fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/init.h>
  32 #include <linux/random.h>
  33 #include <linux/statfs.h>
  34 #include <linux/moduleparam.h>
  35 #include <linux/blkdev.h>
  36 #include <linux/socket.h>
  37 #include <linux/inet.h>
  38 #include <linux/parser.h>
  39 #include <linux/crc32.h>
  40 #include <linux/debugfs.h>
  41 #include <linux/mount.h>
  42 #include <linux/seq_file.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/cleancache.h>
  45
  46 #define CREATE_TRACE_POINTS
  47 #include "ocfs2_trace.h"
  48
  49 #include <cluster/masklog.h>
  50
  51 #include "ocfs2.h"
  52
  53 /* this should be the only file to include a version 1 header */
  54 #include "ocfs1_fs_compat.h"
  55
  56 #include "alloc.h"
  57 #include "aops.h"
  58 #include "blockcheck.h"
  59 #include "dlmglue.h"
  60 #include "export.h"
  61 #include "extent_map.h"
  62 #include "heartbeat.h"
  63 #include "inode.h"
  64 #include "journal.h"
  65 #include "localalloc.h"
  66 #include "namei.h"
  67 #include "slot_map.h"
  68 #include "super.h"
  69 #include "sysfile.h"
  70 #include "uptodate.h"
  71 #include "xattr.h"
  72 #include "quota.h"
  73 #include "refcounttree.h"
  74 #include "suballoc.h"
  75
  76 #include "buffer_head_io.h"
  77 #include "filecheck.h"
  78
  79 static struct kmem_cache *ocfs2_inode_cachep;
  80 struct kmem_cache *ocfs2_dquot_cachep;
  81 struct kmem_cache *ocfs2_qf_chunk_cachep;
  82
  83 /* OCFS2 needs to schedule several different types of work which
  84  * require cluster locking, disk I/O, recovery waits, etc. Since these
  85  * types of work tend to be heavy we avoid using the kernel events
  86  * workqueue and schedule on our own. */
  87 struct workqueue_struct *ocfs2_wq = NULL;
  88
  89 static struct dentry *ocfs2_debugfs_root;
  90
  91 MODULE_AUTHOR("Oracle");
  92 MODULE_LICENSE("GPL");
  93 MODULE_DESCRIPTION("OCFS2 cluster file system");
  94
  95 struct mount_options
  96 {
  97         unsigned long   commit_interval;
  98         unsigned long   mount_opt;
  99         unsigned int    atime_quantum;
 100         signed short    slot;
 101         int             localalloc_opt;
 102         unsigned int    resv_level;
 103         int             dir_resv_level;
 104         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 105 };
 106
 107 static int ocfs2_parse_options(struct super_block *sb, char *options,
 108                                struct mount_options *mopt,
 109                                int is_remount);
 110 static int ocfs2_check_set_options(struct super_block *sb,
 111                                    struct mount_options *options);
 112 static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
 113 static void ocfs2_put_super(struct super_block *sb);
 114 static int ocfs2_mount_volume(struct super_block *sb);
 115 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 116 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 117 static int ocfs2_initialize_mem_caches(void);
 118 static void ocfs2_free_mem_caches(void);
 119 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 120
 121 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 122
 123 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 124
 125 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 126 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 127 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 128 static int ocfs2_check_volume(struct ocfs2_super *osb);
 129 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 130                                struct buffer_head *bh,
 131                                u32 sectsize,
 132                                struct ocfs2_blockcheck_stats *stats);
 133 static int ocfs2_initialize_super(struct super_block *sb,
 134                                   struct buffer_head *bh,
 135                                   int sector_size,
 136                                   struct ocfs2_blockcheck_stats *stats);
 137 static int ocfs2_get_sector(struct super_block *sb,
 138                             struct buffer_head **bh,
 139                             int block,
 140                             int sect_size);
 141 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 142 static void ocfs2_destroy_inode(struct inode *inode);
 143 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 144 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 145 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 146
 147 static struct dquot **ocfs2_get_dquots(struct inode *inode)
 148 {
 149         return OCFS2_I(inode)->i_dquot;
 150 }
 151
 152 static const struct super_operations ocfs2_sops = {
 153         .statfs         = ocfs2_statfs,
 154         .alloc_inode    = ocfs2_alloc_inode,
 155         .destroy_inode  = ocfs2_destroy_inode,
 156         .drop_inode     = ocfs2_drop_inode,
 157         .evict_inode    = ocfs2_evict_inode,
 158         .sync_fs        = ocfs2_sync_fs,
 159         .put_super      = ocfs2_put_super,
 160         .remount_fs     = ocfs2_remount,
 161         .show_options   = ocfs2_show_options,
 162         .quota_read     = ocfs2_quota_read,
 163         .quota_write    = ocfs2_quota_write,
 164         .get_dquots     = ocfs2_get_dquots,
 165 };
 166
 167 enum {
 168         Opt_barrier,
 169         Opt_err_panic,
 170         Opt_err_ro,
 171         Opt_intr,
 172         Opt_nointr,
 173         Opt_hb_none,
 174         Opt_hb_local,
 175         Opt_hb_global,
 176         Opt_data_ordered,
 177         Opt_data_writeback,
 178         Opt_atime_quantum,
 179         Opt_slot,
 180         Opt_commit,
 181         Opt_localalloc,
 182         Opt_localflocks,
 183         Opt_stack,
 184         Opt_user_xattr,
 185         Opt_nouser_xattr,
 186         Opt_inode64,
 187         Opt_acl,
 188         Opt_noacl,
 189         Opt_usrquota,
 190         Opt_grpquota,
 191         Opt_coherency_buffered,
 192         Opt_coherency_full,
 193         Opt_resv_level,
 194         Opt_dir_resv_level,
 195         Opt_journal_async_commit,
 196         Opt_err_cont,
 197         Opt_err,
 198 };
 199
 200 static const match_table_t tokens = {
 201         {Opt_barrier, "barrier=%u"},
 202         {Opt_err_panic, "errors=panic"},
 203         {Opt_err_ro, "errors=remount-ro"},
 204         {Opt_intr, "intr"},
 205         {Opt_nointr, "nointr"},
 206         {Opt_hb_none, OCFS2_HB_NONE},
 207         {Opt_hb_local, OCFS2_HB_LOCAL},
 208         {Opt_hb_global, OCFS2_HB_GLOBAL},
 209         {Opt_data_ordered, "data=ordered"},
 210         {Opt_data_writeback, "data=writeback"},
 211         {Opt_atime_quantum, "atime_quantum=%u"},
 212         {Opt_slot, "preferred_slot=%u"},
 213         {Opt_commit, "commit=%u"},
 214         {Opt_localalloc, "localalloc=%d"},
 215         {Opt_localflocks, "localflocks"},
 216         {Opt_stack, "cluster_stack=%s"},
 217         {Opt_user_xattr, "user_xattr"},
 218         {Opt_nouser_xattr, "nouser_xattr"},
 219         {Opt_inode64, "inode64"},
 220         {Opt_acl, "acl"},
 221         {Opt_noacl, "noacl"},
 222         {Opt_usrquota, "usrquota"},
 223         {Opt_grpquota, "grpquota"},
 224         {Opt_coherency_buffered, "coherency=buffered"},
 225         {Opt_coherency_full, "coherency=full"},
 226         {Opt_resv_level, "resv_level=%u"},
 227         {Opt_dir_resv_level, "dir_resv_level=%u"},
 228         {Opt_journal_async_commit, "journal_async_commit"},
 229         {Opt_err_cont, "errors=continue"},
 230         {Opt_err, NULL}
 231 };
 232
 233 #ifdef CONFIG_DEBUG_FS
 234 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 235 {
 236         struct ocfs2_cluster_connection *cconn = osb->cconn;
 237         struct ocfs2_recovery_map *rm = osb->recovery_map;
 238         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 239         int i, out = 0;
 240         unsigned long flags;
 241
 242         out += snprintf(buf + out, len - out,
 243                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 244                         "Device", osb->dev_str, osb->uuid_str,
 245                         osb->fs_generation, osb->vol_label);
 246
 247         out += snprintf(buf + out, len - out,
 248                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 249                         atomic_read(&osb->vol_state), osb->osb_flags);
 250
 251         out += snprintf(buf + out, len - out,
 252                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 253                         osb->sb->s_blocksize, osb->s_clustersize);
 254
 255         out += snprintf(buf + out, len - out,
 256                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 257                         "ROcompat: 0x%X\n",
 258                         "Features", osb->s_feature_compat,
 259                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 260
 261         out += snprintf(buf + out, len - out,
 262                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 263                         osb->s_mount_opt, osb->s_atime_quantum);
 264
 265         if (cconn) {
 266                 out += snprintf(buf + out, len - out,
 267                                 "%10s => Stack: %s  Name: %*s  "
 268                                 "Version: %d.%d\n", "Cluster",
 269                                 (*osb->osb_cluster_stack == '\0' ?
 270                                  "o2cb" : osb->osb_cluster_stack),
 271                                 cconn->cc_namelen, cconn->cc_name,
 272                                 cconn->cc_version.pv_major,
 273                                 cconn->cc_version.pv_minor);
 274         }
 275
 276         spin_lock_irqsave(&osb->dc_task_lock, flags);
 277         out += snprintf(buf + out, len - out,
 278                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 279                         "WorkSeq: %lu\n", "DownCnvt",
 280                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 281                         osb->blocked_lock_count, osb->dc_wake_sequence,
 282                         osb->dc_work_sequence);
 283         spin_unlock_irqrestore(&osb->dc_task_lock, flags);
 284
 285         spin_lock(&osb->osb_lock);
 286         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 287                         "Recovery",
 288                         (osb->recovery_thread_task ?
