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 #include "filecheck.h"
  78
  79 static struct kmem_cache *ocfs2_inode_cachep;
  80 struct kmem_cache *ocfs2_dquot_cachep;
  81 struct kmem_cache *ocfs2_qf_chunk_cachep;
  82
  83 static struct dentry *ocfs2_debugfs_root;
  84
  85 MODULE_AUTHOR("Oracle");
  86 MODULE_LICENSE("GPL");
  87 MODULE_DESCRIPTION("OCFS2 cluster file system");
  88
  89 struct mount_options
  90 {
  91         unsigned long   commit_interval;
  92         unsigned long   mount_opt;
  93         unsigned int    atime_quantum;
  94         unsigned short  slot;
  95         int             localalloc_opt;
  96         unsigned int    resv_level;
  97         int             dir_resv_level;
  98         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
  99 };
 100
 101 static int ocfs2_parse_options(struct super_block *sb, char *options,
 102                                struct mount_options *mopt,
 103                                int is_remount);
 104 static int ocfs2_check_set_options(struct super_block *sb,
 105                                    struct mount_options *options);
 106 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 107 static void ocfs2_put_super(struct super_block *sb);
 108 static int ocfs2_mount_volume(struct super_block *sb);
 109 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 110 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 111 static int ocfs2_initialize_mem_caches(void);
 112 static void ocfs2_free_mem_caches(void);
 113 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 114
 115 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 116
 117 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 118
 119 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 120 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 121 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 122 static int ocfs2_check_volume(struct ocfs2_super *osb);
 123 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 124                                struct buffer_head *bh,
 125                                u32 sectsize,
 126                                struct ocfs2_blockcheck_stats *stats);
 127 static int ocfs2_initialize_super(struct super_block *sb,
 128                                   struct buffer_head *bh,
 129                                   int sector_size,
 130                                   struct ocfs2_blockcheck_stats *stats);
 131 static int ocfs2_get_sector(struct super_block *sb,
 132                             struct buffer_head **bh,
 133                             int block,
 134                             int sect_size);
 135 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 136 static void ocfs2_destroy_inode(struct inode *inode);
 137 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 138 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 139 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 140
 141 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 142 {
 143         return OCFS2_I(inode)->i_dquot;
 144 }
 145
 146 static const struct super_operations ocfs2_sops = {
 147         .statfs         = ocfs2_statfs,
 148         .alloc_inode    = ocfs2_alloc_inode,
 149         .destroy_inode  = ocfs2_destroy_inode,
 150         .drop_inode     = ocfs2_drop_inode,
 151         .evict_inode    = ocfs2_evict_inode,
 152         .sync_fs        = ocfs2_sync_fs,
 153         .put_super      = ocfs2_put_super,
 154         .remount_fs     = ocfs2_remount,
 155         .show_options   = ocfs2_show_options,
 156         .quota_read     = ocfs2_quota_read,
 157         .quota_write    = ocfs2_quota_write,
 158         .get_dquots     = ocfs2_get_dquots,
 159 };
 160
 161 enum {
 162         Opt_barrier,
 163         Opt_err_panic,
 164         Opt_err_ro,
 165         Opt_intr,
 166         Opt_nointr,
 167         Opt_hb_none,
 168         Opt_hb_local,
 169         Opt_hb_global,
 170         Opt_data_ordered,
 171         Opt_data_writeback,
 172         Opt_atime_quantum,
 173         Opt_slot,
 174         Opt_commit,
 175         Opt_localalloc,
 176         Opt_localflocks,
 177         Opt_stack,
 178         Opt_user_xattr,
 179         Opt_nouser_xattr,
 180         Opt_inode64,
 181         Opt_acl,
 182         Opt_noacl,
 183         Opt_usrquota,
 184         Opt_grpquota,
 185         Opt_coherency_buffered,
 186         Opt_coherency_full,
 187         Opt_resv_level,
 188         Opt_dir_resv_level,
 189         Opt_journal_async_commit,
 190         Opt_err_cont,
 191         Opt_err,
 192 };
 193
 194 static const match_table_t tokens = {
 195         {Opt_barrier, "barrier=%u"},
 196         {Opt_err_panic, "errors=panic"},
 197         {Opt_err_ro, "errors=remount-ro"},
 198         {Opt_intr, "intr"},
 199         {Opt_nointr, "nointr"},
 200         {Opt_hb_none, OCFS2_HB_NONE},
 201         {Opt_hb_local, OCFS2_HB_LOCAL},
 202         {Opt_hb_global, OCFS2_HB_GLOBAL},
 203         {Opt_data_ordered, "data=ordered"},
 204         {Opt_data_writeback, "data=writeback"},
 205         {Opt_atime_quantum, "atime_quantum=%u"},
 206         {Opt_slot, "preferred_slot=%u"},
 207         {Opt_commit, "commit=%u"},
 208         {Opt_localalloc, "localalloc=%d"},
 209         {Opt_localflocks, "localflocks"},
 210         {Opt_stack, "cluster_stack=%s"},
 211         {Opt_user_xattr, "user_xattr"},
 212         {Opt_nouser_xattr, "nouser_xattr"},
 213         {Opt_inode64, "inode64"},
 214         {Opt_acl, "acl"},
 215         {Opt_noacl, "noacl"},
 216         {Opt_usrquota, "usrquota"},
 217         {Opt_grpquota, "grpquota"},
 218         {Opt_coherency_buffered, "coherency=buffered"},
 219         {Opt_coherency_full, "coherency=full"},
 220         {Opt_resv_level, "resv_level=%u"},
 221         {Opt_dir_resv_level, "dir_resv_level=%u"},
 222         {Opt_journal_async_commit, "journal_async_commit"},
 223         {Opt_err_cont, "errors=continue"},
 224         {Opt_err, NULL}
 225 };
 226
 227 #ifdef CONFIG_DEBUG_FS
 228 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 229 {
 230         struct ocfs2_cluster_connection *cconn = osb->cconn;
 231         struct ocfs2_recovery_map *rm = osb->recovery_map;
 232         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 233         int i, out = 0;
 234         unsigned long flags;
 235
 236         out += snprintf(buf + out, len - out,
 237                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 238                         "Device", osb->dev_str, osb->uuid_str,
 239                         osb->fs_generation, osb->vol_label);
 240
 241         out += snprintf(buf + out, len - out,
 242                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 243                         atomic_read(&osb->vol_state), osb->osb_flags);
 244
 245         out += snprintf(buf + out, len - out,
 246                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 247                         osb->sb->s_blocksize, osb->s_clustersize);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 251                         "ROcompat: 0x%X\n",
 252                         "Features", osb->s_feature_compat,
 253                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 254
 255         out += snprintf(buf + out, len - out,
 256                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 257                         osb->s_mount_opt, osb->s_atime_quantum);
 258
 259         if (cconn) {
 260                 out += snprintf(buf + out, len - out,
 261                                 "%10s => Stack: %s  Name: %*s  "
 262                                 "Version: %d.%d\n", "Cluster",
 263                                 (*osb->osb_cluster_stack == '\0' ?
 264                                  "o2cb" : osb->osb_cluster_stack),
 265                                 cconn->cc_namelen, cconn->cc_name,
 266                                 cconn->cc_version.pv_major,
 267                                 cconn->cc_version.pv_minor);
 268         }
 269
 270         spin_lock_irqsave(&osb->dc_task_lock, flags);
 271         out += snprintf(buf + out, len - out,
 272                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 273                         "WorkSeq: %lu\n", "DownCnvt",
 274                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 275                         osb->blocked_lock_count, osb->dc_wake_sequence,
 276                         osb->dc_work_sequence);
 277         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 278
 279         spin_lock(&osb->osb_lock);
 280         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 281                         "Recovery",
 282                         (osb->recovery_thread_task ?
