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 "ver.h"
  72 #include "xattr.h"
  73 #include "quota.h"
  74 #include "refcounttree.h"
  75 #include "suballoc.h"
  76
  77 #include "buffer_head_io.h"
  78
  79 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  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 = NULL;
  90
  91 MODULE_AUTHOR("Oracle");
  92 MODULE_LICENSE("GPL");
  93
  94 struct mount_options
  95 {
  96         unsigned long   commit_interval;
  97         unsigned long   mount_opt;
  98         unsigned int    atime_quantum;
  99         signed short    slot;
 100         int             localalloc_opt;
 101         unsigned int    resv_level;
 102         int             dir_resv_level;
 103         char            cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
 104 };
 105
 106 static int ocfs2_parse_options(struct super_block *sb, char *options,
 107                                struct mount_options *mopt,
 108                                int is_remount);
 109 static int ocfs2_check_set_options(struct super_block *sb,
 110                                    struct mount_options *options);
 111 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
 112 static void ocfs2_put_super(struct super_block *sb);
 113 static int ocfs2_mount_volume(struct super_block *sb);
 114 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 115 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 116 static int ocfs2_initialize_mem_caches(void);
 117 static void ocfs2_free_mem_caches(void);
 118 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 119
 120 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 121
 122 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 123
 124 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 125 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 126 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 127 static int ocfs2_check_volume(struct ocfs2_super *osb);
 128 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 129                                struct buffer_head *bh,
 130                                u32 sectsize,
 131                                struct ocfs2_blockcheck_stats *stats);
 132 static int ocfs2_initialize_super(struct super_block *sb,
 133                                   struct buffer_head *bh,
 134                                   int sector_size,
 135                                   struct ocfs2_blockcheck_stats *stats);
 136 static int ocfs2_get_sector(struct super_block *sb,
 137                             struct buffer_head **bh,
 138                             int block,
 139                             int sect_size);
 140 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 141 static void ocfs2_destroy_inode(struct inode *inode);
 142 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
 143 static int ocfs2_enable_quotas(struct ocfs2_super *osb);
 144 static void ocfs2_disable_quotas(struct ocfs2_super *osb);
 145
 146 static const struct super_operations ocfs2_sops = {
 147         .statfs         = ocfs2_statfs,
 148         .alloc_inode    = ocfs2_alloc_inode,
 149         .destroy_inode  = ocfs2_destroy_inode,
 150         .drop_inode     = ocfs2_drop_inode,
 151         .evict_inode    = ocfs2_evict_inode,
 152         .sync_fs        = ocfs2_sync_fs,
 153         .put_super      = ocfs2_put_super,
 154         .remount_fs     = ocfs2_remount,
 155         .show_options   = ocfs2_show_options,
 156         .quota_read     = ocfs2_quota_read,
 157         .quota_write    = ocfs2_quota_write,
 158 };
 159
 160 enum {
 161         Opt_barrier,
 162         Opt_err_panic,
 163         Opt_err_ro,
 164         Opt_intr,
 165         Opt_nointr,
 166         Opt_hb_none,
 167         Opt_hb_local,
 168         Opt_hb_global,
 169         Opt_data_ordered,
 170         Opt_data_writeback,
 171         Opt_atime_quantum,
 172         Opt_slot,
 173         Opt_commit,
 174         Opt_localalloc,
 175         Opt_localflocks,
 176         Opt_stack,
 177         Opt_user_xattr,
 178         Opt_nouser_xattr,
 179         Opt_inode64,
 180         Opt_acl,
 181         Opt_noacl,
 182         Opt_usrquota,
 183         Opt_grpquota,
 184         Opt_coherency_buffered,
 185         Opt_coherency_full,
 186         Opt_resv_level,
 187         Opt_dir_resv_level,
 188         Opt_err,
 189 };
 190
 191 static const match_table_t tokens = {
 192         {Opt_barrier, "barrier=%u"},
 193         {Opt_err_panic, "errors=panic"},
 194         {Opt_err_ro, "errors=remount-ro"},
 195         {Opt_intr, "intr"},
 196         {Opt_nointr, "nointr"},
 197         {Opt_hb_none, OCFS2_HB_NONE},
 198         {Opt_hb_local, OCFS2_HB_LOCAL},
 199         {Opt_hb_global, OCFS2_HB_GLOBAL},
 200         {Opt_data_ordered, "data=ordered"},
 201         {Opt_data_writeback, "data=writeback"},
 202         {Opt_atime_quantum, "atime_quantum=%u"},
 203         {Opt_slot, "preferred_slot=%u"},
 204         {Opt_commit, "commit=%u"},
 205         {Opt_localalloc, "localalloc=%d"},
 206         {Opt_localflocks, "localflocks"},
 207         {Opt_stack, "cluster_stack=%s"},
 208         {Opt_user_xattr, "user_xattr"},
 209         {Opt_nouser_xattr, "nouser_xattr"},
 210         {Opt_inode64, "inode64"},
 211         {Opt_acl, "acl"},
 212         {Opt_noacl, "noacl"},
 213         {Opt_usrquota, "usrquota"},
 214         {Opt_grpquota, "grpquota"},
 215         {Opt_coherency_buffered, "coherency=buffered"},
 216         {Opt_coherency_full, "coherency=full"},
 217         {Opt_resv_level, "resv_level=%u"},
 218         {Opt_dir_resv_level, "dir_resv_level=%u"},
 219         {Opt_err, NULL}
 220 };
 221
 222 #ifdef CONFIG_DEBUG_FS
 223 static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
 224 {
 225         struct ocfs2_cluster_connection *cconn = osb->cconn;
 226         struct ocfs2_recovery_map *rm = osb->recovery_map;
 227         struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
 228         int i, out = 0;
 229
 230         out += snprintf(buf + out, len - out,
 231                         "%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
 232                         "Device", osb->dev_str, osb->uuid_str,
 233                         osb->fs_generation, osb->vol_label);
 234
 235         out += snprintf(buf + out, len - out,
 236                         "%10s => State: %d  Flags: 0x%lX\n", "Volume",
 237                         atomic_read(&osb->vol_state), osb->osb_flags);
 238
 239         out += snprintf(buf + out, len - out,
 240                         "%10s => Block: %lu  Cluster: %d\n", "Sizes",
 241                         osb->sb->s_blocksize, osb->s_clustersize);
 242
 243         out += snprintf(buf + out, len - out,
 244                         "%10s => Compat: 0x%X  Incompat: 0x%X  "
 245                         "ROcompat: 0x%X\n",
 246                         "Features", osb->s_feature_compat,
 247                         osb->s_feature_incompat, osb->s_feature_ro_compat);
 248
 249         out += snprintf(buf + out, len - out,
 250                         "%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
 251                         osb->s_mount_opt, osb->s_atime_quantum);
 252
 253         if (cconn) {
 254                 out += snprintf(buf + out, len - out,
 255                                 "%10s => Stack: %s  Name: %*s  "
 256                                 "Version: %d.%d\n", "Cluster",
 257                                 (*osb->osb_cluster_stack == '\0' ?
 258                                  "o2cb" : osb->osb_cluster_stack),
 259                                 cconn->cc_namelen, cconn->cc_name,
 260                                 cconn->cc_version.pv_major,
 261                                 cconn->cc_version.pv_minor);
 262         }
 263
 264         spin_lock(&osb->dc_task_lock);
 265         out += snprintf(buf + out, len - out,
 266                         "%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
 267                         "WorkSeq: %lu\n", "DownCnvt",
 268                         (osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
 269                         osb->blocked_lock_count, osb->dc_wake_sequence,
 270                         osb->dc_work_sequence);
 271         spin_unlock(&osb->dc_task_lock);
 272
 273         spin_lock(&osb->osb_lock);
 274         out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
 275                         "Recovery",
 276                         (osb->recovery_thread_task ?
