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 #include <linux/signal.h>
  46
  47 #define CREATE_TRACE_POINTS
  48 #include "ocfs2_trace.h"
  49
  50 #include <cluster/masklog.h>
  51
  52 #include "ocfs2.h"
  53
  54 /* this should be the only file to include a version 1 header */
  55 #include "ocfs1_fs_compat.h"
  56
  57 #include "alloc.h"
  58 #include "aops.h"
  59 #include "blockcheck.h"
  60 #include "dlmglue.h"
  61 #include "export.h"
  62 #include "extent_map.h"
  63 #include "heartbeat.h"
  64 #include "inode.h"
  65 #include "journal.h"
  66 #include "localalloc.h"
  67 #include "namei.h"
  68 #include "slot_map.h"
  69 #include "super.h"
  70 #include "sysfile.h"
  71 #include "uptodate.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78 #include "filecheck.h"
  79
  80 static struct kmem_cache *ocfs2_inode_cachep;
  81 struct kmem_cache *ocfs2_dquot_cachep;
  82 struct kmem_cache *ocfs2_qf_chunk_cachep;
  83
  84 static struct dentry *ocfs2_debugfs_root;
  85
  86 MODULE_AUTHOR("Oracle");
  87 MODULE_LICENSE("GPL");
  88 MODULE_DESCRIPTION("OCFS2 cluster file system");
  89
  90 struct mount_options
  91 {
  92         unsigned long   commit_interval;
  93         unsigned long   mount_opt;
  94         unsigned int    atime_quantum;
  95         unsigned short  slot;
  96         int             localalloc_opt;
  97         unsigned int    resv_level;
  98         int             dir_resv_level;
  99         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 100 };
 101
 102 static int ocfs2_parse_options(struct super_block *sb, char *options,
 103                                struct mount_options *mopt,
 104                                int is_remount);
 105 static int ocfs2_check_set_options(struct super_block *sb,
 106                                    struct mount_options *options);
 107 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 108 static void ocfs2_put_super(struct super_block *sb);
 109 static int ocfs2_mount_volume(struct super_block *sb);
 110 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 111 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 112 static int ocfs2_initialize_mem_caches(void);
 113 static void ocfs2_free_mem_caches(void);
 114 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 115
 116 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 117
 118 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 119
 120 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 121 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 122 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 123 static int ocfs2_check_volume(struct ocfs2_super *osb);
 124 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 125                                struct buffer_head *bh,
 126                                u32 sectsize,
 127                                struct ocfs2_blockcheck_stats *stats);
 128 static int ocfs2_initialize_super(struct super_block *sb,
 129                                   struct buffer_head *bh,
 130                                   int sector_size,
 131                                   struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_get_sector(struct super_block *sb,
 133                             struct buffer_head **bh,
 134                             int block,
 135                             int sect_size);
 136 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 137 static void ocfs2_destroy_inode(struct inode *inode);
 138 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 139 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 140 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 141
 142 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 143 {
 144         return OCFS2_I(inode)->i_dquot;
 145 }
 146
 147 static const struct super_operations ocfs2_sops = {
 148         .statfs         = ocfs2_statfs,
 149         .alloc_inode    = ocfs2_alloc_inode,
 150         .destroy_inode  = ocfs2_destroy_inode,
 151         .drop_inode     = ocfs2_drop_inode,
 152         .evict_inode    = ocfs2_evict_inode,
 153         .sync_fs        = ocfs2_sync_fs,
 154         .put_super      = ocfs2_put_super,
 155         .remount_fs     = ocfs2_remount,
 156         .show_options   = ocfs2_show_options,
 157         .quota_read     = ocfs2_quota_read,
 158         .quota_write    = ocfs2_quota_write,
 159         .get_dquots     = ocfs2_get_dquots,
 160 };
 161
 162 enum {
 163         Opt_barrier,
 164         Opt_err_panic,
 165         Opt_err_ro,
 166         Opt_intr,
 167         Opt_nointr,
 168         Opt_hb_none,
 169         Opt_hb_local,
 170         Opt_hb_global,
 171         Opt_data_ordered,
 172         Opt_data_writeback,
 173         Opt_atime_quantum,
 174         Opt_slot,
 175         Opt_commit,
 176         Opt_localalloc,
 177         Opt_localflocks,
 178         Opt_stack,
 179         Opt_user_xattr,
 180         Opt_nouser_xattr,
 181         Opt_inode64,
 182         Opt_acl,
 183         Opt_noacl,
 184         Opt_usrquota,
 185         Opt_grpquota,
 186         Opt_coherency_buffered,
 187         Opt_coherency_full,
 188         Opt_resv_level,
 189         Opt_dir_resv_level,
 190         Opt_journal_async_commit,
 191         Opt_err_cont,
 192         Opt_err,
 193 };
 194
 195 static const match_table_t tokens = {
 196         {Opt_barrier, "barrier=%u"},
 197         {Opt_err_panic, "errors=panic"},
 198         {Opt_err_ro, "errors=remount-ro"},
 199         {Opt_intr, "intr"},
 200         {Opt_nointr, "nointr"},
 201         {Opt_hb_none, OCFS2_HB_NONE},
 202         {Opt_hb_local, OCFS2_HB_LOCAL},
 203         {Opt_hb_global, OCFS2_HB_GLOBAL},
 204         {Opt_data_ordered, "data=ordered"},
 205         {Opt_data_writeback, "data=writeback"},
 206         {Opt_atime_quantum, "atime_quantum=%u"},
 207         {Opt_slot, "preferred_slot=%u"},
 208         {Opt_commit, "commit=%u"},
 209         {Opt_localalloc, "localalloc=%d"},
 210         {Opt_localflocks, "localflocks"},
 211         {Opt_stack, "cluster_stack=%s"},
 212         {Opt_user_xattr, "user_xattr"},
 213         {Opt_nouser_xattr, "nouser_xattr"},
 214         {Opt_inode64, "inode64"},
 215         {Opt_acl, "acl"},
 216         {Opt_noacl, "noacl"},
 217         {Opt_usrquota, "usrquota"},
 218         {Opt_grpquota, "grpquota"},
 219         {Opt_coherency_buffered, "coherency=buffered"},
 220         {Opt_coherency_full, "coherency=full"},
 221         {Opt_resv_level, "resv_level=%u"},
 222         {Opt_dir_resv_level, "dir_resv_level=%u"},
 223         {Opt_journal_async_commit, "journal_async_commit"},
 224         {Opt_err_cont, "errors=continue"},
 225         {Opt_err, NULL}
 226 };
 227
 228 #ifdef CONFIG_DEBUG_FS
 229 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 230 {
 231         struct ocfs2_cluster_connection *cconn = osb->cconn;
 232         struct ocfs2_recovery_map *rm = osb->recovery_map;
 233         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 234         int i, out = 0;
 235         unsigned long flags;
 236
 237         out += snprintf(buf + out, len - out,
 238                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 239                         "Device", osb->dev_str, osb->uuid_str,
 240                         osb->fs_generation, osb->vol_label);
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 244                         atomic_read(&osb->vol_state), osb->osb_flags);
 245
 246         out += snprintf(buf + out, len - out,
 247                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 248                         osb->sb->s_blocksize, osb->s_clustersize);
 249
 250         out += snprintf(buf + out, len - out,
 251                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 252                         "ROcompat: 0x%X\n",
 253                         "Features", osb->s_feature_compat,
 254                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 255
 256         out += snprintf(buf + out, len - out,
 257                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 258                         osb->s_mount_opt, osb->s_atime_quantum);
 259
 260         if (cconn) {
 261                 out += snprintf(buf + out, len - out,
 262                                 "%10s => Stack: %s  Name: %*s  "
 263                                 "Version: %d.%d\n", "Cluster",
 264                                 (*osb->osb_cluster_stack == '\0' ?
 265                                  "o2cb" : osb->osb_cluster_stack),
 266                                 cconn->cc_namelen, cconn->cc_name,
 267                                 cconn->cc_version.pv_major,
 268                                 cconn->cc_version.pv_minor);
 269         }
 270
 271         spin_lock_irqsave(&osb->dc_task_lock, flags);
 272         out += snprintf(buf + out, len - out,
 273                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 274                         "WorkSeq: %lu\n", "DownCnvt",
 275                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 276                         osb->blocked_lock_count, osb->dc_wake_sequence,
 277                         osb->dc_work_sequence);
 278         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 279
 280         spin_lock(&osb->osb_lock);
 281         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 282                         "Recovery",
 283                         (osb->recovery_thread_task ?