 283                          task_pid_nr(osb->recovery_thread_task) : -1));
 284         if (rm->rm_used == 0)
 285                 out += snprintf(buf + out, len - out, " None\n");
 286         else {
 287                 for (i = 0; i < rm->rm_used; i++)
 288                         out += snprintf(buf + out, len - out, " %d",
 289                                         rm->rm_entries[i]);
 290                 out += snprintf(buf + out, len - out, "\n");
 291         }
 292         spin_unlock(&osb->osb_lock);
 293
 294         out += snprintf(buf + out, len - out,
 295                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 296                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 297                         osb->osb_commit_interval);
 298
 299         out += snprintf(buf + out, len - out,
 300                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 301                         "Journal", osb->journal->j_state,
 302                         osb->journal->j_trans_id,
 303                         atomic_read(&osb->journal->j_num_trans));
 304
 305         out += snprintf(buf + out, len - out,
 306                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 307                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 308                         "Stats",
 309                         atomic_read(&osb->alloc_stats.bitmap_data),
 310                         atomic_read(&osb->alloc_stats.local_data),
 311                         atomic_read(&osb->alloc_stats.bg_allocs),
 312                         atomic_read(&osb->alloc_stats.moves),
 313                         atomic_read(&osb->alloc_stats.bg_extends));
 314
 315         out += snprintf(buf + out, len - out,
 316                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 317                         "Default: %u bits\n",
 318                         "LocalAlloc", osb->local_alloc_state,
 319                         (unsigned long long)osb->la_last_gd,
 320                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 321
 322         spin_lock(&osb->osb_lock);
 323         out += snprintf(buf + out, len - out,
 324                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 325                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 326                         osb->s_inode_steal_slot,
 327                         atomic_read(&osb->s_num_inodes_stolen),
 328                         osb->s_meta_steal_slot,
 329                         atomic_read(&osb->s_num_meta_stolen));
 330         spin_unlock(&osb->osb_lock);
 331
 332         out += snprintf(buf + out, len - out, "OrphanScan => ");
 333         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 334                         os->os_count, os->os_seqno);
 335         out += snprintf(buf + out, len - out, " Last Scan: ");
 336         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 337                 out += snprintf(buf + out, len - out, "Disabled\n");
 338         else
 339                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 340                                 (get_seconds() - os->os_scantime.tv_sec));
 341
 342         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 343                         "Slots", "Num", "RecoGen");
 344         for (i = 0; i < osb->max_slots; ++i) {
 345                 out += snprintf(buf + out, len - out,
 346                                 "%10s  %c %3d  %10d\n",
 347                                 " ",
 348                                 (i == osb->slot_num ? '*' : ' '),
 349                                 i, osb->slot_recovery_generations[i]);
 350         }
 351
 352         return out;
 353 }
 354
 355 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 356 {
 357         struct ocfs2_super *osb = inode->i_private;
 358         char *buf = NULL;
 359
 360         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 361         if (!buf)
 362                 goto bail;
 363
 364         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 365
 366         file->private_data = buf;
 367
 368         return 0;
 369 bail:
 370         return -ENOMEM;
 371 }
 372
 373 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 374 {
 375         kfree(file->private_data);
 376         return 0;
 377 }
 378
 379 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 380                                 size_t nbytes, loff_t *ppos)
 381 {
 382         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 383                                        i_size_read(file->f_mapping->host));
 384 }
 385 #else
 386 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 387 {
 388         return 0;
 389 }
 390 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 391 {
 392         return 0;
 393 }
 394 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 395                                 size_t nbytes, loff_t *ppos)
 396 {
 397         return 0;
 398 }
 399 #endif  /* CONFIG_DEBUG_FS */
 400
 401 static const struct file_operations ocfs2_osb_debug_fops = {
 402         .open =         ocfs2_osb_debug_open,
 403         .release =      ocfs2_debug_release,
 404         .read =         ocfs2_debug_read,
 405         .llseek =       generic_file_llseek,
 406 };
 407
 408 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 409 {
 410         int status;
 411         tid_t target;
 412         struct ocfs2_super *osb = OCFS2_SB(sb);
 413
 414         if (ocfs2_is_hard_readonly(osb))
 415                 return -EROFS;
 416
 417         if (wait) {
 418                 status = ocfs2_flush_truncate_log(osb);
 419                 if (status < 0)
 420                         mlog_errno(status);
 421         } else {
 422                 ocfs2_schedule_truncate_log_flush(osb, 0);
 423         }
 424
 425         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 426                                       &target)) {
 427                 if (wait)
 428                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 429                                              target);
 430         }
 431         return 0;
 432 }
 433
 434 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 435 {
 436         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 437             && (ino == USER_QUOTA_SYSTEM_INODE
 438                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 439                 return 0;
 440         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 441             && (ino == GROUP_QUOTA_SYSTEM_INODE
 442                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 443                 return 0;
 444         return 1;
 445 }
 446
 447 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 448 {
 449         struct inode *new = NULL;
 450         int status = 0;
 451         int i;
 452
 453         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 454         if (IS_ERR(new)) {
 455                 status = PTR_ERR(new);
 456                 mlog_errno(status);
 457                 goto bail;
 458         }
 459         osb->root_inode = new;
 460
 461         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 462         if (IS_ERR(new)) {
 463                 status = PTR_ERR(new);
 464                 mlog_errno(status);
 465                 goto bail;
 466         }
 467         osb->sys_root_inode = new;
 468
 469         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 470              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 471                 if (!ocfs2_need_system_inode(osb, i))
 472                         continue;
 473                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 474                 if (!new) {
 475                         ocfs2_release_system_inodes(osb);
 476                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 477                         mlog_errno(status);
 478                         mlog(ML_ERROR, "Unable to load system inode %d, "
 479                              "possibly corrupt fs?", i);
 480                         goto bail;
 481                 }
 482                 // the array now has one ref, so drop this one
 483                 iput(new);
 484         }
 485
 486 bail:
 487         if (status)
 488                 mlog_errno(status);
 489         return status;
 490 }
 491
 492 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 493 {
 494         struct inode *new = NULL;
 495         int status = 0;
 496         int i;
 497
 498         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 499              i < NUM_SYSTEM_INODES;
 500              i++) {
 501                 if (!ocfs2_need_system_inode(osb, i))
 502                         continue;
 503                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 504                 if (!new) {
 505                         ocfs2_release_system_inodes(osb);
 506                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 507                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 508                              status, i, osb->slot_num);
 509                         goto bail;
 510                 }
 511                 /* the array now has one ref, so drop this one */
 512                 iput(new);
 513         }
 514
 515 bail:
 516         if (status)
 517                 mlog_errno(status);
 518         return status;
 519 }
 520
 521 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 522 {
 523         int i;
 524         struct inode *inode;
 525
 526         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 527                 inode = osb->global_system_inodes[i];
 528                 if (inode) {
 529                         iput(inode);
 530                         osb->global_system_inodes[i] = NULL;
 531                 }
 532         }
 533
 534         inode = osb->sys_root_inode;
 535         if (inode) {
 536                 iput(inode);
 537                 osb->sys_root_inode = NULL;
 538         }
 539
 540         inode = osb->root_inode;
 541         if (inode) {
 542                 iput(inode);
 543                 osb->root_inode = NULL;
 544         }
 545
 546         if (!osb->local_system_inodes)
 547                 return;
 548
 549         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 550                 if (osb->local_system_inodes[i]) {
 551                         iput(osb->local_system_inodes[i]);
 552                         osb->local_system_inodes[i] = NULL;
 553                 }
 554         }
 555
 556         kfree(osb->local_system_inodes);
 557         osb->local_system_inodes = NULL;
 558 }
 559
 560 /* We're allocating fs objects, use GFP_NOFS */
 561 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 562 {
 563         struct ocfs2_inode_info *oi;
 564
 565         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 566         if (!oi)
 567                 return NULL;
 568
 569         oi->i_sync_tid = 0;
 570         oi->i_datasync_tid = 0;
 571         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 572
 573         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 574         return &oi->vfs_inode;
 575 }
 576
 577 static void ocfs2_i_callback(struct rcu_head *head)
 578 {
 579         struct inode *inode = container_of(head, struct inode, i_rcu);
 580         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 581 }
 582
 583 static void ocfs2_destroy_inode(struct inode *inode)
 584 {
 585         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 586 }
 587
 588 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 589                                                 unsigned int cbits)
 590 {
 591         unsigned int bytes = 1 << cbits;
 592         unsigned int trim = bytes;
 593         unsigned int bitshift = 32;
 594
 595         /*
 596          * i_size and all block offsets in ocfs2 are always 64 bits
 597          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 598          * 64 bit platforms, cluster size will be the limiting factor.
 599          */
 600
 601 #if BITS_PER_LONG == 32
 602 # if defined(CONFIG_LBDAF)
 603         BUILD_BUG_ON(sizeof(sector_t) != 8);
 604         /*
 605          * We might be limited by page cache size.
 606          */
 607         if (bytes > PAGE_SIZE) {
 608                 bytes = PAGE_SIZE;
 609                 trim = 1;
 610                 /*
 611                  * Shift by 31 here so that we don't get larger than
 612                  * MAX_LFS_FILESIZE
 613                  */
 614                 bitshift = 31;
 615         }
 616 # else
 617         /*
 618          * We are limited by the size of sector_t. Use block size, as
 619          * that's what we expose to the VFS.
 620          */
 621         bytes = 1 << bbits;
 622         trim = 1;
 623         bitshift = 31;
 624 # endif
 625 #endif
 626
 627         /*
 628          * Trim by a whole cluster when we can actually approach the
 629          * on-disk limits. Otherwise we can overflow i_clusters when
 630          * an extent start is at the max offset.
