
1879         if (status < 0) {
1880                 mlog_errno(status);
1881                 goto leave;
1882         }
1883
1884         status = ocfs2_check_volume(osb);
1885         if (status < 0) {
1886                 mlog_errno(status);
1887                 goto leave;
1888         }
1889
1890         status = ocfs2_truncate_log_init(osb);
1891         if (status < 0)
1892                 mlog_errno(status);
1893
1894 leave:
1895         if (unlock_super)
1896                 ocfs2_super_unlock(osb, 1);
1897
1898         return status;
1899 }
1900
1901 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1902 {
1903         int tmp, hangup_needed = 0;
1904         struct ocfs2_super *osb = NULL;
1905         char nodestr[12];
1906
1907         trace_ocfs2_dismount_volume(sb);
1908
1909         BUG_ON(!sb);
1910         osb = OCFS2_SB(sb);
1911         BUG_ON(!osb);
1912
1913         debugfs_remove(osb->osb_ctxt);
1914
1915         /* Orphan scan should be stopped as early as possible */
1916         ocfs2_orphan_scan_stop(osb);
1917
1918         ocfs2_disable_quotas(osb);
1919
1920         /* All dquots should be freed by now */
1921         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1922         /* Wait for worker to be done with the work structure in osb */
1923         cancel_work_sync(&osb->dquot_drop_work);
1924
1925         ocfs2_shutdown_local_alloc(osb);
1926
1927         ocfs2_truncate_log_shutdown(osb);
1928
1929         /* This will disable recovery and flush any recovery work. */
1930         ocfs2_recovery_exit(osb);
1931
1932         ocfs2_journal_shutdown(osb);
1933
1934         ocfs2_sync_blockdev(sb);
1935
1936         ocfs2_purge_refcount_trees(osb);
1937
1938         /* No cluster connection means we've failed during mount, so skip
1939          * all the steps which depended on that to complete. */
1940         if (osb->cconn) {
1941                 tmp = ocfs2_super_lock(osb, 1);
1942                 if (tmp < 0) {
1943                         mlog_errno(tmp);
1944                         return;
1945                 }
1946         }
1947
1948         if (osb->slot_num != OCFS2_INVALID_SLOT)
1949                 ocfs2_put_slot(osb);
1950
1951         if (osb->cconn)
1952                 ocfs2_super_unlock(osb, 1);
1953
1954         ocfs2_release_system_inodes(osb);
1955
1956         /*
1957          * If we're dismounting due to mount error, mount.ocfs2 will clean
1958          * up heartbeat.  If we're a local mount, there is no heartbeat.
1959          * If we failed before we got a uuid_str yet, we can't stop
1960          * heartbeat.  Otherwise, do it.
1961          */
1962         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1963             !ocfs2_is_hard_readonly(osb))
1964                 hangup_needed = 1;
1965
1966         if (osb->cconn)
1967                 ocfs2_dlm_shutdown(osb, hangup_needed);
1968
1969         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1970         debugfs_remove(osb->osb_debug_root);
1971
1972         if (hangup_needed)
1973                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1974
1975         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1976
1977         if (ocfs2_mount_local(osb))
1978                 snprintf(nodestr, sizeof(nodestr), "local");
1979         else
1980                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1981
1982         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1983                osb->dev_str, nodestr);
1984
1985         ocfs2_delete_osb(osb);
1986         kfree(osb);
1987         sb->s_dev = 0;
1988         sb->s_fs_info = NULL;
1989 }
1990
1991 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1992                                 unsigned uuid_bytes)
1993 {
1994         int i, ret;
1995         char *ptr;
1996
1997         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1998
1999         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2000         if (osb->uuid_str == NULL)
2001                 return -ENOMEM;
2002
2003         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2004                 /* print with null */
2005                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2006                 if (ret != 2) /* drop super cleans up */
2007                         return -EINVAL;
2008                 /* then only advance past the last char */
2009                 ptr += 2;
2010         }
2011
2012         return 0;
2013 }
2014
2015 /* Make sure entire volume is addressable by our journal.  Requires
2016    osb_clusters_at_boot to be valid and for the journal to have been