 284                          task_pid_nr(osb->recovery_thread_task) : -1));
 285         if (rm->rm_used == 0)
 286                 out += snprintf(buf + out, len - out, " None\n");
 287         else {
 288                 for (i = 0; i < rm->rm_used; i++)
 289                         out += snprintf(buf + out, len - out, " %d",
 290                                         rm->rm_entries[i]);
 291                 out += snprintf(buf + out, len - out, "\n");
 292         }
 293         spin_unlock(&osb->osb_lock);
 294
 295         out += snprintf(buf + out, len - out,
 296                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 297                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 298                         osb->osb_commit_interval);
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 302                         "Journal", osb->journal->j_state,
 303                         osb->journal->j_trans_id,
 304                         atomic_read(&osb->journal->j_num_trans));
 305
 306         out += snprintf(buf + out, len - out,
 307                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 308                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 309                         "Stats",
 310                         atomic_read(&osb->alloc_stats.bitmap_data),
 311                         atomic_read(&osb->alloc_stats.local_data),
 312                         atomic_read(&osb->alloc_stats.bg_allocs),
 313                         atomic_read(&osb->alloc_stats.moves),
 314                         atomic_read(&osb->alloc_stats.bg_extends));
 315
 316         out += snprintf(buf + out, len - out,
 317                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 318                         "Default: %u bits\n",
 319                         "LocalAlloc", osb->local_alloc_state,
 320                         (unsigned long long)osb->la_last_gd,
 321                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 322
 323         spin_lock(&osb->osb_lock);
 324         out += snprintf(buf + out, len - out,
 325                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 326                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 327                         osb->s_inode_steal_slot,
 328                         atomic_read(&osb->s_num_inodes_stolen),
 329                         osb->s_meta_steal_slot,
 330                         atomic_read(&osb->s_num_meta_stolen));
 331         spin_unlock(&osb->osb_lock);
 332
 333         out += snprintf(buf + out, len - out, "OrphanScan => ");
 334         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 335                         os->os_count, os->os_seqno);
 336         out += snprintf(buf + out, len - out, " Last Scan: ");
 337         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 338                 out += snprintf(buf + out, len - out, "Disabled\n");
 339         else
 340                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 341                                 (unsigned long)(ktime_get_seconds() - os->os_scantime));
 342
 343         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 344                         "Slots", "Num", "RecoGen");
 345         for (i = 0; i < osb->max_slots; ++i) {
 346                 out += snprintf(buf + out, len - out,
 347                                 "%10s  %c %3d  %10d\n",
 348                                 " ",
 349                                 (i == osb->slot_num ? '*' : ' '),
 350                                 i, osb->slot_recovery_generations[i]);
 351         }
 352
 353         return out;
 354 }
 355
 356 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 357 {
 358         struct ocfs2_super *osb = inode->i_private;
 359         char *buf = NULL;
 360
 361         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 362         if (!buf)
 363                 goto bail;
 364
 365         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 366
 367         file->private_data = buf;
 368
 369         return 0;
 370 bail:
 371         return -ENOMEM;
 372 }
 373
 374 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 375 {
 376         kfree(file->private_data);
 377         return 0;
 378 }
 379
 380 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 381                                 size_t nbytes, loff_t *ppos)
 382 {
 383         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 384                                        i_size_read(file->f_mapping->host));
 385 }
 386 #else
 387 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 388 {
 389         return 0;
 390 }
 391 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 392 {
 393         return 0;
 394 }
 395 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 396                                 size_t nbytes, loff_t *ppos)
 397 {
 398         return 0;
 399 }
 400 #endif  /* CONFIG_DEBUG_FS */
 401
 402 static const struct file_operations ocfs2_osb_debug_fops = {
 403         .open =         ocfs2_osb_debug_open,
 404         .release =      ocfs2_debug_release,
 405         .read =         ocfs2_debug_read,
 406         .llseek =       generic_file_llseek,
 407 };
 408
 409 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 410 {
 411         int status;
 412         tid_t target;
 413         struct ocfs2_super *osb = OCFS2_SB(sb);
 414
 415         if (ocfs2_is_hard_readonly(osb))
 416                 return -EROFS;
 417
 418         if (wait) {
 419                 status = ocfs2_flush_truncate_log(osb);
 420                 if (status < 0)
 421                         mlog_errno(status);
 422         } else {
 423                 ocfs2_schedule_truncate_log_flush(osb, 0);
 424         }
 425
 426         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 427                                       &target)) {
 428                 if (wait)
 429                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 430                                              target);
 431         }
 432         return 0;
 433 }
 434
 435 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 436 {
 437         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 438             && (ino == USER_QUOTA_SYSTEM_INODE
 439                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 440                 return 0;
 441         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 442             && (ino == GROUP_QUOTA_SYSTEM_INODE
 443                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 444                 return 0;
 445         return 1;
 446 }
 447
 448 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 449 {
 450         struct inode *new = NULL;
 451         int status = 0;
 452         int i;
 453
 454         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 455         if (IS_ERR(new)) {
 456                 status = PTR_ERR(new);
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460         osb->root_inode = new;
 461
 462         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 463         if (IS_ERR(new)) {
 464                 status = PTR_ERR(new);
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468         osb->sys_root_inode = new;
 469
 470         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 471              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 472                 if (!ocfs2_need_system_inode(osb, i))
 473                         continue;
 474                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 475                 if (!new) {
 476                         ocfs2_release_system_inodes(osb);
 477                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 478                         mlog_errno(status);
 479                         mlog(ML_ERROR, "Unable to load system inode %d, "
 480                              "possibly corrupt fs?", i);
 481                         goto bail;
 482                 }
 483                 // the array now has one ref, so drop this one
 484                 iput(new);
 485         }
 486
 487 bail:
 488         if (status)
 489                 mlog_errno(status);
 490         return status;
 491 }
 492
 493 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 494 {
 495         struct inode *new = NULL;
 496         int status = 0;
 497         int i;
 498
 499         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 500              i < NUM_SYSTEM_INODES;
 501              i++) {
 502                 if (!ocfs2_need_system_inode(osb, i))
 503                         continue;
 504                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 505                 if (!new) {
 506                         ocfs2_release_system_inodes(osb);
 507                         status = ocfs2_is_soft_readonly(osb) ? -EROFS : -EINVAL;
 508                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 509                              status, i, osb->slot_num);
 510                         goto bail;
 511                 }
 512                 /* the array now has one ref, so drop this one */
 513                 iput(new);
 514         }
 515
 516 bail:
 517         if (status)
 518                 mlog_errno(status);
 519         return status;
 520 }
 521
 522 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 523 {
 524         int i;
 525         struct inode *inode;
 526
 527         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 528                 inode = osb->global_system_inodes[i];
 529                 if (inode) {
 530                         iput(inode);
 531                         osb->global_system_inodes[i] = NULL;
 532                 }
 533         }
 534
 535         inode = osb->sys_root_inode;
 536         if (inode) {
 537                 iput(inode);
 538                 osb->sys_root_inode = NULL;
 539         }
 540
 541         inode = osb->root_inode;
 542         if (inode) {
 543                 iput(inode);
 544                 osb->root_inode = NULL;
 545         }
 546
 547         if (!osb->local_system_inodes)
 548                 return;
 549
 550         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 551                 if (osb->local_system_inodes[i]) {
 552                         iput(osb->local_system_inodes[i]);
 553                         osb->local_system_inodes[i] = NULL;
 554                 }
 555         }
 556
 557         kfree(osb->local_system_inodes);
 558         osb->local_system_inodes = NULL;
 559 }
 560
 561 /* We're allocating fs objects, use GFP_NOFS */
 562 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 563 {
 564         struct ocfs2_inode_info *oi;
 565
 566         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 567         if (!oi)
 568                 return NULL;
 569
 570         oi->i_sync_tid = 0;
 571         oi->i_datasync_tid = 0;
 572         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 573
 574         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 575         return &oi->vfs_inode;
 576 }
 577
 578 static void ocfs2_i_callback(struct rcu_head *head)
 579 {
 580         struct inode *inode = container_of(head, struct inode, i_rcu);
 581         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 582 }
 583
 584 static void ocfs2_destroy_inode(struct inode *inode)
 585 {
 586         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 587 }
 588
 589 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 590                                                 unsigned int cbits)
 591 {
 592         unsigned int bytes = 1 << cbits;
 593         unsigned int trim = bytes;
 594         unsigned int bitshift = 32;
 595
 596         /*
 597          * i_size and all block offsets in ocfs2 are always 64 bits
 598          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 599          * 64 bit platforms, cluster size will be the limiting factor.
 600          */
 601
 602 #if BITS_PER_LONG == 32
 603 # if defined(CONFIG_LBDAF)
 604         BUILD_BUG_ON(sizeof(sector_t) != 8);
 605         /*
 606          * We might be limited by page cache size.
 607          */
 608         if (bytes > PAGE_SIZE) {
 609                 bytes = PAGE_SIZE;
 610                 trim = 1;
 611                 /*
 612                  * Shift by 31 here so that we don't get larger than
 613                  * MAX_LFS_FILESIZE
 614                  */
 615                 bitshift = 31;
 616         }
 617 # else
 618         /*
 619          * We are limited by the size of sector_t. Use block size, as
 620          * that's what we expose to the VFS.
 621          */
 622         bytes = 1 << bbits;
 623         trim = 1;
 624         bitshift = 31;
 625 # endif
 626 #endif
 627
 628         /*
 629          * Trim by a whole cluster when we can actually approach the
 630          * on-disk limits. Otherwise we can overflow i_clusters when
 631          * an extent start is at the max offset.
