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
  45 #define MLOG_MASK_PREFIX ML_SUPER
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 /* this should be the only file to include a version 1 header */
  51 #include "ocfs1_fs_compat.h"
  52
  53 #include "alloc.h"
  54 #include "blockcheck.h"
  55 #include "dlmglue.h"
  56 #include "export.h"
  57 #include "extent_map.h"
  58 #include "heartbeat.h"
  59 #include "inode.h"
  60 #include "journal.h"
  61 #include "localalloc.h"
  62 #include "namei.h"
  63 #include "slot_map.h"
  64 #include "super.h"
  65 #include "sysfile.h"
  66 #include "uptodate.h"
  67 #include "ver.h"
  68 #include "xattr.h"
  69 #include "quota.h"
  70 #include "refcounttree.h"
  71 #include "suballoc.h"
  72
  73 #include "buffer_head_io.h"
  74
  75 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  76 struct kmem_cache *ocfs2_dquot_cachep;
  77 struct kmem_cache *ocfs2_qf_chunk_cachep;
  78
  79 /* OCFS2 needs to schedule several differnt types of work which
  80  * require cluster locking, disk I/O, recovery waits, etc. Since these
  81  * types of work tend to be heavy we avoid using the kernel events
  82  * workqueue and schedule on our own. */
  83 struct workqueue_struct *ocfs2_wq = NULL;
  84
  85 static struct dentry *ocfs2_debugfs_root = NULL;
  86
  87 MODULE_AUTHOR("Oracle");
  88 MODULE_LICENSE("GPL");
  89
  90 struct mount_options
  91 {
  92         unsigned long   commit_interval;
  93         unsigned long   mount_opt;
  94         unsigned int    atime_quantum;
  95         signed 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 vfsmount *mnt);
 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 const struct super_operations ocfs2_sops = {
 143         .statfs         = ocfs2_statfs,
 144         .alloc_inode    = ocfs2_alloc_inode,
 145         .destroy_inode  = ocfs2_destroy_inode,
 146         .drop_inode     = ocfs2_drop_inode,
 147         .evict_inode    = ocfs2_evict_inode,
 148         .sync_fs        = ocfs2_sync_fs,
 149         .put_super      = ocfs2_put_super,
 150         .remount_fs     = ocfs2_remount,
 151         .show_options   = ocfs2_show_options,
 152         .quota_read     = ocfs2_quota_read,
 153         .quota_write    = ocfs2_quota_write,
 154 };
 155
 156 enum {
 157         Opt_barrier,
 158         Opt_err_panic,
 159         Opt_err_ro,
 160         Opt_intr,
 161         Opt_nointr,
 162         Opt_hb_none,
 163         Opt_hb_local,
 164         Opt_hb_global,
 165         Opt_data_ordered,
 166         Opt_data_writeback,
 167         Opt_atime_quantum,
 168         Opt_slot,
 169         Opt_commit,
 170         Opt_localalloc,
 171         Opt_localflocks,
 172         Opt_stack,
 173         Opt_user_xattr,
 174         Opt_nouser_xattr,
 175         Opt_inode64,
 176         Opt_acl,
 177         Opt_noacl,
 178         Opt_usrquota,
 179         Opt_grpquota,
 180         Opt_coherency_buffered,
 181         Opt_coherency_full,
 182         Opt_resv_level,
 183         Opt_dir_resv_level,
 184         Opt_err,
 185 };
 186
 187 static const match_table_t tokens = {
 188         {Opt_barrier, "barrier=%u"},
 189         {Opt_err_panic, "errors=panic"},
 190         {Opt_err_ro, "errors=remount-ro"},
 191         {Opt_intr, "intr"},
 192         {Opt_nointr, "nointr"},
 193         {Opt_hb_none, OCFS2_HB_NONE},
 194         {Opt_hb_local, OCFS2_HB_LOCAL},
 195         {Opt_hb_global, OCFS2_HB_GLOBAL},
 196         {Opt_data_ordered, "data=ordered"},
 197         {Opt_data_writeback, "data=writeback"},
 198         {Opt_atime_quantum, "atime_quantum=%u"},
 199         {Opt_slot, "preferred_slot=%u"},
 200         {Opt_commit, "commit=%u"},
 201         {Opt_localalloc, "localalloc=%d"},
 202         {Opt_localflocks, "localflocks"},
 203         {Opt_stack, "cluster_stack=%s"},
 204         {Opt_user_xattr, "user_xattr"},
 205         {Opt_nouser_xattr, "nouser_xattr"},
 206         {Opt_inode64, "inode64"},
 207         {Opt_acl, "acl"},
 208         {Opt_noacl, "noacl"},
 209         {Opt_usrquota, "usrquota"},
 210         {Opt_grpquota, "grpquota"},
 211         {Opt_coherency_buffered, "coherency=buffered"},
 212         {Opt_coherency_full, "coherency=full"},
 213         {Opt_resv_level, "resv_level=%u"},
 214         {Opt_dir_resv_level, "dir_resv_level=%u"},
 215         {Opt_err, NULL}
 216 };
 217
 218 #ifdef CONFIG_DEBUG_FS
 219 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 220 {
 221         struct ocfs2_cluster_connection *cconn = osb->cconn;
 222         struct ocfs2_recovery_map *rm = osb->recovery_map;
 223         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 224         int i, out = 0;
 225
 226         out += snprintf(buf + out, len - out,
 227                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 228                         "Device", osb->dev_str, osb->uuid_str,
 229                         osb->fs_generation, osb->vol_label);
 230
 231         out += snprintf(buf + out, len - out,
 232                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 233                         atomic_read(&osb->vol_state), osb->osb_flags);
 234
 235         out += snprintf(buf + out, len - out,
 236                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 237                         osb->sb->s_blocksize, osb->s_clustersize);
 238
 239         out += snprintf(buf + out, len - out,
 240                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 241                         "ROcompat: 0x%X\n",
 242                         "Features", osb->s_feature_compat,
 243                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 244
 245         out += snprintf(buf + out, len - out,
 246                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 247                         osb->s_mount_opt, osb->s_atime_quantum);
 248
 249         if (cconn) {
 250                 out += snprintf(buf + out, len - out,
 251                                 "%10s => Stack: %s  Name: %*s  "
 252                                 "Version: %d.%d\n", "Cluster",
 253                                 (*osb->osb_cluster_stack == '\0' ?
 254                                  "o2cb" : osb->osb_cluster_stack),
 255                                 cconn->cc_namelen, cconn->cc_name,
 256                                 cconn->cc_version.pv_major,
 257                                 cconn->cc_version.pv_minor);
 258         }
 259
 260         spin_lock(&osb->dc_task_lock);
 261         out += snprintf(buf + out, len - out,
 262                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 263                         "WorkSeq: %lu\n", "DownCnvt",
 264                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 265                         osb->blocked_lock_count, osb->dc_wake_sequence,
 266                         osb->dc_work_sequence);
 267         spin_unlock(&osb->dc_task_lock);
 268
 269         spin_lock(&osb->osb_lock);
 270         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 271                         "Recovery",
 272                         (osb->recovery_thread_task ?