 289                          task_pid_nr(osb->recovery_thread_task) : -1));
 290         if (rm->rm_used == 0)
 291                 out += snprintf(buf + out, len - out, " None\n");
 292         else {
 293                 for (i = 0; i < rm->rm_used; i++)
 294                         out += snprintf(buf + out, len - out, " %d",
 295                                         rm->rm_entries[i]);
 296                 out += snprintf(buf + out, len - out, "\n");
 297         }
 298         spin_unlock(&osb->osb_lock);
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => Pid: %d  Interval: %lu\n", "Commit",
 302                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 303                         osb->osb_commit_interval);
 304
 305         out += snprintf(buf + out, len - out,
 306                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 307                         "Journal", osb->journal->j_state,
 308                         osb->journal->j_trans_id,
 309                         atomic_read(&osb->journal->j_num_trans));
 310
 311         out += snprintf(buf + out, len - out,
 312                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 313                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 314                         "Stats",
 315                         atomic_read(&osb->alloc_stats.bitmap_data),
 316                         atomic_read(&osb->alloc_stats.local_data),
 317                         atomic_read(&osb->alloc_stats.bg_allocs),
 318                         atomic_read(&osb->alloc_stats.moves),
 319                         atomic_read(&osb->alloc_stats.bg_extends));
 320
 321         out += snprintf(buf + out, len - out,
 322                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 323                         "Default: %u bits\n",
 324                         "LocalAlloc", osb->local_alloc_state,
 325                         (unsigned long long)osb->la_last_gd,
 326                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 327
 328         spin_lock(&osb->osb_lock);
 329         out += snprintf(buf + out, len - out,
 330                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 331                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 332                         osb->s_inode_steal_slot,
 333                         atomic_read(&osb->s_num_inodes_stolen),
 334                         osb->s_meta_steal_slot,
 335                         atomic_read(&osb->s_num_meta_stolen));
 336         spin_unlock(&osb->osb_lock);
 337
 338         out += snprintf(buf + out, len - out, "OrphanScan => ");
 339         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 340                         os->os_count, os->os_seqno);
 341         out += snprintf(buf + out, len - out, " Last Scan: ");
 342         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 343                 out += snprintf(buf + out, len - out, "Disabled\n");
 344         else
 345                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 346                                 (get_seconds() - os->os_scantime.tv_sec));
 347
 348         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 349                         "Slots", "Num", "RecoGen");
 350         for (i = 0; i < osb->max_slots; ++i) {
 351                 out += snprintf(buf + out, len - out,
 352                                 "%10s  %c %3d  %10d\n",
 353                                 " ",
 354                                 (i == osb->slot_num ? '*' : ' '),
 355                                 i, osb->slot_recovery_generations[i]);
 356         }
 357
 358         return out;
 359 }
 360
 361 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 362 {
 363         struct ocfs2_super *osb = inode->i_private;
 364         char *buf = NULL;
 365
 366         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 367         if (!buf)
 368                 goto bail;
 369
 370         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 371
 372         file->private_data = buf;
 373
 374         return 0;
 375 bail:
 376         return -ENOMEM;
 377 }
 378
 379 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 380 {
 381         kfree(file->private_data);
 382         return 0;
 383 }
 384
 385 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 386                                 size_t nbytes, loff_t *ppos)
 387 {
 388         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 389                                        i_size_read(file->f_mapping->host));
 390 }
 391 #else
 392 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 393 {
 394         return 0;
 395 }
 396 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 397 {
 398         return 0;
 399 }
 400 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 401                                 size_t nbytes, loff_t *ppos)
 402 {
 403         return 0;
 404 }
 405 #endif  /* CONFIG_DEBUG_FS */
 406
 407 static const struct file_operations ocfs2_osb_debug_fops = {
 408         .open =         ocfs2_osb_debug_open,
 409         .release =      ocfs2_debug_release,
 410         .read =         ocfs2_debug_read,
 411         .llseek =       generic_file_llseek,
 412 };
 413
 414 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 415 {
 416         int status;
 417         tid_t target;
 418         struct ocfs2_super *osb = OCFS2_SB(sb);
 419
 420         if (ocfs2_is_hard_readonly(osb))
 421                 return -EROFS;
 422
 423         if (wait) {
 424                 status = ocfs2_flush_truncate_log(osb);
 425                 if (status < 0)
 426                         mlog_errno(status);
 427         } else {
 428                 ocfs2_schedule_truncate_log_flush(osb, 0);
 429         }
 430
 431         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 432                                       &target)) {
 433                 if (wait)
 434                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 435                                              target);
 436         }
 437         return 0;
 438 }
 439
 440 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 441 {
 442         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 443             && (ino == USER_QUOTA_SYSTEM_INODE
 444                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 445                 return 0;
 446         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 447             && (ino == GROUP_QUOTA_SYSTEM_INODE
 448                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 449                 return 0;
 450         return 1;
 451 }
 452
 453 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 454 {
 455         struct inode *new = NULL;
 456         int status = 0;
 457         int i;
 458
 459         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 460         if (IS_ERR(new)) {
 461                 status = PTR_ERR(new);
 462                 mlog_errno(status);
 463                 goto bail;
 464         }
 465         osb->root_inode = new;
 466
 467         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 468         if (IS_ERR(new)) {
 469                 status = PTR_ERR(new);
 470                 mlog_errno(status);
 471                 goto bail;
 472         }
 473         osb->sys_root_inode = new;
 474
 475         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 476              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 477                 if (!ocfs2_need_system_inode(osb, i))
 478                         continue;
 479                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 480                 if (!new) {
 481                         ocfs2_release_system_inodes(osb);
 482                         status = -EINVAL;
 483                         mlog_errno(status);
 484                         /* FIXME: Should ERROR_RO_FS */
 485                         mlog(ML_ERROR, "Unable to load system inode %d, "
 486                              "possibly corrupt fs?", i);
 487                         goto bail;
 488                 }
 489                 // the array now has one ref, so drop this one
 490                 iput(new);
 491         }
 492
 493 bail:
 494         if (status)
 495                 mlog_errno(status);
 496         return status;
 497 }
 498
 499 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 500 {
 501         struct inode *new = NULL;
 502         int status = 0;
 503         int i;
 504
 505         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 506              i < NUM_SYSTEM_INODES;
 507              i++) {
 508                 if (!ocfs2_need_system_inode(osb, i))
 509                         continue;
 510                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 511                 if (!new) {
 512                         ocfs2_release_system_inodes(osb);
 513                         status = -EINVAL;
 514                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 515                              status, i, osb->slot_num);
 516                         goto bail;
 517                 }
 518                 /* the array now has one ref, so drop this one */
 519                 iput(new);
 520         }
 521
 522 bail:
 523         if (status)
 524                 mlog_errno(status);
 525         return status;
 526 }
 527
 528 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 529 {
 530         int i;
 531         struct inode *inode;
 532
 533         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 534                 inode = osb->global_system_inodes[i];
 535                 if (inode) {
 536                         iput(inode);
 537                         osb->global_system_inodes[i] = NULL;
 538                 }
 539         }
 540
 541         inode = osb->sys_root_inode;
 542         if (inode) {
 543                 iput(inode);
 544                 osb->sys_root_inode = NULL;
 545         }
 546
 547         inode = osb->root_inode;
 548         if (inode) {
 549                 iput(inode);
 550                 osb->root_inode = NULL;
 551         }
 552
 553         if (!osb->local_system_inodes)
 554                 return;
 555
 556         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 557                 if (osb->local_system_inodes[i]) {
 558                         iput(osb->local_system_inodes[i]);
 559                         osb->local_system_inodes[i] = NULL;
 560                 }
 561         }
 562
 563         kfree(osb->local_system_inodes);
 564         osb->local_system_inodes = NULL;
 565 }
 566
 567 /* We're allocating fs objects, use GFP_NOFS */
 568 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 569 {
 570         struct ocfs2_inode_info *oi;
 571
 572         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 573         if (!oi)
 574                 return NULL;
 575
 576         oi->i_sync_tid = 0;
 577         oi->i_datasync_tid = 0;
 578         memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
 579
 580         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 581         return &oi->vfs_inode;
 582 }
 583
 584 static void ocfs2_i_callback(struct rcu_head *head)
 585 {
 586         struct inode *inode = container_of(head, struct inode, i_rcu);
 587         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 588 }
 589
 590 static void ocfs2_destroy_inode(struct inode *inode)
 591 {
 592         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 593 }
 594
 595 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 596                                                 unsigned int cbits)
 597 {
 598         unsigned int bytes = 1 << cbits;
 599         unsigned int trim = bytes;
 600         unsigned int bitshift = 32;
 601
 602         /*
 603          * i_size and all block offsets in ocfs2 are always 64 bits
 604          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 605          * 64 bit platforms, cluster size will be the limiting factor.
 606          */
 607
 608 #if BITS_PER_LONG == 32
 609 # if defined(CONFIG_LBDAF)
 610         BUILD_BUG_ON(sizeof(sector_t) != 8);
 611         /*
 612          * We might be limited by page cache size.
 613          */
 614         if (bytes > PAGE_CACHE_SIZE) {
 615                 bytes = PAGE_CACHE_SIZE;
 616                 trim = 1;
 617                 /*
 618                  * Shift by 31 here so that we don't get larger than
 619                  * MAX_LFS_FILESIZE
 620                  */
 621                 bitshift = 31;
 622         }
 623 # else
 624         /*
 625          * We are limited by the size of sector_t. Use block size, as
 626          * that's what we expose to the VFS.
 627          */
 628         bytes = 1 << bbits;
 629         trim = 1;
 630         bitshift = 31;
 631 # endif
 632 #endif
 633
 634         /*
 635          * Trim by a whole cluster when we can actually approach the
 636          * on-disk limits. Otherwise we can overflow i_clusters when
 637          * an extent start is at the max offset.