 632          */
 633         return (((unsigned long long)bytes) << bitshift) - trim;
 634 }
 635
 636 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 637 {
 638         int incompat_features;
 639         int ret = 0;
 640         struct mount_options parsed_options;
 641         struct ocfs2_super *osb = OCFS2_SB(sb);
 642         u32 tmp;
 643
 644         sync_filesystem(sb);
 645
 646         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 647             !ocfs2_check_set_options(sb, &parsed_options)) {
 648                 ret = -EINVAL;
 649                 goto out;
 650         }
 651
 652         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 653                 OCFS2_MOUNT_HB_NONE;
 654         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 657                 goto out;
 658         }
 659
 660         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 661             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 662                 ret = -EINVAL;
 663                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 664                 goto out;
 665         }
 666
 667         /* Probably don't want this on remount; it might
 668          * mess with other nodes */
 669         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 670             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 671                 ret = -EINVAL;
 672                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 673                 goto out;
 674         }
 675
 676         /* We're going to/from readonly mode. */
 677         if ((bool)(*flags & MS_RDONLY) != sb_rdonly(sb)) {
 678                 /* Disable quota accounting before remounting RO */
 679                 if (*flags & MS_RDONLY) {
 680                         ret = ocfs2_susp_quotas(osb, 0);
 681                         if (ret < 0)
 682                                 goto out;
 683                 }
 684                 /* Lock here so the check of HARD_RO and the potential
 685                  * setting of SOFT_RO is atomic. */
 686                 spin_lock(&osb->osb_lock);
 687                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 688                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 689                         ret = -EROFS;
 690                         goto unlock_osb;
 691                 }
 692
 693                 if (*flags & MS_RDONLY) {
 694                         sb->s_flags |= MS_RDONLY;
 695                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 696                 } else {
 697                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 698                                 mlog(ML_ERROR, "Cannot remount RDWR "
 699                                      "filesystem due to previous errors.\n");
 700                                 ret = -EROFS;
 701                                 goto unlock_osb;
 702                         }
 703                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 704                         if (incompat_features) {
 705                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 706                                      "of unsupported optional features "
 707                                      "(%x).\n", incompat_features);
 708                                 ret = -EINVAL;
 709                                 goto unlock_osb;
 710                         }
 711                         sb->s_flags &= ~MS_RDONLY;
 712                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 713                 }
 714                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 715 unlock_osb:
 716                 spin_unlock(&osb->osb_lock);
 717                 /* Enable quota accounting after remounting RW */
 718                 if (!ret && !(*flags & MS_RDONLY)) {
 719                         if (sb_any_quota_suspended(sb))
 720                                 ret = ocfs2_susp_quotas(osb, 1);
 721                         else
 722                                 ret = ocfs2_enable_quotas(osb);
 723                         if (ret < 0) {
 724                                 /* Return back changes... */
 725                                 spin_lock(&osb->osb_lock);
 726                                 sb->s_flags |= MS_RDONLY;
 727                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 728                                 spin_unlock(&osb->osb_lock);
 729                                 goto out;
 730                         }
 731                 }
 732         }
 733
 734         if (!ret) {
 735                 /* Only save off the new mount options in case of a successful
 736                  * remount. */
 737                 osb->s_mount_opt = parsed_options.mount_opt;
 738                 osb->s_atime_quantum = parsed_options.atime_quantum;
 739                 osb->preferred_slot = parsed_options.slot;
 740                 if (parsed_options.commit_interval)
 741                         osb->osb_commit_interval = parsed_options.commit_interval;
 742
 743                 if (!ocfs2_is_hard_readonly(osb))
 744                         ocfs2_set_journal_params(osb);
 745
 746                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 747                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 748                                                         MS_POSIXACL : 0);
 749         }
 750 out:
 751         return ret;
 752 }
 753
 754 static int ocfs2_sb_probe(struct super_block *sb,
 755                           struct buffer_head **bh,
 756                           int *sector_size,
 757                           struct ocfs2_blockcheck_stats *stats)
 758 {
 759         int status, tmpstat;
 760         struct ocfs1_vol_disk_hdr *hdr;
 761         struct ocfs2_dinode *di;
 762         int blksize;
 763
 764         *bh = NULL;
 765
 766         /* may be > 512 */
 767         *sector_size = bdev_logical_block_size(sb->s_bdev);
 768         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 769                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 770                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 771                 status = -EINVAL;
 772                 goto bail;
 773         }
 774
 775         /* Can this really happen? */
 776         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 777                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 778
 779         /* check block zero for old format */
 780         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 781         if (status < 0) {
 782                 mlog_errno(status);
 783                 goto bail;
 784         }
 785         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 786         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 787                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 788                      hdr->major_version, hdr->minor_version);
 789                 status = -EINVAL;
 790         }
 791         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 792                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 793                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 794                      hdr->signature);
 795                 status = -EINVAL;
 796         }
 797         brelse(*bh);
 798         *bh = NULL;
 799         if (status < 0) {
 800                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 801                      "upgraded before mounting with ocfs v2\n");
 802                 goto bail;
 803         }
 804
 805         /*
 806          * Now check at magic offset for 512, 1024, 2048, 4096
 807          * blocksizes.  4096 is the maximum blocksize because it is
 808          * the minimum clustersize.
 809          */
 810         status = -EINVAL;
 811         for (blksize = *sector_size;
 812              blksize <= OCFS2_MAX_BLOCKSIZE;
 813              blksize <<= 1) {
 814                 tmpstat = ocfs2_get_sector(sb, bh,
 815                                            OCFS2_SUPER_BLOCK_BLKNO,
 816                                            blksize);
 817                 if (tmpstat < 0) {
 818                         status = tmpstat;
 819                         mlog_errno(status);
 820                         break;
 821                 }
 822                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 823                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 824                 spin_lock_init(&stats->b_lock);
 825                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 826                 if (tmpstat < 0) {
 827                         brelse(*bh);
 828                         *bh = NULL;
 829                 }
 830                 if (tmpstat != -EAGAIN) {
 831                         status = tmpstat;
 832                         break;
 833                 }
 834         }
 835
 836 bail:
 837         return status;
 838 }
 839
 840 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 841 {
 842         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 843
 844         if (osb->s_mount_opt & hb_enabled) {
 845                 if (ocfs2_mount_local(osb)) {
 846                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 847                              "mounted device.\n");
 848                         return -EINVAL;
 849                 }
 850                 if (ocfs2_userspace_stack(osb)) {
 851                         mlog(ML_ERROR, "Userspace stack expected, but "
 852                              "o2cb heartbeat arguments passed to mount\n");
 853                         return -EINVAL;
 854                 }
 855                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 856                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 857                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 858                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 859                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 860                         return -EINVAL;
 861                 }
 862         }
 863
 864         if (!(osb->s_mount_opt & hb_enabled)) {
 865                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 866                     !ocfs2_userspace_stack(osb)) {
 867                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 868                              "a read-write clustered device.\n");
 869                         return -EINVAL;
 870                 }
 871         }
 872
 873         return 0;
 874 }
 875
 876 /*
 877  * If we're using a userspace stack, mount should have passed
 878  * a name that matches the disk.  If not, mount should not
 879  * have passed a stack.
 880  */
 881 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 882                                         struct mount_options *mopt)
 883 {
 884         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 885                 mlog(ML_ERROR,
 886                      "cluster stack passed to mount, but this filesystem "
 887                      "does not support it\n");
 888                 return -EINVAL;
 889         }
 890
 891         if (ocfs2_userspace_stack(osb) &&
 892             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 893                     OCFS2_STACK_LABEL_LEN)) {
 894                 mlog(ML_ERROR,
 895                      "cluster stack passed to mount (\"%s\") does not "
 896                      "match the filesystem (\"%s\")\n",
 897                      mopt->cluster_stack,
 898                      osb->osb_cluster_stack);
 899                 return -EINVAL;
 900         }
 901
 902         return 0;
 903 }
 904
 905 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 906 {
 907         int type;
 908         struct super_block *sb = osb->sb;
 909         unsigned int feature[OCFS2_MAXQUOTAS] = {
 910                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 911                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 912         int status = 0;
 913
 914         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 915                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 916                         continue;
 917                 if (unsuspend)
 918                         status = dquot_resume(sb, type);
 919                 else {
 920                         struct ocfs2_mem_dqinfo *oinfo;
 921
 922                         /* Cancel periodic syncing before suspending */
 923                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 924                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 925                         status = dquot_suspend(sb, type);
 926                 }
 927                 if (status < 0)
 928                         break;
 929         }
 930         if (status < 0)
 931                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 932                      "remount (error = %d).\n", status);
 933         return status;
 934 }
 935
 936 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 937 {
 938         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 939         struct super_block *sb = osb->sb;
 940         unsigned int feature[OCFS2_MAXQUOTAS] = {
 941                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 942                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 943         unsigned int ino[OCFS2_MAXQUOTAS] = {
 944                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 945                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 946         int status;
 947         int type;
 948
 949         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 950         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 951                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 952                         continue;
 953                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 954                                                         osb->slot_num);
 955                 if (!inode[type]) {
 956                         status = -ENOENT;
 957                         goto out_quota_off;
 958                 }
 959                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 960                                       DQUOT_USAGE_ENABLED);
 961                 if (status < 0)
 962                         goto out_quota_off;
 963         }
 964
 965         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 966                 iput(inode[type]);
 967         return 0;
 968 out_quota_off:
 969         ocfs2_disable_quotas(osb);
 970         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 971                 iput(inode[type]);
 972         mlog_errno(status);
 973         return status;
 974 }
 975
 976 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 977 {
 978         int type;
 979         struct inode *inode;
 980         struct super_block *sb = osb->sb;
 981         struct ocfs2_mem_dqinfo *oinfo;
 982
 983         /* We mostly ignore errors in this function because there's not much
 984          * we can do when we see them */
 985         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 986                 if (!sb_has_quota_loaded(sb, type))
 987                         continue;
 988                 if (!sb_has_quota_suspended(sb, type)) {
 989                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 990                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 991                 }
 992                 inode = igrab(sb->s_dquot.files[type]);
 993                 /* Turn off quotas. This will remove all dquot structures from
 994                  * memory and so they will be automatically synced to global
 995                  * quota files */
 996                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 997                                         DQUOT_LIMITS_ENABLED);
 998                 if (!inode)
 999                         continue;
1000                 iput(inode);
1001         }
1002 }
1003
1004 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1005 {
1006         struct dentry *root;
1007         int status, sector_size;
1008         struct mount_options parsed_options;
1009         struct inode *inode = NULL;
1010         struct ocfs2_super *osb = NULL;
1011         struct buffer_head *bh = NULL;
1012         char nodestr[12];
1013         struct ocfs2_blockcheck_stats stats;
1014
1015         trace_ocfs2_fill_super(sb, data, silent);
1016
1017         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1018                 status = -EINVAL;
1019                 goto read_super_error;
1020         }