2017    initialized by ocfs2_journal_init(). */
2018 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2019 {
2020         int status = 0;
2021         u64 max_block =
2022                 ocfs2_clusters_to_blocks(osb->sb,
2023                                          osb->osb_clusters_at_boot) - 1;
2024
2025         /* 32-bit block number is always OK. */
2026         if (max_block <= (u32)~0ULL)
2027                 goto out;
2028
2029         /* Volume is "huge", so see if our journal is new enough to
2030            support it. */
2031         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2032                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2033               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2034                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2035                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2036                      "Enable the 'block64' journal option with tunefs.ocfs2");
2037                 status = -EFBIG;
2038                 goto out;
2039         }
2040
2041  out:
2042         return status;
2043 }
2044
2045 static int ocfs2_initialize_super(struct super_block *sb,
2046                                   struct buffer_head *bh,
2047                                   int sector_size,
2048                                   struct ocfs2_blockcheck_stats *stats)
2049 {
2050         int status;
2051         int i, cbits, bbits;
2052         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2053         struct inode *inode = NULL;
2054         struct ocfs2_journal *journal;
2055         struct ocfs2_super *osb;
2056         u64 total_blocks;
2057
2058         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2059         if (!osb) {
2060                 status = -ENOMEM;
2061                 mlog_errno(status);
2062                 goto bail;
2063         }
2064
2065         sb->s_fs_info = osb;
2066         sb->s_op = &ocfs2_sops;
2067         sb->s_d_op = &ocfs2_dentry_ops;
2068         sb->s_export_op = &ocfs2_export_ops;
2069         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2070         sb->dq_op = &ocfs2_quota_operations;
2071         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2072         sb->s_xattr = ocfs2_xattr_handlers;
2073         sb->s_time_gran = 1;
2074         sb->s_flags |= MS_NOATIME;
2075         /* this is needed to support O_LARGEFILE */
2076         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2077         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2078         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2079         memcpy(sb->s_uuid, di->id2.i_super.s_uuid,
2080                sizeof(di->id2.i_super.s_uuid));
2081
2082         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2083
2084         for (i = 0; i < 3; i++)
2085                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2086         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2087
2088         osb->sb = sb;
2089         /* Save off for ocfs2_rw_direct */
2090         osb->s_sectsize_bits = blksize_bits(sector_size);
2091         BUG_ON(!osb->s_sectsize_bits);
2092
2093         spin_lock_init(&osb->dc_task_lock);
2094         init_waitqueue_head(&osb->dc_event);
2095         osb->dc_work_sequence = 0;
2096         osb->dc_wake_sequence = 0;
2097         INIT_LIST_HEAD(&osb->blocked_lock_list);
2098         osb->blocked_lock_count = 0;
2099         spin_lock_init(&osb->osb_lock);
2100         spin_lock_init(&osb->osb_xattr_lock);
2101         ocfs2_init_steal_slots(osb);
2102
2103         mutex_init(&osb->system_file_mutex);
2104
2105         atomic_set(&osb->alloc_stats.moves, 0);
2106         atomic_set(&osb->alloc_stats.local_data, 0);
2107         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2108         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2109         atomic_set(&osb->alloc_stats.bg_extends, 0);
2110
2111         /* Copy the blockcheck stats from the superblock probe */
2112         osb->osb_ecc_stats = *stats;
2113
2114         ocfs2_init_node_maps(osb);
2115
2116         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2117                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2118
2119         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2120         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2121                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2122                      osb->max_slots);
2123                 status = -EINVAL;
2124                 goto bail;
2125         }
2126
2127         ocfs2_orphan_scan_init(osb);
2128