 638          */
 639         return (((unsigned long long)bytes) << bitshift) - trim;
 640 }
 641
 642 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 643 {
 644         int incompat_features;
 645         int ret = 0;
 646         struct mount_options parsed_options;
 647         struct ocfs2_super *osb = OCFS2_SB(sb);
 648         u32 tmp;
 649
 650         sync_filesystem(sb);
 651
 652         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 653             !ocfs2_check_set_options(sb, &parsed_options)) {
 654                 ret = -EINVAL;
 655                 goto out;
 656         }
 657
 658         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 659                 OCFS2_MOUNT_HB_NONE;
 660         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 661                 ret = -EINVAL;
 662                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 663                 goto out;
 664         }
 665
 666         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 667             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 668                 ret = -EINVAL;
 669                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 670                 goto out;
 671         }
 672
 673         /* Probably don't want this on remount; it might
 674          * mess with other nodes */
 675         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 676             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 677                 ret = -EINVAL;
 678                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 679                 goto out;
 680         }
 681
 682         /* We're going to/from readonly mode. */
 683         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 684                 /* Disable quota accounting before remounting RO */
 685                 if (*flags & MS_RDONLY) {
 686                         ret = ocfs2_susp_quotas(osb, 0);
 687                         if (ret < 0)
 688                                 goto out;
 689                 }
 690                 /* Lock here so the check of HARD_RO and the potential
 691                  * setting of SOFT_RO is atomic. */
 692                 spin_lock(&osb->osb_lock);
 693                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 694                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 695                         ret = -EROFS;
 696                         goto unlock_osb;
 697                 }
 698
 699                 if (*flags & MS_RDONLY) {
 700                         sb->s_flags |= MS_RDONLY;
 701                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 702                 } else {
 703                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 704                                 mlog(ML_ERROR, "Cannot remount RDWR "
 705                                      "filesystem due to previous errors.\n");
 706                                 ret = -EROFS;
 707                                 goto unlock_osb;
 708                         }
 709                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 710                         if (incompat_features) {
 711                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 712                                      "of unsupported optional features "
 713                                      "(%x).\n", incompat_features);
 714                                 ret = -EINVAL;
 715                                 goto unlock_osb;
 716                         }
 717                         sb->s_flags &= ~MS_RDONLY;
 718                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 719                 }
 720                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 721 unlock_osb:
 722                 spin_unlock(&osb->osb_lock);
 723                 /* Enable quota accounting after remounting RW */
 724                 if (!ret && !(*flags & MS_RDONLY)) {
 725                         if (sb_any_quota_suspended(sb))
 726                                 ret = ocfs2_susp_quotas(osb, 1);
 727                         else
 728                                 ret = ocfs2_enable_quotas(osb);
 729                         if (ret < 0) {
 730                                 /* Return back changes... */
 731                                 spin_lock(&osb->osb_lock);
 732                                 sb->s_flags |= MS_RDONLY;
 733                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 734                                 spin_unlock(&osb->osb_lock);
 735                                 goto out;
 736                         }
 737                 }
 738         }
 739
 740         if (!ret) {
 741                 /* Only save off the new mount options in case of a successful
 742                  * remount. */
 743                 osb->s_mount_opt = parsed_options.mount_opt;
 744                 osb->s_atime_quantum = parsed_options.atime_quantum;
 745                 osb->preferred_slot = parsed_options.slot;
 746                 if (parsed_options.commit_interval)
 747                         osb->osb_commit_interval = parsed_options.commit_interval;
 748
 749                 if (!ocfs2_is_hard_readonly(osb))
 750                         ocfs2_set_journal_params(osb);
 751
 752                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 753                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 754                                                         MS_POSIXACL : 0);
 755         }
 756 out:
 757         return ret;
 758 }
 759
 760 static int ocfs2_sb_probe(struct super_block *sb,
 761                           struct buffer_head **bh,
 762                           int *sector_size,
 763                           struct ocfs2_blockcheck_stats *stats)
 764 {
 765         int status, tmpstat;
 766         struct ocfs1_vol_disk_hdr *hdr;
 767         struct ocfs2_dinode *di;
 768         int blksize;
 769
 770         *bh = NULL;
 771
 772         /* may be > 512 */
 773         *sector_size = bdev_logical_block_size(sb->s_bdev);
 774         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 775                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 776                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 777                 status = -EINVAL;
 778                 goto bail;
 779         }
 780
 781         /* Can this really happen? */
 782         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 783                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 784
 785         /* check block zero for old format */
 786         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 787         if (status < 0) {
 788                 mlog_errno(status);
 789                 goto bail;
 790         }
 791         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 792         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 793                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 794                      hdr->major_version, hdr->minor_version);
 795                 status = -EINVAL;
 796         }
 797         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 798                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 799                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 800                      hdr->signature);
 801                 status = -EINVAL;
 802         }
 803         brelse(*bh);
 804         *bh = NULL;
 805         if (status < 0) {
 806                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 807                      "upgraded before mounting with ocfs v2\n");
 808                 goto bail;
 809         }
 810
 811         /*
 812          * Now check at magic offset for 512, 1024, 2048, 4096
 813          * blocksizes.  4096 is the maximum blocksize because it is
 814          * the minimum clustersize.
 815          */
 816         status = -EINVAL;
 817         for (blksize = *sector_size;
 818              blksize <= OCFS2_MAX_BLOCKSIZE;
 819              blksize <<= 1) {
 820                 tmpstat = ocfs2_get_sector(sb, bh,
 821                                            OCFS2_SUPER_BLOCK_BLKNO,
 822                                            blksize);
 823                 if (tmpstat < 0) {
 824                         status = tmpstat;
 825                         mlog_errno(status);
 826                         break;
 827                 }
 828                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 829                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 830                 spin_lock_init(&stats->b_lock);
 831                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 832                 if (tmpstat < 0) {
 833                         brelse(*bh);
 834                         *bh = NULL;
 835                 }
 836                 if (tmpstat != -EAGAIN) {
 837                         status = tmpstat;
 838                         break;
 839                 }
 840         }
 841
 842 bail:
 843         return status;
 844 }
 845
 846 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 847 {
 848         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 849
 850         if (osb->s_mount_opt & hb_enabled) {
 851                 if (ocfs2_mount_local(osb)) {
 852                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 853                              "mounted device.\n");
 854                         return -EINVAL;
 855                 }
 856                 if (ocfs2_userspace_stack(osb)) {
 857                         mlog(ML_ERROR, "Userspace stack expected, but "
 858                              "o2cb heartbeat arguments passed to mount\n");
 859                         return -EINVAL;
 860                 }
 861                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 862                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 863                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 864                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 865                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 866                         return -EINVAL;
 867                 }
 868         }
 869
 870         if (!(osb->s_mount_opt & hb_enabled)) {
 871                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 872                     !ocfs2_userspace_stack(osb)) {
 873                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 874                              "a read-write clustered device.\n");
 875                         return -EINVAL;
 876                 }
 877         }
 878
 879         return 0;
 880 }
 881
 882 /*
 883  * If we're using a userspace stack, mount should have passed
 884  * a name that matches the disk.  If not, mount should not
 885  * have passed a stack.
 886  */
 887 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 888                                         struct mount_options *mopt)
 889 {
 890         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 891                 mlog(ML_ERROR,
 892                      "cluster stack passed to mount, but this filesystem "
 893                      "does not support it\n");
 894                 return -EINVAL;
 895         }
 896
 897         if (ocfs2_userspace_stack(osb) &&
 898             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 899                     OCFS2_STACK_LABEL_LEN)) {
 900                 mlog(ML_ERROR,
 901                      "cluster stack passed to mount (\"%s\") does not "
 902                      "match the filesystem (\"%s\")\n",
 903                      mopt->cluster_stack,
 904                      osb->osb_cluster_stack);
 905                 return -EINVAL;
 906         }
 907
 908         return 0;
 909 }
 910
 911 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 912 {
 913         int type;
 914         struct super_block *sb = osb->sb;
 915         unsigned int feature[OCFS2_MAXQUOTAS] = {
 916                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 917                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 918         int status = 0;
 919
 920         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 921                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 922                         continue;
 923                 if (unsuspend)
 924                         status = dquot_resume(sb, type);
 925                 else {
 926                         struct ocfs2_mem_dqinfo *oinfo;
 927
 928                         /* Cancel periodic syncing before suspending */
 929                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 930                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 931                         status = dquot_suspend(sb, type);
 932                 }
 933                 if (status < 0)
 934                         break;
 935         }
 936         if (status < 0)
 937                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 938                      "remount (error = %d).\n", status);
 939         return status;
 940 }
 941
 942 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 943 {
 944         struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
 945         struct super_block *sb = osb->sb;
 946         unsigned int feature[OCFS2_MAXQUOTAS] = {
 947                                         OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 948                                         OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 949         unsigned int ino[OCFS2_MAXQUOTAS] = {
 950                                         LOCAL_USER_QUOTA_SYSTEM_INODE,
 951                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 952         int status;
 953         int type;
 954
 955         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 956         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 957                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 958                         continue;
 959                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 960                                                         osb->slot_num);
 961                 if (!inode[type]) {
 962                         status = -ENOENT;
 963                         goto out_quota_off;
 964                 }
 965                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 966                                       DQUOT_USAGE_ENABLED);
 967                 if (status < 0)
 968                         goto out_quota_off;
 969         }
 970
 971         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 972                 iput(inode[type]);
 973         return 0;
 974 out_quota_off:
 975         ocfs2_disable_quotas(osb);
 976         for (type = 0; type < OCFS2_MAXQUOTAS; type++)
 977                 iput(inode[type]);
 978         mlog_errno(status);
 979         return status;
 980 }
 981
 982 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 983 {
 984         int type;
 985         struct inode *inode;
 986         struct super_block *sb = osb->sb;
 987         struct ocfs2_mem_dqinfo *oinfo;
 988
 989         /* We mostly ignore errors in this function because there's not much
 990          * we can do when we see them */
 991         for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
 992                 if (!sb_has_quota_loaded(sb, type))
 993                         continue;
 994                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 995                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 996                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 997                 inode = igrab(sb->s_dquot.files[type]);
 998                 /* Turn off quotas. This will remove all dquot structures from
 999                  * memory and so they will be automatically synced to global
1000                  * quota files */
1001                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
1002                                         DQUOT_LIMITS_ENABLED);
1003                 if (!inode)
1004                         continue;
1005                 iput(inode);
1006         }
1007 }
1008
1009 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1010 {
1011         struct dentry *root;
1012         int status, sector_size;
1013         struct mount_options parsed_options;
1014         struct inode *inode = NULL;
1015         struct ocfs2_super *osb = NULL;
1016         struct buffer_head *bh = NULL;
1017         char nodestr[12];
1018         struct ocfs2_blockcheck_stats stats;
1019
1020         trace_ocfs2_fill_super(sb, data, silent);
1021
1022         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1023                 status = -EINVAL;