1021
1022         /* probe for superblock */
1023         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1024         if (status < 0) {
1025                 mlog(ML_ERROR, "superblock probe failed!\n");
1026                 goto read_super_error;
1027         }
1028
1029         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1030         osb = OCFS2_SB(sb);
1031         if (status < 0) {
1032                 mlog_errno(status);
1033                 goto read_super_error;
1034         }
1035         brelse(bh);
1036         bh = NULL;
1037
1038         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1039                 status = -EINVAL;
1040                 goto read_super_error;
1041         }
1042         osb->s_mount_opt = parsed_options.mount_opt;
1043         osb->s_atime_quantum = parsed_options.atime_quantum;
1044         osb->preferred_slot = parsed_options.slot;
1045         osb->osb_commit_interval = parsed_options.commit_interval;
1046
1047         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1048         osb->osb_resv_level = parsed_options.resv_level;
1049         osb->osb_dir_resv_level = parsed_options.resv_level;
1050         if (parsed_options.dir_resv_level == -1)
1051                 osb->osb_dir_resv_level = parsed_options.resv_level;
1052         else
1053                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1054
1055         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1056         if (status)
1057                 goto read_super_error;
1058
1059         sb->s_magic = OCFS2_SUPER_MAGIC;
1060
1061         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1062                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1063
1064         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1065          * heartbeat=none */
1066         if (bdev_read_only(sb->s_bdev)) {
1067                 if (!sb_rdonly(sb)) {
1068                         status = -EACCES;
1069                         mlog(ML_ERROR, "Readonly device detected but readonly "
1070                              "mount was not specified.\n");
1071                         goto read_super_error;
1072                 }
1073
1074                 /* You should not be able to start a local heartbeat
1075                  * on a readonly device. */
1076                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1077                         status = -EROFS;
1078                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1079                              "device.\n");
1080                         goto read_super_error;
1081                 }
1082
1083                 status = ocfs2_check_journals_nolocks(osb);
1084                 if (status < 0) {
1085                         if (status == -EROFS)
1086                                 mlog(ML_ERROR, "Recovery required on readonly "
1087                                      "file system, but write access is "
1088                                      "unavailable.\n");
1089                         else
1090                                 mlog_errno(status);
1091                         goto read_super_error;
1092                 }
1093
1094                 ocfs2_set_ro_flag(osb, 1);
1095
1096                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1097                        "Cluster services will not be used for this mount. "
1098                        "Recovery will be skipped.\n", osb->dev_str);
1099         }
1100
1101         if (!ocfs2_is_hard_readonly(osb)) {
1102                 if (sb_rdonly(sb))
1103                         ocfs2_set_ro_flag(osb, 0);
1104         }
1105
1106         status = ocfs2_verify_heartbeat(osb);
1107         if (status < 0) {
1108                 mlog_errno(status);
1109                 goto read_super_error;
1110         }
1111
1112         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1113                                                  ocfs2_debugfs_root);
1114         if (!osb->osb_debug_root) {
1115                 status = -EINVAL;
1116                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1117                 goto read_super_error;
1118         }
1119
1120         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1121                                             osb->osb_debug_root,
1122                                             osb,
1123                                             &ocfs2_osb_debug_fops);
1124         if (!osb->osb_ctxt) {
1125                 status = -EINVAL;
1126                 mlog_errno(status);
1127                 goto read_super_error;
1128         }
1129
1130         if (ocfs2_meta_ecc(osb)) {
1131                 status = ocfs2_blockcheck_stats_debugfs_install(
1132                                                 &osb->osb_ecc_stats,
1133                                                 osb->osb_debug_root);
1134                 if (status) {
1135                         mlog(ML_ERROR,
1136                              "Unable to create blockcheck statistics "
1137                              "files\n");
1138                         goto read_super_error;
1139                 }
1140         }
1141
1142         status = ocfs2_mount_volume(sb);
1143         if (status < 0)
1144                 goto read_super_error;
1145
1146         if (osb->root_inode)
1147                 inode = igrab(osb->root_inode);
1148
1149         if (!inode) {
1150                 status = -EIO;
1151                 mlog_errno(status);
1152                 goto read_super_error;
1153         }
1154
1155         root = d_make_root(inode);
1156         if (!root) {
1157                 status = -ENOMEM;
1158                 mlog_errno(status);
1159                 goto read_super_error;
1160         }
1161
1162         sb->s_root = root;
1163
1164         ocfs2_complete_mount_recovery(osb);
1165
1166         if (ocfs2_mount_local(osb))
1167                 snprintf(nodestr, sizeof(nodestr), "local");
1168         else
1169                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1170
1171         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1172                "with %s data mode.\n",
1173                osb->dev_str, nodestr, osb->slot_num,
1174                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1175                "ordered");
1176
1177         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1178         wake_up(&osb->osb_mount_event);
1179
1180         /* Now we can initialize quotas because we can afford to wait
1181          * for cluster locks recovery now. That also means that truncation
1182          * log recovery can happen but that waits for proper quota setup */
1183         if (!sb_rdonly(sb)) {
1184                 status = ocfs2_enable_quotas(osb);
1185                 if (status < 0) {
1186                         /* We have to err-out specially here because
1187                          * s_root is already set */
1188                         mlog_errno(status);
1189                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1190                         wake_up(&osb->osb_mount_event);
1191                         return status;
1192                 }
1193         }
1194
1195         ocfs2_complete_quota_recovery(osb);
1196
1197         /* Now we wake up again for processes waiting for quotas */
1198         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1199         wake_up(&osb->osb_mount_event);
1200
1201         /* Start this when the mount is almost sure of being successful */
1202         ocfs2_orphan_scan_start(osb);
1203
1204         /* Create filecheck sysfile /sys/fs/ocfs2/<devname>/filecheck */
1205         ocfs2_filecheck_create_sysfs(sb);
1206
1207         return status;
1208
1209 read_super_error:
1210         brelse(bh);
1211
1212         if (osb) {
1213                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1214                 wake_up(&osb->osb_mount_event);
1215                 ocfs2_dismount_volume(sb, 1);
1216         }
1217
1218         if (status)
1219                 mlog_errno(status);
1220         return status;
1221 }
1222
1223 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1224                         int flags,
1225                         const char *dev_name,
1226                         void *data)
1227 {
1228         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1229 }
1230
1231 static struct file_system_type ocfs2_fs_type = {
1232         .owner          = THIS_MODULE,
1233         .name           = "ocfs2",
1234         .mount          = ocfs2_mount,
1235         .kill_sb        = kill_block_super,
1236         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1237         .next           = NULL
1238 };
1239 MODULE_ALIAS_FS("ocfs2");
1240
1241 static int ocfs2_check_set_options(struct super_block *sb,
1242                                    struct mount_options *options)
1243 {
1244         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1245             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1246                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1247                 mlog(ML_ERROR, "User quotas were requested, but this "
1248                      "filesystem does not have the feature enabled.\n");
1249                 return 0;
1250         }
1251         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1252             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1253                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1254                 mlog(ML_ERROR, "Group quotas were requested, but this "
1255                      "filesystem does not have the feature enabled.\n");
1256                 return 0;
1257         }
1258         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1259             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1260                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1261                      "feature is not enabled\n");
1262                 return 0;
1263         }
1264         /* No ACL setting specified? Use XATTR feature... */
1265         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1266                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1267                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1268                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1269                 else
1270                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1271         }
1272         return 1;
1273 }
1274
1275 static int ocfs2_parse_options(struct super_block *sb,
1276                                char *options,
1277                                struct mount_options *mopt,
1278                                int is_remount)
1279 {
1280         int status, user_stack = 0;
1281         char *p;
1282         u32 tmp;
1283         int token, option;
1284         substring_t args[MAX_OPT_ARGS];
1285
1286         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1287
1288         mopt->commit_interval = 0;
1289         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1290         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1291         mopt->slot = OCFS2_INVALID_SLOT;
1292         mopt->localalloc_opt = -1;
1293         mopt->cluster_stack[0] = '\0';
1294         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1295         mopt->dir_resv_level = -1;
1296
1297         if (!options) {
1298                 status = 1;
1299                 goto bail;
1300         }
1301
1302         while ((p = strsep(&options, ",")) != NULL) {
1303                 if (!*p)
1304                         continue;
1305
1306                 token = match_token(p, tokens, args);
1307                 switch (token) {
1308                 case Opt_hb_local:
1309                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1310                         break;
1311                 case Opt_hb_none:
1312                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1313                         break;
1314                 case Opt_hb_global:
1315                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1316                         break;
1317                 case Opt_barrier:
1318                         if (match_int(&args[0], &option)) {
1319                                 status = 0;
1320                                 goto bail;
1321                         }
1322                         if (option)
1323                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1324                         else
1325                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1326                         break;
1327                 case Opt_intr:
1328                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1329                         break;
1330                 case Opt_nointr:
1331                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1332                         break;
1333                 case Opt_err_panic:
1334                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1335                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1336                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1337                         break;
1338                 case Opt_err_ro:
1339                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1340                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1341                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1342                         break;
1343                 case Opt_err_cont:
1344                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1345                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1346                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1347                         break;
1348                 case Opt_data_ordered:
1349                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1350                         break;
1351                 case Opt_data_writeback:
1352                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1353                         break;
1354                 case Opt_user_xattr:
1355                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1356                         break;
1357                 case Opt_nouser_xattr:
1358                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1359                         break;
1360                 case Opt_atime_quantum:
1361                         if (match_int(&args[0], &option)) {
1362                                 status = 0;
1363                                 goto bail;
1364                         }
1365                         if (option >= 0)
1366                                 mopt->atime_quantum = option;
1367                         break;
1368                 case Opt_slot:
1369                         if (match_int(&args[0], &option)) {
1370                                 status = 0;
1371                                 goto bail;
1372                         }
1373                         if (option)
1374                                 mopt->slot = (u16)option;
1375                         break;
1376                 case Opt_commit:
1377                         if (match_int(&args[0], &option)) {
1378                                 status = 0;
1379                                 goto bail;
1380                         }
1381                         if (option < 0)
1382                                 return 0;
1383                         if (option == 0)
1384                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1385                         mopt->commit_interval = HZ * option;
1386                         break;
1387                 case Opt_localalloc:
1388                         if (match_int(&args[0], &option)) {
1389                                 status = 0;
1390                                 goto bail;
1391                         }
1392                         if (option >= 0)
1393                                 mopt->localalloc_opt = option;
1394                         break;
1395                 case Opt_localflocks:
1396                         /*
1397                          * Changing this during remount could race
1398                          * flock() requests, or "unbalance" existing
1399                          * ones (e.g., a lock is taken in one mode but
1400                          * dropped in the other). If users care enough
1401                          * to flip locking modes during remount, we
1402                          * could add a "local" flag to individual
1403                          * flock structures for proper tracking of
1404                          * state.