2129         status = ocfs2_recovery_init(osb);
2130         if (status) {
2131                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2132                 mlog_errno(status);
2133                 goto bail;
2134         }
2135
2136         init_waitqueue_head(&osb->checkpoint_event);
2137
2138         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2139
2140         osb->slot_num = OCFS2_INVALID_SLOT;
2141
2142         osb->s_xattr_inline_size = le16_to_cpu(
2143                                         di->id2.i_super.s_xattr_inline_size);
2144
2145         osb->local_alloc_state = OCFS2_LA_UNUSED;
2146         osb->local_alloc_bh = NULL;
2147         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2148
2149         init_waitqueue_head(&osb->osb_mount_event);
2150
2151         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2152         if (status) {
2153                 mlog_errno(status);
2154                 goto bail;
2155         }
2156
2157         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2158         if (!osb->vol_label) {
2159                 mlog(ML_ERROR, "unable to alloc vol label\n");
2160                 status = -ENOMEM;
2161                 goto bail;
2162         }
2163
2164         osb->slot_recovery_generations =
2165                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2166                         GFP_KERNEL);
2167         if (!osb->slot_recovery_generations) {
2168                 status = -ENOMEM;
2169                 mlog_errno(status);
2170                 goto bail;
2171         }
2172
2173         init_waitqueue_head(&osb->osb_wipe_event);
2174         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2175                                         sizeof(*osb->osb_orphan_wipes),
2176                                         GFP_KERNEL);
2177         if (!osb->osb_orphan_wipes) {
2178                 status = -ENOMEM;
2179                 mlog_errno(status);
2180                 goto bail;
2181         }
2182
2183         osb->osb_rf_lock_tree = RB_ROOT;
2184
2185         osb->s_feature_compat =
2186                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2187         osb->s_feature_ro_compat =
2188                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2189         osb->s_feature_incompat =
2190                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2191
2192         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2193                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2194                      "optional features (%x).\n", i);
2195                 status = -EINVAL;
2196                 goto bail;
2197         }
2198         if (!(osb->sb->s_flags & MS_RDONLY) &&
2199             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2200                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2201                      "unsupported optional features (%x).\n", i);
2202                 status = -EINVAL;
2203                 goto bail;
2204         }
2205
2206         if (ocfs2_clusterinfo_valid(osb)) {
2207                 osb->osb_stackflags =
2208                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2209                 strlcpy(osb->osb_cluster_stack,
2210                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2211                        OCFS2_STACK_LABEL_LEN + 1);
2212                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2213                         mlog(ML_ERROR,
2214                              "couldn't mount because of an invalid "
2215                              "cluster stack label (%s) \n",
2216                              osb->osb_cluster_stack);
2217                         status = -EINVAL;
2218                         goto bail;
2219                 }
2220                 strlcpy(osb->osb_cluster_name,
2221                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2222                         OCFS2_CLUSTER_NAME_LEN + 1);
2223         } else {
2224                 /* The empty string is identical with classic tools that
2225                  * don't know about s_cluster_info. */
2226                 osb->osb_cluster_stack[0] = '\0';
2227         }
2228
2229         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2230
2231         /* FIXME
2232          * This should be done in ocfs2_journal_init(), but unknown
2233          * ordering issues will cause the filesystem to crash.
2234          * If anyone wants to figure out what part of the code
2235          * refers to osb->journal before ocfs2_journal_init() is run,
2236          * be my guest.