1024                 goto read_super_error;
1025         }
1026
1027         /* probe for superblock */
1028         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1029         if (status < 0) {
1030                 mlog(ML_ERROR, "superblock probe failed!\n");
1031                 goto read_super_error;
1032         }
1033
1034         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1035         osb = OCFS2_SB(sb);
1036         if (status < 0) {
1037                 mlog_errno(status);
1038                 goto read_super_error;
1039         }
1040         brelse(bh);
1041         bh = NULL;
1042
1043         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1044                 status = -EINVAL;
1045                 goto read_super_error;
1046         }
1047         osb->s_mount_opt = parsed_options.mount_opt;
1048         osb->s_atime_quantum = parsed_options.atime_quantum;
1049         osb->preferred_slot = parsed_options.slot;
1050         osb->osb_commit_interval = parsed_options.commit_interval;
1051
1052         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1053         osb->osb_resv_level = parsed_options.resv_level;
1054         osb->osb_dir_resv_level = parsed_options.resv_level;
1055         if (parsed_options.dir_resv_level == -1)
1056                 osb->osb_dir_resv_level = parsed_options.resv_level;
1057         else
1058                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1059
1060         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1061         if (status)
1062                 goto read_super_error;
1063
1064         sb->s_magic = OCFS2_SUPER_MAGIC;
1065
1066         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1067                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1068
1069         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1070          * heartbeat=none */
1071         if (bdev_read_only(sb->s_bdev)) {
1072                 if (!(sb->s_flags & MS_RDONLY)) {
1073                         status = -EACCES;
1074                         mlog(ML_ERROR, "Readonly device detected but readonly "
1075                              "mount was not specified.\n");
1076                         goto read_super_error;
1077                 }
1078
1079                 /* You should not be able to start a local heartbeat
1080                  * on a readonly device. */
1081                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1082                         status = -EROFS;
1083                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1084                              "device.\n");
1085                         goto read_super_error;
1086                 }
1087
1088                 status = ocfs2_check_journals_nolocks(osb);
1089                 if (status < 0) {
1090                         if (status == -EROFS)
1091                                 mlog(ML_ERROR, "Recovery required on readonly "
1092                                      "file system, but write access is "
1093                                      "unavailable.\n");
1094                         else
1095                                 mlog_errno(status);
1096                         goto read_super_error;
1097                 }
1098
1099                 ocfs2_set_ro_flag(osb, 1);
1100
1101                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1102                        "Cluster services will not be used for this mount. "
1103                        "Recovery will be skipped.\n", osb->dev_str);
1104         }
1105
1106         if (!ocfs2_is_hard_readonly(osb)) {
1107                 if (sb->s_flags & MS_RDONLY)
1108                         ocfs2_set_ro_flag(osb, 0);
1109         }
1110
1111         status = ocfs2_verify_heartbeat(osb);
1112         if (status < 0) {
1113                 mlog_errno(status);
1114                 goto read_super_error;
1115         }
1116
1117         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1118                                                  ocfs2_debugfs_root);
1119         if (!osb->osb_debug_root) {
1120                 status = -EINVAL;
1121                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1122                 goto read_super_error;
1123         }
1124
1125         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1126                                             osb->osb_debug_root,
1127                                             osb,
1128                                             &ocfs2_osb_debug_fops);
1129         if (!osb->osb_ctxt) {
1130                 status = -EINVAL;
1131                 mlog_errno(status);
1132                 goto read_super_error;
1133         }
1134
1135         if (ocfs2_meta_ecc(osb)) {
1136                 status = ocfs2_blockcheck_stats_debugfs_install(
1137                                                 &osb->osb_ecc_stats,
1138                                                 osb->osb_debug_root);
1139                 if (status) {
1140                         mlog(ML_ERROR,
1141                              "Unable to create blockcheck statistics "
1142                              "files\n");
1143                         goto read_super_error;
1144                 }
1145         }
1146
1147         status = ocfs2_mount_volume(sb);
1148         if (status < 0)
1149                 goto read_super_error;
1150
1151         if (osb->root_inode)
1152                 inode = igrab(osb->root_inode);
1153
1154         if (!inode) {
1155                 status = -EIO;
1156                 mlog_errno(status);
1157                 goto read_super_error;
1158         }
1159
1160         root = d_make_root(inode);
1161         if (!root) {
1162                 status = -ENOMEM;
1163                 mlog_errno(status);
1164                 goto read_super_error;
1165         }
1166
1167         sb->s_root = root;
1168
1169         ocfs2_complete_mount_recovery(osb);
1170
1171         if (ocfs2_mount_local(osb))
1172                 snprintf(nodestr, sizeof(nodestr), "local");
1173         else
1174                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1175
1176         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1177                "with %s data mode.\n",
1178                osb->dev_str, nodestr, osb->slot_num,
1179                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1180                "ordered");
1181
1182         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1183         wake_up(&osb->osb_mount_event);
1184
1185         /* Now we can initialize quotas because we can afford to wait
1186          * for cluster locks recovery now. That also means that truncation
1187          * log recovery can happen but that waits for proper quota setup */
1188         if (!(sb->s_flags & MS_RDONLY)) {
1189                 status = ocfs2_enable_quotas(osb);
1190                 if (status < 0) {
1191                         /* We have to err-out specially here because
1192                          * s_root is already set */
1193                         mlog_errno(status);
1194                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1195                         wake_up(&osb->osb_mount_event);
1196                         return status;
1197                 }
1198         }
1199
1200         ocfs2_complete_quota_recovery(osb);
1201
1202         /* Now we wake up again for processes waiting for quotas */
1203         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1204         wake_up(&osb->osb_mount_event);
1205
1206         /* Start this when the mount is almost sure of being successful */
1207         ocfs2_orphan_scan_start(osb);
1208
1209         /* Create filecheck sysfile /sys/fs/ocfs2/<devname>/filecheck */
1210         ocfs2_filecheck_create_sysfs(sb);
1211
1212         return status;
1213
1214 read_super_error:
1215         brelse(bh);
1216
1217         if (osb) {
1218                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1219                 wake_up(&osb->osb_mount_event);
1220                 ocfs2_dismount_volume(sb, 1);
1221         }
1222
1223         if (status)
1224                 mlog_errno(status);
1225         return status;
1226 }
1227
1228 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1229                         int flags,
1230                         const char *dev_name,
1231                         void *data)
1232 {
1233         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1234 }
1235
1236 static struct file_system_type ocfs2_fs_type = {
1237         .owner          = THIS_MODULE,
1238         .name           = "ocfs2",
1239         .mount          = ocfs2_mount,
1240         .kill_sb        = kill_block_super,
1241         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1242         .next           = NULL
1243 };
1244 MODULE_ALIAS_FS("ocfs2");
1245
1246 static int ocfs2_check_set_options(struct super_block *sb,
1247                                    struct mount_options *options)
1248 {
1249         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1250             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1251                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1252                 mlog(ML_ERROR, "User quotas were requested, but this "
1253                      "filesystem does not have the feature enabled.\n");
1254                 return 0;
1255         }
1256         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1257             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1258                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1259                 mlog(ML_ERROR, "Group quotas were requested, but this "
1260                      "filesystem does not have the feature enabled.\n");
1261                 return 0;
1262         }
1263         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1264             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1265                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1266                      "feature is not enabled\n");
1267                 return 0;
1268         }
1269         /* No ACL setting specified? Use XATTR feature... */
1270         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1271                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1272                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1273                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1274                 else
1275                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1276         }
1277         return 1;
1278 }
1279
1280 static int ocfs2_parse_options(struct super_block *sb,
1281                                char *options,
1282                                struct mount_options *mopt,
1283                                int is_remount)
1284 {
1285         int status, user_stack = 0;
1286         char *p;
1287         u32 tmp;
1288         int token, option;
1289         substring_t args[MAX_OPT_ARGS];
1290
1291         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1292
1293         mopt->commit_interval = 0;
1294         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1295         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1296         mopt->slot = OCFS2_INVALID_SLOT;
1297         mopt->localalloc_opt = -1;
1298         mopt->cluster_stack[0] = '\0';
1299         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1300         mopt->dir_resv_level = -1;
1301
1302         if (!options) {
1303                 status = 1;
1304                 goto bail;
1305         }
1306
1307         while ((p = strsep(&options, ",")) != NULL) {
1308                 if (!*p)
1309                         continue;
1310
1311                 token = match_token(p, tokens, args);
1312                 switch (token) {
1313                 case Opt_hb_local:
1314                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1315                         break;
1316                 case Opt_hb_none:
1317                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1318                         break;
1319                 case Opt_hb_global:
1320                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1321                         break;
1322                 case Opt_barrier:
1323                         if (match_int(&args[0], &option)) {
1324                                 status = 0;
1325                                 goto bail;
1326                         }
1327                         if (option)
1328                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1329                         else
1330                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1331                         break;
1332                 case Opt_intr:
1333                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1334                         break;
1335                 case Opt_nointr:
1336                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1337                         break;
1338                 case Opt_err_panic:
1339                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1340                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1341                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1342                         break;
1343                 case Opt_err_ro:
1344                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_CONT;
1345                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1346                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_ROFS;
1347                         break;
1348                 case Opt_err_cont:
1349                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_ROFS;
1350                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1351                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_CONT;
1352                         break;
1353                 case Opt_data_ordered:
1354                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1355                         break;
1356                 case Opt_data_writeback:
1357                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1358                         break;
1359                 case Opt_user_xattr:
1360                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1361                         break;
1362                 case Opt_nouser_xattr:
1363                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1364                         break;
1365                 case Opt_atime_quantum:
1366                         if (match_int(&args[0], &option)) {
1367                                 status = 0;
1368                                 goto bail;
1369                         }
1370                         if (option >= 0)
1371                                 mopt->atime_quantum = option;
1372                         break;
1373                 case Opt_slot:
1374                         if (match_int(&args[0], &option)) {
1375                                 status = 0;
1376                                 goto bail;
1377                         }
1378                         if (option)
1379                                 mopt->slot = (s16)option;
1380                         break;
1381                 case Opt_commit:
1382                         if (match_int(&args[0], &option)) {
1383                                 status = 0;
1384                                 goto bail;
1385                         }
1386                         if (option < 0)
1387                                 return 0;
1388                         if (option == 0)
1389                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1390                         mopt->commit_interval = HZ * option;
1391                         break;
1392                 case Opt_localalloc:
1393                         if (match_int(&args[0], &option)) {
1394                                 status = 0;
1395                                 goto bail;
1396                         }
1397                         if (option >= 0)
1398                                 mopt->localalloc_opt = option;
1399                         break;
1400                 case Opt_localflocks:
1401                         /*
1402                          * Changing this during remount could race
1403                          * flock() requests, or "unbalance" existing
1404                          * ones (e.g., a lock is taken in one mode but
1405                          * dropped in the other). If users care enough
1406                          * to flip locking modes during remount, we
1407                          * could add a "local" flag to individual
1408                          * flock structures for proper tracking of
1409                          * state.