1405                          */
1406                         if (!is_remount)
1407                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1408                         break;
1409                 case Opt_stack:
1410                         /* Check both that the option we were passed
1411                          * is of the right length and that it is a proper
1412                          * string of the right length.
1413                          */
1414                         if (((args[0].to - args[0].from) !=
1415                              OCFS2_STACK_LABEL_LEN) ||
1416                             (strnlen(args[0].from,
1417                                      OCFS2_STACK_LABEL_LEN) !=
1418                              OCFS2_STACK_LABEL_LEN)) {
1419                                 mlog(ML_ERROR,
1420                                      "Invalid cluster_stack option\n");
1421                                 status = 0;
1422                                 goto bail;
1423                         }
1424                         memcpy(mopt->cluster_stack, args[0].from,
1425                                OCFS2_STACK_LABEL_LEN);
1426                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1427                         /*
1428                          * Open code the memcmp here as we don't have
1429                          * an osb to pass to
1430                          * ocfs2_userspace_stack().
1431                          */
1432                         if (memcmp(mopt->cluster_stack,
1433                                    OCFS2_CLASSIC_CLUSTER_STACK,
1434                                    OCFS2_STACK_LABEL_LEN))
1435                                 user_stack = 1;
1436                         break;
1437                 case Opt_inode64:
1438                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1439                         break;
1440                 case Opt_usrquota:
1441                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1442                         break;
1443                 case Opt_grpquota:
1444                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1445                         break;
1446                 case Opt_coherency_buffered:
1447                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1448                         break;
1449                 case Opt_coherency_full:
1450                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1451                         break;
1452                 case Opt_acl:
1453                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1454                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1455                         break;
1456                 case Opt_noacl:
1457                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1458                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1459                         break;
1460                 case Opt_resv_level:
1461                         if (is_remount)
1462                                 break;
1463                         if (match_int(&args[0], &option)) {
1464                                 status = 0;
1465                                 goto bail;
1466                         }
1467                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1468                             option < OCFS2_MAX_RESV_LEVEL)
1469                                 mopt->resv_level = option;
1470                         break;
1471                 case Opt_dir_resv_level:
1472                         if (is_remount)
1473                                 break;
1474                         if (match_int(&args[0], &option)) {
1475                                 status = 0;
1476                                 goto bail;
1477                         }
1478                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1479                             option < OCFS2_MAX_RESV_LEVEL)
1480                                 mopt->dir_resv_level = option;
1481                         break;
1482                 case Opt_journal_async_commit:
1483                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1484                         break;
1485                 default:
1486                         mlog(ML_ERROR,
1487                              "Unrecognized mount option \"%s\" "
1488                              "or missing value\n", p);
1489                         status = 0;
1490                         goto bail;
1491                 }
1492         }
1493
1494         if (user_stack == 0) {
1495                 /* Ensure only one heartbeat mode */
1496                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1497                                          OCFS2_MOUNT_HB_GLOBAL |
1498                                          OCFS2_MOUNT_HB_NONE);
1499                 if (hweight32(tmp) != 1) {
1500                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1501                         status = 0;
1502                         goto bail;
1503                 }
1504         }
1505
1506         status = 1;
1507
1508 bail:
1509         return status;
1510 }
1511
1512 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1513 {
1514         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1515         unsigned long opts = osb->s_mount_opt;
1516         unsigned int local_alloc_megs;
1517
1518         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1519                 seq_printf(s, ",_netdev");
1520                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1521                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1522                 else
1523                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1524         } else
1525                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1526
1527         if (opts & OCFS2_MOUNT_NOINTR)
1528                 seq_printf(s, ",nointr");
1529
1530         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1531                 seq_printf(s, ",data=writeback");
1532         else
1533                 seq_printf(s, ",data=ordered");
1534
1535         if (opts & OCFS2_MOUNT_BARRIER)
1536                 seq_printf(s, ",barrier=1");
1537
1538         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1539                 seq_printf(s, ",errors=panic");
1540         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1541                 seq_printf(s, ",errors=continue");
1542         else
1543                 seq_printf(s, ",errors=remount-ro");
1544
1545         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1546                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1547
1548         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1549
1550         if (osb->osb_commit_interval)
1551                 seq_printf(s, ",commit=%u",
1552                            (unsigned) (osb->osb_commit_interval / HZ));
1553
1554         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1555         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1556                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1557
1558         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1559                 seq_printf(s, ",localflocks,");
1560
1561         if (osb->osb_cluster_stack[0])
1562                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1563                                   OCFS2_STACK_LABEL_LEN);
1564         if (opts & OCFS2_MOUNT_USRQUOTA)
1565                 seq_printf(s, ",usrquota");
1566         if (opts & OCFS2_MOUNT_GRPQUOTA)
1567                 seq_printf(s, ",grpquota");
1568
1569         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1570                 seq_printf(s, ",coherency=buffered");
1571         else
1572                 seq_printf(s, ",coherency=full");
1573
1574         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1575                 seq_printf(s, ",nouser_xattr");
1576         else
1577                 seq_printf(s, ",user_xattr");
1578
1579         if (opts & OCFS2_MOUNT_INODE64)
1580                 seq_printf(s, ",inode64");
1581
1582         if (opts & OCFS2_MOUNT_POSIX_ACL)
1583                 seq_printf(s, ",acl");
1584         else
1585                 seq_printf(s, ",noacl");
1586
1587         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1588                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1589
1590         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1591                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1592
1593         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1594                 seq_printf(s, ",journal_async_commit");
1595
1596         return 0;
1597 }
1598
1599 static int __init ocfs2_init(void)
1600 {
1601         int status;
1602
1603         status = init_ocfs2_uptodate_cache();
1604         if (status < 0)
1605                 goto out1;
1606
1607         status = ocfs2_initialize_mem_caches();
1608         if (status < 0)
1609                 goto out2;
1610
1611         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1612         if (!ocfs2_debugfs_root) {
1613                 status = -ENOMEM;
1614                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1615                 goto out3;
1616         }
1617
1618         ocfs2_set_locking_protocol();
1619
1620         status = register_quota_format(&ocfs2_quota_format);
1621         if (status < 0)
1622                 goto out3;
1623         status = register_filesystem(&ocfs2_fs_type);
1624         if (!status)
1625                 return 0;
1626
1627         unregister_quota_format(&ocfs2_quota_format);
1628 out3:
1629         debugfs_remove(ocfs2_debugfs_root);
1630         ocfs2_free_mem_caches();
1631 out2:
1632         exit_ocfs2_uptodate_cache();
1633 out1:
1634         mlog_errno(status);
1635         return status;
1636 }
1637
1638 static void __exit ocfs2_exit(void)
1639 {
1640         unregister_quota_format(&ocfs2_quota_format);
1641
1642         debugfs_remove(ocfs2_debugfs_root);
1643
1644         ocfs2_free_mem_caches();
1645
1646         unregister_filesystem(&ocfs2_fs_type);
1647
1648         exit_ocfs2_uptodate_cache();
1649 }
1650
1651 static void ocfs2_put_super(struct super_block *sb)
1652 {
1653         trace_ocfs2_put_super(sb);
1654
1655         ocfs2_sync_blockdev(sb);