 277                          task_pid_nr(osb->recovery_thread_task) : -1));
 278         if (rm->rm_used == 0)
 279                 out += snprintf(buf + out, len - out, " None\n");
 280         else {
 281                 for (i = 0; i < rm->rm_used; i++)
 282                         out += snprintf(buf + out, len - out, " %d",
 283                                         rm->rm_entries[i]);
 284                 out += snprintf(buf + out, len - out, "\n");
 285         }
 286         spin_unlock(&osb->osb_lock);
 287
 288         out += snprintf(buf + out, len - out,
 289                         "%10s => Pid: %d  Interval: %lu  Needs: %d\n", "Commit",
 290                         (osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
 291                         osb->osb_commit_interval,
 292                         atomic_read(&osb->needs_checkpoint));
 293
 294         out += snprintf(buf + out, len - out,
 295                         "%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
 296                         "Journal", osb->journal->j_state,
 297                         osb->journal->j_trans_id,
 298                         atomic_read(&osb->journal->j_num_trans));
 299
 300         out += snprintf(buf + out, len - out,
 301                         "%10s => GlobalAllocs: %d  LocalAllocs: %d  "
 302                         "SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
 303                         "Stats",
 304                         atomic_read(&osb->alloc_stats.bitmap_data),
 305                         atomic_read(&osb->alloc_stats.local_data),
 306                         atomic_read(&osb->alloc_stats.bg_allocs),
 307                         atomic_read(&osb->alloc_stats.moves),
 308                         atomic_read(&osb->alloc_stats.bg_extends));
 309
 310         out += snprintf(buf + out, len - out,
 311                         "%10s => State: %u  Descriptor: %llu  Size: %u bits  "
 312                         "Default: %u bits\n",
 313                         "LocalAlloc", osb->local_alloc_state,
 314                         (unsigned long long)osb->la_last_gd,
 315                         osb->local_alloc_bits, osb->local_alloc_default_bits);
 316
 317         spin_lock(&osb->osb_lock);
 318         out += snprintf(buf + out, len - out,
 319                         "%10s => InodeSlot: %d  StolenInodes: %d, "
 320                         "MetaSlot: %d  StolenMeta: %d\n", "Steal",
 321                         osb->s_inode_steal_slot,
 322                         atomic_read(&osb->s_num_inodes_stolen),
 323                         osb->s_meta_steal_slot,
 324                         atomic_read(&osb->s_num_meta_stolen));
 325         spin_unlock(&osb->osb_lock);
 326
 327         out += snprintf(buf + out, len - out, "OrphanScan => ");
 328         out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
 329                         os->os_count, os->os_seqno);
 330         out += snprintf(buf + out, len - out, " Last Scan: ");
 331         if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
 332                 out += snprintf(buf + out, len - out, "Disabled\n");
 333         else
 334                 out += snprintf(buf + out, len - out, "%lu seconds ago\n",
 335                                 (get_seconds() - os->os_scantime.tv_sec));
 336
 337         out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
 338                         "Slots", "Num", "RecoGen");
 339         for (i = 0; i < osb->max_slots; ++i) {
 340                 out += snprintf(buf + out, len - out,
 341                                 "%10s  %c %3d  %10d\n",
 342                                 " ",
 343                                 (i == osb->slot_num ? '*' : ' '),
 344                                 i, osb->slot_recovery_generations[i]);
 345         }
 346
 347         return out;
 348 }
 349
 350 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 351 {
 352         struct ocfs2_super *osb = inode->i_private;
 353         char *buf = NULL;
 354
 355         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 356         if (!buf)
 357                 goto bail;
 358
 359         i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
 360
 361         file->private_data = buf;
 362
 363         return 0;
 364 bail:
 365         return -ENOMEM;
 366 }
 367
 368 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 369 {
 370         kfree(file->private_data);
 371         return 0;
 372 }
 373
 374 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 375                                 size_t nbytes, loff_t *ppos)
 376 {
 377         return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
 378                                        i_size_read(file->f_mapping->host));
 379 }
 380 #else
 381 static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
 382 {
 383         return 0;
 384 }
 385 static int ocfs2_debug_release(struct inode *inode, struct file *file)
 386 {
 387         return 0;
 388 }
 389 static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
 390                                 size_t nbytes, loff_t *ppos)
 391 {
 392         return 0;
 393 }
 394 #endif  /* CONFIG_DEBUG_FS */
 395
 396 static const struct file_operations ocfs2_osb_debug_fops = {
 397         .open =         ocfs2_osb_debug_open,
 398         .release =      ocfs2_debug_release,
 399         .read =         ocfs2_debug_read,
 400         .llseek =       generic_file_llseek,
 401 };
 402
 403 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 404 {
 405         int status;
 406         tid_t target;
 407         struct ocfs2_super *osb = OCFS2_SB(sb);
 408
 409         if (ocfs2_is_hard_readonly(osb))
 410                 return -EROFS;
 411
 412         if (wait) {
 413                 status = ocfs2_flush_truncate_log(osb);
 414                 if (status < 0)
 415                         mlog_errno(status);
 416         } else {
 417                 ocfs2_schedule_truncate_log_flush(osb, 0);
 418         }
 419
 420         if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
 421                                       &target)) {
 422                 if (wait)
 423                         jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 424                                              target);
 425         }
 426         return 0;
 427 }
 428
 429 static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
 430 {
 431         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
 432             && (ino == USER_QUOTA_SYSTEM_INODE
 433                 || ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
 434                 return 0;
 435         if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
 436             && (ino == GROUP_QUOTA_SYSTEM_INODE
 437                 || ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
 438                 return 0;
 439         return 1;
 440 }
 441
 442 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 443 {
 444         struct inode *new = NULL;
 445         int status = 0;
 446         int i;
 447
 448         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 449         if (IS_ERR(new)) {
 450                 status = PTR_ERR(new);
 451                 mlog_errno(status);
 452                 goto bail;
 453         }
 454         osb->root_inode = new;
 455
 456         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 457         if (IS_ERR(new)) {
 458                 status = PTR_ERR(new);
 459                 mlog_errno(status);
 460                 goto bail;
 461         }
 462         osb->sys_root_inode = new;
 463
 464         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 465              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 466                 if (!ocfs2_need_system_inode(osb, i))
 467                         continue;
 468                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 469                 if (!new) {
 470                         ocfs2_release_system_inodes(osb);
 471                         status = -EINVAL;
 472                         mlog_errno(status);
 473                         /* FIXME: Should ERROR_RO_FS */
 474                         mlog(ML_ERROR, "Unable to load system inode %d, "
 475                              "possibly corrupt fs?", i);
 476                         goto bail;
 477                 }
 478                 // the array now has one ref, so drop this one
 479                 iput(new);
 480         }
 481
 482 bail:
 483         if (status)
 484                 mlog_errno(status);
 485         return status;
 486 }
 487
 488 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 489 {
 490         struct inode *new = NULL;
 491         int status = 0;
 492         int i;
 493
 494         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 495              i < NUM_SYSTEM_INODES;
 496              i++) {
 497                 if (!ocfs2_need_system_inode(osb, i))
 498                         continue;
 499                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 500                 if (!new) {
 501                         ocfs2_release_system_inodes(osb);
 502                         status = -EINVAL;
 503                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 504                              status, i, osb->slot_num);
 505                         goto bail;
 506                 }
 507                 /* the array now has one ref, so drop this one */
 508                 iput(new);
 509         }
 510
 511 bail:
 512         if (status)
 513                 mlog_errno(status);
 514         return status;
 515 }
 516
 517 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 518 {
 519         int i;
 520         struct inode *inode;
 521
 522         for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
 523                 inode = osb->global_system_inodes[i];
 524                 if (inode) {
 525                         iput(inode);
 526                         osb->global_system_inodes[i] = NULL;
 527                 }
 528         }
 529
 530         inode = osb->sys_root_inode;
 531         if (inode) {
 532                 iput(inode);
 533                 osb->sys_root_inode = NULL;
 534         }
 535
 536         inode = osb->root_inode;
 537         if (inode) {
 538                 iput(inode);
 539                 osb->root_inode = NULL;
 540         }
 541
 542         if (!osb->local_system_inodes)
 543                 return;
 544
 545         for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
 546                 if (osb->local_system_inodes[i]) {
 547                         iput(osb->local_system_inodes[i]);
 548                         osb->local_system_inodes[i] = NULL;
 549                 }
 550         }
 551
 552         kfree(osb->local_system_inodes);
 553         osb->local_system_inodes = NULL;
 554 }
 555
 556 /* We're allocating fs objects, use GFP_NOFS */
 557 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 558 {
 559         struct ocfs2_inode_info *oi;
 560
 561         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 562         if (!oi)
 563                 return NULL;
 564
 565         jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
 566         return &oi->vfs_inode;
 567 }
 568
 569 static void ocfs2_i_callback(struct rcu_head *head)
 570 {
 571         struct inode *inode = container_of(head, struct inode, i_rcu);
 572         INIT_LIST_HEAD(&inode->i_dentry);
 573         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 574 }
 575
 576 static void ocfs2_destroy_inode(struct inode *inode)
 577 {
 578         call_rcu(&inode->i_rcu, ocfs2_i_callback);
 579 }
 580
 581 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 582                                                 unsigned int cbits)
 583 {
 584         unsigned int bytes = 1 << cbits;
 585         unsigned int trim = bytes;
 586         unsigned int bitshift = 32;
 587
 588         /*
 589          * i_size and all block offsets in ocfs2 are always 64 bits
 590          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 591          * 64 bit platforms, cluster size will be the limiting factor.
 592          */
 593
 594 #if BITS_PER_LONG == 32
 595 # if defined(CONFIG_LBDAF)
 596         BUILD_BUG_ON(sizeof(sector_t) != 8);
 597         /*
 598          * We might be limited by page cache size.
 599          */
 600         if (bytes > PAGE_CACHE_SIZE) {
 601                 bytes = PAGE_CACHE_SIZE;
 602                 trim = 1;
 603                 /*
 604                  * Shift by 31 here so that we don't get larger than
 605                  * MAX_LFS_FILESIZE
 606                  */
 607                 bitshift = 31;
 608         }
 609 # else
 610         /*
 611          * We are limited by the size of sector_t. Use block size, as
 612          * that's what we expose to the VFS.
 613          */
 614         bytes = 1 << bbits;
 615         trim = 1;
 616         bitshift = 31;
 617 # endif
 618 #endif
 619
 620         /*
 621          * Trim by a whole cluster when we can actually approach the
 622          * on-disk limits. Otherwise we can overflow i_clusters when
 623          * an extent start is at the max offset.