 273                          task_pid_nr(osb->recovery_thread_task) : -1));
 274         if (rm->rm_used == 0)
 275                 out += snprintf(buf + out, len - out, " None\n");
 276         else {
 277                 for (i = 0; i < rm->rm_used; i++)
 278                         out += snprintf(buf + out, len - out, " %d",
 279                                         rm->rm_entries[i]);
 280                 out += snprintf(buf + out, len - out, "\n");
 281         }
 282         spin_unlock(&osb->osb_lock);
 283
 284         out += snprintf(buf + out, len - out,
 285                         "%10s => Pid: %d  Interval: %lu  Needs: %d\n", "Commit",
 286                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 287                         osb->osb_commit_interval,
 288                         atomic_read(&osb->needs_checkpoint));
 289
 290         out += snprintf(buf + out, len - out,
 291                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 292                         "Journal", osb->journal->j_state,
 293                         osb->journal->j_trans_id,
 294                         atomic_read(&osb->journal->j_num_trans));
 295
 296         out += snprintf(buf + out, len - out,
 297                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 298                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 299                         "Stats",
 300                         atomic_read(&osb->alloc_stats.bitmap_data),
 301                         atomic_read(&osb->alloc_stats.local_data),
 302                         atomic_read(&osb->alloc_stats.bg_allocs),
 303                         atomic_read(&osb->alloc_stats.moves),
 304                         atomic_read(&osb->alloc_stats.bg_extends));
 305
 306         out += snprintf(buf + out, len - out,
 307                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 308                         "Default: %u bits\n",
 309                         "LocalAlloc", osb->local_alloc_state,
 310                         (unsigned long long)osb->la_last_gd,
 311                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 312
 313         spin_lock(&osb->osb_lock);
 314         out += snprintf(buf + out, len - out,
 315                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 316                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 317                         osb->s_inode_steal_slot,
 318                         atomic_read(&osb->s_num_inodes_stolen),
 319                         osb->s_meta_steal_slot,
 320                         atomic_read(&osb->s_num_meta_stolen));
 321         spin_unlock(&osb->osb_lock);
 322
 323         out += snprintf(buf + out, len - out, "OrphanScan => ");
 324         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 325                         os->os_count, os->os_seqno);
 326         out += snprintf(buf + out, len - out, " Last Scan: ");
 327         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 328                 out += snprintf(buf + out, len - out, "Disabled\n");
 329         else
 330                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 331                                 (get_seconds() - os->os_scantime.tv_sec));
 332
 333         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 334                         "Slots", "Num", "RecoGen");
 335         for (i = 0; i < osb->max_slots; ++i) {
 336                 out += snprintf(buf + out, len - out,
 337                                 "%10s  %c %3d  %10d\n",
 338                                 " ",
 339                                 (i == osb->slot_num ? '*' : ' '),
 340                                 i, osb->slot_recovery_generations[i]);
 341         }
 342
 343         return out;
 344 }
 345
 346 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 347 {
 348         struct ocfs2_super *osb = inode->i_private;
 349         char *buf = NULL;
 350
 351         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 352         if (!buf)
 353                 goto bail;
 354
 355         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 356
 357         file->private_data = buf;
 358
 359         return 0;
 360 bail:
 361         return -ENOMEM;
 362 }
 363
 364 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 365 {
 366         kfree(file->private_data);
 367         return 0;
 368 }
 369
 370 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 371                                 size_t nbytes, loff_t *ppos)
 372 {
 373         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 374                                        i_size_read(file->f_mapping->host));
 375 }
 376 #else
 377 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 378 {
 379         return 0;
 380 }
 381 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 382 {
 383         return 0;
 384 }
 385 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 386                                 size_t nbytes, loff_t *ppos)
 387 {
 388         return 0;
 389 }
 390 #endif  /* CONFIG_DEBUG_FS */
 391
 392 static const struct file_operations ocfs2_osb_debug_fops = {
 393         .open =         ocfs2_osb_debug_open,
 394         .release =      ocfs2_debug_release,
 395         .read =         ocfs2_debug_read,
 396         .llseek =       generic_file_llseek,
 397 };
 398
 399 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 400 {
 401         int status;
 402         tid_t target;
 403         struct ocfs2_super *osb = OCFS2_SB(sb);
 404
 405         if (ocfs2_is_hard_readonly(osb))
 406                 return -EROFS;
 407
 408         if (wait) {
 409                 status = ocfs2_flush_truncate_log(osb);
 410                 if (status < 0)
 411                         mlog_errno(status);
 412         } else {
 413                 ocfs2_schedule_truncate_log_flush(osb, 0);
 414         }
 415
 416         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 417                                       &target)) {
 418                 if (wait)
 419                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 420                                              target);
 421         }
 422         return 0;
 423 }
 424
 425 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 426 {
 427         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 428             && (ino == USER_QUOTA_SYSTEM_INODE
 429                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 430                 return 0;
 431         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 432             && (ino == GROUP_QUOTA_SYSTEM_INODE
 433                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 434                 return 0;
 435         return 1;
 436 }
 437
 438 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 439 {
 440         struct inode *new = NULL;
 441         int status = 0;
 442         int i;
 443
 444         mlog_entry_void();
 445
 446         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 447         if (IS_ERR(new)) {
 448                 status = PTR_ERR(new);
 449                 mlog_errno(status);
 450                 goto bail;
 451         }
 452         osb->root_inode = new;
 453
 454         new = ocfs2_iget(osb, osb->system_dir_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->sys_root_inode = new;
 461
 462         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 463              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 464                 if (!ocfs2_need_system_inode(osb, i))
 465                         continue;
 466                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 467                 if (!new) {
 468                         ocfs2_release_system_inodes(osb);
 469                         status = -EINVAL;
 470                         mlog_errno(status);
 471                         /* FIXME: Should ERROR_RO_FS */
 472                         mlog(ML_ERROR, "Unable to load system inode %d, "
 473                              "possibly corrupt fs?", i);
 474                         goto bail;
 475                 }
 476                 // the array now has one ref, so drop this one
 477                 iput(new);
 478         }
 479
 480 bail:
 481         mlog_exit(status);
 482         return status;
 483 }
 484
 485 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 486 {
 487         struct inode *new = NULL;
 488         int status = 0;
 489         int i;
 490
 491         mlog_entry_void();
 492
 493         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 494              i < NUM_SYSTEM_INODES;
 495              i++) {
 496                 if (!ocfs2_need_system_inode(osb, i))
 497                         continue;
 498                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 499                 if (!new) {
 500                         ocfs2_release_system_inodes(osb);
 501                         status = -EINVAL;
 502                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 503                              status, i, osb->slot_num);
 504                         goto bail;
 505                 }
 506                 /* the array now has one ref, so drop this one */
 507                 iput(new);
 508         }
 509
 510 bail:
 511         mlog_exit(status);
 512         return status;
 513 }
 514
 515 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 516 {
 517         int i;
 518         struct inode *inode;
 519
 520         mlog_entry_void();
 521
 522         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 523                 inode = osb->global_system_inodes[i];
 524                 if (inode) {
 525                         iput(inode);
 526                         osb->global_system_inodes[i] = NULL;
 527                 }
 528         }
 529
 530         inode = osb->sys_root_inode;
 531         if (inode) {
 532                 iput(inode);
 533                 osb->sys_root_inode = NULL;
 534         }
 535
 536         inode = osb->root_inode;
 537         if (inode) {
 538                 iput(inode);
 539                 osb->root_inode = NULL;
 540         }
 541
 542         if (!osb->local_system_inodes)
 543                 goto out;
 544
 545         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 546                 if (osb->local_system_inodes[i]) {
 547                         iput(osb->local_system_inodes[i]);
 548                         osb->local_system_inodes[i] = NULL;
 549                 }
 550         }
 551
 552         kfree(osb->local_system_inodes);
 553         osb->local_system_inodes = NULL;
 554
 555 out:
 556         mlog_exit(0);
 557 }
 558
 559 /* We're allocating fs objects, use GFP_NOFS */
 560 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 561 {
 562         struct ocfs2_inode_info *oi;
 563
 564         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 565         if (!oi)
 566                 return NULL;
 567
 568         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 569         return &oi->vfs_inode;
 570 }
 571
 572 static void ocfs2_i_callback(struct rcu_head *head)
 573 {
 574         struct inode *inode = container_of(head, struct inode, i_rcu);
 575         INIT_LIST_HEAD(&inode->i_dentry);
 576         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 577 }
 578
 579 static void ocfs2_destroy_inode(struct inode *inode)
 580 {
 581         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 582 }
 583
 584 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 585                                                 unsigned int cbits)
 586 {
 587         unsigned int bytes = 1 << cbits;
 588         unsigned int trim = bytes;
 589         unsigned int bitshift = 32;
 590
 591         /*
 592          * i_size and all block offsets in ocfs2 are always 64 bits
 593          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 594          * 64 bit platforms, cluster size will be the limiting factor.
 595          */
 596
 597 #if BITS_PER_LONG == 32
 598 # if defined(CONFIG_LBDAF)
 599         BUILD_BUG_ON(sizeof(sector_t) != 8);
 600         /*
 601          * We might be limited by page cache size.
 602          */
 603         if (bytes > PAGE_CACHE_SIZE) {
 604                 bytes = PAGE_CACHE_SIZE;
 605                 trim = 1;
 606                 /*
 607                  * Shift by 31 here so that we don't get larger than
 608                  * MAX_LFS_FILESIZE
 609                  */
 610                 bitshift = 31;
 611         }
 612 # else
 613         /*
 614          * We are limited by the size of sector_t. Use block size, as
 615          * that's what we expose to the VFS.
 616          */
 617         bytes = 1 << bbits;
 618         trim = 1;
 619         bitshift = 31;
 620 # endif
 621 #endif
 622
 623         /*
 624          * Trim by a whole cluster when we can actually approach the
 625          * on-disk limits. Otherwise we can overflow i_clusters when
 626          * an extent start is at the max offset.