1410                          */
1411                         if (!is_remount)
1412                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1413                         break;
1414                 case Opt_stack:
1415                         /* Check both that the option we were passed
1416                          * is of the right length and that it is a proper
1417                          * string of the right length.
1418                          */
1419                         if (((args[0].to - args[0].from) !=
1420                              OCFS2_STACK_LABEL_LEN) ||
1421                             (strnlen(args[0].from,
1422                                      OCFS2_STACK_LABEL_LEN) !=
1423                              OCFS2_STACK_LABEL_LEN)) {
1424                                 mlog(ML_ERROR,
1425                                      "Invalid cluster_stack option\n");
1426                                 status = 0;
1427                                 goto bail;
1428                         }
1429                         memcpy(mopt->cluster_stack, args[0].from,
1430                                OCFS2_STACK_LABEL_LEN);
1431                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1432                         /*
1433                          * Open code the memcmp here as we don't have
1434                          * an osb to pass to
1435                          * ocfs2_userspace_stack().
1436                          */
1437                         if (memcmp(mopt->cluster_stack,
1438                                    OCFS2_CLASSIC_CLUSTER_STACK,
1439                                    OCFS2_STACK_LABEL_LEN))
1440                                 user_stack = 1;
1441                         break;
1442                 case Opt_inode64:
1443                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1444                         break;
1445                 case Opt_usrquota:
1446                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1447                         break;
1448                 case Opt_grpquota:
1449                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1450                         break;
1451                 case Opt_coherency_buffered:
1452                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1453                         break;
1454                 case Opt_coherency_full:
1455                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1456                         break;
1457                 case Opt_acl:
1458                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1459                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1460                         break;
1461                 case Opt_noacl:
1462                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1463                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1464                         break;
1465                 case Opt_resv_level:
1466                         if (is_remount)
1467                                 break;
1468                         if (match_int(&args[0], &option)) {
1469                                 status = 0;
1470                                 goto bail;
1471                         }
1472                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1473                             option < OCFS2_MAX_RESV_LEVEL)
1474                                 mopt->resv_level = option;
1475                         break;
1476                 case Opt_dir_resv_level:
1477                         if (is_remount)
1478                                 break;
1479                         if (match_int(&args[0], &option)) {
1480                                 status = 0;
1481                                 goto bail;
1482                         }
1483                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1484                             option < OCFS2_MAX_RESV_LEVEL)
1485                                 mopt->dir_resv_level = option;
1486                         break;
1487                 case Opt_journal_async_commit:
1488                         mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
1489                         break;
1490                 default:
1491                         mlog(ML_ERROR,
1492                              "Unrecognized mount option \"%s\" "
1493                              "or missing value\n", p);
1494                         status = 0;
1495                         goto bail;
1496                 }
1497         }
1498
1499         if (user_stack == 0) {
1500                 /* Ensure only one heartbeat mode */
1501                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1502                                          OCFS2_MOUNT_HB_GLOBAL |
1503                                          OCFS2_MOUNT_HB_NONE);
1504                 if (hweight32(tmp) != 1) {
1505                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1506                         status = 0;
1507                         goto bail;
1508                 }
1509         }
1510
1511         status = 1;
1512
1513 bail:
1514         return status;
1515 }
1516
1517 static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
1518 {
1519         struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
1520         unsigned long opts = osb->s_mount_opt;
1521         unsigned int local_alloc_megs;
1522
1523         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1524                 seq_printf(s, ",_netdev");
1525                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1526                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1527                 else
1528                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1529         } else
1530                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1531
1532         if (opts & OCFS2_MOUNT_NOINTR)
1533                 seq_printf(s, ",nointr");
1534
1535         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1536                 seq_printf(s, ",data=writeback");
1537         else
1538                 seq_printf(s, ",data=ordered");
1539
1540         if (opts & OCFS2_MOUNT_BARRIER)
1541                 seq_printf(s, ",barrier=1");
1542
1543         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1544                 seq_printf(s, ",errors=panic");
1545         else if (opts & OCFS2_MOUNT_ERRORS_CONT)
1546                 seq_printf(s, ",errors=continue");
1547         else
1548                 seq_printf(s, ",errors=remount-ro");
1549
1550         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1551                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1552
1553         seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1554
1555         if (osb->osb_commit_interval)
1556                 seq_printf(s, ",commit=%u",
1557                            (unsigned) (osb->osb_commit_interval / HZ));
1558
1559         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1560         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1561                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1562
1563         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1564                 seq_printf(s, ",localflocks,");
1565
1566         if (osb->osb_cluster_stack[0])
1567                 seq_show_option_n(s, "cluster_stack", osb->osb_cluster_stack,
1568                                   OCFS2_STACK_LABEL_LEN);
1569         if (opts & OCFS2_MOUNT_USRQUOTA)
1570                 seq_printf(s, ",usrquota");
1571         if (opts & OCFS2_MOUNT_GRPQUOTA)
1572                 seq_printf(s, ",grpquota");
1573
1574         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1575                 seq_printf(s, ",coherency=buffered");
1576         else
1577                 seq_printf(s, ",coherency=full");
1578
1579         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1580                 seq_printf(s, ",nouser_xattr");
1581         else
1582                 seq_printf(s, ",user_xattr");
1583
1584         if (opts & OCFS2_MOUNT_INODE64)
1585                 seq_printf(s, ",inode64");
1586
1587         if (opts & OCFS2_MOUNT_POSIX_ACL)
1588                 seq_printf(s, ",acl");
1589         else
1590                 seq_printf(s, ",noacl");
1591
1592         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1593                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1594
1595         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1596                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1597
1598         if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
1599                 seq_printf(s, ",journal_async_commit");
1600
1601         return 0;
1602 }
1603
1604 static int __init ocfs2_init(void)
1605 {
1606         int status;
1607
1608         status = init_ocfs2_uptodate_cache();
1609         if (status < 0)
1610                 goto out1;
1611
1612         status = ocfs2_initialize_mem_caches();
1613         if (status < 0)
1614                 goto out2;
1615
1616         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1617         if (!ocfs2_wq) {
1618                 status = -ENOMEM;
1619                 goto out3;
1620         }
1621
1622         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1623         if (!ocfs2_debugfs_root) {
1624                 status = -ENOMEM;
1625                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1626                 goto out4;
1627         }
1628
1629         ocfs2_set_locking_protocol();
1630
1631         status = register_quota_format(&ocfs2_quota_format);
1632         if (status < 0)
1633                 goto out4;
1634         status = register_filesystem(&ocfs2_fs_type);
1635         if (!status)
1636                 return 0;
1637
1638         unregister_quota_format(&ocfs2_quota_format);
1639 out4:
1640         destroy_workqueue(ocfs2_wq);
1641         debugfs_remove(ocfs2_debugfs_root);
1642 out3:
1643         ocfs2_free_mem_caches();
1644 out2:
1645         exit_ocfs2_uptodate_cache();
1646 out1:
1647         mlog_errno(status);
1648         return status;
1649 }
1650
1651 static void __exit ocfs2_exit(void)
1652 {
1653         if (ocfs2_wq) {
1654                 flush_workqueue(ocfs2_wq);
1655                 destroy_workqueue(ocfs2_wq);
1656         }
1657
1658         unregister_quota_format(&ocfs2_quota_format);
1659
1660         debugfs_remove(ocfs2_debugfs_root);
1661
1662         ocfs2_free_mem_caches();
1663
1664         unregister_filesystem(&ocfs2_fs_type);