 624          */
 625         return (((unsigned long long)bytes) << bitshift) - trim;
 626 }
 627
 628 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 629 {
 630         int incompat_features;
 631         int ret = 0;
 632         struct mount_options parsed_options;
 633         struct ocfs2_super *osb = OCFS2_SB(sb);
 634         u32 tmp;
 635
 636         if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
 637             !ocfs2_check_set_options(sb, &parsed_options)) {
 638                 ret = -EINVAL;
 639                 goto out;
 640         }
 641
 642         tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
 643                 OCFS2_MOUNT_HB_NONE;
 644         if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
 645                 ret = -EINVAL;
 646                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 647                 goto out;
 648         }
 649
 650         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 651             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 652                 ret = -EINVAL;
 653                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 654                 goto out;
 655         }
 656
 657         /* Probably don't want this on remount; it might
 658          * mess with other nodes */
 659         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
 660             (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
 661                 ret = -EINVAL;
 662                 mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
 663                 goto out;
 664         }
 665
 666         /* We're going to/from readonly mode. */
 667         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 668                 /* Disable quota accounting before remounting RO */
 669                 if (*flags & MS_RDONLY) {
 670                         ret = ocfs2_susp_quotas(osb, 0);
 671                         if (ret < 0)
 672                                 goto out;
 673                 }
 674                 /* Lock here so the check of HARD_RO and the potential
 675                  * setting of SOFT_RO is atomic. */
 676                 spin_lock(&osb->osb_lock);
 677                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 678                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 679                         ret = -EROFS;
 680                         goto unlock_osb;
 681                 }
 682
 683                 if (*flags & MS_RDONLY) {
 684                         sb->s_flags |= MS_RDONLY;
 685                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 686                 } else {
 687                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 688                                 mlog(ML_ERROR, "Cannot remount RDWR "
 689                                      "filesystem due to previous errors.\n");
 690                                 ret = -EROFS;
 691                                 goto unlock_osb;
 692                         }
 693                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 694                         if (incompat_features) {
 695                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 696                                      "of unsupported optional features "
 697                                      "(%x).\n", incompat_features);
 698                                 ret = -EINVAL;
 699                                 goto unlock_osb;
 700                         }
 701                         sb->s_flags &= ~MS_RDONLY;
 702                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 703                 }
 704                 trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
 705 unlock_osb:
 706                 spin_unlock(&osb->osb_lock);
 707                 /* Enable quota accounting after remounting RW */
 708                 if (!ret && !(*flags & MS_RDONLY)) {
 709                         if (sb_any_quota_suspended(sb))
 710                                 ret = ocfs2_susp_quotas(osb, 1);
 711                         else
 712                                 ret = ocfs2_enable_quotas(osb);
 713                         if (ret < 0) {
 714                                 /* Return back changes... */
 715                                 spin_lock(&osb->osb_lock);
 716                                 sb->s_flags |= MS_RDONLY;
 717                                 osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 718                                 spin_unlock(&osb->osb_lock);
 719                                 goto out;
 720                         }
 721                 }
 722         }
 723
 724         if (!ret) {
 725                 /* Only save off the new mount options in case of a successful
 726                  * remount. */
 727                 osb->s_mount_opt = parsed_options.mount_opt;
 728                 osb->s_atime_quantum = parsed_options.atime_quantum;
 729                 osb->preferred_slot = parsed_options.slot;
 730                 if (parsed_options.commit_interval)
 731                         osb->osb_commit_interval = parsed_options.commit_interval;
 732
 733                 if (!ocfs2_is_hard_readonly(osb))
 734                         ocfs2_set_journal_params(osb);
 735
 736                 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
 737                         ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
 738                                                         MS_POSIXACL : 0);
 739         }
 740 out:
 741         return ret;
 742 }
 743
 744 static int ocfs2_sb_probe(struct super_block *sb,
 745                           struct buffer_head **bh,
 746                           int *sector_size,
 747                           struct ocfs2_blockcheck_stats *stats)
 748 {
 749         int status, tmpstat;
 750         struct ocfs1_vol_disk_hdr *hdr;
 751         struct ocfs2_dinode *di;
 752         int blksize;
 753
 754         *bh = NULL;
 755
 756         /* may be > 512 */
 757         *sector_size = bdev_logical_block_size(sb->s_bdev);
 758         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 759                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 760                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 761                 status = -EINVAL;
 762                 goto bail;
 763         }
 764
 765         /* Can this really happen? */
 766         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 767                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 768
 769         /* check block zero for old format */
 770         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 771         if (status < 0) {
 772                 mlog_errno(status);
 773                 goto bail;
 774         }
 775         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 776         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 777                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 778                      hdr->major_version, hdr->minor_version);
 779                 status = -EINVAL;
 780         }
 781         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 782                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 783                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 784                      hdr->signature);
 785                 status = -EINVAL;
 786         }
 787         brelse(*bh);
 788         *bh = NULL;
 789         if (status < 0) {
 790                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 791                      "upgraded before mounting with ocfs v2\n");
 792                 goto bail;
 793         }
 794
 795         /*
 796          * Now check at magic offset for 512, 1024, 2048, 4096
 797          * blocksizes.  4096 is the maximum blocksize because it is
 798          * the minimum clustersize.
 799          */
 800         status = -EINVAL;
 801         for (blksize = *sector_size;
 802              blksize <= OCFS2_MAX_BLOCKSIZE;
 803              blksize <<= 1) {
 804                 tmpstat = ocfs2_get_sector(sb, bh,
 805                                            OCFS2_SUPER_BLOCK_BLKNO,
 806                                            blksize);
 807                 if (tmpstat < 0) {
 808                         status = tmpstat;
 809                         mlog_errno(status);
 810                         break;
 811                 }
 812                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 813                 memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
 814                 spin_lock_init(&stats->b_lock);
 815                 tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
 816                 if (tmpstat < 0) {
 817                         brelse(*bh);
 818                         *bh = NULL;
 819                 }
 820                 if (tmpstat != -EAGAIN) {
 821                         status = tmpstat;
 822                         break;
 823                 }
 824         }
 825
 826 bail:
 827         return status;
 828 }
 829
 830 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 831 {
 832         u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
 833
 834         if (osb->s_mount_opt & hb_enabled) {
 835                 if (ocfs2_mount_local(osb)) {
 836                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 837                              "mounted device.\n");
 838                         return -EINVAL;
 839                 }
 840                 if (ocfs2_userspace_stack(osb)) {
 841                         mlog(ML_ERROR, "Userspace stack expected, but "
 842                              "o2cb heartbeat arguments passed to mount\n");
 843                         return -EINVAL;
 844                 }
 845                 if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
 846                      !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
 847                     ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
 848                      ocfs2_cluster_o2cb_global_heartbeat(osb))) {
 849                         mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
 850                         return -EINVAL;
 851                 }
 852         }
 853
 854         if (!(osb->s_mount_opt & hb_enabled)) {
 855                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
 856                     !ocfs2_userspace_stack(osb)) {
 857                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 858                              "a read-write clustered device.\n");
 859                         return -EINVAL;
 860                 }
 861         }
 862
 863         return 0;
 864 }
 865
 866 /*
 867  * If we're using a userspace stack, mount should have passed
 868  * a name that matches the disk.  If not, mount should not
 869  * have passed a stack.