2237          */
2238         /* initialize our journal structure */
2239
2240         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2241         if (!journal) {
2242                 mlog(ML_ERROR, "unable to alloc journal\n");
2243                 status = -ENOMEM;
2244                 goto bail;
2245         }
2246         osb->journal = journal;
2247         journal->j_osb = osb;
2248
2249         atomic_set(&journal->j_num_trans, 0);
2250         init_rwsem(&journal->j_trans_barrier);
2251         init_waitqueue_head(&journal->j_checkpointed);
2252         spin_lock_init(&journal->j_lock);
2253         journal->j_trans_id = (unsigned long) 1;
2254         INIT_LIST_HEAD(&journal->j_la_cleanups);
2255         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2256         journal->j_state = OCFS2_JOURNAL_FREE;
2257
2258         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2259         init_llist_head(&osb->dquot_drop_list);
2260
2261         /* get some pseudo constants for clustersize bits */
2262         osb->s_clustersize_bits =
2263                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2264         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2265
2266         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2267             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2268                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2269                      osb->s_clustersize);
2270                 status = -EINVAL;
2271                 goto bail;
2272         }
2273
2274         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2275                                                 le32_to_cpu(di->i_clusters));
2276
2277         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2278                                            total_blocks);
2279         if (status) {
2280                 mlog(ML_ERROR, "Volume too large "
2281                      "to mount safely on this system");
2282                 status = -EFBIG;
2283                 goto bail;
2284         }
2285
2286         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2287                                  sizeof(di->id2.i_super.s_uuid))) {
2288                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2289                 status = -ENOMEM;
2290                 goto bail;
2291         }
2292
2293         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2294                 OCFS2_MAX_VOL_LABEL_LEN);
2295         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2296         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2297         osb->first_cluster_group_blkno =
2298                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2299         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2300         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2301         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2302                                      (unsigned long long)osb->root_blkno,
2303                                      (unsigned long long)osb->system_dir_blkno,
2304                                      osb->s_clustersize_bits);
2305
2306         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2307         if (!osb->osb_dlm_debug) {
2308                 status = -ENOMEM;
2309                 mlog_errno(status);
2310                 goto bail;
2311         }
2312
2313         atomic_set(&osb->vol_state, VOLUME_INIT);
2314
2315         /* load root, system_dir, and all global system inodes */
2316         status = ocfs2_init_global_system_inodes(osb);
2317         if (status < 0) {
2318                 mlog_errno(status);
2319                 goto bail;
2320         }
2321
2322         /*
2323          * global bitmap
2324          */
2325         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2326                                             OCFS2_INVALID_SLOT);
2327         if (!inode) {
2328                 status = -EINVAL;
2329                 mlog_errno(status);
2330                 goto bail;
2331         }
2332
2333         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2334         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2335         iput(inode);
2336
2337         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2338                                  osb->s_feature_incompat) * 8;
2339
2340         status = ocfs2_init_slot_info(osb);
2341         if (status < 0) {
2342                 mlog_errno(status);
2343                 goto bail;
2344         }
2345         cleancache_init_shared_fs(sb);
2346
2347 bail:
2348         return status;
2349 }
2350
2351 /*
2352  * will return: -EAGAIN if it is ok to keep searching for superblocks
2353  *              -EINVAL if there is a bad superblock
2354  *              0 on success
2355  */
2356 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2357                                struct buffer_head *bh,
2358                                u32 blksz,
2359                                struct ocfs2_blockcheck_stats *stats)
2360 {
2361         int status = -EAGAIN;
2362
2363         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2364                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2365                 /* We have to do a raw check of the feature here */
2366                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2367                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2368                         status = ocfs2_block_check_validate(bh->b_data,
2369                                                             bh->b_size,
2370                                                             &di->i_check,
2371                                                             stats);
2372                         if (status)
2373                                 goto out;
2374                 }
2375                 status = -EINVAL;
2376                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2377                         mlog(ML_ERROR, "found superblock with incorrect block "
2378                              "size: found %u, should be %u\n",
2379                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2380                                blksz);
2381                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2382                            OCFS2_MAJOR_REV_LEVEL ||
2383                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2384                            OCFS2_MINOR_REV_LEVEL) {