1656         ocfs2_dismount_volume(sb, 0);
1657         ocfs2_filecheck_remove_sysfs(sb);
1658 }
1659
1660 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1661 {
1662         struct ocfs2_super *osb;
1663         u32 numbits, freebits;
1664         int status;
1665         struct ocfs2_dinode *bm_lock;
1666         struct buffer_head *bh = NULL;
1667         struct inode *inode = NULL;
1668
1669         trace_ocfs2_statfs(dentry->d_sb, buf);
1670
1671         osb = OCFS2_SB(dentry->d_sb);
1672
1673         inode = ocfs2_get_system_file_inode(osb,
1674                                             GLOBAL_BITMAP_SYSTEM_INODE,
1675                                             OCFS2_INVALID_SLOT);
1676         if (!inode) {
1677                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1678                 status = -EIO;
1679                 goto bail;
1680         }
1681
1682         status = ocfs2_inode_lock(inode, &bh, 0);
1683         if (status < 0) {
1684                 mlog_errno(status);
1685                 goto bail;
1686         }
1687
1688         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1689
1690         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1691         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1692
1693         buf->f_type = OCFS2_SUPER_MAGIC;
1694         buf->f_bsize = dentry->d_sb->s_blocksize;
1695         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1696         buf->f_blocks = ((sector_t) numbits) *
1697                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1698         buf->f_bfree = ((sector_t) freebits) *
1699                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1700         buf->f_bavail = buf->f_bfree;
1701         buf->f_files = numbits;
1702         buf->f_ffree = freebits;
1703         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1704                                 & 0xFFFFFFFFUL;
1705         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1706                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1707
1708         brelse(bh);
1709
1710         ocfs2_inode_unlock(inode, 0);
1711         status = 0;
1712 bail:
1713         iput(inode);
1714
1715         if (status)
1716                 mlog_errno(status);
1717
1718         return status;
1719 }
1720
1721 static void ocfs2_inode_init_once(void *data)
1722 {
1723         struct ocfs2_inode_info *oi = data;
1724
1725         oi->ip_flags = 0;
1726         oi->ip_open_count = 0;
1727         spin_lock_init(&oi->ip_lock);
1728         ocfs2_extent_map_init(&oi->vfs_inode);
1729         INIT_LIST_HEAD(&oi->ip_io_markers);
1730         INIT_LIST_HEAD(&oi->ip_unwritten_list);
1731         oi->ip_dir_start_lookup = 0;
1732         init_rwsem(&oi->ip_alloc_sem);
1733         init_rwsem(&oi->ip_xattr_sem);
1734         mutex_init(&oi->ip_io_mutex);
1735
1736         oi->ip_blkno = 0ULL;
1737         oi->ip_clusters = 0;
1738         oi->ip_next_orphan = NULL;
1739
1740         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1741
1742         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1743         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1744         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1745
1746         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1747                                   &ocfs2_inode_caching_ops);
1748
1749         inode_init_once(&oi->vfs_inode);
1750 }
1751
1752 static int ocfs2_initialize_mem_caches(void)
1753 {
1754         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1755                                        sizeof(struct ocfs2_inode_info),
1756                                        0,
1757                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1758                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1759                                        ocfs2_inode_init_once);
1760         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1761                                         sizeof(struct ocfs2_dquot),
1762                                         0,
1763                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1764                                                 SLAB_MEM_SPREAD),
1765                                         NULL);
1766         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1767                                         sizeof(struct ocfs2_quota_chunk),
1768                                         0,
1769                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1770                                         NULL);
1771         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1772             !ocfs2_qf_chunk_cachep) {
1773                 if (ocfs2_inode_cachep)
1774                         kmem_cache_destroy(ocfs2_inode_cachep);
1775                 if (ocfs2_dquot_cachep)
1776                         kmem_cache_destroy(ocfs2_dquot_cachep);
1777                 if (ocfs2_qf_chunk_cachep)
1778                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1779                 return -ENOMEM;
1780         }
1781
1782         return 0;
1783 }
1784
1785 static void ocfs2_free_mem_caches(void)
1786 {
1787         /*
1788          * Make sure all delayed rcu free inodes are flushed before we
1789          * destroy cache.
1790          */
1791         rcu_barrier();
1792         if (ocfs2_inode_cachep)
1793                 kmem_cache_destroy(ocfs2_inode_cachep);
1794         ocfs2_inode_cachep = NULL;
1795
1796         if (ocfs2_dquot_cachep)
1797                 kmem_cache_destroy(ocfs2_dquot_cachep);
1798         ocfs2_dquot_cachep = NULL;
1799
1800         if (ocfs2_qf_chunk_cachep)
1801                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1802         ocfs2_qf_chunk_cachep = NULL;
1803 }
1804
1805 static int ocfs2_get_sector(struct super_block *sb,
1806                             struct buffer_head **bh,
1807                             int block,
1808                             int sect_size)
1809 {
1810         if (!sb_set_blocksize(sb, sect_size)) {
1811                 mlog(ML_ERROR, "unable to set blocksize\n");
1812                 return -EIO;
1813         }
1814
1815         *bh = sb_getblk(sb, block);
1816         if (!*bh) {
1817                 mlog_errno(-ENOMEM);
1818                 return -ENOMEM;
1819         }
1820         lock_buffer(*bh);
1821         if (!buffer_dirty(*bh))
1822                 clear_buffer_uptodate(*bh);
1823         unlock_buffer(*bh);
1824         ll_rw_block(REQ_OP_READ, 0, 1, bh);
1825         wait_on_buffer(*bh);
1826         if (!buffer_uptodate(*bh)) {
1827                 mlog_errno(-EIO);
1828                 brelse(*bh);
1829                 *bh = NULL;
1830                 return -EIO;
1831         }
1832
1833         return 0;
1834 }
1835
1836 static int ocfs2_mount_volume(struct super_block *sb)
1837 {
1838         int status = 0;
1839         int unlock_super = 0;
1840         struct ocfs2_super *osb = OCFS2_SB(sb);
1841
1842         if (ocfs2_is_hard_readonly(osb))
1843                 goto leave;
1844
1845         status = ocfs2_dlm_init(osb);
1846         if (status < 0) {
1847                 mlog_errno(status);
1848                 goto leave;
1849         }
1850
1851         status = ocfs2_super_lock(osb, 1);
1852         if (status < 0) {
1853                 mlog_errno(status);
1854                 goto leave;
1855         }
1856         unlock_super = 1;
1857
1858         /* This will load up the node map and add ourselves to it. */
1859         status = ocfs2_find_slot(osb);
1860         if (status < 0) {
1861                 mlog_errno(status);
1862                 goto leave;
1863         }
1864
1865         /* load all node-local system inodes */
1866         status = ocfs2_init_local_system_inodes(osb);
1867         if (status < 0) {
1868                 mlog_errno(status);
1869                 goto leave;
1870         }
1871
1872         status = ocfs2_check_volume(osb);
1873         if (status < 0) {
1874                 mlog_errno(status);
1875                 goto leave;
1876         }
1877
1878         status = ocfs2_truncate_log_init(osb);
1879         if (status < 0)
1880                 mlog_errno(status);
1881
1882 leave:
1883         if (unlock_super)
1884                 ocfs2_super_unlock(osb, 1);
1885
1886         return status;
1887 }
1888
1889 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1890 {
1891         int tmp, hangup_needed = 0;
1892         struct ocfs2_super *osb = NULL;
1893         char nodestr[12];
1894
1895         trace_ocfs2_dismount_volume(sb);
1896
1897         BUG_ON(!sb);
1898         osb = OCFS2_SB(sb);
1899         BUG_ON(!osb);
1900
1901         debugfs_remove(osb->osb_ctxt);
1902
1903         /* Orphan scan should be stopped as early as possible */
1904         ocfs2_orphan_scan_stop(osb);
1905
1906         ocfs2_disable_quotas(osb);
1907
1908         /* All dquots should be freed by now */
1909         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1910         /* Wait for worker to be done with the work structure in osb */
1911         cancel_work_sync(&osb->dquot_drop_work);
1912
1913         ocfs2_shutdown_local_alloc(osb);
1914
1915         ocfs2_truncate_log_shutdown(osb);
1916
1917         /* This will disable recovery and flush any recovery work. */
1918         ocfs2_recovery_exit(osb);
1919
1920         ocfs2_journal_shutdown(osb);
1921
1922         ocfs2_sync_blockdev(sb);
1923
1924         ocfs2_purge_refcount_trees(osb);
1925
1926         /* No cluster connection means we've failed during mount, so skip
1927          * all the steps which depended on that to complete. */
1928         if (osb->cconn) {
1929                 tmp = ocfs2_super_lock(osb, 1);
1930                 if (tmp < 0) {
1931                         mlog_errno(tmp);
1932                         return;
1933                 }
1934         }
1935
1936         if (osb->slot_num != OCFS2_INVALID_SLOT)
1937                 ocfs2_put_slot(osb);
1938
1939         if (osb->cconn)
1940                 ocfs2_super_unlock(osb, 1);
1941
1942         ocfs2_release_system_inodes(osb);
1943
1944         /*
1945          * If we're dismounting due to mount error, mount.ocfs2 will clean
1946          * up heartbeat.  If we're a local mount, there is no heartbeat.
1947          * If we failed before we got a uuid_str yet, we can't stop
1948          * heartbeat.  Otherwise, do it.