1665
1666         exit_ocfs2_uptodate_cache();
1667 }
1668
1669 static void ocfs2_put_super(struct super_block *sb)
1670 {
1671         trace_ocfs2_put_super(sb);
1672
1673         ocfs2_sync_blockdev(sb);
1674         ocfs2_dismount_volume(sb, 0);
1675         ocfs2_filecheck_remove_sysfs(sb);
1676 }
1677
1678 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1679 {
1680         struct ocfs2_super *osb;
1681         u32 numbits, freebits;
1682         int status;
1683         struct ocfs2_dinode *bm_lock;
1684         struct buffer_head *bh = NULL;
1685         struct inode *inode = NULL;
1686
1687         trace_ocfs2_statfs(dentry->d_sb, buf);
1688
1689         osb = OCFS2_SB(dentry->d_sb);
1690
1691         inode = ocfs2_get_system_file_inode(osb,
1692                                             GLOBAL_BITMAP_SYSTEM_INODE,
1693                                             OCFS2_INVALID_SLOT);
1694         if (!inode) {
1695                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1696                 status = -EIO;
1697                 goto bail;
1698         }
1699
1700         status = ocfs2_inode_lock(inode, &bh, 0);
1701         if (status < 0) {
1702                 mlog_errno(status);
1703                 goto bail;
1704         }
1705
1706         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1707
1708         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1709         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1710
1711         buf->f_type = OCFS2_SUPER_MAGIC;
1712         buf->f_bsize = dentry->d_sb->s_blocksize;
1713         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1714         buf->f_blocks = ((sector_t) numbits) *
1715                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1716         buf->f_bfree = ((sector_t) freebits) *
1717                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1718         buf->f_bavail = buf->f_bfree;
1719         buf->f_files = numbits;
1720         buf->f_ffree = freebits;
1721         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1722                                 & 0xFFFFFFFFUL;
1723         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1724                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1725
1726         brelse(bh);
1727
1728         ocfs2_inode_unlock(inode, 0);
1729         status = 0;
1730 bail:
1731         iput(inode);
1732
1733         if (status)
1734                 mlog_errno(status);
1735
1736         return status;
1737 }
1738
1739 static void ocfs2_inode_init_once(void *data)
1740 {
1741         struct ocfs2_inode_info *oi = data;
1742
1743         oi->ip_flags = 0;
1744         oi->ip_open_count = 0;
1745         spin_lock_init(&oi->ip_lock);
1746         ocfs2_extent_map_init(&oi->vfs_inode);
1747         INIT_LIST_HEAD(&oi->ip_io_markers);
1748         oi->ip_dir_start_lookup = 0;
1749         mutex_init(&oi->ip_unaligned_aio);
1750         init_rwsem(&oi->ip_alloc_sem);
1751         init_rwsem(&oi->ip_xattr_sem);
1752         mutex_init(&oi->ip_io_mutex);
1753
1754         oi->ip_blkno = 0ULL;
1755         oi->ip_clusters = 0;
1756
1757         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1758
1759         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1760         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1761         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1762
1763         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1764                                   &ocfs2_inode_caching_ops);
1765
1766         inode_init_once(&oi->vfs_inode);
1767 }
1768
1769 static int ocfs2_initialize_mem_caches(void)
1770 {
1771         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1772                                        sizeof(struct ocfs2_inode_info),
1773                                        0,
1774                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1775                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1776                                        ocfs2_inode_init_once);
1777         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1778                                         sizeof(struct ocfs2_dquot),
1779                                         0,
1780                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1781                                                 SLAB_MEM_SPREAD),
1782                                         NULL);
1783         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1784                                         sizeof(struct ocfs2_quota_chunk),
1785                                         0,
1786                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1787                                         NULL);
1788         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1789             !ocfs2_qf_chunk_cachep) {
1790                 if (ocfs2_inode_cachep)
1791                         kmem_cache_destroy(ocfs2_inode_cachep);
1792                 if (ocfs2_dquot_cachep)
1793                         kmem_cache_destroy(ocfs2_dquot_cachep);
1794                 if (ocfs2_qf_chunk_cachep)
1795                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1796                 return -ENOMEM;
1797         }
1798
1799         return 0;
1800 }
1801
1802 static void ocfs2_free_mem_caches(void)
1803 {
1804         /*
1805          * Make sure all delayed rcu free inodes are flushed before we
1806          * destroy cache.
1807          */
1808         rcu_barrier();
1809         if (ocfs2_inode_cachep)
1810                 kmem_cache_destroy(ocfs2_inode_cachep);
1811         ocfs2_inode_cachep = NULL;
1812
1813         if (ocfs2_dquot_cachep)
1814                 kmem_cache_destroy(ocfs2_dquot_cachep);
1815         ocfs2_dquot_cachep = NULL;
1816
1817         if (ocfs2_qf_chunk_cachep)
1818                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1819         ocfs2_qf_chunk_cachep = NULL;
1820 }
1821
1822 static int ocfs2_get_sector(struct super_block *sb,
1823                             struct buffer_head **bh,
1824                             int block,
1825                             int sect_size)
1826 {
1827         if (!sb_set_blocksize(sb, sect_size)) {
1828                 mlog(ML_ERROR, "unable to set blocksize\n");
1829                 return -EIO;
1830         }
1831
1832         *bh = sb_getblk(sb, block);
1833         if (!*bh) {
1834                 mlog_errno(-ENOMEM);
1835                 return -ENOMEM;
1836         }
1837         lock_buffer(*bh);
1838         if (!buffer_dirty(*bh))
1839                 clear_buffer_uptodate(*bh);
1840         unlock_buffer(*bh);
1841         ll_rw_block(READ, 1, bh);
1842         wait_on_buffer(*bh);
1843         if (!buffer_uptodate(*bh)) {
1844                 mlog_errno(-EIO);
1845                 brelse(*bh);
1846                 *bh = NULL;
1847                 return -EIO;
1848         }
1849
1850         return 0;
1851 }
1852
1853 static int ocfs2_mount_volume(struct super_block *sb)
1854 {
1855         int status = 0;
1856         int unlock_super = 0;
1857         struct ocfs2_super *osb = OCFS2_SB(sb);
1858
1859         if (ocfs2_is_hard_readonly(osb))
1860                 goto leave;
1861
1862         status = ocfs2_dlm_init(osb);
1863         if (status < 0) {
1864                 mlog_errno(status);
1865                 goto leave;
1866         }
1867
1868         status = ocfs2_super_lock(osb, 1);
1869         if (status < 0) {
1870                 mlog_errno(status);
1871                 goto leave;
1872         }
1873         unlock_super = 1;
1874
1875         /* This will load up the node map and add ourselves to it. */
1876         status = ocfs2_find_slot(osb);
1877         if (status < 0) {
1878                 mlog_errno(status);
1879                 goto leave;
1880         }
1881
1882         /* load all node-local system inodes */
1883         status = ocfs2_init_local_system_inodes(osb);
1884         if (status < 0) {
1885                 mlog_errno(status);
1886                 goto leave;
1887         }
1888
1889         status = ocfs2_check_volume(osb);
1890         if (status < 0) {
1891                 mlog_errno(status);
1892                 goto leave;
1893         }
1894
1895         status = ocfs2_truncate_log_init(osb);
1896         if (status < 0)
1897                 mlog_errno(status);
1898
1899 leave:
1900         if (unlock_super)
1901                 ocfs2_super_unlock(osb, 1);
1902
1903         return status;
1904 }
1905
1906 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1907 {
1908         int tmp, hangup_needed = 0;
1909         struct ocfs2_super *osb = NULL;
1910         char nodestr[12];
1911
1912         trace_ocfs2_dismount_volume(sb);
1913
1914         BUG_ON(!sb);
1915         osb = OCFS2_SB(sb);
1916         BUG_ON(!osb);
1917
1918         debugfs_remove(osb->osb_ctxt);
1919
1920         /* Orphan scan should be stopped as early as possible */
1921         ocfs2_orphan_scan_stop(osb);
1922
1923         ocfs2_disable_quotas(osb);
1924
1925         /* All dquots should be freed by now */
1926         WARN_ON(!llist_empty(&osb->dquot_drop_list));
1927         /* Wait for worker to be done with the work structure in osb */
1928         cancel_work_sync(&osb->dquot_drop_work);
1929
1930         ocfs2_shutdown_local_alloc(osb);
1931
1932         ocfs2_truncate_log_shutdown(osb);
1933
1934         /* This will disable recovery and flush any recovery work. */
1935         ocfs2_recovery_exit(osb);
1936
1937         ocfs2_journal_shutdown(osb);
1938
1939         ocfs2_sync_blockdev(sb);
1940
1941         ocfs2_purge_refcount_trees(osb);
1942
1943         /* No cluster connection means we've failed during mount, so skip
1944          * all the steps which depended on that to complete. */
1945         if (osb->cconn) {
1946                 tmp = ocfs2_super_lock(osb, 1);
1947                 if (tmp < 0) {
1948                         mlog_errno(tmp);
1949                         return;
1950                 }
1951         }
1952
1953         if (osb->slot_num != OCFS2_INVALID_SLOT)
1954                 ocfs2_put_slot(osb);
1955
1956         if (osb->cconn)
1957                 ocfs2_super_unlock(osb, 1);
1958
1959         ocfs2_release_system_inodes(osb);
1960
1961         /*
1962          * If we're dismounting due to mount error, mount.ocfs2 will clean
1963          * up heartbeat.  If we're a local mount, there is no heartbeat.
1964          * If we failed before we got a uuid_str yet, we can't stop
1965          * heartbeat.  Otherwise, do it.