 631          */
 632         return (((unsigned long long)bytes) << bitshift) - trim;
 633 }
 634
 635 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 636 {
 637         int incompat_features;
 638         int ret = 0;
 639         struct mount_options parsed_options;
 640         struct ocfs2_super *osb = OCFS2_SB(sb);
 641         u32 tmp;
 642
 643         sync_filesystem(sb);
 644
 645         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 646             !ocfs2_check_set_options(sb, &parsed_options)) {
 647                 ret = -EINVAL;
 648                 goto out;
 649         }
 650
 651         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 652                 OCFS2_MOUNT_HB_NONE;
 653         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 654                 ret = -EINVAL;
 655                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 656                 goto out;
 657         }
 658
 659         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 660             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 661                 ret = -EINVAL;
 662                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 663                 goto out;
 664         }
 665
 666         /* Probably don't want this on remount; it might
 667          * mess with other nodes */
 668         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 669             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 670                 ret = -EINVAL;
 671                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 672                 goto out;
 673         }
 674
 675         /* We're going to/from readonly mode. */
 676         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 677                 /* Disable quota accounting before remounting RO */
 678                 if (*flags & MS_RDONLY) {
 679                         ret = ocfs2_susp_quotas(osb, 0);
 680                         if (ret < 0)
 681                                 goto out;
 682                 }
 683                 /* Lock here so the check of HARD_RO and the potential
 684                  * setting of SOFT_RO is atomic. */
 685                 spin_lock(&osb->osb_lock);
 686                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 687                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 688                         ret = -EROFS;
 689                         goto unlock_osb;
 690                 }
 691
 692                 if (*flags & MS_RDONLY) {
 693                         sb->s_flags |= MS_RDONLY;
 694                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 695                 } else {
 696                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 697                                 mlog(ML_ERROR, "Cannot remount RDWR "
 698                                      "filesystem due to previous errors.\n");
 699                                 ret = -EROFS;
 700                                 goto unlock_osb;
 701                         }
 702                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 703                         if (incompat_features) {
 704                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 705                                      "of unsupported optional features "
 706                                      "(%x).\n", incompat_features);
 707                                 ret = -EINVAL;
 708                                 goto unlock_osb;
 709                         }
 710                         sb->s_flags &= ~MS_RDONLY;
 711                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 712                 }
 713                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 714 unlock_osb:
 715                 spin_unlock(&osb->osb_lock);
 716                 /* Enable quota accounting after remounting RW */
 717                 if (!ret && !(*flags & MS_RDONLY)) {
 718                         if (sb_any_quota_suspended(sb))
 719                                 ret = ocfs2_susp_quotas(osb, 1);
 720                         else
 721                                 ret = ocfs2_enable_quotas(osb);
 722                         if (ret < 0) {
 723                                 /* Return back changes... */
 724                                 spin_lock(&osb->osb_lock);
 725                                 sb->s_flags |= MS_RDONLY;
 726                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 727                                 spin_unlock(&osb->osb_lock);
 728                                 goto out;
 729                         }
 730                 }
 731         }
 732
 733         if (!ret) {
 734                 /* Only save off the new mount options in case of a successful
 735                  * remount. */
 736                 osb->s_mount_opt = parsed_options.mount_opt;
 737                 osb->s_atime_quantum = parsed_options.atime_quantum;
 738                 osb->preferred_slot = parsed_options.slot;
 739                 if (parsed_options.commit_interval)
 740                         osb->osb_commit_interval = parsed_options.commit_interval;
 741
 742                 if (!ocfs2_is_hard_readonly(osb))
 743                         ocfs2_set_journal_params(osb);
 744
 745                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 746                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 747                                                         MS_POSIXACL : 0);
 748         }
 749 out:
 750         return ret;
 751 }
 752
 753 static int ocfs2_sb_probe(struct super_block *sb,
 754                           struct buffer_head **bh,
 755                           int *sector_size,
 756                           struct ocfs2_blockcheck_stats *stats)
 757 {
 758         int status, tmpstat;
 759         struct ocfs1_vol_disk_hdr *hdr;
 760         struct ocfs2_dinode *di;
 761         int blksize;
 762
 763         *bh = NULL;
 764
 765         /* may be > 512 */
 766         *sector_size = bdev_logical_block_size(sb->s_bdev);
 767         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 768                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 769                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 770                 status = -EINVAL;
 771                 goto bail;
 772         }
 773
 774         /* Can this really happen? */
 775         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 776                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 777
 778         /* check block zero for old format */
 779         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 780         if (status < 0) {
 781                 mlog_errno(status);
 782                 goto bail;
 783         }
 784         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 785         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 786                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 787                      hdr->major_version, hdr->minor_version);
 788                 status = -EINVAL;
 789         }
 790         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 791                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 792                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 793                      hdr->signature);
 794                 status = -EINVAL;
 795         }
 796         brelse(*bh);
 797         *bh = NULL;
 798         if (status < 0) {
 799                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 800                      "upgraded before mounting with ocfs v2\n");
 801                 goto bail;
 802         }
 803
 804         /*
 805          * Now check at magic offset for 512, 1024, 2048, 4096
 806          * blocksizes.  4096 is the maximum blocksize because it is
 807          * the minimum clustersize.
 808          */
 809         status = -EINVAL;
 810         for (blksize = *sector_size;
 811              blksize <= OCFS2_MAX_BLOCKSIZE;
 812              blksize <<= 1) {
 813                 tmpstat = ocfs2_get_sector(sb, bh,
 814                                            OCFS2_SUPER_BLOCK_BLKNO,
 815                                            blksize);
 816                 if (tmpstat < 0) {
 817                         status = tmpstat;
 818                         mlog_errno(status);
 819                         break;
 820                 }
 821                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 822                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 823                 spin_lock_init(&stats->b_lock);
 824                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 825                 if (tmpstat < 0) {
 826                         brelse(*bh);
 827                         *bh = NULL;
 828                 }
 829                 if (tmpstat != -EAGAIN) {
 830                         status = tmpstat;
 831                         break;
 832                 }
 833         }
 834
 835 bail:
 836         return status;
 837 }
 838
 839 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 840 {
 841         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 842
 843         if (osb->s_mount_opt & hb_enabled) {
 844                 if (ocfs2_mount_local(osb)) {
 845                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 846                              "mounted device.\n");
 847                         return -EINVAL;
 848                 }
 849                 if (ocfs2_userspace_stack(osb)) {
 850                         mlog(ML_ERROR, "Userspace stack expected, but "
 851                              "o2cb heartbeat arguments passed to mount\n");
 852                         return -EINVAL;
 853                 }
 854                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 855                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 856                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 857                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 858                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 859                         return -EINVAL;
 860                 }
 861         }
 862
 863         if (!(osb->s_mount_opt & hb_enabled)) {
 864                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 865                     !ocfs2_userspace_stack(osb)) {
 866                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 867                              "a read-write clustered device.\n");
 868                         return -EINVAL;
 869                 }
 870         }
 871
 872         return 0;
 873 }
 874
 875 /*
 876  * If we're using a userspace stack, mount should have passed
 877  * a name that matches the disk.  If not, mount should not
 878  * have passed a stack.
 879  */
 880 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 881                                         struct mount_options *mopt)
 882 {
 883         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 884                 mlog(ML_ERROR,
 885                      "cluster stack passed to mount, but this filesystem "
 886                      "does not support it\n");
 887                 return -EINVAL;
 888         }
 889
 890         if (ocfs2_userspace_stack(osb) &&
 891             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 892                     OCFS2_STACK_LABEL_LEN)) {
 893                 mlog(ML_ERROR,
 894                      "cluster stack passed to mount (\"%s\") does not "
 895                      "match the filesystem (\"%s\")\n",
 896                      mopt->cluster_stack,
 897                      osb->osb_cluster_stack);
 898                 return -EINVAL;
 899         }
 900
 901         return 0;
 902 }
 903
 904 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 905 {
 906         int type;
 907         struct super_block *sb = osb->sb;
 908         unsigned int feature[OCFS2_MAXQUOTAS] = {
 909                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 910                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 911         int status = 0;
 912
 913         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 914                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 915                         continue;
 916                 if (unsuspend)
 917                         status = dquot_resume(sb, type);
 918                 else {
 919                         struct ocfs2_mem_dqinfo *oinfo;
 920
 921                         /* Cancel periodic syncing before suspending */
 922                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 923                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 924                         status = dquot_suspend(sb, type);
 925                 }
 926                 if (status < 0)
 927                         break;
 928         }
 929         if (status < 0)
 930                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 931                      "remount (error = %d).\n", status);
 932         return status;
 933 }
 934
 935 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 936 {
 937         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 938         struct super_block *sb = osb->sb;
 939         unsigned int feature[OCFS2_MAXQUOTAS] = {
 940                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 941                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 942         unsigned int ino[OCFS2_MAXQUOTAS] = {
 943                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 944                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 945         int status;
 946         int type;
 947
 948         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 949         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 950                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 951                         continue;
 952                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 953                                                         osb->slot_num);
 954                 if (!inode[type]) {
 955                         status = -ENOENT;
 956                         goto out_quota_off;
 957                 }
 958                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 959                                       DQUOT_USAGE_ENABLED);
 960                 if (status < 0)
 961                         goto out_quota_off;
 962         }
 963
 964         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 965                 iput(inode[type]);
 966         return 0;
 967 out_quota_off:
 968         ocfs2_disable_quotas(osb);
 969         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 970                 iput(inode[type]);
 971         mlog_errno(status);
 972         return status;
 973 }
 974
 975 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 976 {
 977         int type;
 978         struct inode *inode;
 979         struct super_block *sb = osb->sb;
 980         struct ocfs2_mem_dqinfo *oinfo;
 981
 982         /* We mostly ignore errors in this function because there's not much
 983          * we can do when we see them */
 984         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 985                 if (!sb_has_quota_loaded(sb, type))
 986                         continue;
 987                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 988                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 989                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 990                 inode = igrab(sb->s_dquot.files[type]);
 991                 /* Turn off quotas. This will remove all dquot structures from
 992                  * memory and so they will be automatically synced to global
 993                  * quota files */
 994                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 995                                         DQUOT_LIMITS_ENABLED);
 996                 if (!inode)
 997                         continue;
 998                 iput(inode);
 999         }
1000 }
1001
1002 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1003 {
1004         struct dentry *root;
1005         int status, sector_size;
1006         struct mount_options parsed_options;
1007         struct inode *inode = NULL;
1008         struct ocfs2_super *osb = NULL;
1009         struct buffer_head *bh = NULL;
1010         char nodestr[12];
1011         struct ocfs2_blockcheck_stats stats;
1012
1013         trace_ocfs2_fill_super(sb, data, silent);
1014
1015         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1016                 status = -EINVAL;
1017                 goto read_super_error;