 870  */
 871 static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
 872                                         struct mount_options *mopt)
 873 {
 874         if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
 875                 mlog(ML_ERROR,
 876                      "cluster stack passed to mount, but this filesystem "
 877                      "does not support it\n");
 878                 return -EINVAL;
 879         }
 880
 881         if (ocfs2_userspace_stack(osb) &&
 882             strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
 883                     OCFS2_STACK_LABEL_LEN)) {
 884                 mlog(ML_ERROR,
 885                      "cluster stack passed to mount (\"%s\") does not "
 886                      "match the filesystem (\"%s\")\n",
 887                      mopt->cluster_stack,
 888                      osb->osb_cluster_stack);
 889                 return -EINVAL;
 890         }
 891
 892         return 0;
 893 }
 894
 895 static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
 896 {
 897         int type;
 898         struct super_block *sb = osb->sb;
 899         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 900                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 901         int status = 0;
 902
 903         for (type = 0; type < MAXQUOTAS; type++) {
 904                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 905                         continue;
 906                 if (unsuspend)
 907                         status = dquot_resume(sb, type);
 908                 else {
 909                         struct ocfs2_mem_dqinfo *oinfo;
 910
 911                         /* Cancel periodic syncing before suspending */
 912                         oinfo = sb_dqinfo(sb, type)->dqi_priv;
 913                         cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 914                         status = dquot_suspend(sb, type);
 915                 }
 916                 if (status < 0)
 917                         break;
 918         }
 919         if (status < 0)
 920                 mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
 921                      "remount (error = %d).\n", status);
 922         return status;
 923 }
 924
 925 static int ocfs2_enable_quotas(struct ocfs2_super *osb)
 926 {
 927         struct inode *inode[MAXQUOTAS] = { NULL, NULL };
 928         struct super_block *sb = osb->sb;
 929         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 930                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 931         unsigned int ino[MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
 932                                         LOCAL_GROUP_QUOTA_SYSTEM_INODE };
 933         int status;
 934         int type;
 935
 936         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
 937         for (type = 0; type < MAXQUOTAS; type++) {
 938                 if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 939                         continue;
 940                 inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
 941                                                         osb->slot_num);
 942                 if (!inode[type]) {
 943                         status = -ENOENT;
 944                         goto out_quota_off;
 945                 }
 946                 status = dquot_enable(inode[type], type, QFMT_OCFS2,
 947                                       DQUOT_USAGE_ENABLED);
 948                 if (status < 0)
 949                         goto out_quota_off;
 950         }
 951
 952         for (type = 0; type < MAXQUOTAS; type++)
 953                 iput(inode[type]);
 954         return 0;
 955 out_quota_off:
 956         ocfs2_disable_quotas(osb);
 957         for (type = 0; type < MAXQUOTAS; type++)
 958                 iput(inode[type]);
 959         mlog_errno(status);
 960         return status;
 961 }
 962
 963 static void ocfs2_disable_quotas(struct ocfs2_super *osb)
 964 {
 965         int type;
 966         struct inode *inode;
 967         struct super_block *sb = osb->sb;
 968         struct ocfs2_mem_dqinfo *oinfo;
 969
 970         /* We mostly ignore errors in this function because there's not much
 971          * we can do when we see them */
 972         for (type = 0; type < MAXQUOTAS; type++) {
 973                 if (!sb_has_quota_loaded(sb, type))
 974                         continue;
 975                 /* Cancel periodic syncing before we grab dqonoff_mutex */
 976                 oinfo = sb_dqinfo(sb, type)->dqi_priv;
 977                 cancel_delayed_work_sync(&oinfo->dqi_sync_work);
 978                 inode = igrab(sb->s_dquot.files[type]);
 979                 /* Turn off quotas. This will remove all dquot structures from
 980                  * memory and so they will be automatically synced to global
 981                  * quota files */
 982                 dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
 983                                         DQUOT_LIMITS_ENABLED);
 984                 if (!inode)
 985                         continue;
 986                 iput(inode);
 987         }
 988 }
 989
 990 /* Handle quota on quotactl */
 991 static int ocfs2_quota_on(struct super_block *sb, int type, int format_id)
 992 {
 993         unsigned int feature[MAXQUOTAS] = { OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
 994                                              OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
 995
 996         if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
 997                 return -EINVAL;
 998
 999         return dquot_enable(sb_dqopt(sb)->files[type], type,
1000                             format_id, DQUOT_LIMITS_ENABLED);
1001 }
1002
1003 /* Handle quota off quotactl */
1004 static int ocfs2_quota_off(struct super_block *sb, int type)
1005 {
1006         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
1007 }
1008
1009 static const struct quotactl_ops ocfs2_quotactl_ops = {
1010         .quota_on_meta  = ocfs2_quota_on,
1011         .quota_off      = ocfs2_quota_off,
1012         .quota_sync     = dquot_quota_sync,
1013         .get_info       = dquot_get_dqinfo,
1014         .set_info       = dquot_set_dqinfo,
1015         .get_dqblk      = dquot_get_dqblk,
1016         .set_dqblk      = dquot_set_dqblk,
1017 };
1018
1019 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
1020 {
1021         struct dentry *root;
1022         int status, sector_size;
1023         struct mount_options parsed_options;
1024         struct inode *inode = NULL;
1025         struct ocfs2_super *osb = NULL;
1026         struct buffer_head *bh = NULL;
1027         char nodestr[8];
1028         struct ocfs2_blockcheck_stats stats;
1029
1030         trace_ocfs2_fill_super(sb, data, silent);
1031
1032         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
1033                 status = -EINVAL;
1034                 goto read_super_error;
1035         }
1036
1037         /* probe for superblock */
1038         status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
1039         if (status < 0) {
1040                 mlog(ML_ERROR, "superblock probe failed!\n");
1041                 goto read_super_error;
1042         }
1043
1044         status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
1045         osb = OCFS2_SB(sb);
1046         if (status < 0) {
1047                 mlog_errno(status);
1048                 goto read_super_error;
1049         }
1050         brelse(bh);
1051         bh = NULL;
1052
1053         if (!ocfs2_check_set_options(sb, &parsed_options)) {
1054                 status = -EINVAL;
1055                 goto read_super_error;
1056         }
1057         osb->s_mount_opt = parsed_options.mount_opt;
1058         osb->s_atime_quantum = parsed_options.atime_quantum;
1059         osb->preferred_slot = parsed_options.slot;
1060         osb->osb_commit_interval = parsed_options.commit_interval;
1061
1062         ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
1063         osb->osb_resv_level = parsed_options.resv_level;
1064         osb->osb_dir_resv_level = parsed_options.resv_level;
1065         if (parsed_options.dir_resv_level == -1)
1066                 osb->osb_dir_resv_level = parsed_options.resv_level;
1067         else
1068                 osb->osb_dir_resv_level = parsed_options.dir_resv_level;
1069
1070         status = ocfs2_verify_userspace_stack(osb, &parsed_options);
1071         if (status)
1072                 goto read_super_error;
1073
1074         sb->s_magic = OCFS2_SUPER_MAGIC;
1075
1076         sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
1077                 ((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1078
1079         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
1080          * heartbeat=none */
1081         if (bdev_read_only(sb->s_bdev)) {
1082                 if (!(sb->s_flags & MS_RDONLY)) {
1083                         status = -EACCES;
1084                         mlog(ML_ERROR, "Readonly device detected but readonly "
1085                              "mount was not specified.\n");
1086                         goto read_super_error;
1087                 }
1088
1089                 /* You should not be able to start a local heartbeat
1090                  * on a readonly device. */
1091                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
1092                         status = -EROFS;
1093                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
1094                              "device.\n");
1095                         goto read_super_error;
1096                 }
1097
1098                 status = ocfs2_check_journals_nolocks(osb);
1099                 if (status < 0) {
1100                         if (status == -EROFS)
1101                                 mlog(ML_ERROR, "Recovery required on readonly "
1102                                      "file system, but write access is "
1103                                      "unavailable.\n");
1104                         else
1105                                 mlog_errno(status);
1106                         goto read_super_error;
1107                 }
1108
1109                 ocfs2_set_ro_flag(osb, 1);
1110
1111                 printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
1112                        "Cluster services will not be used for this mount. "
1113                        "Recovery will be skipped.\n", osb->dev_str);
1114         }
1115
1116         if (!ocfs2_is_hard_readonly(osb)) {
1117                 if (sb->s_flags & MS_RDONLY)
1118                         ocfs2_set_ro_flag(osb, 0);
1119         }
1120
1121         status = ocfs2_verify_heartbeat(osb);
1122         if (status < 0) {
1123                 mlog_errno(status);
1124                 goto read_super_error;
1125         }
1126
1127         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
1128                                                  ocfs2_debugfs_root);
1129         if (!osb->osb_debug_root) {
1130                 status = -EINVAL;
1131                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
1132                 goto read_super_error;
1133         }
1134
1135         osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
1136                                             osb->osb_debug_root,
1137                                             osb,
1138                                             &ocfs2_osb_debug_fops);
1139         if (!osb->osb_ctxt) {
1140                 status = -EINVAL;
1141                 mlog_errno(status);
1142                 goto read_super_error;
1143         }
1144
1145         if (ocfs2_meta_ecc(osb)) {
1146                 status = ocfs2_blockcheck_stats_debugfs_install(
1147                                                 &osb->osb_ecc_stats,
1148                                                 osb->osb_debug_root);
1149                 if (status) {
1150                         mlog(ML_ERROR,
1151                              "Unable to create blockcheck statistics "
1152                              "files\n");
1153                         goto read_super_error;
1154                 }
1155         }
1156
1157         status = ocfs2_mount_volume(sb);
1158         if (osb->root_inode)
1159                 inode = igrab(osb->root_inode);
1160
1161         if (status < 0)
1162                 goto read_super_error;
1163
1164         if (!inode) {
1165                 status = -EIO;
1166                 mlog_errno(status);
1167                 goto read_super_error;
1168         }
1169
1170         root = d_alloc_root(inode);
1171         if (!root) {
1172                 status = -ENOMEM;
1173                 mlog_errno(status);
1174                 goto read_super_error;