1966          */
1967         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1968             !ocfs2_is_hard_readonly(osb))
1969                 hangup_needed = 1;
1970
1971         if (osb->cconn)
1972                 ocfs2_dlm_shutdown(osb, hangup_needed);
1973
1974         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1975         debugfs_remove(osb->osb_debug_root);
1976
1977         if (hangup_needed)
1978                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1979
1980         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1981
1982         if (ocfs2_mount_local(osb))
1983                 snprintf(nodestr, sizeof(nodestr), "local");
1984         else
1985                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1986
1987         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1988                osb->dev_str, nodestr);
1989
1990         ocfs2_delete_osb(osb);
1991         kfree(osb);
1992         sb->s_dev = 0;
1993         sb->s_fs_info = NULL;
1994 }
1995
1996 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1997                                 unsigned uuid_bytes)
1998 {
1999         int i, ret;
2000         char *ptr;
2001
2002         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2003
2004         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2005         if (osb->uuid_str == NULL)
2006                 return -ENOMEM;
2007
2008         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2009                 /* print with null */
2010                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2011                 if (ret != 2) /* drop super cleans up */
2012                         return -EINVAL;
2013                 /* then only advance past the last char */
2014                 ptr += 2;
2015         }
2016
2017         return 0;
2018 }
2019
2020 /* Make sure entire volume is addressable by our journal.  Requires
2021    osb_clusters_at_boot to be valid and for the journal to have been
2022    initialized by ocfs2_journal_init(). */
2023 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2024 {
2025         int status = 0;
2026         u64 max_block =
2027                 ocfs2_clusters_to_blocks(osb->sb,
2028                                          osb->osb_clusters_at_boot) - 1;
2029
2030         /* 32-bit block number is always OK. */
2031         if (max_block <= (u32)~0ULL)
2032                 goto out;
2033
2034         /* Volume is "huge", so see if our journal is new enough to
2035            support it. */
2036         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2037                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2038               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2039                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2040                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2041                      "Enable the 'block64' journal option with tunefs.ocfs2");
2042                 status = -EFBIG;
2043                 goto out;
2044         }
2045
2046  out:
2047         return status;
2048 }
2049
2050 static int ocfs2_initialize_super(struct super_block *sb,
2051                                   struct buffer_head *bh,
2052                                   int sector_size,
2053                                   struct ocfs2_blockcheck_stats *stats)
2054 {
2055         int status;
2056         int i, cbits, bbits;
2057         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2058         struct inode *inode = NULL;
2059         struct ocfs2_journal *journal;
2060         struct ocfs2_super *osb;
2061         u64 total_blocks;
2062
2063         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2064         if (!osb) {
2065                 status = -ENOMEM;
2066                 mlog_errno(status);
2067                 goto bail;
2068         }
2069
2070         sb->s_fs_info = osb;
2071         sb->s_op = &ocfs2_sops;
2072         sb->s_d_op = &ocfs2_dentry_ops;
2073         sb->s_export_op = &ocfs2_export_ops;
2074         sb->s_qcop = &dquot_quotactl_sysfile_ops;
2075         sb->dq_op = &ocfs2_quota_operations;
2076         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
2077         sb->s_xattr = ocfs2_xattr_handlers;
2078         sb->s_time_gran = 1;
2079         sb->s_flags |= MS_NOATIME;
2080         /* this is needed to support O_LARGEFILE */
2081         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2082         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2083         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2084         memcpy(sb->s_uuid, di->id2.i_super.s_uuid,
2085                sizeof(di->id2.i_super.s_uuid));
2086
2087         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2088
2089         for (i = 0; i < 3; i++)
2090                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2091         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2092
2093         osb->sb = sb;
2094         /* Save off for ocfs2_rw_direct */
2095         osb->s_sectsize_bits = blksize_bits(sector_size);
2096         BUG_ON(!osb->s_sectsize_bits);
2097
2098         spin_lock_init(&osb->dc_task_lock);
2099         init_waitqueue_head(&osb->dc_event);
2100         osb->dc_work_sequence = 0;
2101         osb->dc_wake_sequence = 0;
2102         INIT_LIST_HEAD(&osb->blocked_lock_list);
2103         osb->blocked_lock_count = 0;
2104         spin_lock_init(&osb->osb_lock);
2105         spin_lock_init(&osb->osb_xattr_lock);
2106         ocfs2_init_steal_slots(osb);
2107
2108         mutex_init(&osb->system_file_mutex);
2109
2110         atomic_set(&osb->alloc_stats.moves, 0);
2111         atomic_set(&osb->alloc_stats.local_data, 0);
2112         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2113         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2114         atomic_set(&osb->alloc_stats.bg_extends, 0);
2115
2116         /* Copy the blockcheck stats from the superblock probe */
2117         osb->osb_ecc_stats = *stats;
2118
2119         ocfs2_init_node_maps(osb);
2120
2121         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2122                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2123
2124         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2125         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2126                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2127                      osb->max_slots);
2128                 status = -EINVAL;
2129                 goto bail;
2130         }
2131
2132         ocfs2_orphan_scan_init(osb);
2133
2134         status = ocfs2_recovery_init(osb);
2135         if (status) {
2136                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2137                 mlog_errno(status);
2138                 goto bail;
2139         }
2140
2141         init_waitqueue_head(&osb->checkpoint_event);
2142
2143         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2144
2145         osb->slot_num = OCFS2_INVALID_SLOT;
2146
2147         osb->s_xattr_inline_size = le16_to_cpu(
2148                                         di->id2.i_super.s_xattr_inline_size);
2149
2150         osb->local_alloc_state = OCFS2_LA_UNUSED;
2151         osb->local_alloc_bh = NULL;
2152         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2153
2154         init_waitqueue_head(&osb->osb_mount_event);
2155
2156         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2157         if (status) {
2158                 mlog_errno(status);
2159                 goto bail;
2160         }
2161
2162         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2163         if (!osb->vol_label) {
2164                 mlog(ML_ERROR, "unable to alloc vol label\n");
2165                 status = -ENOMEM;
2166                 goto bail;
2167         }
2168
2169         osb->slot_recovery_generations =
2170                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2171                         GFP_KERNEL);
2172         if (!osb->slot_recovery_generations) {
2173                 status = -ENOMEM;
2174                 mlog_errno(status);
2175                 goto bail;
2176         }
2177
2178         init_waitqueue_head(&osb->osb_wipe_event);
2179         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2180                                         sizeof(*osb->osb_orphan_wipes),
2181                                         GFP_KERNEL);
2182         if (!osb->osb_orphan_wipes) {
2183                 status = -ENOMEM;
2184                 mlog_errno(status);
2185                 goto bail;
2186         }
2187
2188         osb->osb_rf_lock_tree = RB_ROOT;
2189
2190         osb->s_feature_compat =
2191                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2192         osb->s_feature_ro_compat =
2193                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2194         osb->s_feature_incompat =
2195                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2196
2197         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2198                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2199                      "optional features (%x).\n", i);
2200                 status = -EINVAL;
2201                 goto bail;
2202         }
2203         if (!(osb->sb->s_flags & MS_RDONLY) &&
2204             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2205                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2206                      "unsupported optional features (%x).\n", i);
2207                 status = -EINVAL;
2208                 goto bail;
2209         }
2210
2211         if (ocfs2_clusterinfo_valid(osb)) {
2212                 osb->osb_stackflags =
2213                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2214                 strlcpy(osb->osb_cluster_stack,
2215                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2216                        OCFS2_STACK_LABEL_LEN + 1);
2217                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2218                         mlog(ML_ERROR,
2219                              "couldn't mount because of an invalid "
2220                              "cluster stack label (%s) \n",
2221                              osb->osb_cluster_stack);
2222                         status = -EINVAL;
2223                         goto bail;
2224                 }
2225                 strlcpy(osb->osb_cluster_name,
2226                         OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
2227                         OCFS2_CLUSTER_NAME_LEN + 1);
2228         } else {
2229                 /* The empty string is identical with classic tools that
2230                  * don't know about s_cluster_info. */
2231                 osb->osb_cluster_stack[0] = '\0';
2232         }
2233
2234         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2235
2236         /* FIXME
2237          * This should be done in ocfs2_journal_init(), but unknown
2238          * ordering issues will cause the filesystem to crash.
2239          * If anyone wants to figure out what part of the code
2240          * refers to osb->journal before ocfs2_journal_init() is run,
2241          * be my guest.