1018         }
1019
1020         /* probe for superblock */
1021         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1022         if (status < 0) {
1023                 mlog(ML_ERROR, "superblock probe failed!\n");
1024                 goto read_super_error;
1025         }
1026
1027         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1028         osb = OCFS2_SB(sb);
1029         if (status < 0) {
1030                 mlog_errno(status);
1031                 goto read_super_error;
1032         }
1033         brelse(bh);
1034         bh = NULL;
1035
1036         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1037                 status = -EINVAL;
1038                 goto read_super_error;
1039         }
1040         osb->s_mount_opt = parsed_options.mount_opt;
1041         osb->s_atime_quantum = parsed_options.atime_quantum;
1042         osb->preferred_slot = parsed_options.slot;
1043         osb->osb_commit_interval = parsed_options.commit_interval;
1044
1045         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1046         osb->osb_resv_level = parsed_options.resv_level;
1047         osb->osb_dir_resv_level = parsed_options.resv_level;
1048         if (parsed_options.dir_resv_level == -1)
1049                 osb->osb_dir_resv_level = parsed_options.resv_level;
1050         else
1051                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1052
1053         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1054         if (status)
1055                 goto read_super_error;
1056
1057         sb->s_magic = OCFS2_SUPER_MAGIC;
1058
1059         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1060                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1061
1062         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1063          * heartbeat=none */
1064         if (bdev_read_only(sb->s_bdev)) {
1065                 if (!(sb->s_flags & MS_RDONLY)) {
1066                         status = -EACCES;
1067                         mlog(ML_ERROR, "Readonly device detected but readonly "
1068                              "mount was not specified.\n");
1069                         goto read_super_error;
1070                 }
1071
1072                 /* You should not be able to start a local heartbeat
1073                  * on a readonly device. */
1074                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1075                         status = -EROFS;
1076                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1077                              "device.\n");
1078                         goto read_super_error;
1079                 }
1080
1081                 status = ocfs2_check_journals_nolocks(osb);
1082                 if (status < 0) {
1083                         if (status == -EROFS)
1084                                 mlog(ML_ERROR, "Recovery required on readonly "
1085                                      "file system, but write access is "
1086                                      "unavailable.\n");
1087                         else
1088                                 mlog_errno(status);
1089                         goto read_super_error;
1090                 }
1091
1092                 ocfs2_set_ro_flag(osb, 1);
1093
1094                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1095                        "Cluster services will not be used for this mount. "
1096                        "Recovery will be skipped.\n", osb->dev_str);
1097         }
1098
1099         if (!ocfs2_is_hard_readonly(osb)) {
1100                 if (sb->s_flags & MS_RDONLY)
1101                         ocfs2_set_ro_flag(osb, 0);
1102         }
1103
1104         status = ocfs2_verify_heartbeat(osb);
1105         if (status < 0) {
1106                 mlog_errno(status);
1107                 goto read_super_error;
1108         }
1109
1110         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1111                                                  ocfs2_debugfs_root);
1112         if (!osb->osb_debug_root) {
1113                 status = -EINVAL;
1114                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1115                 goto read_super_error;
1116         }
1117
1118         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1119                                             osb->osb_debug_root,
1120                                             osb,
1121                                             &ocfs2_osb_debug_fops);
1122         if (!osb->osb_ctxt) {
1123                 status = -EINVAL;
1124                 mlog_errno(status);
1125                 goto read_super_error;
1126         }
1127
1128         if (ocfs2_meta_ecc(osb)) {
1129                 status = ocfs2_blockcheck_stats_debugfs_install(
1130                                                 &osb->osb_ecc_stats,
1131                                                 osb->osb_debug_root);
1132                 if (status) {
1133                         mlog(ML_ERROR,
1134                              "Unable to create blockcheck statistics "
1135                              "files\n");
1136                         goto read_super_error;
1137                 }
1138         }
1139
1140         status = ocfs2_mount_volume(sb);
1141         if (status < 0)
1142                 goto read_super_error;
1143
1144         if (osb->root_inode)
1145                 inode = igrab(osb->root_inode);
1146
1147         if (!inode) {
1148                 status = -EIO;
1149                 mlog_errno(status);
1150                 goto read_super_error;
1151         }
1152
1153         root = d_make_root(inode);
1154         if (!root) {
1155                 status = -ENOMEM;
1156                 mlog_errno(status);
1157                 goto read_super_error;
1158         }
1159
1160         sb->s_root = root;
1161
1162         ocfs2_complete_mount_recovery(osb);
1163
1164         if (ocfs2_mount_local(osb))
1165                 snprintf(nodestr, sizeof(nodestr), "local");
1166         else
1167                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1168
1169         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1170                "with %s data mode.\n",
1171                osb->dev_str, nodestr, osb->slot_num,
1172                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1173                "ordered");
1174
1175         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1176         wake_up(&osb->osb_mount_event);
1177
1178         /* Now we can initialize quotas because we can afford to wait
1179          * for cluster locks recovery now. That also means that truncation
1180          * log recovery can happen but that waits for proper quota setup */
1181         if (!(sb->s_flags & MS_RDONLY)) {
1182                 status = ocfs2_enable_quotas(osb);
1183                 if (status < 0) {
1184                         /* We have to err-out specially here because
1185                          * s_root is already set */
1186                         mlog_errno(status);
1187                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1188                         wake_up(&osb->osb_mount_event);
1189                         return status;
1190                 }
1191         }
1192
1193         ocfs2_complete_quota_recovery(osb);
1194
1195         /* Now we wake up again for processes waiting for quotas */
1196         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1197         wake_up(&osb->osb_mount_event);
1198
1199         /* Start this when the mount is almost sure of being successful */
1200         ocfs2_orphan_scan_start(osb);
1201
1202         /* Create filecheck sysfile /sys/fs/ocfs2/<devname>/filecheck */
1203         ocfs2_filecheck_create_sysfs(sb);
1204
1205         return status;
1206
1207 read_super_error:
1208         brelse(bh);
1209
1210         if (osb) {
1211                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1212                 wake_up(&osb->osb_mount_event);
1213                 ocfs2_dismount_volume(sb, 1);
1214         }
1215
1216         if (status)
1217                 mlog_errno(status);
1218         return status;
1219 }
1220
1221 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1222                         int flags,
1223                         const char *dev_name,
1224                         void *data)
1225 {
1226         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1227 }
1228
1229 static struct file_system_type ocfs2_fs_type = {
1230         .owner          = THIS_MODULE,
1231         .name           = "ocfs2",
1232         .mount          = ocfs2_mount,
1233         .kill_sb        = kill_block_super,
1234         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1235         .next           = NULL
1236 };
1237 MODULE_ALIAS_FS("ocfs2");
1238
1239 static int ocfs2_check_set_options(struct super_block *sb,
1240                                    struct mount_options *options)
1241 {
1242         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1243             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1244                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1245                 mlog(ML_ERROR, "User quotas were requested, but this "
1246                      "filesystem does not have the feature enabled.\n");
1247                 return 0;
1248         }
1249         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1250             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1251                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1252                 mlog(ML_ERROR, "Group quotas were requested, but this "
1253                      "filesystem does not have the feature enabled.\n");
1254                 return 0;
1255         }
1256         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1257             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1258                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1259                      "feature is not enabled\n");
1260                 return 0;
1261         }
1262         /* No ACL setting specified? Use XATTR feature... */
1263         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1264                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1265                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1266                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1267                 else
1268                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1269         }
1270         return 1;
1271 }
1272
1273 static int ocfs2_parse_options(struct super_block *sb,
1274                                char *options,
1275                                struct mount_options *mopt,
1276                                int is_remount)
1277 {
1278         int status, user_stack = 0;
1279         char *p;
1280         u32 tmp;
1281         int token, option;
1282         substring_t args[MAX_OPT_ARGS];
1283
1284         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1285
1286         mopt->commit_interval = 0;
1287         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1288         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1289         mopt->slot = OCFS2_INVALID_SLOT;
1290         mopt->localalloc_opt = -1;
1291         mopt->cluster_stack[0] = '\0';
1292         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1293         mopt->dir_resv_level = -1;
1294
1295         if (!options) {
1296                 status = 1;
1297                 goto bail;
1298         }
1299
1300         while ((p = strsep(&options, ",")) != NULL) {
1301                 if (!*p)
1302                         continue;
1303
1304                 token = match_token(p, tokens, args);
1305                 switch (token) {
1306                 case Opt_hb_local:
1307                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1308                         break;
1309                 case Opt_hb_none:
1310                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1311                         break;
1312                 case Opt_hb_global:
1313                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1314                         break;
1315                 case Opt_barrier:
1316                         if (match_int(&args[0], &option)) {
1317                                 status = 0;
1318                                 goto bail;
1319                         }
1320                         if (option)
1321                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1322                         else
1323                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1324                         break;
1325                 case Opt_intr:
1326                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1327                         break;
1328                 case Opt_nointr:
1329                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1330                         break;
1331                 case Opt_err_panic:
1332                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1333                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1334                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1335                         break;
1336                 case Opt_err_ro:
1337                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1338                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1339                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1340                         break;
1341                 case Opt_err_cont:
1342                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1343                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1344                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1345                         break;
1346                 case Opt_data_ordered:
1347                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1348                         break;
1349                 case Opt_data_writeback:
1350                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1351                         break;
1352                 case Opt_user_xattr:
1353                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1354                         break;
1355                 case Opt_nouser_xattr:
1356                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1357                         break;
1358                 case Opt_atime_quantum:
1359                         if (match_int(&args[0], &option)) {
1360                                 status = 0;
1361                                 goto bail;
1362                         }
1363                         if (option >= 0)
1364                                 mopt->atime_quantum = option;
1365                         break;
1366                 case Opt_slot:
1367                         if (match_int(&args[0], &option)) {
1368                                 status = 0;
1369                                 goto bail;
1370                         }
1371                         if (option)
1372                                 mopt->slot = (u16)option;
1373                         break;
1374                 case Opt_commit:
1375                         if (match_int(&args[0], &option)) {
1376                                 status = 0;
1377                                 goto bail;
1378                         }
1379                         if (option < 0)
1380                                 return 0;
1381                         if (option == 0)
1382                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1383                         mopt->commit_interval = HZ * option;
1384                         break;
1385                 case Opt_localalloc:
1386                         if (match_int(&args[0], &option)) {
1387                                 status = 0;
1388                                 goto bail;
1389                         }
1390                         if (option >= 0)
1391                                 mopt->localalloc_opt = option;
1392                         break;
1393                 case Opt_localflocks:
1394                         /*
1395                          * Changing this during remount could race
1396                          * flock() requests, or "unbalance" existing
1397                          * ones (e.g., a lock is taken in one mode but
1398                          * dropped in the other). If users care enough
1399                          * to flip locking modes during remount, we
1400                          * could add a "local" flag to individual
1401                          * flock structures for proper tracking of
1402                          * state.