1175         }
1176
1177         sb->s_root = root;
1178
1179         ocfs2_complete_mount_recovery(osb);
1180
1181         if (ocfs2_mount_local(osb))
1182                 snprintf(nodestr, sizeof(nodestr), "local");
1183         else
1184                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1185
1186         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
1187                "with %s data mode.\n",
1188                osb->dev_str, nodestr, osb->slot_num,
1189                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
1190                "ordered");
1191
1192         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
1193         wake_up(&osb->osb_mount_event);
1194
1195         /* Now we can initialize quotas because we can afford to wait
1196          * for cluster locks recovery now. That also means that truncation
1197          * log recovery can happen but that waits for proper quota setup */
1198         if (!(sb->s_flags & MS_RDONLY)) {
1199                 status = ocfs2_enable_quotas(osb);
1200                 if (status < 0) {
1201                         /* We have to err-out specially here because
1202                          * s_root is already set */
1203                         mlog_errno(status);
1204                         atomic_set(&osb->vol_state, VOLUME_DISABLED);
1205                         wake_up(&osb->osb_mount_event);
1206                         return status;
1207                 }
1208         }
1209
1210         ocfs2_complete_quota_recovery(osb);
1211
1212         /* Now we wake up again for processes waiting for quotas */
1213         atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
1214         wake_up(&osb->osb_mount_event);
1215
1216         /* Start this when the mount is almost sure of being successful */
1217         ocfs2_orphan_scan_start(osb);
1218
1219         return status;
1220
1221 read_super_error:
1222         brelse(bh);
1223
1224         if (inode)
1225                 iput(inode);
1226
1227         if (osb) {
1228                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
1229                 wake_up(&osb->osb_mount_event);
1230                 ocfs2_dismount_volume(sb, 1);
1231         }
1232
1233         if (status)
1234                 mlog_errno(status);
1235         return status;
1236 }
1237
1238 static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
1239                         int flags,
1240                         const char *dev_name,
1241                         void *data)
1242 {
1243         return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
1244 }
1245
1246 static void ocfs2_kill_sb(struct super_block *sb)
1247 {
1248         struct ocfs2_super *osb = OCFS2_SB(sb);
1249
1250         /* Failed mount? */
1251         if (!osb || atomic_read(&osb->vol_state) == VOLUME_DISABLED)
1252                 goto out;
1253
1254         /* Prevent further queueing of inode drop events */
1255         spin_lock(&dentry_list_lock);
1256         ocfs2_set_osb_flag(osb, OCFS2_OSB_DROP_DENTRY_LOCK_IMMED);
1257         spin_unlock(&dentry_list_lock);
1258         /* Wait for work to finish and/or remove it */
1259         cancel_work_sync(&osb->dentry_lock_work);
1260 out:
1261         kill_block_super(sb);
1262 }
1263
1264 static struct file_system_type ocfs2_fs_type = {
1265         .owner          = THIS_MODULE,
1266         .name           = "ocfs2",
1267         .mount          = ocfs2_mount,
1268         .kill_sb        = ocfs2_kill_sb,
1269
1270         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
1271         .next           = NULL
1272 };
1273
1274 static int ocfs2_check_set_options(struct super_block *sb,
1275                                    struct mount_options *options)
1276 {
1277         if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
1278             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1279                                          OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1280                 mlog(ML_ERROR, "User quotas were requested, but this "
1281                      "filesystem does not have the feature enabled.\n");
1282                 return 0;
1283         }
1284         if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
1285             !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1286                                          OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1287                 mlog(ML_ERROR, "Group quotas were requested, but this "
1288                      "filesystem does not have the feature enabled.\n");
1289                 return 0;
1290         }
1291         if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
1292             !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
1293                 mlog(ML_ERROR, "ACL support requested but extended attributes "
1294                      "feature is not enabled\n");
1295                 return 0;
1296         }
1297         /* No ACL setting specified? Use XATTR feature... */
1298         if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
1299                                     OCFS2_MOUNT_NO_POSIX_ACL))) {
1300                 if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
1301                         options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1302                 else
1303                         options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1304         }
1305         return 1;
1306 }
1307
1308 static int ocfs2_parse_options(struct super_block *sb,
1309                                char *options,
1310                                struct mount_options *mopt,
1311                                int is_remount)
1312 {
1313         int status, user_stack = 0;
1314         char *p;
1315         u32 tmp;
1316
1317         trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
1318
1319         mopt->commit_interval = 0;
1320         mopt->mount_opt = OCFS2_MOUNT_NOINTR;
1321         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1322         mopt->slot = OCFS2_INVALID_SLOT;
1323         mopt->localalloc_opt = -1;
1324         mopt->cluster_stack[0] = '\0';
1325         mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
1326         mopt->dir_resv_level = -1;
1327
1328         if (!options) {
1329                 status = 1;
1330                 goto bail;
1331         }
1332
1333         while ((p = strsep(&options, ",")) != NULL) {
1334                 int token, option;
1335                 substring_t args[MAX_OPT_ARGS];
1336
1337                 if (!*p)
1338                         continue;
1339
1340                 token = match_token(p, tokens, args);
1341                 switch (token) {
1342                 case Opt_hb_local:
1343                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
1344                         break;
1345                 case Opt_hb_none:
1346                         mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
1347                         break;
1348                 case Opt_hb_global:
1349                         mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
1350                         break;
1351                 case Opt_barrier:
1352                         if (match_int(&args[0], &option)) {
1353                                 status = 0;
1354                                 goto bail;
1355                         }
1356                         if (option)
1357                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
1358                         else
1359                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
1360                         break;
1361                 case Opt_intr:
1362                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
1363                         break;
1364                 case Opt_nointr:
1365                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
1366                         break;
1367                 case Opt_err_panic:
1368                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1369                         break;
1370                 case Opt_err_ro:
1371                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
1372                         break;
1373                 case Opt_data_ordered:
1374                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
1375                         break;
1376                 case Opt_data_writeback:
1377                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
1378                         break;
1379                 case Opt_user_xattr:
1380                         mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
1381                         break;
1382                 case Opt_nouser_xattr:
1383                         mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
1384                         break;
1385                 case Opt_atime_quantum:
1386                         if (match_int(&args[0], &option)) {
1387                                 status = 0;
1388                                 goto bail;
1389                         }
1390                         if (option >= 0)
1391                                 mopt->atime_quantum = option;
1392                         break;
1393                 case Opt_slot:
1394                         option = 0;
1395                         if (match_int(&args[0], &option)) {
1396                                 status = 0;
1397                                 goto bail;
1398                         }
1399                         if (option)
1400                                 mopt->slot = (s16)option;
1401                         break;
1402                 case Opt_commit:
1403                         option = 0;
1404                         if (match_int(&args[0], &option)) {
1405                                 status = 0;
1406                                 goto bail;
1407                         }
1408                         if (option < 0)
1409                                 return 0;
1410                         if (option == 0)
1411                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1412                         mopt->commit_interval = HZ * option;
1413                         break;
1414                 case Opt_localalloc:
1415                         option = 0;
1416                         if (match_int(&args[0], &option)) {
1417                                 status = 0;
1418                                 goto bail;
1419                         }
1420                         if (option >= 0)
1421                                 mopt->localalloc_opt = option;
1422                         break;
1423                 case Opt_localflocks:
1424                         /*
1425                          * Changing this during remount could race
1426                          * flock() requests, or "unbalance" existing
1427                          * ones (e.g., a lock is taken in one mode but
1428                          * dropped in the other). If users care enough
1429                          * to flip locking modes during remount, we
1430                          * could add a "local" flag to individual
1431                          * flock structures for proper tracking of
1432                          * state.
1433                          */
1434                         if (!is_remount)
1435                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
1436                         break;
1437                 case Opt_stack:
1438                         /* Check both that the option we were passed
1439                          * is of the right length and that it is a proper
1440                          * string of the right length.
1441                          */
1442                         if (((args[0].to - args[0].from) !=
1443                              OCFS2_STACK_LABEL_LEN) ||
1444                             (strnlen(args[0].from,
1445                                      OCFS2_STACK_LABEL_LEN) !=
1446                              OCFS2_STACK_LABEL_LEN)) {
1447                                 mlog(ML_ERROR,
1448                                      "Invalid cluster_stack option\n");
1449                                 status = 0;
1450                                 goto bail;
1451                         }
1452                         memcpy(mopt->cluster_stack, args[0].from,
1453                                OCFS2_STACK_LABEL_LEN);
1454                         mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
1455                         /*
1456                          * Open code the memcmp here as we don't have
1457                          * an osb to pass to
1458                          * ocfs2_userspace_stack().