2242          */
2243         /* initialize our journal structure */
2244
2245         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2246         if (!journal) {
2247                 mlog(ML_ERROR, "unable to alloc journal\n");
2248                 status = -ENOMEM;
2249                 goto bail;
2250         }
2251         osb->journal = journal;
2252         journal->j_osb = osb;
2253
2254         atomic_set(&journal->j_num_trans, 0);
2255         init_rwsem(&journal->j_trans_barrier);
2256         init_waitqueue_head(&journal->j_checkpointed);
2257         spin_lock_init(&journal->j_lock);
2258         journal->j_trans_id = (unsigned long) 1;
2259         INIT_LIST_HEAD(&journal->j_la_cleanups);
2260         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2261         journal->j_state = OCFS2_JOURNAL_FREE;
2262
2263         INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
2264         init_llist_head(&osb->dquot_drop_list);
2265
2266         /* get some pseudo constants for clustersize bits */
2267         osb->s_clustersize_bits =
2268                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2269         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2270
2271         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2272             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2273                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2274                      osb->s_clustersize);
2275                 status = -EINVAL;
2276                 goto bail;
2277         }
2278
2279         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2280                                                 le32_to_cpu(di->i_clusters));
2281
2282         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2283                                            total_blocks);
2284         if (status) {
2285                 mlog(ML_ERROR, "Volume too large "
2286                      "to mount safely on this system");
2287                 status = -EFBIG;
2288                 goto bail;
2289         }
2290
2291         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2292                                  sizeof(di->id2.i_super.s_uuid))) {
2293                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2294                 status = -ENOMEM;
2295                 goto bail;
2296         }
2297
2298         strlcpy(osb->vol_label, di->id2.i_super.s_label,
2299                 OCFS2_MAX_VOL_LABEL_LEN);
2300         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2301         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2302         osb->first_cluster_group_blkno =
2303                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2304         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2305         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2306         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2307                                      (unsigned long long)osb->root_blkno,
2308                                      (unsigned long long)osb->system_dir_blkno,
2309                                      osb->s_clustersize_bits);
2310
2311         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2312         if (!osb->osb_dlm_debug) {
2313                 status = -ENOMEM;
2314                 mlog_errno(status);
2315                 goto bail;
2316         }
2317
2318         atomic_set(&osb->vol_state, VOLUME_INIT);
2319
2320         /* load root, system_dir, and all global system inodes */
2321         status = ocfs2_init_global_system_inodes(osb);
2322         if (status < 0) {
2323                 mlog_errno(status);
2324                 goto bail;
2325         }
2326
2327         /*
2328          * global bitmap
2329          */
2330         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2331                                             OCFS2_INVALID_SLOT);
2332         if (!inode) {
2333                 status = -EINVAL;
2334                 mlog_errno(status);
2335                 goto bail;
2336         }
2337
2338         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2339         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2340         iput(inode);
2341
2342         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2343                                  osb->s_feature_incompat) * 8;
2344
2345         status = ocfs2_init_slot_info(osb);
2346         if (status < 0) {
2347                 mlog_errno(status);
2348                 goto bail;
2349         }
2350         cleancache_init_shared_fs(sb);
2351
2352 bail:
2353         return status;
2354 }
2355
2356 /*
2357  * will return: -EAGAIN if it is ok to keep searching for superblocks
2358  *              -EINVAL if there is a bad superblock
2359  *              0 on success
2360  */
2361 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2362                                struct buffer_head *bh,
2363                                u32 blksz,
2364                                struct ocfs2_blockcheck_stats *stats)
2365 {
2366         int status = -EAGAIN;
2367
2368         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2369                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2370                 /* We have to do a raw check of the feature here */
2371                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2372                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2373                         status = ocfs2_block_check_validate(bh->b_data,
2374                                                             bh->b_size,
2375                                                             &di->i_check,
2376                                                             stats);
2377                         if (status)
2378                                 goto out;
2379                 }
2380                 status = -EINVAL;
2381                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2382                         mlog(ML_ERROR, "found superblock with incorrect block "
2383                              "size: found %u, should be %u\n",
2384                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2385                                blksz);
2386                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2387                            OCFS2_MAJOR_REV_LEVEL ||
2388                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2389                            OCFS2_MINOR_REV_LEVEL) {
2390                         mlog(ML_ERROR, "found superblock with bad version: "
2391                              "found %u.%u, should be %u.%u\n",
2392                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2393                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2394                              OCFS2_MAJOR_REV_LEVEL,
2395                              OCFS2_MINOR_REV_LEVEL);
2396                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2397                         mlog(ML_ERROR, "bad block number on superblock: "
2398                              "found %llu, should be %llu\n",
2399                              (unsigned long long)le64_to_cpu(di->i_blkno),
2400                              (unsigned long long)bh->b_blocknr);
2401                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2402                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2403                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2404                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2405                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2406                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2407                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2408                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2409                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2410                         mlog(ML_ERROR,
2411                              "Superblock slots found greater than file system "
2412                              "maximum: found %u, max %u\n",
2413                              le16_to_cpu(di->id2.i_super.s_max_slots),
2414                              OCFS2_MAX_SLOTS);
2415                 } else {
2416                         /* found it! */
2417                         status = 0;
2418                 }
2419         }
2420
2421 out:
2422         if (status && status != -EAGAIN)
2423                 mlog_errno(status);
2424         return status;
2425 }
2426
2427 static int ocfs2_check_volume(struct ocfs2_super *osb)
2428 {
2429         int status;
2430         int dirty;
2431         int local;
2432         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2433                                                   * recover
2434                                                   * ourselves. */
2435
2436         /* Init our journal object. */
2437         status = ocfs2_journal_init(osb->journal, &dirty);
2438         if (status < 0) {
2439                 mlog(ML_ERROR, "Could not initialize journal!\n");
2440                 goto finally;
2441         }
2442
2443         /* Now that journal has been initialized, check to make sure
2444            entire volume is addressable. */
2445         status = ocfs2_journal_addressable(osb);
2446         if (status)
2447                 goto finally;
2448
2449         /* If the journal was unmounted cleanly then we don't want to
2450          * recover anything. Otherwise, journal_load will do that
2451          * dirty work for us :) */
2452         if (!dirty) {
2453                 status = ocfs2_journal_wipe(osb->journal, 0);
2454                 if (status < 0) {
2455                         mlog_errno(status);
2456                         goto finally;
2457                 }
2458         } else {
2459                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2460                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2461         }
2462
2463         local = ocfs2_mount_local(osb);
2464
2465         /* will play back anything left in the journal. */
2466         status = ocfs2_journal_load(osb->journal, local, dirty);
2467         if (status < 0) {
2468                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2469                 goto finally;
2470         }
2471
2472         if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
2473                 jbd2_journal_set_features(osb->journal->j_journal,
2474                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2475                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2476         else
2477                 jbd2_journal_clear_features(osb->journal->j_journal,
2478                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2479                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2480
2481         if (dirty) {
2482                 /* recover my local alloc if we didn't unmount cleanly. */
2483                 status = ocfs2_begin_local_alloc_recovery(osb,
2484                                                           osb->slot_num,
2485                                                           &local_alloc);
2486                 if (status < 0) {
2487                         mlog_errno(status);
2488                         goto finally;
2489                 }
2490                 /* we complete the recovery process after we've marked
2491                  * ourselves as mounted. */
2492         }
2493
2494         status = ocfs2_load_local_alloc(osb);
2495         if (status < 0) {
2496                 mlog_errno(status);
2497                 goto finally;
2498         }
2499
2500         if (dirty) {
2501                 /* Recovery will be completed after we've mounted the
2502                  * rest of the volume. */
2503                 osb->dirty = 1;
2504                 osb->local_alloc_copy = local_alloc;
2505                 local_alloc = NULL;
2506         }
2507
2508         /* go through each journal, trylock it and if you get the
2509          * lock, and it's marked as dirty, set the bit in the recover
2510          * map and launch a recovery thread for it. */
2511         status = ocfs2_mark_dead_nodes(osb);
2512         if (status < 0) {
2513                 mlog_errno(status);
2514                 goto finally;
2515         }
2516
2517         status = ocfs2_compute_replay_slots(osb);
2518         if (status < 0)
2519                 mlog_errno(status);
2520
2521 finally:
2522         kfree(local_alloc);
2523
2524         if (status)
2525                 mlog_errno(status);
2526         return status;
2527 }
2528
2529 /*
2530  * The routine gets called from dismount or close whenever a dismount on
2531  * volume is requested and the osb open count becomes 1.
2532  * It will remove the osb from the global list and also free up all the
2533  * initialized resources and fileobject.
2534  */
2535 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2536 {
2537         /* This function assumes that the caller has the main osb resource */
2538
2539         ocfs2_free_slot_info(osb);
2540
2541         kfree(osb->osb_orphan_wipes);
2542         kfree(osb->slot_recovery_generations);
2543         /* FIXME
2544          * This belongs in journal shutdown, but because we have to
2545          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2546          * we free it here.
2547          */
2548         kfree(osb->journal);
2549         kfree(osb->local_alloc_copy);
2550         kfree(osb->uuid_str);
2551         kfree(osb->vol_label);
2552         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2553         memset(osb, 0, sizeof(struct ocfs2_super));
2554 }
2555
2556 /* Depending on the mount option passed, perform one of the following:
2557  * Put OCFS2 into a readonly state (default)
2558  * Return EIO so that only the process errs
2559  * Fix the error as if fsck.ocfs2 -y
2560  * panic
2561  */
2562 static int ocfs2_handle_error(struct super_block *sb)
2563 {
2564         struct ocfs2_super *osb = OCFS2_SB(sb);
2565         int rv = 0;
2566
2567         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2568         pr_crit("On-disk corruption discovered. "
2569                 "Please run fsck.ocfs2 once the filesystem is unmounted.\n");
2570
2571         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC) {
2572                 panic("OCFS2: (device %s): panic forced after error\n",
2573                       sb->s_id);
2574         } else if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_CONT) {
2575                 pr_crit("OCFS2: Returning error to the calling process.\n");
2576                 rv = -EIO;
2577         } else { /* default option */
2578                 rv = -EROFS;
2579                 if (sb->s_flags & MS_RDONLY &&
2580                                 (ocfs2_is_soft_readonly(osb) ||
2581                                  ocfs2_is_hard_readonly(osb)))
2582                         return rv;
2583
2584                 pr_crit("OCFS2: File system is now read-only.\n");
2585                 sb->s_flags |= MS_RDONLY;
2586                 ocfs2_set_ro_flag(osb, 0);
2587         }
2588
2589         return rv;
2590 }
2591
2592 int __ocfs2_error(struct super_block *sb, const char *function,
2593                   const char *fmt, ...)
2594 {
2595         struct va_format vaf;
2596         va_list args;
2597
2598         va_start(args, fmt);
2599         vaf.fmt = fmt;
2600         vaf.va = &args;
2601
2602         /* Not using mlog here because we want to show the actual
2603          * function the error came from. */
2604         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV",
2605                sb->s_id, function, &vaf);
2606
2607         va_end(args);
2608
2609         return ocfs2_handle_error(sb);
2610 }
2611
2612 /* Handle critical errors. This is intentionally more drastic than
2613  * ocfs2_handle_error, so we only use for things like journal errors,
2614  * etc. */
2615 void __ocfs2_abort(struct super_block *sb, const char *function,
2616                    const char *fmt, ...)
2617 {
2618         struct va_format vaf;
2619         va_list args;
2620
2621         va_start(args, fmt);
2622
2623         vaf.fmt = fmt;
2624         vaf.va = &args;
2625
2626         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV",
2627                sb->s_id, function, &vaf);
2628
2629         va_end(args);
2630
2631         /* We don't have the cluster support yet to go straight to
2632          * hard readonly in here. Until then, we want to keep
2633          * ocfs2_abort() so that we can at least mark critical
2634          * errors.
2635          *
2636          * TODO: This should abort the journal and alert other nodes
2637          * that our slot needs recovery. */
2638
2639         /* Force a panic(). This stinks, but it's better than letting
2640          * things continue without having a proper hard readonly
2641          * here. */
2642         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2643                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2644         ocfs2_handle_error(sb);
2645 }
2646
2647 /*
2648  * Void signal blockers, because in-kernel sigprocmask() only fails
2649  * when SIG_* is wrong.
2650  */
2651 void ocfs2_block_signals(sigset_t *oldset)
2652 {
2653         int rc;
2654         sigset_t blocked;
2655
2656         sigfillset(&blocked);
2657         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2658         BUG_ON(rc);
2659 }
2660
2661 void ocfs2_unblock_signals(sigset_t *oldset)
2662 {
2663         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2664         BUG_ON(rc);
2665 }
2666
2667 module_init(ocfs2_init);
2668 module_exit(ocfs2_exit);