1949          */
1950         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1951             !ocfs2_is_hard_readonly(osb))
1952                 hangup_needed = 1;
1953
1954         if (osb->cconn)
1955                 ocfs2_dlm_shutdown(osb, hangup_needed);
1956
1957         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1958         debugfs_remove(osb->osb_debug_root);
1959
1960         if (hangup_needed)
1961                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1962
1963         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1964
1965         if (ocfs2_mount_local(osb))
1966                 snprintf(nodestr, sizeof(nodestr), "local");
1967         else
1968                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1969
1970         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1971                osb->dev_str, nodestr);
1972
1973         ocfs2_delete_osb(osb);
1974         kfree(osb);
1975         sb->s_dev = 0;
1976         sb->s_fs_info = NULL;
1977 }
1978
1979 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1980                                 unsigned uuid_bytes)
1981 {
1982         int i, ret;
1983         char *ptr;
1984
1985         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1986
1987         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1988         if (osb->uuid_str == NULL)
1989                 return -ENOMEM;
1990
1991         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1992                 /* print with null */
1993                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1994                 if (ret != 2) /* drop super cleans up */
1995                         return -EINVAL;
1996                 /* then only advance past the last char */
1997                 ptr += 2;
1998         }
1999
2000         return 0;
2001 }
2002
2003 /* Make sure entire volume is addressable by our journal.  Requires
2004    osb_clusters_at_boot to be valid and for the journal to have been
2005    initialized by ocfs2_journal_init(). */
2006 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2007 {
2008         int status = 0;
2009         u64 max_block =
2010                 ocfs2_clusters_to_blocks(osb->sb,
2011                                          osb->osb_clusters_at_boot) - 1;
2012
2013         /* 32-bit block number is always OK. */
2014         if (max_block <= (u32)~0ULL)
2015                 goto out;
2016
2017         /* Volume is "huge", so see if our journal is new enough to
2018            support it. */
2019         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2020                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2021               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2022                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2023                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2024                      "Enable the 'block64' journal option with tunefs.ocfs2");
2025                 status = -EFBIG;
2026                 goto out;
2027         }
2028
2029  out:
2030         return status;
2031 }
2032
2033 static int ocfs2_initialize_super(struct super_block *sb,
2034                                   struct buffer_head *bh,
2035                                   int sector_size,
2036                                   struct ocfs2_blockcheck_stats *stats)
2037 {
2038         int status;
2039         int i, cbits, bbits;
2040         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2041         struct inode *inode = NULL;
2042         struct ocfs2_journal *journal;
2043         struct ocfs2_super *osb;
2044         u64 total_blocks;
2045
2046         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2047         if (!osb) {
2048                 status = -ENOMEM;
2049                 mlog_errno(status);
2050                 goto bail;
2051         }
2052
2053         sb->s_fs_info = osb;
2054         sb->s_op = &ocfs2_sops;
2055         sb->s_d_op = &ocfs2_dentry_ops;
2056         sb->s_export_op = &ocfs2_export_ops;
2057         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2058         sb->dq_op = &ocfs2_quota_operations;
2059         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2060         sb->s_xattr = ocfs2_xattr_handlers;
2061         sb->s_time_gran = 1;
2062         sb->s_flags |= MS_NOATIME;
2063         /* this is needed to support O_LARGEFILE */
2064         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2065         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2066         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2067         memcpy(&sb->s_uuid, di->id2.i_super.s_uuid,
2068                sizeof(di->id2.i_super.s_uuid));
2069
2070         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2071
2072         for (i = 0; i < 3; i++)
2073                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2074         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2075
2076         osb->sb = sb;
2077         osb->s_sectsize_bits = blksize_bits(sector_size);
2078         BUG_ON(!osb->s_sectsize_bits);
2079
2080         spin_lock_init(&osb->dc_task_lock);
2081         init_waitqueue_head(&osb->dc_event);
2082         osb->dc_work_sequence = 0;
2083         osb->dc_wake_sequence = 0;
2084         INIT_LIST_HEAD(&osb->blocked_lock_list);
2085         osb->blocked_lock_count = 0;
2086         spin_lock_init(&osb->osb_lock);
2087         spin_lock_init(&osb->osb_xattr_lock);
2088         ocfs2_init_steal_slots(osb);
2089
2090         mutex_init(&osb->system_file_mutex);
2091
2092         atomic_set(&osb->alloc_stats.moves, 0);
2093         atomic_set(&osb->alloc_stats.local_data, 0);
2094         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2095         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2096         atomic_set(&osb->alloc_stats.bg_extends, 0);
2097
2098         /* Copy the blockcheck stats from the superblock probe */
2099         osb->osb_ecc_stats = *stats;
2100
2101         ocfs2_init_node_maps(osb);
2102
2103         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2104                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2105
2106         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2107         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2108                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2109                      osb->max_slots);
2110                 status = -EINVAL;
2111                 goto bail;
2112         }
2113
2114         ocfs2_orphan_scan_init(osb);
2115
2116         status = ocfs2_recovery_init(osb);
2117         if (status) {
2118                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2119                 mlog_errno(status);
2120                 goto bail;
2121         }
2122
2123         init_waitqueue_head(&osb->checkpoint_event);
2124
2125         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2126
2127         osb->slot_num = OCFS2_INVALID_SLOT;
2128
2129         osb->s_xattr_inline_size = le16_to_cpu(
2130                                         di->id2.i_super.s_xattr_inline_size);
2131
2132         osb->local_alloc_state = OCFS2_LA_UNUSED;
2133         osb->local_alloc_bh = NULL;
2134         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2135
2136         init_waitqueue_head(&osb->osb_mount_event);
2137
2138         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2139         if (status) {
2140                 mlog_errno(status);
2141                 goto bail;
2142         }
2143
2144         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2145         if (!osb->vol_label) {
2146                 mlog(ML_ERROR, "unable to alloc vol label\n");
2147                 status = -ENOMEM;
2148                 goto bail;
2149         }
2150
2151         osb->slot_recovery_generations =
2152                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2153                         GFP_KERNEL);
2154         if (!osb->slot_recovery_generations) {
2155                 status = -ENOMEM;
2156                 mlog_errno(status);
2157                 goto bail;
2158         }
2159
2160         init_waitqueue_head(&osb->osb_wipe_event);
2161         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2162                                         sizeof(*osb->osb_orphan_wipes),
2163                                         GFP_KERNEL);
2164         if (!osb->osb_orphan_wipes) {
2165                 status = -ENOMEM;
2166                 mlog_errno(status);
2167                 goto bail;
2168         }
2169
2170         osb->osb_rf_lock_tree = RB_ROOT;
2171
2172         osb->s_feature_compat =
2173                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2174         osb->s_feature_ro_compat =
2175                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2176         osb->s_feature_incompat =
2177                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2178
2179         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2180                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2181                      "optional features (%x).\n", i);
2182                 status = -EINVAL;
2183                 goto bail;
2184         }
2185         if (!sb_rdonly(osb->sb) && (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2186                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2187                      "unsupported optional features (%x).\n", i);
2188                 status = -EINVAL;
2189                 goto bail;
2190         }
2191
2192         if (ocfs2_clusterinfo_valid(osb)) {
2193                 /*
2194                  * ci_stack and ci_cluster in ocfs2_cluster_info may not be null
2195                  * terminated, so make sure no overflow happens here by using
2196                  * memcpy. Destination strings will always be null terminated
2197                  * because osb is allocated using kzalloc.
2198                  */
2199                 osb->osb_stackflags =
2200                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2201                 memcpy(osb->osb_cluster_stack,
2202                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2203                        OCFS2_STACK_LABEL_LEN);
2204                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2205                         mlog(ML_ERROR,
2206                              "couldn't mount because of an invalid "
2207                              "cluster stack label (%s) \n",
2208                              osb->osb_cluster_stack);
2209                         status = -EINVAL;
2210                         goto bail;
2211                 }
2212                 memcpy(osb->osb_cluster_name,
2213                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2214                         OCFS2_CLUSTER_NAME_LEN);
2215         } else {
2216                 /* The empty string is identical with classic tools that
2217                  * don't know about s_cluster_info. */
2218                 osb->osb_cluster_stack[0] = '\0';
2219         }
2220
2221         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2222
2223         /* FIXME
2224          * This should be done in ocfs2_journal_init(), but unknown
2225          * ordering issues will cause the filesystem to crash.
2226          * If anyone wants to figure out what part of the code
2227          * refers to osb->journal before ocfs2_journal_init() is run,
2228          * be my guest.