 627          */
 628         return (((unsigned long long)bytes) << bitshift) - trim;
 629 }
 630
 631 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 632 {
 633         int incompat_features;
 634         int ret = 0;
 635         struct mount_options parsed_options;
 636         struct ocfs2_super *osb = OCFS2_SB(sb);
 637         u32 tmp;
 638
 639         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 640             !ocfs2_check_set_options(sb, &parsed_options)) {
 641                 ret = -EINVAL;
 642                 goto out;
 643         }
 644
 645         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 646                 OCFS2_MOUNT_HB_NONE;
 647         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 648                 ret = -EINVAL;
 649                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 650                 goto out;
 651         }
 652
 653         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 654             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 655                 ret = -EINVAL;
 656                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 657                 goto out;
 658         }
 659
 660         /* Probably don't want this on remount; it might
 661          * mess with other nodes */
 662         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 663             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 664                 ret = -EINVAL;
 665                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 666                 goto out;
 667         }
 668
 669         /* We're going to/from readonly mode. */
 670         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 671                 /* Disable quota accounting before remounting RO */
 672                 if (*flags & MS_RDONLY) {
 673                         ret = ocfs2_susp_quotas(osb, 0);
 674                         if (ret < 0)
 675                                 goto out;
 676                 }
 677                 /* Lock here so the check of HARD_RO and the potential
 678                  * setting of SOFT_RO is atomic. */
 679                 spin_lock(&osb->osb_lock);
 680                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 681                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 682                         ret = -EROFS;
 683                         goto unlock_osb;
 684                 }
 685
 686                 if (*flags & MS_RDONLY) {
 687                         mlog(0, "Going to ro mode.\n");
 688                         sb->s_flags |= MS_RDONLY;
 689                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 690                 } else {
 691                         mlog(0, "Making ro filesystem writeable.\n");
 692
 693                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 694                                 mlog(ML_ERROR, "Cannot remount RDWR "
 695                                      "filesystem due to previous errors.\n");
 696                                 ret = -EROFS;
 697                                 goto unlock_osb;
 698                         }
 699                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 700                         if (incompat_features) {
 701                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 702                                      "of unsupported optional features "
 703                                      "(%x).\n", incompat_features);
 704                                 ret = -EINVAL;
 705                                 goto unlock_osb;
 706                         }
 707                         sb->s_flags &= ~MS_RDONLY;
 708                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 709                 }
 710 unlock_osb:
 711                 spin_unlock(&osb->osb_lock);
 712                 /* Enable quota accounting after remounting RW */
 713                 if (!ret && !(*flags & MS_RDONLY)) {
 714                         if (sb_any_quota_suspended(sb))
 715                                 ret = ocfs2_susp_quotas(osb, 1);
 716                         else
 717                                 ret = ocfs2_enable_quotas(osb);
 718                         if (ret < 0) {
 719                                 /* Return back changes... */
 720                                 spin_lock(&osb->osb_lock);
 721                                 sb->s_flags |= MS_RDONLY;
 722                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 723                                 spin_unlock(&osb->osb_lock);
 724                                 goto out;
 725                         }
 726                 }
 727         }
 728
 729         if (!ret) {
 730                 /* Only save off the new mount options in case of a successful
 731                  * remount. */
 732                 osb->s_mount_opt = parsed_options.mount_opt;
 733                 osb->s_atime_quantum = parsed_options.atime_quantum;
 734                 osb->preferred_slot = parsed_options.slot;
 735                 if (parsed_options.commit_interval)
 736                         osb->osb_commit_interval = parsed_options.commit_interval;
 737
 738                 if (!ocfs2_is_hard_readonly(osb))
 739                         ocfs2_set_journal_params(osb);
 740
 741                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 742                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 743                                                         MS_POSIXACL : 0);
 744         }
 745 out:
 746         return ret;
 747 }
 748
 749 static int ocfs2_sb_probe(struct super_block *sb,
 750                           struct buffer_head **bh,
 751                           int *sector_size,
 752                           struct ocfs2_blockcheck_stats *stats)
 753 {
 754         int status, tmpstat;
 755         struct ocfs1_vol_disk_hdr *hdr;
 756         struct ocfs2_dinode *di;
 757         int blksize;
 758
 759         *bh = NULL;
 760
 761         /* may be > 512 */
 762         *sector_size = bdev_logical_block_size(sb->s_bdev);
 763         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 764                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 765                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 766                 status = -EINVAL;
 767                 goto bail;
 768         }
 769
 770         /* Can this really happen? */
 771         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 772                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 773
 774         /* check block zero for old format */
 775         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 776         if (status < 0) {
 777                 mlog_errno(status);
 778                 goto bail;
 779         }
 780         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 781         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 782                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 783                      hdr->major_version, hdr->minor_version);
 784                 status = -EINVAL;
 785         }
 786         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 787                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 788                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 789                      hdr->signature);
 790                 status = -EINVAL;
 791         }
 792         brelse(*bh);
 793         *bh = NULL;
 794         if (status < 0) {
 795                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 796                      "upgraded before mounting with ocfs v2\n");
 797                 goto bail;
 798         }
 799
 800         /*
 801          * Now check at magic offset for 512, 1024, 2048, 4096
 802          * blocksizes.  4096 is the maximum blocksize because it is
 803          * the minimum clustersize.
 804          */
 805         status = -EINVAL;
 806         for (blksize = *sector_size;
 807              blksize <= OCFS2_MAX_BLOCKSIZE;
 808              blksize <<= 1) {
 809                 tmpstat = ocfs2_get_sector(sb, bh,
 810                                            OCFS2_SUPER_BLOCK_BLKNO,
 811                                            blksize);
 812                 if (tmpstat < 0) {
 813                         status = tmpstat;
 814                         mlog_errno(status);
 815                         break;
 816                 }
 817                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 818                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 819                 spin_lock_init(&stats->b_lock);
 820                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 821                 if (tmpstat < 0) {
 822                         brelse(*bh);
 823                         *bh = NULL;
 824                 }
 825                 if (tmpstat != -EAGAIN) {
 826                         status = tmpstat;
 827                         break;
 828                 }
 829         }
 830
 831 bail:
 832         return status;
 833 }
 834
 835 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 836 {
 837         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 838
 839         if (osb->s_mount_opt & hb_enabled) {
 840                 if (ocfs2_mount_local(osb)) {
 841                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 842                              "mounted device.\n");
 843                         return -EINVAL;
 844                 }
 845                 if (ocfs2_userspace_stack(osb)) {
 846                         mlog(ML_ERROR, "Userspace stack expected, but "
 847                              "o2cb heartbeat arguments passed to mount\n");
 848                         return -EINVAL;
 849                 }
 850                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 851                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 852                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 853                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 854                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 855                         return -EINVAL;
 856                 }
 857         }
 858
 859         if (!(osb->s_mount_opt & hb_enabled)) {
 860                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 861                     !ocfs2_userspace_stack(osb)) {
 862                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 863                              "a read-write clustered device.\n");
 864                         return -EINVAL;
 865                 }
 866         }
 867
 868         return 0;
 869 }
 870
 871 /*
 872  * If we're using a userspace stack, mount should have passed
 873  * a name that matches the disk.  If not, mount should not
 874  * have passed a stack.
 875  */
 876 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 877                                         struct mount_options *mopt)
 878 {
 879         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 880                 mlog(ML_ERROR,
 881                      "cluster stack passed to mount, but this filesystem "
 882                      "does not support it\n");
 883                 return -EINVAL;
 884         }
 885
 886         if (ocfs2_userspace_stack(osb) &&
 887             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 888                     OCFS2_STACK_LABEL_LEN)) {
 889                 mlog(ML_ERROR,
 890                      "cluster stack passed to mount (\"%s\") does not "
 891                      "match the filesystem (\"%s\")\n",
 892                      mopt->cluster_stack,
 893                      osb->osb_cluster_stack);
 894                 return -EINVAL;
 895         }
 896
 897         return 0;
 898 }
 899
 900 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 901 {
 902         int type;
 903         struct super_block *sb = osb->sb;
 904         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 905                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 906         int status = 0;
 907
 908         for (type = 0; type < MAXQUOTAS; type++) {
 909                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 910                         continue;
 911                 if (unsuspend)
 912                         status = dquot_resume(sb, type);
 913                 else {
 914                         struct ocfs2_mem_dqinfo *oinfo;
 915
 916                         /* Cancel periodic syncing before suspending */
 917                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 918                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 919                         status = dquot_suspend(sb, type);
 920                 }
 921                 if (status < 0)
 922                         break;
 923         }
 924         if (status < 0)
 925                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 926                      "remount (error = %d).\n", status);
 927         return status;
 928 }
 929
 930 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 931 {
 932         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 933         struct super_block *sb = osb->sb;
 934         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 935                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 936         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 937                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 938         int status;
 939         int type;
 940
 941         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 942         for (type = 0; type < MAXQUOTAS; type++) {
 943                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 944                         continue;
 945                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 946                                                         osb->slot_num);
 947                 if (!inode[type]) {
 948                         status = -ENOENT;
 949                         goto out_quota_off;
 950                 }
 951                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 952                                       DQUOT_USAGE_ENABLED);
 953                 if (status < 0)
 954                         goto out_quota_off;
 955         }
 956
 957         for (type = 0; type < MAXQUOTAS; type++)
 958                 iput(inode[type]);
 959         return 0;
 960 out_quota_off:
 961         ocfs2_disable_quotas(osb);
 962         for (type = 0; type < MAXQUOTAS; type++)
 963                 iput(inode[type]);
 964         mlog_errno(status);
 965         return status;
 966 }
 967
 968 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 969 {
 970         int type;
 971         struct inode *inode;
 972         struct super_block *sb = osb->sb;
 973         struct ocfs2_mem_dqinfo *oinfo;
 974
 975         /* We mostly ignore errors in this function because there's not much
 976          * we can do when we see them */
 977         for (type = 0; type < MAXQUOTAS; type++) {
 978                 if (!sb_has_quota_loaded(sb, type))
 979                         continue;
 980                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 981                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 982                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 983                 inode = igrab(sb->s_dquot.files[type]);
 984                 /* Turn off quotas. This will remove all dquot structures from
 985                  * memory and so they will be automatically synced to global
 986                  * quota files */
 987                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 988                                         DQUOT_LIMITS_ENABLED);
 989                 if (!inode)
 990                         continue;
 991                 iput(inode);
 992         }
 993 }
 994
 995 /* Handle quota on quotactl */
 996 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 997 {
 998         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 999                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
1000
1001         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
1002                 return -EINVAL;
1003
1004         return dquot_enable(sb_dqopt(sb)->files[type], type,
1005                             format_id, DQUOT_LIMITS_ENABLED);
1006 }
1007
1008 /* Handle quota off quotactl */
1009 static int ocfs2_quota_off(struct super_block *sb, int type)
1010 {
1011         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1012 }
1013
1014 static const struct quotactl_ops ocfs2_quotactl_ops = {
1015         .quota_on_meta  = ocfs2_quota_on,
1016         .quota_off      = ocfs2_quota_off,
1017         .quota_sync     = dquot_quota_sync,
1018         .get_info       = dquot_get_dqinfo,
1019         .set_info       = dquot_set_dqinfo,
1020         .get_dqblk      = dquot_get_dqblk,
1021         .set_dqblk      = dquot_set_dqblk,
1022 };
1023
1024 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1025 {
1026         struct dentry *root;
1027         int status, sector_size;
1028         struct mount_options parsed_options;
1029         struct inode *inode = NULL;