1403                          */
1404                         if (!is_remount)
1405                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1406                         break;
1407                 case Opt_stack:
1408                         /* Check both that the option we were passed
1409                          * is of the right length and that it is a proper
1410                          * string of the right length.
1411                          */
1412                         if (((args[0].to - args[0].from) !=
1413                              OCFS2_STACK_LABEL_LEN) ||
1414                             (strnlen(args[0].from,
1415                                      OCFS2_STACK_LABEL_LEN) !=
1416                              OCFS2_STACK_LABEL_LEN)) {
1417                                 mlog(ML_ERROR,
1418                                      "Invalid cluster_stack option\n");
1419                                 status = 0;
1420                                 goto bail;
1421                         }
1422                         memcpy(mopt->cluster_stack, args[0].from,
1423                                OCFS2_STACK_LABEL_LEN);
1424                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1425                         /*
1426                          * Open code the memcmp here as we don't have
1427                          * an osb to pass to
1428                          * ocfs2_userspace_stack().
1429                          */
1430                         if (memcmp(mopt->cluster_stack,
1431                                    OCFS2_CLASSIC_CLUSTER_STACK,
1432                                    OCFS2_STACK_LABEL_LEN))
1433                                 user_stack = 1;
1434                         break;
1435                 case Opt_inode64:
1436                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1437                         break;
1438                 case Opt_usrquota:
1439                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1440                         break;
1441                 case Opt_grpquota:
1442                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1443                         break;
1444                 case Opt_coherency_buffered:
1445                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1446                         break;
1447                 case Opt_coherency_full:
1448                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1449                         break;
1450                 case Opt_acl:
1451                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1452                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1453                         break;
1454                 case Opt_noacl:
1455                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1456                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1457                         break;
1458                 case Opt_resv_level:
1459                         if (is_remount)
1460                                 break;
1461                         if (match_int(&args[0], &option)) {
1462                                 status = 0;
1463                                 goto bail;
1464                         }
1465                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1466                             option < OCFS2_MAX_RESV_LEVEL)
1467                                 mopt->resv_level = option;
1468                         break;
1469                 case Opt_dir_resv_level:
1470                         if (is_remount)
1471                                 break;
1472                         if (match_int(&args[0], &option)) {
1473                                 status = 0;
1474                                 goto bail;
1475                         }
1476                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1477                             option < OCFS2_MAX_RESV_LEVEL)
1478                                 mopt->dir_resv_level = option;
1479                         break;
1480                 case Opt_journal_async_commit:
1481                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1482                         break;
1483                 default:
1484                         mlog(ML_ERROR,
1485                              "Unrecognized mount option \"%s\" "
1486                              "or missing value\n", p);
1487                         status = 0;
1488                         goto bail;
1489                 }
1490         }
1491
1492         if (user_stack == 0) {
1493                 /* Ensure only one heartbeat mode */
1494                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1495                                          OCFS2_MOUNT_HB_GLOBAL |
1496                                          OCFS2_MOUNT_HB_NONE);
1497                 if (hweight32(tmp) != 1) {
1498                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1499                         status = 0;
1500                         goto bail;
1501                 }
1502         }
1503
1504         status = 1;
1505
1506 bail:
1507         return status;
1508 }
1509
1510 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1511 {
1512         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1513         unsigned long opts = osb->s_mount_opt;
1514         unsigned int local_alloc_megs;
1515
1516         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1517                 seq_printf(s, ",_netdev");
1518                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1519                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1520                 else
1521                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1522         } else
1523                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1524
1525         if (opts & OCFS2_MOUNT_NOINTR)
1526                 seq_printf(s, ",nointr");
1527
1528         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1529                 seq_printf(s, ",data=writeback");
1530         else
1531                 seq_printf(s, ",data=ordered");
1532
1533         if (opts & OCFS2_MOUNT_BARRIER)
1534                 seq_printf(s, ",barrier=1");
1535
1536         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1537                 seq_printf(s, ",errors=panic");
1538         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1539                 seq_printf(s, ",errors=continue");
1540         else
1541                 seq_printf(s, ",errors=remount-ro");
1542
1543         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1544                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1545
1546         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1547
1548         if (osb->osb_commit_interval)
1549                 seq_printf(s, ",commit=%u",
1550                            (unsigned) (osb->osb_commit_interval / HZ));
1551
1552         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1553         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1554                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1555
1556         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1557                 seq_printf(s, ",localflocks,");
1558
1559         if (osb->osb_cluster_stack[0])
1560                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1561                                   OCFS2_STACK_LABEL_LEN);
1562         if (opts & OCFS2_MOUNT_USRQUOTA)
1563                 seq_printf(s, ",usrquota");
1564         if (opts & OCFS2_MOUNT_GRPQUOTA)
1565                 seq_printf(s, ",grpquota");
1566
1567         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1568                 seq_printf(s, ",coherency=buffered");
1569         else
1570                 seq_printf(s, ",coherency=full");
1571
1572         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1573                 seq_printf(s, ",nouser_xattr");
1574         else
1575                 seq_printf(s, ",user_xattr");
1576
1577         if (opts & OCFS2_MOUNT_INODE64)
1578                 seq_printf(s, ",inode64");
1579
1580         if (opts & OCFS2_MOUNT_POSIX_ACL)
1581                 seq_printf(s, ",acl");
1582         else
1583                 seq_printf(s, ",noacl");
1584
1585         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1586                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1587
1588         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1589                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1590
1591         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1592                 seq_printf(s, ",journal_async_commit");
1593
1594         return 0;
1595 }
1596
1597 static int __init ocfs2_init(void)
1598 {
1599         int status;
1600
1601         status = init_ocfs2_uptodate_cache();
1602         if (status < 0)
1603                 goto out1;
1604
1605         status = ocfs2_initialize_mem_caches();
1606         if (status < 0)
1607                 goto out2;
1608
1609         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1610         if (!ocfs2_debugfs_root) {
1611                 status = -ENOMEM;
1612                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1613                 goto out3;
1614         }
1615
1616         ocfs2_set_locking_protocol();
1617
1618         status = register_quota_format(&ocfs2_quota_format);
1619         if (status < 0)
1620                 goto out3;
1621         status = register_filesystem(&ocfs2_fs_type);
1622         if (!status)
1623                 return 0;
1624
1625         unregister_quota_format(&ocfs2_quota_format);
1626 out3:
1627         debugfs_remove(ocfs2_debugfs_root);
1628         ocfs2_free_mem_caches();
1629 out2:
1630         exit_ocfs2_uptodate_cache();
1631 out1:
1632         mlog_errno(status);
1633         return status;
1634 }
1635
1636 static void __exit ocfs2_exit(void)
1637 {
1638         unregister_quota_format(&ocfs2_quota_format);
1639
1640         debugfs_remove(ocfs2_debugfs_root);
1641
1642         ocfs2_free_mem_caches();
1643
1644         unregister_filesystem(&ocfs2_fs_type);
1645
1646         exit_ocfs2_uptodate_cache();
1647 }
1648
1649 static void ocfs2_put_super(struct super_block *sb)
1650 {
1651         trace_ocfs2_put_super(sb);
1652
1653         ocfs2_sync_blockdev(sb);
1654         ocfs2_dismount_volume(sb, 0);
1655         ocfs2_filecheck_remove_sysfs(sb);
1656 }
1657
1658 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1659 {
1660         struct ocfs2_super *osb;
1661         u32 numbits, freebits;
1662         int status;
1663         struct ocfs2_dinode *bm_lock;
1664         struct buffer_head *bh = NULL;
1665         struct inode *inode = NULL;
1666
1667         trace_ocfs2_statfs(dentry->d_sb, buf);
1668
1669         osb = OCFS2_SB(dentry->d_sb);
1670
1671         inode = ocfs2_get_system_file_inode(osb,
1672                                             GLOBAL_BITMAP_SYSTEM_INODE,
1673                                             OCFS2_INVALID_SLOT);
1674         if (!inode) {
1675                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1676                 status = -EIO;
1677                 goto bail;
1678         }
1679
1680         status = ocfs2_inode_lock(inode, &bh, 0);
1681         if (status < 0) {
1682                 mlog_errno(status);
1683                 goto bail;
1684         }
1685
1686         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1687
1688         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1689         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1690
1691         buf->f_type = OCFS2_SUPER_MAGIC;
1692         buf->f_bsize = dentry->d_sb->s_blocksize;
1693         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1694         buf->f_blocks = ((sector_t) numbits) *
1695                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1696         buf->f_bfree = ((sector_t) freebits) *
1697                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1698         buf->f_bavail = buf->f_bfree;
1699         buf->f_files = numbits;
1700         buf->f_ffree = freebits;
1701         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1702                                 & 0xFFFFFFFFUL;
1703         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1704                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1705
1706         brelse(bh);
1707
1708         ocfs2_inode_unlock(inode, 0);
1709         status = 0;
1710 bail:
1711         iput(inode);
1712
1713         if (status)
1714                 mlog_errno(status);
1715
1716         return status;
1717 }
1718
1719 static void ocfs2_inode_init_once(void *data)
1720 {
1721         struct ocfs2_inode_info *oi = data;
1722
1723         oi->ip_flags = 0;
1724         oi->ip_open_count = 0;
1725         spin_lock_init(&oi->ip_lock);
1726         ocfs2_extent_map_init(&oi->vfs_inode);
1727         INIT_LIST_HEAD(&oi->ip_io_markers);
1728         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1729         oi->ip_dir_start_lookup = 0;
1730         init_rwsem(&oi->ip_alloc_sem);
1731         init_rwsem(&oi->ip_xattr_sem);
1732         mutex_init(&oi->ip_io_mutex);
1733
1734         oi->ip_blkno = 0ULL;
1735         oi->ip_clusters = 0;
1736         oi->ip_next_orphan = NULL;
1737
1738         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1739
1740         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1741         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1742         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1743
1744         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1745                                   &ocfs2_inode_caching_ops);
1746
1747         inode_init_once(&oi->vfs_inode);
1748 }
1749
1750 static int ocfs2_initialize_mem_caches(void)
1751 {
1752         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1753                                        sizeof(struct ocfs2_inode_info),
1754                                        0,
1755                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1756                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1757                                        ocfs2_inode_init_once);
1758         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1759                                         sizeof(struct ocfs2_dquot),
1760                                         0,
1761                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1762                                                 SLAB_MEM_SPREAD),
1763                                         NULL);
1764         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1765                                         sizeof(struct ocfs2_quota_chunk),
1766                                         0,
1767                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1768                                         NULL);
1769         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1770             !ocfs2_qf_chunk_cachep) {
1771                 if (ocfs2_inode_cachep)
1772                         kmem_cache_destroy(ocfs2_inode_cachep);
1773                 if (ocfs2_dquot_cachep)
1774                         kmem_cache_destroy(ocfs2_dquot_cachep);
1775                 if (ocfs2_qf_chunk_cachep)
1776                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1777                 return -ENOMEM;
1778         }
1779
1780         return 0;
1781 }
1782
1783 static void ocfs2_free_mem_caches(void)
1784 {
1785         /*
1786          * Make sure all delayed rcu free inodes are flushed before we
1787          * destroy cache.