2385                         mlog(ML_ERROR, "found superblock with bad version: "
2386                              "found %u.%u, should be %u.%u\n",
2387                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2388                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2389                              OCFS2_MAJOR_REV_LEVEL,
2390                              OCFS2_MINOR_REV_LEVEL);
2391                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2392                         mlog(ML_ERROR, "bad block number on superblock: "
2393                              "found %llu, should be %llu\n",
2394                              (unsigned long long)le64_to_cpu(di->i_blkno),
2395                              (unsigned long long)bh->b_blocknr);
2396                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2397                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2398                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2399                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2400                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2401                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2402                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2403                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2404                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2405                         mlog(ML_ERROR,
2406                              "Superblock slots found greater than file system "
2407                              "maximum: found %u, max %u\n",
2408                              le16_to_cpu(di->id2.i_super.s_max_slots),
2409                              OCFS2_MAX_SLOTS);
2410                 } else {
2411                         /* found it! */
2412                         status = 0;
2413                 }
2414         }
2415
2416 out:
2417         if (status && status != -EAGAIN)
2418                 mlog_errno(status);
2419         return status;
2420 }
2421
2422 static int ocfs2_check_volume(struct ocfs2_super *osb)
2423 {
2424         int status;
2425         int dirty;
2426         int local;
2427         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2428                                                   * recover
2429                                                   * ourselves. */
2430
2431         /* Init our journal object. */
2432         status = ocfs2_journal_init(osb->journal, &dirty);
2433         if (status < 0) {
2434                 mlog(ML_ERROR, "Could not initialize journal!\n");
2435                 goto finally;
2436         }
2437
2438         /* Now that journal has been initialized, check to make sure
2439            entire volume is addressable. */
2440         status = ocfs2_journal_addressable(osb);
2441         if (status)
2442                 goto finally;
2443
2444         /* If the journal was unmounted cleanly then we don't want to
2445          * recover anything. Otherwise, journal_load will do that
2446          * dirty work for us :) */
2447         if (!dirty) {
2448                 status = ocfs2_journal_wipe(osb->journal, 0);
2449                 if (status < 0) {
2450                         mlog_errno(status);
2451                         goto finally;
2452                 }
2453         } else {
2454                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2455                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2456         }
2457
2458         local = ocfs2_mount_local(osb);
2459
2460         /* will play back anything left in the journal. */
2461         status = ocfs2_journal_load(osb->journal, local, dirty);
2462         if (status < 0) {
2463                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2464                 goto finally;
2465         }
2466
2467         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2468                 jbd2_journal_set_features(osb->journal->j_journal,
2469                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2470                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2471         else
2472                 jbd2_journal_clear_features(osb->journal->j_journal,
2473                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2474                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2475
2476         if (dirty) {
2477                 /* recover my local alloc if we didn't unmount cleanly. */
2478                 status = ocfs2_begin_local_alloc_recovery(osb,
2479                                                           osb->slot_num,
2480                                                           &local_alloc);
2481                 if (status < 0) {
2482                         mlog_errno(status);
2483                         goto finally;
2484                 }
2485                 /* we complete the recovery process after we've marked
2486                  * ourselves as mounted. */
2487         }
2488
2489         status = ocfs2_load_local_alloc(osb);
2490         if (status < 0) {
2491                 mlog_errno(status);
2492                 goto finally;
2493         }
2494
2495         if (dirty) {
2496                 /* Recovery will be completed after we've mounted the
2497                  * rest of the volume. */
2498                 osb->dirty = 1;
2499                 osb->local_alloc_copy = local_alloc;
2500                 local_alloc = NULL;
2501         }
2502
2503         /* go through each journal, trylock it and if you get the
2504          * lock, and it's marked as dirty, set the bit in the recover
2505          * map and launch a recovery thread for it. */
2506         status = ocfs2_mark_dead_nodes(osb);
2507         if (status < 0) {
2508                 mlog_errno(status);
2509                 goto finally;
2510         }
2511
2512         status = ocfs2_compute_replay_slots(osb);
2513         if (status < 0)
2514                 mlog_errno(status);
2515
2516 finally:
2517         kfree(local_alloc);
2518
2519         if (status)
2520                 mlog_errno(status);
2521         return status;
2522 }
2523
2524 /*
2525  * The routine gets called from dismount or close whenever a dismount on
2526  * volume is requested and the osb open count becomes 1.
2527  * It will remove the osb from the global list and also free up all the
2528  * initialized resources and fileobject.