1030         struct ocfs2_super *osb = NULL;
1031         struct buffer_head *bh = NULL;
1032         char nodestr[8];
1033         struct ocfs2_blockcheck_stats stats;
1034
1035         mlog_entry("%p, %p, %i", sb, data, silent);
1036
1037         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1038                 status = -EINVAL;
1039                 goto read_super_error;
1040         }
1041
1042         /* probe for superblock */
1043         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1044         if (status < 0) {
1045                 mlog(ML_ERROR, "superblock probe failed!\n");
1046                 goto read_super_error;
1047         }
1048
1049         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1050         osb = OCFS2_SB(sb);
1051         if (status < 0) {
1052                 mlog_errno(status);
1053                 goto read_super_error;
1054         }
1055         brelse(bh);
1056         bh = NULL;
1057
1058         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1059                 status = -EINVAL;
1060                 goto read_super_error;
1061         }
1062         osb->s_mount_opt = parsed_options.mount_opt;
1063         osb->s_atime_quantum = parsed_options.atime_quantum;
1064         osb->preferred_slot = parsed_options.slot;
1065         osb->osb_commit_interval = parsed_options.commit_interval;
1066
1067         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1068         osb->osb_resv_level = parsed_options.resv_level;
1069         osb->osb_dir_resv_level = parsed_options.resv_level;
1070         if (parsed_options.dir_resv_level == -1)
1071                 osb->osb_dir_resv_level = parsed_options.resv_level;
1072         else
1073                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1074
1075         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1076         if (status)
1077                 goto read_super_error;
1078
1079         sb->s_magic = OCFS2_SUPER_MAGIC;
1080
1081         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1082                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1083
1084         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1085          * heartbeat=none */
1086         if (bdev_read_only(sb->s_bdev)) {
1087                 if (!(sb->s_flags & MS_RDONLY)) {
1088                         status = -EACCES;
1089                         mlog(ML_ERROR, "Readonly device detected but readonly "
1090                              "mount was not specified.\n");
1091                         goto read_super_error;
1092                 }
1093
1094                 /* You should not be able to start a local heartbeat
1095                  * on a readonly device. */
1096                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1097                         status = -EROFS;
1098                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1099                              "device.\n");
1100                         goto read_super_error;
1101                 }
1102
1103                 status = ocfs2_check_journals_nolocks(osb);
1104                 if (status < 0) {
1105                         if (status == -EROFS)
1106                                 mlog(ML_ERROR, "Recovery required on readonly "
1107                                      "file system, but write access is "
1108                                      "unavailable.\n");
1109                         else
1110                                 mlog_errno(status);
1111                         goto read_super_error;
1112                 }
1113
1114                 ocfs2_set_ro_flag(osb, 1);
1115
1116                 printk(KERN_NOTICE "Readonly device detected. No cluster "
1117                        "services will be utilized for this mount. Recovery "
1118                        "will be skipped.\n");
1119         }
1120
1121         if (!ocfs2_is_hard_readonly(osb)) {
1122                 if (sb->s_flags & MS_RDONLY)
1123                         ocfs2_set_ro_flag(osb, 0);
1124         }
1125
1126         status = ocfs2_verify_heartbeat(osb);
1127         if (status < 0) {
1128                 mlog_errno(status);
1129                 goto read_super_error;
1130         }
1131
1132         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1133                                                  ocfs2_debugfs_root);
1134         if (!osb->osb_debug_root) {
1135                 status = -EINVAL;
1136                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1137                 goto read_super_error;
1138         }
1139
1140         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1141                                             osb->osb_debug_root,
1142                                             osb,
1143                                             &ocfs2_osb_debug_fops);
1144         if (!osb->osb_ctxt) {
1145                 status = -EINVAL;
1146                 mlog_errno(status);
1147                 goto read_super_error;
1148         }
1149
1150         if (ocfs2_meta_ecc(osb)) {
1151                 status = ocfs2_blockcheck_stats_debugfs_install(
1152                                                 &osb->osb_ecc_stats,
1153                                                 osb->osb_debug_root);
1154                 if (status) {
1155                         mlog(ML_ERROR,
1156                              "Unable to create blockcheck statistics "
1157                              "files\n");
1158                         goto read_super_error;
1159                 }
1160         }
1161
1162         status = ocfs2_mount_volume(sb);
1163         if (osb->root_inode)
1164                 inode = igrab(osb->root_inode);
1165
1166         if (status < 0)
1167                 goto read_super_error;
1168
1169         if (!inode) {
1170                 status = -EIO;
1171                 mlog_errno(status);
1172                 goto read_super_error;
1173         }
1174
1175         root = d_alloc_root(inode);
1176         if (!root) {
1177                 status = -ENOMEM;
1178                 mlog_errno(status);
1179                 goto read_super_error;
1180         }
1181
1182         sb->s_root = root;
1183
1184         ocfs2_complete_mount_recovery(osb);
1185
1186         if (ocfs2_mount_local(osb))
1187                 snprintf(nodestr, sizeof(nodestr), "local");
1188         else
1189                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1190
1191         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1192                "with %s data mode.\n",
1193                osb->dev_str, nodestr, osb->slot_num,
1194                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1195                "ordered");
1196
1197         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1198         wake_up(&osb->osb_mount_event);
1199
1200         /* Now we can initialize quotas because we can afford to wait
1201          * for cluster locks recovery now. That also means that truncation
1202          * log recovery can happen but that waits for proper quota setup */
1203         if (!(sb->s_flags & MS_RDONLY)) {
1204                 status = ocfs2_enable_quotas(osb);
1205                 if (status < 0) {
1206                         /* We have to err-out specially here because
1207                          * s_root is already set */
1208                         mlog_errno(status);
1209                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1210                         wake_up(&osb->osb_mount_event);
1211                         mlog_exit(status);
1212                         return status;
1213                 }
1214         }
1215
1216         ocfs2_complete_quota_recovery(osb);
1217
1218         /* Now we wake up again for processes waiting for quotas */
1219         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1220         wake_up(&osb->osb_mount_event);
1221
1222         /* Start this when the mount is almost sure of being successful */
1223         ocfs2_orphan_scan_start(osb);
1224
1225         mlog_exit(status);
1226         return status;
1227
1228 read_super_error:
1229         brelse(bh);
1230
1231         if (inode)
1232                 iput(inode);
1233
1234         if (osb) {
1235                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1236                 wake_up(&osb->osb_mount_event);
1237                 ocfs2_dismount_volume(sb, 1);
1238         }
1239
1240         mlog_exit(status);
1241         return status;
1242 }
1243
1244 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1245                         int flags,
1246                         const char *dev_name,
1247                         void *data)
1248 {
1249         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1250 }
1251
1252 static void ocfs2_kill_sb(struct super_block *sb)
1253 {
1254         struct ocfs2_super *osb = OCFS2_SB(sb);
1255
1256         /* Failed mount? */
1257         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1258                 goto out;
1259
1260         /* Prevent further queueing of inode drop events */
1261         spin_lock(&dentry_list_lock);
1262         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1263         spin_unlock(&dentry_list_lock);
1264         /* Wait for work to finish and/or remove it */
1265         cancel_work_sync(&osb->dentry_lock_work);
1266 out:
1267         kill_block_super(sb);
1268 }
1269
1270 static struct file_system_type ocfs2_fs_type = {
1271         .owner          = THIS_MODULE,
1272         .name           = "ocfs2",
1273         .mount          = ocfs2_mount,
1274         .kill_sb        = ocfs2_kill_sb,
1275
1276         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1277         .next           = NULL
1278 };
1279
1280 static int ocfs2_check_set_options(struct super_block *sb,
1281                                    struct mount_options *options)
1282 {
1283         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1284             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1285                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1286                 mlog(ML_ERROR, "User quotas were requested, but this "
1287                      "filesystem does not have the feature enabled.\n");
1288                 return 0;
1289         }
1290         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1291             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1292                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1293                 mlog(ML_ERROR, "Group quotas were requested, but this "
1294                      "filesystem does not have the feature enabled.\n");
1295                 return 0;
1296         }
1297         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1298             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1299                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1300                      "feature is not enabled\n");
1301                 return 0;
1302         }
1303         /* No ACL setting specified? Use XATTR feature... */
1304         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1305                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1306                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1307                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1308                 else
1309                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1310         }
1311         return 1;
1312 }
1313
1314 static int ocfs2_parse_options(struct super_block *sb,
1315                                char *options,
1316                                struct mount_options *mopt,
1317                                int is_remount)
1318 {
1319         int status, user_stack = 0;
1320         char *p;
1321         u32 tmp;
1322
1323         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
1324                    options ? options : "(none)");
1325
1326         mopt->commit_interval = 0;
1327         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1328         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1329         mopt->slot = OCFS2_INVALID_SLOT;
1330         mopt->localalloc_opt = -1;
1331         mopt->cluster_stack[0] = '\0';
1332         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1333         mopt->dir_resv_level = -1;
1334
1335         if (!options) {
1336                 status = 1;
1337                 goto bail;
1338         }
1339
1340         while ((p = strsep(&options, ",")) != NULL) {
1341                 int token, option;
1342                 substring_t args[MAX_OPT_ARGS];
1343
1344                 if (!*p)
1345                         continue;
1346
1347                 token = match_token(p, tokens, args);
1348                 switch (token) {
1349                 case Opt_hb_local:
1350                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1351                         break;
1352                 case Opt_hb_none:
1353                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1354                         break;
1355                 case Opt_hb_global:
1356                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1357                         break;
1358                 case Opt_barrier:
1359                         if (match_int(&args[0], &option)) {
1360                                 status = 0;
1361                                 goto bail;
1362                         }
1363                         if (option)
1364                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1365                         else
1366                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1367                         break;
1368                 case Opt_intr:
1369                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1370                         break;
1371                 case Opt_nointr:
1372                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1373                         break;
1374                 case Opt_err_panic:
1375                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1376                         break;
1377                 case Opt_err_ro:
1378                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1379                         break;
1380                 case Opt_data_ordered:
1381                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1382                         break;
1383                 case Opt_data_writeback:
1384                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1385                         break;
1386                 case Opt_user_xattr:
1387                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1388                         break;
1389                 case Opt_nouser_xattr:
1390                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1391                         break;
1392                 case Opt_atime_quantum:
1393                         if (match_int(&args[0], &option)) {
1394                                 status = 0;
1395                                 goto bail;
1396                         }
1397                         if (option >= 0)
1398                                 mopt->atime_quantum = option;
1399                         break;
1400                 case Opt_slot:
1401                         option = 0;
1402                         if (match_int(&args[0], &option)) {
1403                                 status = 0;
1404                                 goto bail;
1405                         }
1406                         if (option)
1407                                 mopt->slot = (s16)option;
1408                         break;
1409                 case Opt_commit:
1410                         option = 0;
1411                         if (match_int(&args[0], &option)) {
1412                                 status = 0;
1413                                 goto bail;
1414                         }
1415                         if (option < 0)
1416                                 return 0;
1417                         if (option == 0)
1418                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1419                         mopt->commit_interval = HZ * option;
1420                         break;
1421                 case Opt_localalloc:
1422                         option = 0;
1423                         if (match_int(&args[0], &option)) {
1424                                 status = 0;
1425                                 goto bail;
1426                         }
1427                         if (option >= 0)
1428                                 mopt->localalloc_opt = option;
1429                         break;
1430                 case Opt_localflocks:
1431                         /*
1432                          * Changing this during remount could race
1433                          * flock() requests, or "unbalance" existing
1434                          * ones (e.g., a lock is taken in one mode but
1435                          * dropped in the other). If users care enough
1436                          * to flip locking modes during remount, we
1437                          * could add a "local" flag to individual
1438                          * flock structures for proper tracking of
1439                          * state.