1788          */
1789         rcu_barrier();
1790         if (ocfs2_inode_cachep)
1791                 kmem_cache_destroy(ocfs2_inode_cachep);
1792         ocfs2_inode_cachep = NULL;
1793
1794         if (ocfs2_dquot_cachep)
1795                 kmem_cache_destroy(ocfs2_dquot_cachep);
1796         ocfs2_dquot_cachep = NULL;
1797
1798         if (ocfs2_qf_chunk_cachep)
1799                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1800         ocfs2_qf_chunk_cachep = NULL;
1801 }
1802
1803 static int ocfs2_get_sector(struct super_block *sb,
1804                             struct buffer_head **bh,
1805                             int block,
1806                             int sect_size)
1807 {
1808         if (!sb_set_blocksize(sb, sect_size)) {
1809                 mlog(ML_ERROR, "unable to set blocksize\n");
1810                 return -EIO;
1811         }
1812
1813         *bh = sb_getblk(sb, block);
1814         if (!*bh) {
1815                 mlog_errno(-ENOMEM);
1816                 return -ENOMEM;
1817         }
1818         lock_buffer(*bh);
1819         if (!buffer_dirty(*bh))
1820                 clear_buffer_uptodate(*bh);
1821         unlock_buffer(*bh);
1822         ll_rw_block(REQ_OP_READ, 0, 1, bh);
1823         wait_on_buffer(*bh);
1824         if (!buffer_uptodate(*bh)) {
1825                 mlog_errno(-EIO);
1826                 brelse(*bh);
1827                 *bh = NULL;
1828                 return -EIO;
1829         }
1830
1831         return 0;
1832 }
1833
1834 static int ocfs2_mount_volume(struct super_block *sb)
1835 {
1836         int status = 0;
1837         int unlock_super = 0;
1838         struct ocfs2_super *osb = OCFS2_SB(sb);
1839
1840         if (ocfs2_is_hard_readonly(osb))
1841                 goto leave;
1842
1843         status = ocfs2_dlm_init(osb);
1844         if (status < 0) {
1845                 mlog_errno(status);
1846                 goto leave;
1847         }
1848
1849         status = ocfs2_super_lock(osb, 1);
1850         if (status < 0) {
1851                 mlog_errno(status);
1852                 goto leave;
1853         }
1854         unlock_super = 1;
1855
1856         /* This will load up the node map and add ourselves to it. */
1857         status = ocfs2_find_slot(osb);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto leave;
1861         }
1862
1863         /* load all node-local system inodes */
1864         status = ocfs2_init_local_system_inodes(osb);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 goto leave;
1868         }
1869
1870         status = ocfs2_check_volume(osb);
1871         if (status < 0) {
1872                 mlog_errno(status);
1873                 goto leave;
1874         }
1875
1876         status = ocfs2_truncate_log_init(osb);
1877         if (status < 0)
1878                 mlog_errno(status);
1879
1880 leave:
1881         if (unlock_super)
1882                 ocfs2_super_unlock(osb, 1);
1883
1884         return status;
1885 }
1886
1887 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1888 {
1889         int tmp, hangup_needed = 0;
1890         struct ocfs2_super *osb = NULL;
1891         char nodestr[12];
1892
1893         trace_ocfs2_dismount_volume(sb);
1894
1895         BUG_ON(!sb);
1896         osb = OCFS2_SB(sb);
1897         BUG_ON(!osb);
1898
1899         debugfs_remove(osb->osb_ctxt);
1900
1901         /* Orphan scan should be stopped as early as possible */
1902         ocfs2_orphan_scan_stop(osb);
1903
1904         ocfs2_disable_quotas(osb);
1905
1906         /* All dquots should be freed by now */
1907         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1908         /* Wait for worker to be done with the work structure in osb */
1909         cancel_work_sync(&osb->dquot_drop_work);
1910
1911         ocfs2_shutdown_local_alloc(osb);
1912
1913         ocfs2_truncate_log_shutdown(osb);
1914
1915         /* This will disable recovery and flush any recovery work. */
1916         ocfs2_recovery_exit(osb);
1917
1918         ocfs2_journal_shutdown(osb);
1919
1920         ocfs2_sync_blockdev(sb);
1921
1922         ocfs2_purge_refcount_trees(osb);
1923
1924         /* No cluster connection means we've failed during mount, so skip
1925          * all the steps which depended on that to complete. */
1926         if (osb->cconn) {
1927                 tmp = ocfs2_super_lock(osb, 1);
1928                 if (tmp < 0) {
1929                         mlog_errno(tmp);
1930                         return;
1931                 }
1932         }
1933
1934         if (osb->slot_num != OCFS2_INVALID_SLOT)
1935                 ocfs2_put_slot(osb);
1936
1937         if (osb->cconn)
1938                 ocfs2_super_unlock(osb, 1);
1939
1940         ocfs2_release_system_inodes(osb);
1941
1942         /*
1943          * If we're dismounting due to mount error, mount.ocfs2 will clean
1944          * up heartbeat.  If we're a local mount, there is no heartbeat.
1945          * If we failed before we got a uuid_str yet, we can't stop
1946          * heartbeat.  Otherwise, do it.