2229          */
2230         /* initialize our journal structure */
2231
2232         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2233         if (!journal) {
2234                 mlog(ML_ERROR, "unable to alloc journal\n");
2235                 status = -ENOMEM;
2236                 goto bail;
2237         }
2238         osb->journal = journal;
2239         journal->j_osb = osb;
2240
2241         atomic_set(&journal->j_num_trans, 0);
2242         init_rwsem(&journal->j_trans_barrier);
2243         init_waitqueue_head(&journal->j_checkpointed);
2244         spin_lock_init(&journal->j_lock);
2245         journal->j_trans_id = (unsigned long) 1;
2246         INIT_LIST_HEAD(&journal->j_la_cleanups);
2247         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2248         journal->j_state = OCFS2_JOURNAL_FREE;
2249
2250         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2251         init_llist_head(&osb->dquot_drop_list);
2252
2253         /* get some pseudo constants for clustersize bits */
2254         osb->s_clustersize_bits =
2255                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2256         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2257
2258         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2259             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2260                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2261                      osb->s_clustersize);
2262                 status = -EINVAL;
2263                 goto bail;
2264         }
2265
2266         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2267                                                 le32_to_cpu(di->i_clusters));
2268
2269         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2270                                            total_blocks);
2271         if (status) {
2272                 mlog(ML_ERROR, "Volume too large "
2273                      "to mount safely on this system");
2274                 status = -EFBIG;
2275                 goto bail;
2276         }
2277
2278         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2279                                  sizeof(di->id2.i_super.s_uuid))) {
2280                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2281                 status = -ENOMEM;
2282                 goto bail;
2283         }
2284
2285         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2286                 OCFS2_MAX_VOL_LABEL_LEN);
2287         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2288         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2289         osb->first_cluster_group_blkno =
2290                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2291         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2292         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2293         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2294                                      (unsigned long long)osb->root_blkno,
2295                                      (unsigned long long)osb->system_dir_blkno,
2296                                      osb->s_clustersize_bits);
2297
2298         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2299         if (!osb->osb_dlm_debug) {
2300                 status = -ENOMEM;
2301                 mlog_errno(status);
2302                 goto bail;
2303         }
2304
2305         atomic_set(&osb->vol_state, VOLUME_INIT);
2306
2307         /* load root, system_dir, and all global system inodes */
2308         status = ocfs2_init_global_system_inodes(osb);
2309         if (status < 0) {
2310                 mlog_errno(status);
2311                 goto bail;
2312         }
2313
2314         /*
2315          * global bitmap
2316          */
2317         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2318                                             OCFS2_INVALID_SLOT);
2319         if (!inode) {
2320                 status = -EINVAL;
2321                 mlog_errno(status);
2322                 goto bail;
2323         }
2324
2325         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2326         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2327         iput(inode);
2328
2329         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2330                                  osb->s_feature_incompat) * 8;
2331
2332         status = ocfs2_init_slot_info(osb);
2333         if (status < 0) {
2334                 mlog_errno(status);
2335                 goto bail;
2336         }
2337         cleancache_init_shared_fs(sb);
2338
2339         osb->ocfs2_wq = alloc_ordered_workqueue("ocfs2_wq", WQ_MEM_RECLAIM);
2340         if (!osb->ocfs2_wq) {
2341                 status = -ENOMEM;
2342                 mlog_errno(status);
2343         }
2344
2345 bail:
2346         return status;
2347 }
2348
2349 /*
2350  * will return: -EAGAIN if it is ok to keep searching for superblocks
2351  *              -EINVAL if there is a bad superblock
2352  *              0 on success
2353  */
2354 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2355                                struct buffer_head *bh,
2356                                u32 blksz,
2357                                struct ocfs2_blockcheck_stats *stats)
2358 {
2359         int status = -EAGAIN;
2360
2361         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2362                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2363                 /* We have to do a raw check of the feature here */
2364                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2365                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2366                         status = ocfs2_block_check_validate(bh->b_data,
2367                                                             bh->b_size,
2368                                                             &di->i_check,
2369                                                             stats);
2370                         if (status)
2371                                 goto out;
2372                 }
2373                 status = -EINVAL;
2374                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2375                         mlog(ML_ERROR, "found superblock with incorrect block "
2376                              "size: found %u, should be %u\n",
2377                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2378                                blksz);
2379                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2380                            OCFS2_MAJOR_REV_LEVEL ||
2381                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2382                            OCFS2_MINOR_REV_LEVEL) {
2383                         mlog(ML_ERROR, "found superblock with bad version: "
2384                              "found %u.%u, should be %u.%u\n",
2385                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2386                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2387                              OCFS2_MAJOR_REV_LEVEL,
2388                              OCFS2_MINOR_REV_LEVEL);
2389                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2390                         mlog(ML_ERROR, "bad block number on superblock: "
2391                              "found %llu, should be %llu\n",
2392                              (unsigned long long)le64_to_cpu(di->i_blkno),
2393                              (unsigned long long)bh->b_blocknr);
2394                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2395                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2396                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2397                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2398                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2399                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2400                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2401                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2402                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2403                         mlog(ML_ERROR,
2404                              "Superblock slots found greater than file system "
2405                              "maximum: found %u, max %u\n",
2406                              le16_to_cpu(di->id2.i_super.s_max_slots),
2407                              OCFS2_MAX_SLOTS);
2408                 } else {
2409                         /* found it! */
2410                         status = 0;
2411                 }
2412         }
2413
2414 out:
2415         if (status && status != -EAGAIN)
2416                 mlog_errno(status);
2417         return status;
2418 }
2419
2420 static int ocfs2_check_volume(struct ocfs2_super *osb)
2421 {
2422         int status;
2423         int dirty;
2424         int local;
2425         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2426                                                   * recover
2427                                                   * ourselves. */
2428
2429         /* Init our journal object. */
2430         status = ocfs2_journal_init(osb->journal, &dirty);
2431         if (status < 0) {
2432                 mlog(ML_ERROR, "Could not initialize journal!\n");
2433                 goto finally;
2434         }
2435
2436         /* Now that journal has been initialized, check to make sure
2437            entire volume is addressable. */
2438         status = ocfs2_journal_addressable(osb);
2439         if (status)
2440                 goto finally;
2441
2442         /* If the journal was unmounted cleanly then we don't want to
2443          * recover anything. Otherwise, journal_load will do that
2444          * dirty work for us :) */
2445         if (!dirty) {
2446                 status = ocfs2_journal_wipe(osb->journal, 0);
2447                 if (status < 0) {
2448                         mlog_errno(status);
2449                         goto finally;
2450                 }
2451         } else {
2452                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2453                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2454         }
2455
2456         local = ocfs2_mount_local(osb);
2457
2458         /* will play back anything left in the journal. */
2459         status = ocfs2_journal_load(osb->journal, local, dirty);
2460         if (status < 0) {
2461                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2462                 goto finally;
2463         }
2464
2465         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2466                 jbd2_journal_set_features(osb->journal->j_journal,
2467                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2468                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2469         else
2470                 jbd2_journal_clear_features(osb->journal->j_journal,
2471                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2472                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2473
2474         if (dirty) {
2475                 /* recover my local alloc if we didn't unmount cleanly. */
2476                 status = ocfs2_begin_local_alloc_recovery(osb,
2477                                                           osb->slot_num,
2478                                                           &local_alloc);
2479                 if (status < 0) {
2480                         mlog_errno(status);
2481                         goto finally;
2482                 }
2483                 /* we complete the recovery process after we've marked
2484                  * ourselves as mounted. */
2485         }
2486
2487         status = ocfs2_load_local_alloc(osb);
2488         if (status < 0) {
2489                 mlog_errno(status);
2490                 goto finally;
2491         }
2492
2493         if (dirty) {
2494                 /* Recovery will be completed after we've mounted the
2495                  * rest of the volume. */
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_rdonly(sb) && (ocfs2_is_soft_readonly(osb) || ocfs2_is_hard_readonly(osb)))
2578                         return rv;
2579
2580                 pr_crit("OCFS2: File system is now read-only.\n");
2581                 sb->s_flags |= MS_RDONLY;
2582                 ocfs2_set_ro_flag(osb, 0);
2583         }
2584
2585         return rv;
2586 }
2587
2588 int __ocfs2_error(struct super_block *sb, const char *function,
2589                   const char *fmt, ...)
2590 {
2591         struct va_format vaf;
2592         va_list args;
2593
2594         va_start(args, fmt);
2595         vaf.fmt = fmt;
2596         vaf.va = &args;
2597
2598         /* Not using mlog here because we want to show the actual
2599          * function the error came from. */
2600         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2601                sb->s_id, function, &vaf);
2602
2603         va_end(args);
2604
2605         return ocfs2_handle_error(sb);
2606 }
2607
2608 /* Handle critical errors. This is intentionally more drastic than
2609  * ocfs2_handle_error, so we only use for things like journal errors,
2610  * etc. */
2611 void __ocfs2_abort(struct super_block *sb, const char *function,
2612                    const char *fmt, ...)
2613 {
2614         struct va_format vaf;
2615         va_list args;
2616
2617         va_start(args, fmt);
2618
2619         vaf.fmt = fmt;
2620         vaf.va = &args;
2621
2622         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2623                sb->s_id, function, &vaf);
2624
2625         va_end(args);
2626
2627         /* We don't have the cluster support yet to go straight to
2628          * hard readonly in here. Until then, we want to keep
2629          * ocfs2_abort() so that we can at least mark critical
2630          * errors.
2631          *
2632          * TODO: This should abort the journal and alert other nodes
2633          * that our slot needs recovery. */
2634
2635         /* Force a panic(). This stinks, but it's better than letting
2636          * things continue without having a proper hard readonly
2637          * here. */
2638         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2639                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2640         ocfs2_handle_error(sb);
2641 }
2642
2643 /*
2644  * Void signal blockers, because in-kernel sigprocmask() only fails
2645  * when SIG_* is wrong.
2646  */
2647 void ocfs2_block_signals(sigset_t *oldset)
2648 {
2649         int rc;
2650         sigset_t blocked;
2651
2652         sigfillset(&blocked);
2653         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2654         BUG_ON(rc);
2655 }
2656
2657 void ocfs2_unblock_signals(sigset_t *oldset)
2658 {
2659         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2660         BUG_ON(rc);
2661 }
2662
2663 module_init(ocfs2_init);
2664 module_exit(ocfs2_exit);