1440                          */
1441                         if (!is_remount)
1442                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1443                         break;
1444                 case Opt_stack:
1445                         /* Check both that the option we were passed
1446                          * is of the right length and that it is a proper
1447                          * string of the right length.
1448                          */
1449                         if (((args[0].to - args[0].from) !=
1450                              OCFS2_STACK_LABEL_LEN) ||
1451                             (strnlen(args[0].from,
1452                                      OCFS2_STACK_LABEL_LEN) !=
1453                              OCFS2_STACK_LABEL_LEN)) {
1454                                 mlog(ML_ERROR,
1455                                      "Invalid cluster_stack option\n");
1456                                 status = 0;
1457                                 goto bail;
1458                         }
1459                         memcpy(mopt->cluster_stack, args[0].from,
1460                                OCFS2_STACK_LABEL_LEN);
1461                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1462                         /*
1463                          * Open code the memcmp here as we don't have
1464                          * an osb to pass to
1465                          * ocfs2_userspace_stack().
1466                          */
1467                         if (memcmp(mopt->cluster_stack,
1468                                    OCFS2_CLASSIC_CLUSTER_STACK,
1469                                    OCFS2_STACK_LABEL_LEN))
1470                                 user_stack = 1;
1471                         break;
1472                 case Opt_inode64:
1473                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1474                         break;
1475                 case Opt_usrquota:
1476                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1477                         break;
1478                 case Opt_grpquota:
1479                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1480                         break;
1481                 case Opt_coherency_buffered:
1482                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1483                         break;
1484                 case Opt_coherency_full:
1485                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1486                         break;
1487                 case Opt_acl:
1488                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1489                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1490                         break;
1491                 case Opt_noacl:
1492                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1493                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1494                         break;
1495                 case Opt_resv_level:
1496                         if (is_remount)
1497                                 break;
1498                         if (match_int(&args[0], &option)) {
1499                                 status = 0;
1500                                 goto bail;
1501                         }
1502                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1503                             option < OCFS2_MAX_RESV_LEVEL)
1504                                 mopt->resv_level = option;
1505                         break;
1506                 case Opt_dir_resv_level:
1507                         if (is_remount)
1508                                 break;
1509                         if (match_int(&args[0], &option)) {
1510                                 status = 0;
1511                                 goto bail;
1512                         }
1513                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1514                             option < OCFS2_MAX_RESV_LEVEL)
1515                                 mopt->dir_resv_level = option;
1516                         break;
1517                 default:
1518                         mlog(ML_ERROR,
1519                              "Unrecognized mount option \"%s\" "
1520                              "or missing value\n", p);
1521                         status = 0;
1522                         goto bail;
1523                 }
1524         }
1525
1526         if (user_stack == 0) {
1527                 /* Ensure only one heartbeat mode */
1528                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1529                                          OCFS2_MOUNT_HB_GLOBAL |
1530                                          OCFS2_MOUNT_HB_NONE);
1531                 if (hweight32(tmp) != 1) {
1532                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1533                         status = 0;
1534                         goto bail;
1535                 }
1536         }
1537
1538         status = 1;
1539
1540 bail:
1541         mlog_exit(status);
1542         return status;
1543 }
1544
1545 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1546 {
1547         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
1548         unsigned long opts = osb->s_mount_opt;
1549         unsigned int local_alloc_megs;
1550
1551         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1552                 seq_printf(s, ",_netdev");
1553                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1554                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1555                 else
1556                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1557         } else
1558                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1559
1560         if (opts & OCFS2_MOUNT_NOINTR)
1561                 seq_printf(s, ",nointr");
1562
1563         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1564                 seq_printf(s, ",data=writeback");
1565         else
1566                 seq_printf(s, ",data=ordered");
1567
1568         if (opts & OCFS2_MOUNT_BARRIER)
1569                 seq_printf(s, ",barrier=1");
1570
1571         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1572                 seq_printf(s, ",errors=panic");
1573         else
1574                 seq_printf(s, ",errors=remount-ro");
1575
1576         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1577                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1578
1579         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
1580                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1581
1582         if (osb->osb_commit_interval)
1583                 seq_printf(s, ",commit=%u",
1584                            (unsigned) (osb->osb_commit_interval / HZ));
1585
1586         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1587         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1588                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1589
1590         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1591                 seq_printf(s, ",localflocks,");
1592
1593         if (osb->osb_cluster_stack[0])
1594                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1595                            osb->osb_cluster_stack);
1596         if (opts & OCFS2_MOUNT_USRQUOTA)
1597                 seq_printf(s, ",usrquota");
1598         if (opts & OCFS2_MOUNT_GRPQUOTA)
1599                 seq_printf(s, ",grpquota");
1600
1601         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1602                 seq_printf(s, ",coherency=buffered");
1603         else
1604                 seq_printf(s, ",coherency=full");
1605
1606         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1607                 seq_printf(s, ",nouser_xattr");
1608         else
1609                 seq_printf(s, ",user_xattr");
1610
1611         if (opts & OCFS2_MOUNT_INODE64)
1612                 seq_printf(s, ",inode64");
1613
1614         if (opts & OCFS2_MOUNT_POSIX_ACL)
1615                 seq_printf(s, ",acl");
1616         else
1617                 seq_printf(s, ",noacl");
1618
1619         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1620                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1621
1622         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1623                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1624
1625         return 0;
1626 }
1627
1628 static int __init ocfs2_init(void)
1629 {
1630         int status;
1631
1632         mlog_entry_void();
1633
1634         ocfs2_print_version();
1635
1636         status = init_ocfs2_uptodate_cache();
1637         if (status < 0) {
1638                 mlog_errno(status);
1639                 goto leave;
1640         }
1641
1642         status = ocfs2_initialize_mem_caches();
1643         if (status < 0) {
1644                 mlog_errno(status);
1645                 goto leave;
1646         }
1647
1648         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1649         if (!ocfs2_wq) {
1650                 status = -ENOMEM;
1651                 goto leave;
1652         }
1653
1654         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1655         if (!ocfs2_debugfs_root) {
1656                 status = -EFAULT;
1657                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1658         }
1659
1660         ocfs2_set_locking_protocol();
1661
1662         status = register_quota_format(&ocfs2_quota_format);
1663 leave:
1664         if (status < 0) {
1665                 ocfs2_free_mem_caches();
1666                 exit_ocfs2_uptodate_cache();
1667         }
1668
1669         mlog_exit(status);
1670
1671         if (status >= 0) {
1672                 return register_filesystem(&ocfs2_fs_type);
1673         } else
1674                 return -1;
1675 }
1676
1677 static void __exit ocfs2_exit(void)
1678 {
1679         mlog_entry_void();
1680
1681         if (ocfs2_wq) {
1682                 flush_workqueue(ocfs2_wq);
1683                 destroy_workqueue(ocfs2_wq);
1684         }
1685
1686         unregister_quota_format(&ocfs2_quota_format);
1687
1688         debugfs_remove(ocfs2_debugfs_root);
1689
1690         ocfs2_free_mem_caches();
1691
1692         unregister_filesystem(&ocfs2_fs_type);
1693
1694         exit_ocfs2_uptodate_cache();
1695
1696         mlog_exit_void();