1947          */
1948         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1949             !ocfs2_is_hard_readonly(osb))
1950                 hangup_needed = 1;
1951
1952         if (osb->cconn)
1953                 ocfs2_dlm_shutdown(osb, hangup_needed);
1954
1955         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1956         debugfs_remove(osb->osb_debug_root);
1957
1958         if (hangup_needed)
1959                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1960
1961         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1962
1963         if (ocfs2_mount_local(osb))
1964                 snprintf(nodestr, sizeof(nodestr), "local");
1965         else
1966                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1967
1968         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1969                osb->dev_str, nodestr);
1970
1971         ocfs2_delete_osb(osb);
1972         kfree(osb);
1973         sb->s_dev = 0;
1974         sb->s_fs_info = NULL;
1975 }
1976
1977 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1978                                 unsigned uuid_bytes)
1979 {
1980         int i, ret;
1981         char *ptr;
1982
1983         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1984
1985         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1986         if (osb->uuid_str == NULL)
1987                 return -ENOMEM;
1988
1989         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1990                 /* print with null */
1991                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1992                 if (ret != 2) /* drop super cleans up */
1993                         return -EINVAL;
1994                 /* then only advance past the last char */
1995                 ptr += 2;
1996         }
1997
1998         return 0;
1999 }
2000
2001 /* Make sure entire volume is addressable by our journal.  Requires
2002    osb_clusters_at_boot to be valid and for the journal to have been
2003    initialized by ocfs2_journal_init(). */
2004 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2005 {
2006         int status = 0;
2007         u64 max_block =
2008                 ocfs2_clusters_to_blocks(osb->sb,
2009                                          osb->osb_clusters_at_boot) - 1;
2010
2011         /* 32-bit block number is always OK. */
2012         if (max_block <= (u32)~0ULL)
2013                 goto out;
2014
2015         /* Volume is "huge", so see if our journal is new enough to
2016            support it. */
2017         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2018                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2019               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2020                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2021                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2022                      "Enable the 'block64' journal option with tunefs.ocfs2");
2023                 status = -EFBIG;
2024                 goto out;
2025         }
2026
2027  out:
2028         return status;
2029 }
2030
2031 static int ocfs2_initialize_super(struct super_block *sb,
2032                                   struct buffer_head *bh,
2033                                   int sector_size,
2034                                   struct ocfs2_blockcheck_stats *stats)
2035 {
2036         int status;
2037         int i, cbits, bbits;
2038         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2039         struct inode *inode = NULL;
2040         struct ocfs2_journal *journal;
2041         struct ocfs2_super *osb;
2042         u64 total_blocks;
2043
2044         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2045         if (!osb) {
2046                 status = -ENOMEM;
2047                 mlog_errno(status);
2048                 goto bail;
2049         }
2050
2051         sb->s_fs_info = osb;
2052         sb->s_op = &ocfs2_sops;
2053         sb->s_d_op = &ocfs2_dentry_ops;
2054         sb->s_export_op = &ocfs2_export_ops;
2055         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2056         sb->dq_op = &ocfs2_quota_operations;
2057         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2058         sb->s_xattr = ocfs2_xattr_handlers;
2059         sb->s_time_gran = 1;
2060         sb->s_flags |= MS_NOATIME;
2061         /* this is needed to support O_LARGEFILE */
2062         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2063         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2064         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2065         memcpy(sb->s_uuid, di->id2.i_super.s_uuid,
2066                sizeof(di->id2.i_super.s_uuid));
2067
2068         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2069
2070         for (i = 0; i < 3; i++)
2071                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2072         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2073
2074         osb->sb = sb;
2075         osb->s_sectsize_bits = blksize_bits(sector_size);
2076         BUG_ON(!osb->s_sectsize_bits);
2077
2078         spin_lock_init(&osb->dc_task_lock);
2079         init_waitqueue_head(&osb->dc_event);
2080         osb->dc_work_sequence = 0;
2081         osb->dc_wake_sequence = 0;
2082         INIT_LIST_HEAD(&osb->blocked_lock_list);
2083         osb->blocked_lock_count = 0;
2084         spin_lock_init(&osb->osb_lock);
2085         spin_lock_init(&osb->osb_xattr_lock);
2086         ocfs2_init_steal_slots(osb);
2087
2088         mutex_init(&osb->system_file_mutex);
2089
2090         atomic_set(&osb->alloc_stats.moves, 0);
2091         atomic_set(&osb->alloc_stats.local_data, 0);
2092         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2093         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2094         atomic_set(&osb->alloc_stats.bg_extends, 0);
2095
2096         /* Copy the blockcheck stats from the superblock probe */
2097         osb->osb_ecc_stats = *stats;
2098
2099         ocfs2_init_node_maps(osb);
2100
2101         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2102                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2103
2104         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2105         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2106                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2107                      osb->max_slots);
2108                 status = -EINVAL;
2109                 goto bail;
2110         }
2111
2112         ocfs2_orphan_scan_init(osb);
2113
2114         status = ocfs2_recovery_init(osb);
2115         if (status) {
2116                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2117                 mlog_errno(status);
2118                 goto bail;
2119         }
2120
2121         init_waitqueue_head(&osb->checkpoint_event);
2122
2123         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2124
2125         osb->slot_num = OCFS2_INVALID_SLOT;
2126
2127         osb->s_xattr_inline_size = le16_to_cpu(
2128                                         di->id2.i_super.s_xattr_inline_size);
2129
2130         osb->local_alloc_state = OCFS2_LA_UNUSED;
2131         osb->local_alloc_bh = NULL;
2132         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2133
2134         init_waitqueue_head(&osb->osb_mount_event);
2135
2136         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2137         if (status) {
2138                 mlog_errno(status);
2139                 goto bail;
2140         }
2141
2142         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2143         if (!osb->vol_label) {
2144                 mlog(ML_ERROR, "unable to alloc vol label\n");
2145                 status = -ENOMEM;
2146                 goto bail;
2147         }
2148
2149         osb->slot_recovery_generations =
2150                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2151                         GFP_KERNEL);
2152         if (!osb->slot_recovery_generations) {
2153                 status = -ENOMEM;
2154                 mlog_errno(status);
2155                 goto bail;
2156         }
2157
2158         init_waitqueue_head(&osb->osb_wipe_event);
2159         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2160                                         sizeof(*osb->osb_orphan_wipes),
2161                                         GFP_KERNEL);
2162         if (!osb->osb_orphan_wipes) {
2163                 status = -ENOMEM;
2164                 mlog_errno(status);
2165                 goto bail;
2166         }
2167
2168         osb->osb_rf_lock_tree = RB_ROOT;
2169
2170         osb->s_feature_compat =
2171                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2172         osb->s_feature_ro_compat =
2173                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2174         osb->s_feature_incompat =
2175                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2176
2177         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2178                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2179                      "optional features (%x).\n", i);
2180                 status = -EINVAL;
2181                 goto bail;
2182         }
2183         if (!(osb->sb->s_flags & MS_RDONLY) &&
2184             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2185                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2186                      "unsupported optional features (%x).\n", i);
2187                 status = -EINVAL;
2188                 goto bail;
2189         }
2190
2191         if (ocfs2_clusterinfo_valid(osb)) {
2192                 /*
2193                  * ci_stack and ci_cluster in ocfs2_cluster_info may not be null
2194                  * terminated, so make sure no overflow happens here by using
2195                  * memcpy. Destination strings will always be null terminated
2196                  * because osb is allocated using kzalloc.
2197                  */
2198                 osb->osb_stackflags =
2199                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2200                 memcpy(osb->osb_cluster_stack,
2201                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2202                        OCFS2_STACK_LABEL_LEN);
2203                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2204                         mlog(ML_ERROR,
2205                              "couldn't mount because of an invalid "
2206                              "cluster stack label (%s) \n",
2207                              osb->osb_cluster_stack);
2208                         status = -EINVAL;
2209                         goto bail;
2210                 }
2211                 memcpy(osb->osb_cluster_name,
2212                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2213                         OCFS2_CLUSTER_NAME_LEN);
2214         } else {
2215                 /* The empty string is identical with classic tools that
2216                  * don't know about s_cluster_info. */
2217                 osb->osb_cluster_stack[0] = '\0';
2218         }
2219
2220         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2221
2222         /* FIXME
2223          * This should be done in ocfs2_journal_init(), but unknown
2224          * ordering issues will cause the filesystem to crash.
2225          * If anyone wants to figure out what part of the code
2226          * refers to osb->journal before ocfs2_journal_init() is run,
2227          * be my guest.
2228          */
2229         /* initialize our journal structure */
2230
2231         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2232         if (!journal) {
2233                 mlog(ML_ERROR, "unable to alloc journal\n");
2234                 status = -ENOMEM;
2235                 goto bail;
2236         }
2237         osb->journal = journal;
2238         journal->j_osb = osb;
2239
2240         atomic_set(&journal->j_num_trans, 0);
2241         init_rwsem(&journal->j_trans_barrier);
2242         init_waitqueue_head(&journal->j_checkpointed);
2243         spin_lock_init(&journal->j_lock);
2244         journal->j_trans_id = (unsigned long) 1;
2245         INIT_LIST_HEAD(&journal->j_la_cleanups);
2246         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2247         journal->j_state = OCFS2_JOURNAL_FREE;
2248
2249         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2250         init_llist_head(&osb->dquot_drop_list);
2251
2252         /* get some pseudo constants for clustersize bits */
2253         osb->s_clustersize_bits =
2254                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2255         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2256
2257         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2258             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2259                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2260                      osb->s_clustersize);
2261                 status = -EINVAL;
2262                 goto bail;
2263         }
2264
2265         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2266                                                 le32_to_cpu(di->i_clusters));
2267
2268         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2269                                            total_blocks);
2270         if (status) {
2271                 mlog(ML_ERROR, "Volume too large "
2272                      "to mount safely on this system");
2273                 status = -EFBIG;
2274                 goto bail;
2275         }
2276
2277         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2278                                  sizeof(di->id2.i_super.s_uuid))) {
2279                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2280                 status = -ENOMEM;
2281                 goto bail;
2282         }
2283
2284         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2285                 OCFS2_MAX_VOL_LABEL_LEN);
2286         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2287         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2288         osb->first_cluster_group_blkno =
2289                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2290         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2291         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2292         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2293                                      (unsigned long long)osb->root_blkno,
2294                                      (unsigned long long)osb->system_dir_blkno,
2295                                      osb->s_clustersize_bits);
2296
2297         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2298         if (!osb->osb_dlm_debug) {
2299                 status = -ENOMEM;