1459                          */
1460                         if (memcmp(mopt->cluster_stack,
1461                                    OCFS2_CLASSIC_CLUSTER_STACK,
1462                                    OCFS2_STACK_LABEL_LEN))
1463                                 user_stack = 1;
1464                         break;
1465                 case Opt_inode64:
1466                         mopt->mount_opt |= OCFS2_MOUNT_INODE64;
1467                         break;
1468                 case Opt_usrquota:
1469                         mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
1470                         break;
1471                 case Opt_grpquota:
1472                         mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
1473                         break;
1474                 case Opt_coherency_buffered:
1475                         mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
1476                         break;
1477                 case Opt_coherency_full:
1478                         mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
1479                         break;
1480                 case Opt_acl:
1481                         mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
1482                         mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
1483                         break;
1484                 case Opt_noacl:
1485                         mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
1486                         mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
1487                         break;
1488                 case Opt_resv_level:
1489                         if (is_remount)
1490                                 break;
1491                         if (match_int(&args[0], &option)) {
1492                                 status = 0;
1493                                 goto bail;
1494                         }
1495                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1496                             option < OCFS2_MAX_RESV_LEVEL)
1497                                 mopt->resv_level = option;
1498                         break;
1499                 case Opt_dir_resv_level:
1500                         if (is_remount)
1501                                 break;
1502                         if (match_int(&args[0], &option)) {
1503                                 status = 0;
1504                                 goto bail;
1505                         }
1506                         if (option >= OCFS2_MIN_RESV_LEVEL &&
1507                             option < OCFS2_MAX_RESV_LEVEL)
1508                                 mopt->dir_resv_level = option;
1509                         break;
1510                 default:
1511                         mlog(ML_ERROR,
1512                              "Unrecognized mount option \"%s\" "
1513                              "or missing value\n", p);
1514                         status = 0;
1515                         goto bail;
1516                 }
1517         }
1518
1519         if (user_stack == 0) {
1520                 /* Ensure only one heartbeat mode */
1521                 tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
1522                                          OCFS2_MOUNT_HB_GLOBAL |
1523                                          OCFS2_MOUNT_HB_NONE);
1524                 if (hweight32(tmp) != 1) {
1525                         mlog(ML_ERROR, "Invalid heartbeat mount options\n");
1526                         status = 0;
1527                         goto bail;
1528                 }
1529         }
1530
1531         status = 1;
1532
1533 bail:
1534         return status;
1535 }
1536
1537 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
1538 {
1539         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
1540         unsigned long opts = osb->s_mount_opt;
1541         unsigned int local_alloc_megs;
1542
1543         if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
1544                 seq_printf(s, ",_netdev");
1545                 if (opts & OCFS2_MOUNT_HB_LOCAL)
1546                         seq_printf(s, ",%s", OCFS2_HB_LOCAL);
1547                 else
1548                         seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
1549         } else
1550                 seq_printf(s, ",%s", OCFS2_HB_NONE);
1551
1552         if (opts & OCFS2_MOUNT_NOINTR)
1553                 seq_printf(s, ",nointr");
1554
1555         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
1556                 seq_printf(s, ",data=writeback");
1557         else
1558                 seq_printf(s, ",data=ordered");
1559
1560         if (opts & OCFS2_MOUNT_BARRIER)
1561                 seq_printf(s, ",barrier=1");
1562
1563         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
1564                 seq_printf(s, ",errors=panic");
1565         else
1566                 seq_printf(s, ",errors=remount-ro");
1567
1568         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
1569                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
1570
1571         if (!(mnt->mnt_flags & MNT_NOATIME) && !(mnt->mnt_flags & MNT_RELATIME))
1572                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
1573
1574         if (osb->osb_commit_interval)
1575                 seq_printf(s, ",commit=%u",
1576                            (unsigned) (osb->osb_commit_interval / HZ));
1577
1578         local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
1579         if (local_alloc_megs != ocfs2_la_default_mb(osb))
1580                 seq_printf(s, ",localalloc=%d", local_alloc_megs);
1581
1582         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
1583                 seq_printf(s, ",localflocks,");
1584
1585         if (osb->osb_cluster_stack[0])
1586                 seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
1587                            osb->osb_cluster_stack);
1588         if (opts & OCFS2_MOUNT_USRQUOTA)
1589                 seq_printf(s, ",usrquota");
1590         if (opts & OCFS2_MOUNT_GRPQUOTA)
1591                 seq_printf(s, ",grpquota");
1592
1593         if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
1594                 seq_printf(s, ",coherency=buffered");
1595         else
1596                 seq_printf(s, ",coherency=full");
1597
1598         if (opts & OCFS2_MOUNT_NOUSERXATTR)
1599                 seq_printf(s, ",nouser_xattr");
1600         else
1601                 seq_printf(s, ",user_xattr");
1602
1603         if (opts & OCFS2_MOUNT_INODE64)
1604                 seq_printf(s, ",inode64");
1605
1606         if (opts & OCFS2_MOUNT_POSIX_ACL)
1607                 seq_printf(s, ",acl");
1608         else
1609                 seq_printf(s, ",noacl");
1610
1611         if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
1612                 seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
1613
1614         if (osb->osb_dir_resv_level != osb->osb_resv_level)
1615                 seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
1616
1617         return 0;
1618 }
1619
1620 wait_queue_head_t ocfs2__ioend_wq[OCFS2_IOEND_WQ_HASH_SZ];
1621
1622 static int __init ocfs2_init(void)
1623 {
1624         int status, i;
1625
1626         ocfs2_print_version();
1627
1628         for (i = 0; i < OCFS2_IOEND_WQ_HASH_SZ; i++)
1629                 init_waitqueue_head(&ocfs2__ioend_wq[i]);
1630
1631         status = init_ocfs2_uptodate_cache();
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto leave;
1635         }
1636
1637         status = ocfs2_initialize_mem_caches();
1638         if (status < 0) {
1639                 mlog_errno(status);
1640                 goto leave;
1641         }
1642
1643         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
1644         if (!ocfs2_wq) {
1645                 status = -ENOMEM;
1646                 goto leave;
1647         }
1648
1649         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
1650         if (!ocfs2_debugfs_root) {
1651                 status = -EFAULT;
1652                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
1653         }
1654
1655         ocfs2_set_locking_protocol();
1656
1657         status = register_quota_format(&ocfs2_quota_format);
1658 leave:
1659         if (status < 0) {
1660                 ocfs2_free_mem_caches();
1661                 exit_ocfs2_uptodate_cache();
1662                 mlog_errno(status);
1663         }
1664
1665         if (status >= 0) {
1666                 return register_filesystem(&ocfs2_fs_type);
1667         } else
1668                 return -1;
1669 }
1670
1671 static void __exit ocfs2_exit(void)
1672 {
1673         if (ocfs2_wq) {
1674                 flush_workqueue(ocfs2_wq);
1675                 destroy_workqueue(ocfs2_wq);
1676         }
1677
1678         unregister_quota_format(&ocfs2_quota_format);
1679
1680         debugfs_remove(ocfs2_debugfs_root);
1681
1682         ocfs2_free_mem_caches();
1683
1684         unregister_filesystem(&ocfs2_fs_type);
1685
1686         exit_ocfs2_uptodate_cache();
1687 }
1688
1689 static void ocfs2_put_super(struct super_block *sb)
1690 {
1691         trace_ocfs2_put_super(sb);
1692
1693         ocfs2_sync_blockdev(sb);
1694         ocfs2_dismount_volume(sb, 0);
1695 }
1696
1697 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1698 {
1699         struct ocfs2_super *osb;
1700         u32 numbits, freebits;
1701         int status;
1702         struct ocfs2_dinode *bm_lock;
1703         struct buffer_head *bh = NULL;
1704         struct inode *inode = NULL;
1705
1706         trace_ocfs2_statfs(dentry->d_sb, buf);
1707
1708         osb = OCFS2_SB(dentry->d_sb);
1709
1710         inode = ocfs2_get_system_file_inode(osb,
1711                                             GLOBAL_BITMAP_SYSTEM_INODE,
1712                                             OCFS2_INVALID_SLOT);
1713         if (!inode) {
1714                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1715                 status = -EIO;
1716                 goto bail;
1717         }
1718
1719         status = ocfs2_inode_lock(inode, &bh, 0);
1720         if (status < 0) {
1721                 mlog_errno(status);
1722                 goto bail;
1723         }
1724
1725         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1726
1727         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1728         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1729
1730         buf->f_type = OCFS2_SUPER_MAGIC;
1731         buf->f_bsize = dentry->d_sb->s_blocksize;
1732         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1733         buf->f_blocks = ((sector_t) numbits) *
1734                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1735         buf->f_bfree = ((sector_t) freebits) *
1736                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1737         buf->f_bavail = buf->f_bfree;
1738         buf->f_files = numbits;
1739         buf->f_ffree = freebits;
1740         buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
1741                                 & 0xFFFFFFFFUL;
1742         buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
1743                                 OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
1744
1745         brelse(bh);
1746
1747         ocfs2_inode_unlock(inode, 0);
1748         status = 0;
1749 bail:
1750         if (inode)
1751                 iput(inode);
1752
1753         if (status)
1754                 mlog_errno(status);
1755
1756         return status;
1757 }
1758
1759 static void ocfs2_inode_init_once(void *data)
1760 {
1761         struct ocfs2_inode_info *oi = data;
1762
1763         oi->ip_flags = 0;
1764         oi->ip_open_count = 0;
1765         spin_lock_init(&oi->ip_lock);
1766         ocfs2_extent_map_init(&oi->vfs_inode);
1767         INIT_LIST_HEAD(&oi->ip_io_markers);