1697 }
1698
1699 static void ocfs2_put_super(struct super_block *sb)
1700 {
1701         mlog_entry("(0x%p)\n", sb);
1702
1703         ocfs2_sync_blockdev(sb);
1704         ocfs2_dismount_volume(sb, 0);
1705
1706         mlog_exit_void();
1707 }
1708
1709 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1710 {
1711         struct ocfs2_super *osb;
1712         u32 numbits, freebits;
1713         int status;
1714         struct ocfs2_dinode *bm_lock;
1715         struct buffer_head *bh = NULL;
1716         struct inode *inode = NULL;
1717
1718         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1719
1720         osb = OCFS2_SB(dentry->d_sb);
1721
1722         inode = ocfs2_get_system_file_inode(osb,
1723                                             GLOBAL_BITMAP_SYSTEM_INODE,
1724                                             OCFS2_INVALID_SLOT);
1725         if (!inode) {
1726                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1727                 status = -EIO;
1728                 goto bail;
1729         }
1730
1731         status = ocfs2_inode_lock(inode, &bh, 0);
1732         if (status < 0) {
1733                 mlog_errno(status);
1734                 goto bail;
1735         }
1736
1737         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1738
1739         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1740         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1741
1742         buf->f_type = OCFS2_SUPER_MAGIC;
1743         buf->f_bsize = dentry->d_sb->s_blocksize;
1744         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1745         buf->f_blocks = ((sector_t) numbits) *
1746                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1747         buf->f_bfree = ((sector_t) freebits) *
1748                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1749         buf->f_bavail = buf->f_bfree;
1750         buf->f_files = numbits;
1751         buf->f_ffree = freebits;
1752         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1753                                 & 0xFFFFFFFFUL;
1754         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1755                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1756
1757         brelse(bh);
1758
1759         ocfs2_inode_unlock(inode, 0);
1760         status = 0;
1761 bail:
1762         if (inode)
1763                 iput(inode);
1764
1765         mlog_exit(status);
1766
1767         return status;
1768 }
1769
1770 static void ocfs2_inode_init_once(void *data)
1771 {
1772         struct ocfs2_inode_info *oi = data;
1773
1774         oi->ip_flags = 0;
1775         oi->ip_open_count = 0;
1776         spin_lock_init(&oi->ip_lock);
1777         ocfs2_extent_map_init(&oi->vfs_inode);
1778         INIT_LIST_HEAD(&oi->ip_io_markers);
1779         oi->ip_dir_start_lookup = 0;
1780
1781         init_rwsem(&oi->ip_alloc_sem);
1782         init_rwsem(&oi->ip_xattr_sem);
1783         mutex_init(&oi->ip_io_mutex);
1784
1785         oi->ip_blkno = 0ULL;
1786         oi->ip_clusters = 0;
1787
1788         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1789
1790         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1791         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1792         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1793
1794         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1795                                   &ocfs2_inode_caching_ops);
1796
1797         inode_init_once(&oi->vfs_inode);
1798 }
1799
1800 static int ocfs2_initialize_mem_caches(void)
1801 {
1802         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1803                                        sizeof(struct ocfs2_inode_info),
1804                                        0,
1805                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1806                                                 SLAB_MEM_SPREAD),
1807                                        ocfs2_inode_init_once);
1808         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1809                                         sizeof(struct ocfs2_dquot),
1810                                         0,
1811                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1812                                                 SLAB_MEM_SPREAD),
1813                                         NULL);
1814         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1815                                         sizeof(struct ocfs2_quota_chunk),
1816                                         0,
1817                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1818                                         NULL);
1819         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1820             !ocfs2_qf_chunk_cachep) {
1821                 if (ocfs2_inode_cachep)
1822                         kmem_cache_destroy(ocfs2_inode_cachep);
1823                 if (ocfs2_dquot_cachep)
1824                         kmem_cache_destroy(ocfs2_dquot_cachep);
1825                 if (ocfs2_qf_chunk_cachep)
1826                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1827                 return -ENOMEM;
1828         }
1829
1830         return 0;
1831 }
1832
1833 static void ocfs2_free_mem_caches(void)
1834 {
1835         if (ocfs2_inode_cachep)
1836                 kmem_cache_destroy(ocfs2_inode_cachep);
1837         ocfs2_inode_cachep = NULL;
1838
1839         if (ocfs2_dquot_cachep)
1840                 kmem_cache_destroy(ocfs2_dquot_cachep);
1841         ocfs2_dquot_cachep = NULL;
1842
1843         if (ocfs2_qf_chunk_cachep)
1844                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1845         ocfs2_qf_chunk_cachep = NULL;
1846 }
1847
1848 static int ocfs2_get_sector(struct super_block *sb,
1849                             struct buffer_head **bh,
1850                             int block,
1851                             int sect_size)
1852 {
1853         if (!sb_set_blocksize(sb, sect_size)) {
1854                 mlog(ML_ERROR, "unable to set blocksize\n");
1855                 return -EIO;
1856         }
1857
1858         *bh = sb_getblk(sb, block);
1859         if (!*bh) {
1860                 mlog_errno(-EIO);
1861                 return -EIO;
1862         }
1863         lock_buffer(*bh);
1864         if (!buffer_dirty(*bh))
1865                 clear_buffer_uptodate(*bh);
1866         unlock_buffer(*bh);
1867         ll_rw_block(READ, 1, bh);
1868         wait_on_buffer(*bh);
1869         if (!buffer_uptodate(*bh)) {
1870                 mlog_errno(-EIO);
1871                 brelse(*bh);
1872                 *bh = NULL;
1873                 return -EIO;
1874         }
1875
1876         return 0;
1877 }
1878
1879 static int ocfs2_mount_volume(struct super_block *sb)
1880 {
1881         int status = 0;
1882         int unlock_super = 0;
1883         struct ocfs2_super *osb = OCFS2_SB(sb);
1884
1885         mlog_entry_void();
1886
1887         if (ocfs2_is_hard_readonly(osb))
1888                 goto leave;
1889
1890         status = ocfs2_dlm_init(osb);
1891         if (status < 0) {
1892                 mlog_errno(status);
1893                 goto leave;
1894         }
1895
1896         status = ocfs2_super_lock(osb, 1);
1897         if (status < 0) {
1898                 mlog_errno(status);
1899                 goto leave;
1900         }
1901         unlock_super = 1;
1902
1903         /* This will load up the node map and add ourselves to it. */
1904         status = ocfs2_find_slot(osb);
1905         if (status < 0) {
1906                 mlog_errno(status);
1907                 goto leave;
1908         }
1909
1910         /* load all node-local system inodes */
1911         status = ocfs2_init_local_system_inodes(osb);
1912         if (status < 0) {
1913                 mlog_errno(status);
1914                 goto leave;
1915         }
1916
1917         status = ocfs2_check_volume(osb);
1918         if (status < 0) {
1919                 mlog_errno(status);
1920                 goto leave;
1921         }
1922
1923         status = ocfs2_truncate_log_init(osb);
1924         if (status < 0)
1925                 mlog_errno(status);
1926
1927 leave:
1928         if (unlock_super)
1929                 ocfs2_super_unlock(osb, 1);
1930
1931         mlog_exit(status);
1932         return status;
1933 }
1934
1935 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1936 {
1937         int tmp, hangup_needed = 0;
1938         struct ocfs2_super *osb = NULL;
1939         char nodestr[8];
1940
1941         mlog_entry("(0x%p)\n", sb);
1942
1943         BUG_ON(!sb);
1944         osb = OCFS2_SB(sb);
1945         BUG_ON(!osb);
1946
1947         debugfs_remove(osb->osb_ctxt);
1948
1949         /*
1950          * Flush inode dropping work queue so that deletes are
1951          * performed while the filesystem is still working
1952          */
1953         ocfs2_drop_all_dl_inodes(osb);
1954
1955         /* Orphan scan should be stopped as early as possible */
1956         ocfs2_orphan_scan_stop(osb);
1957
1958         ocfs2_disable_quotas(osb);
1959
1960         ocfs2_shutdown_local_alloc(osb);
1961
1962         ocfs2_truncate_log_shutdown(osb);
1963
1964         /* This will disable recovery and flush any recovery work. */
1965         ocfs2_recovery_exit(osb);
1966
1967         ocfs2_journal_shutdown(osb);
1968
1969         ocfs2_sync_blockdev(sb);
1970
1971         ocfs2_purge_refcount_trees(osb);
1972
1973         /* No cluster connection means we've failed during mount, so skip
1974          * all the steps which depended on that to complete. */
1975         if (osb->cconn) {
1976                 tmp = ocfs2_super_lock(osb, 1);
1977                 if (tmp < 0) {
1978                         mlog_errno(tmp);
1979                         return;
1980                 }
1981         }
1982
1983         if (osb->slot_num != OCFS2_INVALID_SLOT)
1984                 ocfs2_put_slot(osb);
1985
1986         if (osb->cconn)
1987                 ocfs2_super_unlock(osb, 1);
1988
1989         ocfs2_release_system_inodes(osb);
1990
1991         /*
1992          * If we're dismounting due to mount error, mount.ocfs2 will clean
1993          * up heartbeat.  If we're a local mount, there is no heartbeat.
1994          * If we failed before we got a uuid_str yet, we can't stop
1995          * heartbeat.  Otherwise, do it.