2300                 mlog_errno(status);
2301                 goto bail;
2302         }
2303
2304         atomic_set(&osb->vol_state, VOLUME_INIT);
2305
2306         /* load root, system_dir, and all global system inodes */
2307         status = ocfs2_init_global_system_inodes(osb);
2308         if (status < 0) {
2309                 mlog_errno(status);
2310                 goto bail;
2311         }
2312
2313         /*
2314          * global bitmap
2315          */
2316         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2317                                             OCFS2_INVALID_SLOT);
2318         if (!inode) {
2319                 status = -EINVAL;
2320                 mlog_errno(status);
2321                 goto bail;
2322         }
2323
2324         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2325         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2326         iput(inode);
2327
2328         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2329                                  osb->s_feature_incompat) * 8;
2330
2331         status = ocfs2_init_slot_info(osb);
2332         if (status < 0) {
2333                 mlog_errno(status);
2334                 goto bail;
2335         }
2336         cleancache_init_shared_fs(sb);
2337
2338         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2339         if (!osb->ocfs2_wq) {
2340                 status = -ENOMEM;
2341                 mlog_errno(status);
2342         }
2343
2344 bail:
2345         return status;
2346 }
2347
2348 /*
2349  * will return: -EAGAIN if it is ok to keep searching for superblocks
2350  *              -EINVAL if there is a bad superblock
2351  *              0 on success
2352  */
2353 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2354                                struct buffer_head *bh,
2355                                u32 blksz,
2356                                struct ocfs2_blockcheck_stats *stats)
2357 {
2358         int status = -EAGAIN;
2359
2360         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2361                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2362                 /* We have to do a raw check of the feature here */
2363                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2364                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2365                         status = ocfs2_block_check_validate(bh->b_data,
2366                                                             bh->b_size,
2367                                                             &di->i_check,
2368                                                             stats);
2369                         if (status)
2370                                 goto out;
2371                 }
2372                 status = -EINVAL;
2373                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2374                         mlog(ML_ERROR, "found superblock with incorrect block "
2375                              "size: found %u, should be %u\n",
2376                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2377                                blksz);
2378                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2379                            OCFS2_MAJOR_REV_LEVEL ||
2380                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2381                            OCFS2_MINOR_REV_LEVEL) {
2382                         mlog(ML_ERROR, "found superblock with bad version: "
2383                              "found %u.%u, should be %u.%u\n",
2384                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2385                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2386                              OCFS2_MAJOR_REV_LEVEL,
2387                              OCFS2_MINOR_REV_LEVEL);
2388                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2389                         mlog(ML_ERROR, "bad block number on superblock: "
2390                              "found %llu, should be %llu\n",
2391                              (unsigned long long)le64_to_cpu(di->i_blkno),
2392                              (unsigned long long)bh->b_blocknr);
2393                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2394                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2395                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2396                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2397                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2398                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2399                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2400                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2401                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2402                         mlog(ML_ERROR,
2403                              "Superblock slots found greater than file system "
2404                              "maximum: found %u, max %u\n",
2405                              le16_to_cpu(di->id2.i_super.s_max_slots),
2406                              OCFS2_MAX_SLOTS);
2407                 } else {
2408                         /* found it! */
2409                         status = 0;
2410                 }
2411         }
2412
2413 out:
2414         if (status && status != -EAGAIN)
2415                 mlog_errno(status);
2416         return status;
2417 }
2418
2419 static int ocfs2_check_volume(struct ocfs2_super *osb)
2420 {
2421         int status;
2422         int dirty;
2423         int local;
2424         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2425                                                   * recover
2426                                                   * ourselves. */
2427
2428         /* Init our journal object. */
2429         status = ocfs2_journal_init(osb->journal, &dirty);
2430         if (status < 0) {
2431                 mlog(ML_ERROR, "Could not initialize journal!\n");
2432                 goto finally;
2433         }
2434
2435         /* Now that journal has been initialized, check to make sure
2436            entire volume is addressable. */
2437         status = ocfs2_journal_addressable(osb);
2438         if (status)
2439                 goto finally;
2440
2441         /* If the journal was unmounted cleanly then we don't want to
2442          * recover anything. Otherwise, journal_load will do that
2443          * dirty work for us :) */
2444         if (!dirty) {
2445                 status = ocfs2_journal_wipe(osb->journal, 0);
2446                 if (status < 0) {
2447                         mlog_errno(status);
2448                         goto finally;
2449                 }
2450         } else {
2451                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2452                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2453         }
2454
2455         local = ocfs2_mount_local(osb);
2456
2457         /* will play back anything left in the journal. */
2458         status = ocfs2_journal_load(osb->journal, local, dirty);
2459         if (status < 0) {
2460                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2461                 goto finally;
2462         }
2463
2464         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2465                 jbd2_journal_set_features(osb->journal->j_journal,
2466                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2467                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2468         else
2469                 jbd2_journal_clear_features(osb->journal->j_journal,
2470                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2471                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2472
2473         if (dirty) {
2474                 /* recover my local alloc if we didn't unmount cleanly. */
2475                 status = ocfs2_begin_local_alloc_recovery(osb,
2476                                                           osb->slot_num,
2477                                                           &local_alloc);
2478                 if (status < 0) {
2479                         mlog_errno(status);
2480                         goto finally;
2481                 }
2482                 /* we complete the recovery process after we've marked
2483                  * ourselves as mounted. */
2484         }
2485
2486         status = ocfs2_load_local_alloc(osb);
2487         if (status < 0) {
2488                 mlog_errno(status);
2489                 goto finally;
2490         }
2491
2492         if (dirty) {
2493                 /* Recovery will be completed after we've mounted the
2494                  * rest of the volume. */
2495                 osb->dirty = 1;
2496                 osb->local_alloc_copy = local_alloc;
2497                 local_alloc = NULL;
2498         }
2499
2500         /* go through each journal, trylock it and if you get the
2501          * lock, and it's marked as dirty, set the bit in the recover
2502          * map and launch a recovery thread for it. */
2503         status = ocfs2_mark_dead_nodes(osb);
2504         if (status < 0) {
2505                 mlog_errno(status);
2506                 goto finally;
2507         }
2508
2509         status = ocfs2_compute_replay_slots(osb);
2510         if (status < 0)
2511                 mlog_errno(status);
2512
2513 finally:
2514         kfree(local_alloc);
2515
2516         if (status)
2517                 mlog_errno(status);
2518         return status;
2519 }
2520
2521 /*
2522  * The routine gets called from dismount or close whenever a dismount on
2523  * volume is requested and the osb open count becomes 1.
2524  * It will remove the osb from the global list and also free up all the
2525  * initialized resources and fileobject.
2526  */
2527 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2528 {
2529         /* This function assumes that the caller has the main osb resource */
2530
2531         /* ocfs2_initializer_super have already created this workqueue */
2532         if (osb->ocfs2_wq) {
2533                 flush_workqueue(osb->ocfs2_wq);
2534                 destroy_workqueue(osb->ocfs2_wq);
2535         }
2536
2537         ocfs2_free_slot_info(osb);
2538
2539         kfree(osb->osb_orphan_wipes);
2540         kfree(osb->slot_recovery_generations);
2541         /* FIXME
2542          * This belongs in journal shutdown, but because we have to
2543          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2544          * we free it here.
2545          */
2546         kfree(osb->journal);
2547         kfree(osb->local_alloc_copy);
2548         kfree(osb->uuid_str);
2549         kfree(osb->vol_label);
2550         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2551         memset(osb, 0, sizeof(struct ocfs2_super));
2552 }
2553
2554 /* Depending on the mount option passed, perform one of the following:
2555  * Put OCFS2 into a readonly state (default)
2556  * Return EIO so that only the process errs
2557  * Fix the error as if fsck.ocfs2 -y
2558  * panic
2559  */
2560 static int ocfs2_handle_error(struct super_block *sb)
2561 {
2562         struct ocfs2_super *osb = OCFS2_SB(sb);
2563         int rv = 0;
2564
2565         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2566         pr_crit("On-disk corruption discovered. "
2567                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2568
2569         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2570                 panic("OCFS2: (device %s): panic forced after error\n",
2571                       sb->s_id);
2572         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2573                 pr_crit("OCFS2: Returning error to the calling process.\n");
2574                 rv = -EIO;
2575         } else { /* default option */
2576                 rv = -EROFS;
2577                 if (sb->s_flags & MS_RDONLY &&
2578                                 (ocfs2_is_soft_readonly(osb) ||
2579                                  ocfs2_is_hard_readonly(osb)))
2580                         return rv;
2581
2582                 pr_crit("OCFS2: File system is now read-only.\n");
2583                 sb->s_flags |= MS_RDONLY;
2584                 ocfs2_set_ro_flag(osb, 0);
2585         }
2586
2587         return rv;
2588 }
2589
2590 int __ocfs2_error(struct super_block *sb, const char *function,
2591                   const char *fmt, ...)
2592 {
2593         struct va_format vaf;
2594         va_list args;
2595
2596         va_start(args, fmt);
2597         vaf.fmt = fmt;
2598         vaf.va = &args;
2599
2600         /* Not using mlog here because we want to show the actual
2601          * function the error came from. */
2602         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2603                sb->s_id, function, &vaf);
2604
2605         va_end(args);
2606
2607         return ocfs2_handle_error(sb);
2608 }
2609
2610 /* Handle critical errors. This is intentionally more drastic than
2611  * ocfs2_handle_error, so we only use for things like journal errors,
2612  * etc. */
2613 void __ocfs2_abort(struct super_block *sb, const char *function,
2614                    const char *fmt, ...)
2615 {
2616         struct va_format vaf;
2617         va_list args;
2618
2619         va_start(args, fmt);
2620
2621         vaf.fmt = fmt;
2622         vaf.va = &args;
2623
2624         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2625                sb->s_id, function, &vaf);
2626
2627         va_end(args);
2628
2629         /* We don't have the cluster support yet to go straight to
2630          * hard readonly in here. Until then, we want to keep
2631          * ocfs2_abort() so that we can at least mark critical
2632          * errors.
2633          *
2634          * TODO: This should abort the journal and alert other nodes
2635          * that our slot needs recovery. */
2636
2637         /* Force a panic(). This stinks, but it's better than letting
2638          * things continue without having a proper hard readonly
2639          * here. */
2640         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2641                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2642         ocfs2_handle_error(sb);
2643 }
2644
2645 /*
2646  * Void signal blockers, because in-kernel sigprocmask() only fails
2647  * when SIG_* is wrong.
2648  */
2649 void ocfs2_block_signals(sigset_t *oldset)
2650 {
2651         int rc;
2652         sigset_t blocked;
2653
2654         sigfillset(&blocked);
2655         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2656         BUG_ON(rc);
2657 }
2658
2659 void ocfs2_unblock_signals(sigset_t *oldset)
2660 {
2661         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2662         BUG_ON(rc);
2663 }
2664
2665 module_init(ocfs2_init);
2666 module_exit(ocfs2_exit);