1768         oi->ip_dir_start_lookup = 0;
1769         atomic_set(&oi->ip_unaligned_aio, 0);
1770         init_rwsem(&oi->ip_alloc_sem);
1771         init_rwsem(&oi->ip_xattr_sem);
1772         mutex_init(&oi->ip_io_mutex);
1773
1774         oi->ip_blkno = 0ULL;
1775         oi->ip_clusters = 0;
1776
1777         ocfs2_resv_init_once(&oi->ip_la_data_resv);
1778
1779         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1780         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1781         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1782
1783         ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
1784                                   &ocfs2_inode_caching_ops);
1785
1786         inode_init_once(&oi->vfs_inode);
1787 }
1788
1789 static int ocfs2_initialize_mem_caches(void)
1790 {
1791         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1792                                        sizeof(struct ocfs2_inode_info),
1793                                        0,
1794                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1795                                                 SLAB_MEM_SPREAD),
1796                                        ocfs2_inode_init_once);
1797         ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
1798                                         sizeof(struct ocfs2_dquot),
1799                                         0,
1800                                         (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1801                                                 SLAB_MEM_SPREAD),
1802                                         NULL);
1803         ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
1804                                         sizeof(struct ocfs2_quota_chunk),
1805                                         0,
1806                                         (SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
1807                                         NULL);
1808         if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
1809             !ocfs2_qf_chunk_cachep) {
1810                 if (ocfs2_inode_cachep)
1811                         kmem_cache_destroy(ocfs2_inode_cachep);
1812                 if (ocfs2_dquot_cachep)
1813                         kmem_cache_destroy(ocfs2_dquot_cachep);
1814                 if (ocfs2_qf_chunk_cachep)
1815                         kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1816                 return -ENOMEM;
1817         }
1818
1819         return 0;
1820 }
1821
1822 static void ocfs2_free_mem_caches(void)
1823 {
1824         if (ocfs2_inode_cachep)
1825                 kmem_cache_destroy(ocfs2_inode_cachep);
1826         ocfs2_inode_cachep = NULL;
1827
1828         if (ocfs2_dquot_cachep)
1829                 kmem_cache_destroy(ocfs2_dquot_cachep);
1830         ocfs2_dquot_cachep = NULL;
1831
1832         if (ocfs2_qf_chunk_cachep)
1833                 kmem_cache_destroy(ocfs2_qf_chunk_cachep);
1834         ocfs2_qf_chunk_cachep = NULL;
1835 }
1836
1837 static int ocfs2_get_sector(struct super_block *sb,
1838                             struct buffer_head **bh,
1839                             int block,
1840                             int sect_size)
1841 {
1842         if (!sb_set_blocksize(sb, sect_size)) {
1843                 mlog(ML_ERROR, "unable to set blocksize\n");
1844                 return -EIO;
1845         }
1846
1847         *bh = sb_getblk(sb, block);
1848         if (!*bh) {
1849                 mlog_errno(-EIO);
1850                 return -EIO;
1851         }
1852         lock_buffer(*bh);
1853         if (!buffer_dirty(*bh))
1854                 clear_buffer_uptodate(*bh);
1855         unlock_buffer(*bh);
1856         ll_rw_block(READ, 1, bh);
1857         wait_on_buffer(*bh);
1858         if (!buffer_uptodate(*bh)) {
1859                 mlog_errno(-EIO);
1860                 brelse(*bh);
1861                 *bh = NULL;
1862                 return -EIO;
1863         }
1864
1865         return 0;
1866 }
1867
1868 static int ocfs2_mount_volume(struct super_block *sb)
1869 {
1870         int status = 0;
1871         int unlock_super = 0;
1872         struct ocfs2_super *osb = OCFS2_SB(sb);
1873
1874         if (ocfs2_is_hard_readonly(osb))
1875                 goto leave;
1876
1877         status = ocfs2_dlm_init(osb);
1878         if (status < 0) {
1879                 mlog_errno(status);
1880                 goto leave;
1881         }
1882
1883         status = ocfs2_super_lock(osb, 1);
1884         if (status < 0) {
1885                 mlog_errno(status);
1886                 goto leave;
1887         }
1888         unlock_super = 1;
1889
1890         /* This will load up the node map and add ourselves to it. */
1891         status = ocfs2_find_slot(osb);
1892         if (status < 0) {
1893                 mlog_errno(status);
1894                 goto leave;
1895         }
1896
1897         /* load all node-local system inodes */
1898         status = ocfs2_init_local_system_inodes(osb);
1899         if (status < 0) {
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         status = ocfs2_check_volume(osb);
1905         if (status < 0) {
1906                 mlog_errno(status);
1907                 goto leave;
1908         }
1909
1910         status = ocfs2_truncate_log_init(osb);
1911         if (status < 0)
1912                 mlog_errno(status);
1913
1914 leave:
1915         if (unlock_super)
1916                 ocfs2_super_unlock(osb, 1);
1917
1918         return status;
1919 }
1920
1921 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1922 {
1923         int tmp, hangup_needed = 0;
1924         struct ocfs2_super *osb = NULL;
1925         char nodestr[8];
1926
1927         trace_ocfs2_dismount_volume(sb);
1928
1929         BUG_ON(!sb);
1930         osb = OCFS2_SB(sb);
1931         BUG_ON(!osb);
1932
1933         debugfs_remove(osb->osb_ctxt);
1934
1935         /*
1936          * Flush inode dropping work queue so that deletes are
1937          * performed while the filesystem is still working
1938          */
1939         ocfs2_drop_all_dl_inodes(osb);
1940
1941         /* Orphan scan should be stopped as early as possible */
1942         ocfs2_orphan_scan_stop(osb);
1943
1944         ocfs2_disable_quotas(osb);
1945
1946         ocfs2_shutdown_local_alloc(osb);
1947
1948         ocfs2_truncate_log_shutdown(osb);
1949
1950         /* This will disable recovery and flush any recovery work. */
1951         ocfs2_recovery_exit(osb);
1952
1953         ocfs2_journal_shutdown(osb);
1954
1955         ocfs2_sync_blockdev(sb);
1956
1957         ocfs2_purge_refcount_trees(osb);
1958
1959         /* No cluster connection means we've failed during mount, so skip
1960          * all the steps which depended on that to complete. */
1961         if (osb->cconn) {
1962                 tmp = ocfs2_super_lock(osb, 1);
1963                 if (tmp < 0) {
1964                         mlog_errno(tmp);
1965                         return;
1966                 }
1967         }
1968
1969         if (osb->slot_num != OCFS2_INVALID_SLOT)
1970                 ocfs2_put_slot(osb);
1971
1972         if (osb->cconn)
1973                 ocfs2_super_unlock(osb, 1);
1974
1975         ocfs2_release_system_inodes(osb);
1976
1977         /*
1978          * If we're dismounting due to mount error, mount.ocfs2 will clean
1979          * up heartbeat.  If we're a local mount, there is no heartbeat.
1980          * If we failed before we got a uuid_str yet, we can't stop
1981          * heartbeat.  Otherwise, do it.
1982          */
1983         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
1984             !ocfs2_is_hard_readonly(osb))
1985                 hangup_needed = 1;
1986
1987         if (osb->cconn)
1988                 ocfs2_dlm_shutdown(osb, hangup_needed);
1989
1990         ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
1991         debugfs_remove(osb->osb_debug_root);
1992
1993         if (hangup_needed)
1994                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1995
1996         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1997
1998         if (ocfs2_mount_local(osb))
1999                 snprintf(nodestr, sizeof(nodestr), "local");
2000         else
2001                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
2002
2003         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
2004                osb->dev_str, nodestr);
2005
2006         ocfs2_delete_osb(osb);
2007         kfree(osb);
2008         sb->s_dev = 0;
2009         sb->s_fs_info = NULL;
2010 }
2011
2012 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
2013                                 unsigned uuid_bytes)
2014 {
2015         int i, ret;
2016         char *ptr;
2017
2018         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
2019
2020         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
2021         if (osb->uuid_str == NULL)
2022                 return -ENOMEM;
2023
2024         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
2025                 /* print with null */
2026                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
2027                 if (ret != 2) /* drop super cleans up */
2028                         return -EINVAL;
2029                 /* then only advance past the last char */
2030                 ptr += 2;
2031         }
2032
2033         return 0;
2034 }
2035
2036 /* Make sure entire volume is addressable by our journal.  Requires
2037    osb_clusters_at_boot to be valid and for the journal to have been
2038    initialized by ocfs2_journal_init(). */
2039 static int ocfs2_journal_addressable(struct ocfs2_super *osb)
2040 {
2041         int status = 0;
2042         u64 max_block =
2043                 ocfs2_clusters_to_blocks(osb->sb,
2044                                          osb->osb_clusters_at_boot) - 1;
2045
2046         /* 32-bit block number is always OK. */
2047         if (max_block <= (u32)~0ULL)
2048                 goto out;
2049
2050         /* Volume is "huge", so see if our journal is new enough to
2051            support it. */
2052         if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
2053                                        OCFS2_FEATURE_COMPAT_JBD2_SB) &&
2054               jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
2055                                                JBD2_FEATURE_INCOMPAT_64BIT))) {
2056                 mlog(ML_ERROR, "The journal cannot address the entire volume. "
2057                      "Enable the 'block64' journal option with tunefs.ocfs2");
2058                 status = -EFBIG;
2059                 goto out;
2060         }
2061
2062  out:
2063         return status;
2064 }
2065
2066 static int ocfs2_initialize_super(struct super_block *sb,
2067                                   struct buffer_head *bh,
2068                                   int sector_size,
2069                                   struct ocfs2_blockcheck_stats *stats)
2070 {
2071         int status;
2072         int i, cbits, bbits;
2073         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2074         struct inode *inode = NULL;
2075         struct ocfs2_journal *journal;
2076         __le32 uuid_net_key;
2077         struct ocfs2_super *osb;
2078         u64 total_blocks;
2079
2080         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
2081         if (!osb) {
2082                 status = -ENOMEM;
2083                 mlog_errno(status);
2084                 goto bail;
2085         }
2086
2087         sb->s_fs_info = osb;
2088         sb->s_op = &ocfs2_sops;
2089         sb->s_d_op = &ocfs2_dentry_ops;
2090         sb->s_export_op = &ocfs2_export_ops;
2091         sb->s_qcop = &ocfs2_quotactl_ops;
2092         sb->dq_op = &ocfs2_quota_operations;
2093         sb->s_xattr = ocfs2_xattr_handlers;
2094         sb->s_time_gran = 1;
2095         sb->s_flags |= MS_NOATIME;