1996          */
1997         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1998                 hangup_needed = 1;
1999
2000         if (osb->cconn)
2001                 ocfs2_dlm_shutdown(osb, hangup_needed);
2002
2003         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
2004         debugfs_remove(osb->osb_debug_root);
2005
2006         if (hangup_needed)
2007                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
2008
2009         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
2010
2011         if (ocfs2_mount_local(osb))
2012                 snprintf(nodestr, sizeof(nodestr), "local");
2013         else
2014                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2015
2016         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2017                osb->dev_str, nodestr);
2018
2019         ocfs2_delete_osb(osb);
2020         kfree(osb);
2021         sb->s_dev = 0;
2022         sb->s_fs_info = NULL;
2023 }
2024
2025 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2026                                 unsigned uuid_bytes)
2027 {
2028         int i, ret;
2029         char *ptr;
2030
2031         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2032
2033         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2034         if (osb->uuid_str == NULL)
2035                 return -ENOMEM;
2036
2037         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2038                 /* print with null */
2039                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2040                 if (ret != 2) /* drop super cleans up */
2041                         return -EINVAL;
2042                 /* then only advance past the last char */
2043                 ptr += 2;
2044         }
2045
2046         return 0;
2047 }
2048
2049 /* Make sure entire volume is addressable by our journal.  Requires
2050    osb_clusters_at_boot to be valid and for the journal to have been
2051    initialized by ocfs2_journal_init(). */
2052 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2053 {
2054         int status = 0;
2055         u64 max_block =
2056                 ocfs2_clusters_to_blocks(osb->sb,
2057                                          osb->osb_clusters_at_boot) - 1;
2058
2059         /* 32-bit block number is always OK. */
2060         if (max_block <= (u32)~0ULL)
2061                 goto out;
2062
2063         /* Volume is "huge", so see if our journal is new enough to
2064            support it. */
2065         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2066                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2067               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2068                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2069                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2070                      "Enable the 'block64' journal option with tunefs.ocfs2");
2071                 status = -EFBIG;
2072                 goto out;
2073         }
2074
2075  out:
2076         return status;
2077 }
2078
2079 static int ocfs2_initialize_super(struct super_block *sb,
2080                                   struct buffer_head *bh,
2081                                   int sector_size,
2082                                   struct ocfs2_blockcheck_stats *stats)
2083 {
2084         int status;
2085         int i, cbits, bbits;
2086         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2087         struct inode *inode = NULL;
2088         struct ocfs2_journal *journal;
2089         __le32 uuid_net_key;
2090         struct ocfs2_super *osb;
2091         u64 total_blocks;
2092
2093         mlog_entry_void();
2094
2095         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2096         if (!osb) {
2097                 status = -ENOMEM;
2098                 mlog_errno(status);
2099                 goto bail;
2100         }
2101
2102         sb->s_fs_info = osb;
2103         sb->s_op = &ocfs2_sops;
2104         sb->s_d_op = &ocfs2_dentry_ops;
2105         sb->s_export_op = &ocfs2_export_ops;
2106         sb->s_qcop = &ocfs2_quotactl_ops;
2107         sb->dq_op = &ocfs2_quota_operations;
2108         sb->s_xattr = ocfs2_xattr_handlers;
2109         sb->s_time_gran = 1;
2110         sb->s_flags |= MS_NOATIME;
2111         /* this is needed to support O_LARGEFILE */
2112         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2113         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2114         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2115
2116         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2117
2118         for (i = 0; i < 3; i++)
2119                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2120         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2121
2122         osb->sb = sb;
2123         /* Save off for ocfs2_rw_direct */
2124         osb->s_sectsize_bits = blksize_bits(sector_size);
2125         BUG_ON(!osb->s_sectsize_bits);
2126
2127         spin_lock_init(&osb->dc_task_lock);
2128         init_waitqueue_head(&osb->dc_event);
2129         osb->dc_work_sequence = 0;
2130         osb->dc_wake_sequence = 0;
2131         INIT_LIST_HEAD(&osb->blocked_lock_list);
2132         osb->blocked_lock_count = 0;
2133         spin_lock_init(&osb->osb_lock);
2134         spin_lock_init(&osb->osb_xattr_lock);
2135         ocfs2_init_steal_slots(osb);
2136
2137         atomic_set(&osb->alloc_stats.moves, 0);
2138         atomic_set(&osb->alloc_stats.local_data, 0);
2139         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2140         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2141         atomic_set(&osb->alloc_stats.bg_extends, 0);
2142
2143         /* Copy the blockcheck stats from the superblock probe */
2144         osb->osb_ecc_stats = *stats;
2145
2146         ocfs2_init_node_maps(osb);
2147
2148         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2149                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2150
2151         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2152         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2153                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2154                      osb->max_slots);
2155                 status = -EINVAL;
2156                 goto bail;
2157         }
2158         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
2159
2160         ocfs2_orphan_scan_init(osb);
2161
2162         status = ocfs2_recovery_init(osb);
2163         if (status) {
2164                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2165                 mlog_errno(status);
2166                 goto bail;
2167         }
2168
2169         init_waitqueue_head(&osb->checkpoint_event);
2170         atomic_set(&osb->needs_checkpoint, 0);
2171
2172         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2173
2174         osb->slot_num = OCFS2_INVALID_SLOT;
2175
2176         osb->s_xattr_inline_size = le16_to_cpu(
2177                                         di->id2.i_super.s_xattr_inline_size);
2178
2179         osb->local_alloc_state = OCFS2_LA_UNUSED;
2180         osb->local_alloc_bh = NULL;
2181         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2182
2183         init_waitqueue_head(&osb->osb_mount_event);
2184
2185         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2186         if (status) {
2187                 mlog_errno(status);
2188                 goto bail;
2189         }
2190
2191         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2192         if (!osb->vol_label) {
2193                 mlog(ML_ERROR, "unable to alloc vol label\n");
2194                 status = -ENOMEM;
2195                 goto bail;
2196         }
2197
2198         osb->slot_recovery_generations =
2199                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2200                         GFP_KERNEL);
2201         if (!osb->slot_recovery_generations) {
2202                 status = -ENOMEM;
2203                 mlog_errno(status);
2204                 goto bail;
2205         }
2206
2207         init_waitqueue_head(&osb->osb_wipe_event);
2208         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2209                                         sizeof(*osb->osb_orphan_wipes),
2210                                         GFP_KERNEL);
2211         if (!osb->osb_orphan_wipes) {
2212                 status = -ENOMEM;
2213                 mlog_errno(status);
2214                 goto bail;
2215         }
2216
2217         osb->osb_rf_lock_tree = RB_ROOT;
2218
2219         osb->s_feature_compat =
2220                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2221         osb->s_feature_ro_compat =
2222                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2223         osb->s_feature_incompat =
2224                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2225
2226         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2227                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2228                      "optional features (%x).\n", i);
2229                 status = -EINVAL;
2230                 goto bail;
2231         }
2232         if (!(osb->sb->s_flags & MS_RDONLY) &&
2233             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2234                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2235                      "unsupported optional features (%x).\n", i);
2236                 status = -EINVAL;
2237                 goto bail;
2238         }
2239
2240         if (ocfs2_clusterinfo_valid(osb)) {
2241                 osb->osb_stackflags =
2242                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2243                 memcpy(osb->osb_cluster_stack,
2244                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2245                        OCFS2_STACK_LABEL_LEN);
2246                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
2247                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2248                         mlog(ML_ERROR,
2249                              "couldn't mount because of an invalid "
2250                              "cluster stack label (%s) \n",
2251                              osb->osb_cluster_stack);
2252                         status = -EINVAL;
2253                         goto bail;
2254                 }
2255         } else {
2256                 /* The empty string is identical with classic tools that
2257                  * don't know about s_cluster_info. */
2258                 osb->osb_cluster_stack[0] = '\0';
2259         }
2260
2261         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2262
2263         /* FIXME
2264          * This should be done in ocfs2_journal_init(), but unknown
2265          * ordering issues will cause the filesystem to crash.
2266          * If anyone wants to figure out what part of the code
2267          * refers to osb->journal before ocfs2_journal_init() is run,
2268          * be my guest.