2529  */
2530 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2531 {
2532         /* This function assumes that the caller has the main osb resource */
2533
2534         ocfs2_free_slot_info(osb);
2535
2536         kfree(osb->osb_orphan_wipes);
2537         kfree(osb->slot_recovery_generations);
2538         /* FIXME
2539          * This belongs in journal shutdown, but because we have to
2540          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2541          * we free it here.
2542          */
2543         kfree(osb->journal);
2544         kfree(osb->local_alloc_copy);
2545         kfree(osb->uuid_str);
2546         kfree(osb->vol_label);
2547         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2548         memset(osb, 0, sizeof(struct ocfs2_super));
2549 }
2550
2551 /* Depending on the mount option passed, perform one of the following:
2552  * Put OCFS2 into a readonly state (default)
2553  * Return EIO so that only the process errs
2554  * Fix the error as if fsck.ocfs2 -y
2555  * panic
2556  */
2557 static int ocfs2_handle_error(struct super_block *sb)
2558 {
2559         struct ocfs2_super *osb = OCFS2_SB(sb);
2560         int rv = 0;
2561
2562         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2563         pr_crit("On-disk corruption discovered. "
2564                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2565
2566         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2567                 panic("OCFS2: (device %s): panic forced after error\n",
2568                       sb->s_id);
2569         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2570                 pr_crit("OCFS2: Returning error to the calling process.\n");
2571                 rv = -EIO;
2572         } else { /* default option */
2573                 rv = -EROFS;
2574                 if (sb->s_flags & MS_RDONLY &&
2575                                 (ocfs2_is_soft_readonly(osb) ||
2576                                  ocfs2_is_hard_readonly(osb)))
2577                         return rv;
2578
2579                 pr_crit("OCFS2: File system is now read-only.\n");
2580                 sb->s_flags |= MS_RDONLY;
2581                 ocfs2_set_ro_flag(osb, 0);
2582         }
2583
2584         return rv;
2585 }
2586
2587 int __ocfs2_error(struct super_block *sb, const char *function,
2588                   const char *fmt, ...)
2589 {
2590         struct va_format vaf;
2591         va_list args;
2592
2593         va_start(args, fmt);
2594         vaf.fmt = fmt;
2595         vaf.va = &args;
2596
2597         /* Not using mlog here because we want to show the actual
2598          * function the error came from. */
2599         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2600                sb->s_id, function, &vaf);
2601
2602         va_end(args);
2603
2604         return 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, const char *function,
2611                    const char *fmt, ...)
2612 {
2613         struct va_format vaf;
2614         va_list args;
2615
2616         va_start(args, fmt);
2617
2618         vaf.fmt = fmt;
2619         vaf.va = &args;
2620
2621         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2622                sb->s_id, function, &vaf);
2623
2624         va_end(args);
2625
2626         /* We don't have the cluster support yet to go straight to
2627          * hard readonly in here. Until then, we want to keep
2628          * ocfs2_abort() so that we can at least mark critical
2629          * errors.
2630          *
2631          * TODO: This should abort the journal and alert other nodes
2632          * that our slot needs recovery. */
2633
2634         /* Force a panic(). This stinks, but it's better than letting
2635          * things continue without having a proper hard readonly
2636          * here. */
2637         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2638                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2639         ocfs2_handle_error(sb);
2640 }
2641
2642 /*
2643  * Void signal blockers, because in-kernel sigprocmask() only fails
2644  * when SIG_* is wrong.
2645  */
2646 void ocfs2_block_signals(sigset_t *oldset)
2647 {
2648         int rc;
2649         sigset_t blocked;
2650
2651         sigfillset(&blocked);
2652         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2653         BUG_ON(rc);
2654 }
2655
2656 void ocfs2_unblock_signals(sigset_t *oldset)
2657 {
2658         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2659         BUG_ON(rc);
2660 }
2661
2662 module_init(ocfs2_init);
2663 module_exit(ocfs2_exit);