2269          */
2270         /* initialize our journal structure */
2271
2272         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2273         if (!journal) {
2274                 mlog(ML_ERROR, "unable to alloc journal\n");
2275                 status = -ENOMEM;
2276                 goto bail;
2277         }
2278         osb->journal = journal;
2279         journal->j_osb = osb;
2280
2281         atomic_set(&journal->j_num_trans, 0);
2282         init_rwsem(&journal->j_trans_barrier);
2283         init_waitqueue_head(&journal->j_checkpointed);
2284         spin_lock_init(&journal->j_lock);
2285         journal->j_trans_id = (unsigned long) 1;
2286         INIT_LIST_HEAD(&journal->j_la_cleanups);
2287         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2288         journal->j_state = OCFS2_JOURNAL_FREE;
2289
2290         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2291         osb->dentry_lock_list = NULL;
2292
2293         /* get some pseudo constants for clustersize bits */
2294         osb->s_clustersize_bits =
2295                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2296         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2297         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
2298
2299         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2300             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2301                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2302                      osb->s_clustersize);
2303                 status = -EINVAL;
2304                 goto bail;
2305         }
2306
2307         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2308                                                 le32_to_cpu(di->i_clusters));
2309
2310         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2311                                            total_blocks);
2312         if (status) {
2313                 mlog(ML_ERROR, "Volume too large "
2314                      "to mount safely on this system");
2315                 status = -EFBIG;
2316                 goto bail;
2317         }
2318
2319         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2320                                  sizeof(di->id2.i_super.s_uuid))) {
2321                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2322                 status = -ENOMEM;
2323                 goto bail;
2324         }
2325
2326         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2327
2328         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2329         osb->vol_label[63] = '\0';
2330         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2331         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2332         osb->first_cluster_group_blkno =
2333                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2334         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2335         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2336         mlog(0, "vol_label: %s\n", osb->vol_label);
2337         mlog(0, "uuid: %s\n", osb->uuid_str);
2338         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
2339              (unsigned long long)osb->root_blkno,
2340              (unsigned long long)osb->system_dir_blkno);
2341
2342         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2343         if (!osb->osb_dlm_debug) {
2344                 status = -ENOMEM;
2345                 mlog_errno(status);
2346                 goto bail;
2347         }
2348
2349         atomic_set(&osb->vol_state, VOLUME_INIT);
2350
2351         /* load root, system_dir, and all global system inodes */
2352         status = ocfs2_init_global_system_inodes(osb);
2353         if (status < 0) {
2354                 mlog_errno(status);
2355                 goto bail;
2356         }
2357
2358         /*
2359          * global bitmap
2360          */
2361         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2362                                             OCFS2_INVALID_SLOT);
2363         if (!inode) {
2364                 status = -EINVAL;
2365                 mlog_errno(status);
2366                 goto bail;
2367         }
2368
2369         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2370         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2371         iput(inode);
2372
2373         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2374                                  osb->s_feature_incompat) * 8;
2375
2376         status = ocfs2_init_slot_info(osb);
2377         if (status < 0) {
2378                 mlog_errno(status);
2379                 goto bail;
2380         }
2381
2382 bail:
2383         mlog_exit(status);
2384         return status;
2385 }
2386
2387 /*
2388  * will return: -EAGAIN if it is ok to keep searching for superblocks
2389  *              -EINVAL if there is a bad superblock
2390  *              0 on success
2391  */
2392 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2393                                struct buffer_head *bh,
2394                                u32 blksz,
2395                                struct ocfs2_blockcheck_stats *stats)
2396 {
2397         int status = -EAGAIN;
2398
2399         mlog_entry_void();
2400
2401         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2402                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2403                 /* We have to do a raw check of the feature here */
2404                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2405                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2406                         status = ocfs2_block_check_validate(bh->b_data,
2407                                                             bh->b_size,
2408                                                             &di->i_check,
2409                                                             stats);
2410                         if (status)
2411                                 goto out;
2412                 }
2413                 status = -EINVAL;
2414                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2415                         mlog(ML_ERROR, "found superblock with incorrect block "
2416                              "size: found %u, should be %u\n",
2417                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2418                                blksz);
2419                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2420                            OCFS2_MAJOR_REV_LEVEL ||
2421                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2422                            OCFS2_MINOR_REV_LEVEL) {
2423                         mlog(ML_ERROR, "found superblock with bad version: "
2424                              "found %u.%u, should be %u.%u\n",
2425                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2426                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2427                              OCFS2_MAJOR_REV_LEVEL,
2428                              OCFS2_MINOR_REV_LEVEL);
2429                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2430                         mlog(ML_ERROR, "bad block number on superblock: "
2431                              "found %llu, should be %llu\n",
2432                              (unsigned long long)le64_to_cpu(di->i_blkno),
2433                              (unsigned long long)bh->b_blocknr);
2434                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2435                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2436                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2437                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2438                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2439                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2440                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2441                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2442                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2443                         mlog(ML_ERROR,
2444                              "Superblock slots found greater than file system "
2445                              "maximum: found %u, max %u\n",
2446                              le16_to_cpu(di->id2.i_super.s_max_slots),
2447                              OCFS2_MAX_SLOTS);
2448                 } else {
2449                         /* found it! */
2450                         status = 0;
2451                 }
2452         }
2453
2454 out:
2455         mlog_exit(status);
2456         return status;
2457 }
2458
2459 static int ocfs2_check_volume(struct ocfs2_super *osb)
2460 {
2461         int status;
2462         int dirty;
2463         int local;
2464         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2465                                                   * recover
2466                                                   * ourselves. */
2467
2468         mlog_entry_void();
2469
2470         /* Init our journal object. */
2471         status = ocfs2_journal_init(osb->journal, &dirty);
2472         if (status < 0) {
2473                 mlog(ML_ERROR, "Could not initialize journal!\n");
2474                 goto finally;
2475         }
2476
2477         /* Now that journal has been initialized, check to make sure
2478            entire volume is addressable. */
2479         status = ocfs2_journal_addressable(osb);
2480         if (status)
2481                 goto finally;
2482
2483         /* If the journal was unmounted cleanly then we don't want to
2484          * recover anything. Otherwise, journal_load will do that
2485          * dirty work for us :) */
2486         if (!dirty) {
2487                 status = ocfs2_journal_wipe(osb->journal, 0);
2488                 if (status < 0) {
2489                         mlog_errno(status);
2490                         goto finally;
2491                 }
2492         } else {
2493                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
2494                      "recovering volume.\n");
2495         }
2496
2497         local = ocfs2_mount_local(osb);
2498
2499         /* will play back anything left in the journal. */
2500         status = ocfs2_journal_load(osb->journal, local, dirty);
2501         if (status < 0) {
2502                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2503                 goto finally;
2504         }
2505
2506         if (dirty) {
2507                 /* recover my local alloc if we didn't unmount cleanly. */
2508                 status = ocfs2_begin_local_alloc_recovery(osb,
2509                                                           osb->slot_num,
2510                                                           &local_alloc);
2511                 if (status < 0) {
2512                         mlog_errno(status);
2513                         goto finally;
2514                 }
2515                 /* we complete the recovery process after we've marked
2516                  * ourselves as mounted. */
2517         }
2518
2519         mlog(0, "Journal loaded.\n");
2520
2521         status = ocfs2_load_local_alloc(osb);
2522         if (status < 0) {
2523                 mlog_errno(status);
2524                 goto finally;
2525         }
2526
2527         if (dirty) {
2528                 /* Recovery will be completed after we've mounted the
2529                  * rest of the volume. */
2530                 osb->dirty = 1;
2531                 osb->local_alloc_copy = local_alloc;
2532                 local_alloc = NULL;
2533         }
2534
2535         /* go through each journal, trylock it and if you get the
2536          * lock, and it's marked as dirty, set the bit in the recover
2537          * map and launch a recovery thread for it. */
2538         status = ocfs2_mark_dead_nodes(osb);
2539         if (status < 0) {
2540                 mlog_errno(status);
2541                 goto finally;
2542         }
2543
2544         status = ocfs2_compute_replay_slots(osb);
2545         if (status < 0)
2546                 mlog_errno(status);
2547
2548 finally:
2549         if (local_alloc)
2550                 kfree(local_alloc);
2551
2552         mlog_exit(status);
2553         return status;
2554 }
2555
2556 /*
2557  * The routine gets called from dismount or close whenever a dismount on
2558  * volume is requested and the osb open count becomes 1.
2559  * It will remove the osb from the global list and also free up all the
2560  * initialized resources and fileobject.
2561  */
2562 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2563 {
2564         mlog_entry_void();
2565
2566         /* This function assumes that the caller has the main osb resource */
2567
2568         ocfs2_free_slot_info(osb);
2569
2570         kfree(osb->osb_orphan_wipes);
2571         kfree(osb->slot_recovery_generations);
2572         /* FIXME
2573          * This belongs in journal shutdown, but because we have to
2574          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2575          * we free it here.
2576          */
2577         kfree(osb->journal);
2578         if (osb->local_alloc_copy)
2579                 kfree(osb->local_alloc_copy);
2580         kfree(osb->uuid_str);
2581         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2582         memset(osb, 0, sizeof(struct ocfs2_super));
2583
2584         mlog_exit_void();
2585 }
2586
2587 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2588  * panic(). We do not support continue-on-error operation. */
2589 static void ocfs2_handle_error(struct super_block *sb)
2590 {
2591         struct ocfs2_super *osb = OCFS2_SB(sb);
2592
2593         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2594                 panic("OCFS2: (device %s): panic forced after error\n",
2595                       sb->s_id);
2596
2597         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2598
2599         if (sb->s_flags & MS_RDONLY &&
2600             (ocfs2_is_soft_readonly(osb) ||
2601              ocfs2_is_hard_readonly(osb)))
2602                 return;
2603
2604         printk(KERN_CRIT "File system is now read-only due to the potential "
2605                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2606                "system is unmounted.\n");
2607         sb->s_flags |= MS_RDONLY;
2608         ocfs2_set_ro_flag(osb, 0);
2609 }
2610
2611 static char error_buf[1024];
2612
2613 void __ocfs2_error(struct super_block *sb,
2614                    const char *function,
2615                    const char *fmt, ...)
2616 {
2617         va_list args;
2618
2619         va_start(args, fmt);
2620         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2621         va_end(args);
2622
2623         /* Not using mlog here because we want to show the actual
2624          * function the error came from. */
2625         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2626                sb->s_id, function, error_buf);
2627
2628         ocfs2_handle_error(sb);
2629 }
2630
2631 /* Handle critical errors. This is intentionally more drastic than
2632  * ocfs2_handle_error, so we only use for things like journal errors,
2633  * etc. */
2634 void __ocfs2_abort(struct super_block* sb,
2635                    const char *function,
2636                    const char *fmt, ...)
2637 {
2638         va_list args;
2639
2640         va_start(args, fmt);
2641         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2642         va_end(args);
2643
2644         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2645                sb->s_id, function, error_buf);
2646
2647         /* We don't have the cluster support yet to go straight to
2648          * hard readonly in here. Until then, we want to keep
2649          * ocfs2_abort() so that we can at least mark critical
2650          * errors.
2651          *
2652          * TODO: This should abort the journal and alert other nodes
2653          * that our slot needs recovery. */
2654
2655         /* Force a panic(). This stinks, but it's better than letting
2656          * things continue without having a proper hard readonly
2657          * here. */
2658         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2659                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2660         ocfs2_handle_error(sb);
2661 }
2662
2663 /*
2664  * Void signal blockers, because in-kernel sigprocmask() only fails
2665  * when SIG_* is wrong.
2666  */
2667 void ocfs2_block_signals(sigset_t *oldset)
2668 {
2669         int rc;
2670         sigset_t blocked;
2671
2672         sigfillset(&blocked);
2673         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2674         BUG_ON(rc);
2675 }
2676
2677 void ocfs2_unblock_signals(sigset_t *oldset)
2678 {
2679         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2680         BUG_ON(rc);
2681 }
2682
2683 module_init(ocfs2_init);
2684 module_exit(ocfs2_exit);