2096         /* this is needed to support O_LARGEFILE */
2097         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2098         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
2099         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
2100
2101         osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
2102
2103         for (i = 0; i < 3; i++)
2104                 osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
2105         osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2106
2107         osb->sb = sb;
2108         /* Save off for ocfs2_rw_direct */
2109         osb->s_sectsize_bits = blksize_bits(sector_size);
2110         BUG_ON(!osb->s_sectsize_bits);
2111
2112         spin_lock_init(&osb->dc_task_lock);
2113         init_waitqueue_head(&osb->dc_event);
2114         osb->dc_work_sequence = 0;
2115         osb->dc_wake_sequence = 0;
2116         INIT_LIST_HEAD(&osb->blocked_lock_list);
2117         osb->blocked_lock_count = 0;
2118         spin_lock_init(&osb->osb_lock);
2119         spin_lock_init(&osb->osb_xattr_lock);
2120         ocfs2_init_steal_slots(osb);
2121
2122         atomic_set(&osb->alloc_stats.moves, 0);
2123         atomic_set(&osb->alloc_stats.local_data, 0);
2124         atomic_set(&osb->alloc_stats.bitmap_data, 0);
2125         atomic_set(&osb->alloc_stats.bg_allocs, 0);
2126         atomic_set(&osb->alloc_stats.bg_extends, 0);
2127
2128         /* Copy the blockcheck stats from the superblock probe */
2129         osb->osb_ecc_stats = *stats;
2130
2131         ocfs2_init_node_maps(osb);
2132
2133         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
2134                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
2135
2136         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
2137         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
2138                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
2139                      osb->max_slots);
2140                 status = -EINVAL;
2141                 goto bail;
2142         }
2143
2144         ocfs2_orphan_scan_init(osb);
2145
2146         status = ocfs2_recovery_init(osb);
2147         if (status) {
2148                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
2149                 mlog_errno(status);
2150                 goto bail;
2151         }
2152
2153         init_waitqueue_head(&osb->checkpoint_event);
2154         atomic_set(&osb->needs_checkpoint, 0);
2155
2156         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
2157
2158         osb->slot_num = OCFS2_INVALID_SLOT;
2159
2160         osb->s_xattr_inline_size = le16_to_cpu(
2161                                         di->id2.i_super.s_xattr_inline_size);
2162
2163         osb->local_alloc_state = OCFS2_LA_UNUSED;
2164         osb->local_alloc_bh = NULL;
2165         INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
2166
2167         init_waitqueue_head(&osb->osb_mount_event);
2168
2169         status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
2170         if (status) {
2171                 mlog_errno(status);
2172                 goto bail;
2173         }
2174
2175         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
2176         if (!osb->vol_label) {
2177                 mlog(ML_ERROR, "unable to alloc vol label\n");
2178                 status = -ENOMEM;
2179                 goto bail;
2180         }
2181
2182         osb->slot_recovery_generations =
2183                 kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
2184                         GFP_KERNEL);
2185         if (!osb->slot_recovery_generations) {
2186                 status = -ENOMEM;
2187                 mlog_errno(status);
2188                 goto bail;
2189         }
2190
2191         init_waitqueue_head(&osb->osb_wipe_event);
2192         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
2193                                         sizeof(*osb->osb_orphan_wipes),
2194                                         GFP_KERNEL);
2195         if (!osb->osb_orphan_wipes) {
2196                 status = -ENOMEM;
2197                 mlog_errno(status);
2198                 goto bail;
2199         }
2200
2201         osb->osb_rf_lock_tree = RB_ROOT;
2202
2203         osb->s_feature_compat =
2204                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
2205         osb->s_feature_ro_compat =
2206                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
2207         osb->s_feature_incompat =
2208                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
2209
2210         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
2211                 mlog(ML_ERROR, "couldn't mount because of unsupported "
2212                      "optional features (%x).\n", i);
2213                 status = -EINVAL;
2214                 goto bail;
2215         }
2216         if (!(osb->sb->s_flags & MS_RDONLY) &&
2217             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
2218                 mlog(ML_ERROR, "couldn't mount RDWR because of "
2219                      "unsupported optional features (%x).\n", i);
2220                 status = -EINVAL;
2221                 goto bail;
2222         }
2223
2224         if (ocfs2_clusterinfo_valid(osb)) {
2225                 osb->osb_stackflags =
2226                         OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
2227                 memcpy(osb->osb_cluster_stack,
2228                        OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
2229                        OCFS2_STACK_LABEL_LEN);
2230                 osb->osb_cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
2231                 if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
2232                         mlog(ML_ERROR,
2233                              "couldn't mount because of an invalid "
2234                              "cluster stack label (%s) \n",
2235                              osb->osb_cluster_stack);
2236                         status = -EINVAL;
2237                         goto bail;
2238                 }
2239         } else {
2240                 /* The empty string is identical with classic tools that
2241                  * don't know about s_cluster_info. */
2242                 osb->osb_cluster_stack[0] = '\0';
2243         }
2244
2245         get_random_bytes(&osb->s_next_generation, sizeof(u32));
2246
2247         /* FIXME
2248          * This should be done in ocfs2_journal_init(), but unknown
2249          * ordering issues will cause the filesystem to crash.
2250          * If anyone wants to figure out what part of the code
2251          * refers to osb->journal before ocfs2_journal_init() is run,
2252          * be my guest.
2253          */
2254         /* initialize our journal structure */
2255
2256         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
2257         if (!journal) {
2258                 mlog(ML_ERROR, "unable to alloc journal\n");
2259                 status = -ENOMEM;
2260                 goto bail;
2261         }
2262         osb->journal = journal;
2263         journal->j_osb = osb;
2264
2265         atomic_set(&journal->j_num_trans, 0);
2266         init_rwsem(&journal->j_trans_barrier);
2267         init_waitqueue_head(&journal->j_checkpointed);
2268         spin_lock_init(&journal->j_lock);
2269         journal->j_trans_id = (unsigned long) 1;
2270         INIT_LIST_HEAD(&journal->j_la_cleanups);
2271         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
2272         journal->j_state = OCFS2_JOURNAL_FREE;
2273
2274         INIT_WORK(&osb->dentry_lock_work, ocfs2_drop_dl_inodes);
2275         osb->dentry_lock_list = NULL;
2276
2277         /* get some pseudo constants for clustersize bits */
2278         osb->s_clustersize_bits =
2279                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
2280         osb->s_clustersize = 1 << osb->s_clustersize_bits;
2281
2282         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
2283             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
2284                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
2285                      osb->s_clustersize);
2286                 status = -EINVAL;
2287                 goto bail;
2288         }
2289
2290         total_blocks = ocfs2_clusters_to_blocks(osb->sb,
2291                                                 le32_to_cpu(di->i_clusters));
2292
2293         status = generic_check_addressable(osb->sb->s_blocksize_bits,
2294                                            total_blocks);
2295         if (status) {
2296                 mlog(ML_ERROR, "Volume too large "
2297                      "to mount safely on this system");
2298                 status = -EFBIG;
2299                 goto bail;
2300         }
2301
2302         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
2303                                  sizeof(di->id2.i_super.s_uuid))) {
2304                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
2305                 status = -ENOMEM;
2306                 goto bail;
2307         }
2308
2309         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
2310
2311         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
2312         osb->vol_label[63] = '\0';
2313         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
2314         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
2315         osb->first_cluster_group_blkno =
2316                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
2317         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
2318         osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
2319         trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
2320                                      (unsigned long long)osb->root_blkno,
2321                                      (unsigned long long)osb->system_dir_blkno,
2322                                      osb->s_clustersize_bits);
2323
2324         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
2325         if (!osb->osb_dlm_debug) {
2326                 status = -ENOMEM;
2327                 mlog_errno(status);
2328                 goto bail;
2329         }
2330
2331         atomic_set(&osb->vol_state, VOLUME_INIT);
2332
2333         /* load root, system_dir, and all global system inodes */
2334         status = ocfs2_init_global_system_inodes(osb);
2335         if (status < 0) {
2336                 mlog_errno(status);
2337                 goto bail;
2338         }
2339
2340         /*
2341          * global bitmap
2342          */
2343         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
2344                                             OCFS2_INVALID_SLOT);
2345         if (!inode) {
2346                 status = -EINVAL;
2347                 mlog_errno(status);
2348                 goto bail;
2349         }
2350
2351         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
2352         osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
2353         iput(inode);
2354
2355         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
2356                                  osb->s_feature_incompat) * 8;
2357
2358         status = ocfs2_init_slot_info(osb);
2359         if (status < 0) {
2360                 mlog_errno(status);
2361                 goto bail;
2362         }
2363         cleancache_init_shared_fs((char *)&di->id2.i_super.s_uuid, sb);
2364
2365 bail:
2366         return status;
2367 }
2368
2369 /*
2370  * will return: -EAGAIN if it is ok to keep searching for superblocks
2371  *              -EINVAL if there is a bad superblock
2372  *              0 on success
2373  */
2374 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
2375                                struct buffer_head *bh,
2376                                u32 blksz,
2377                                struct ocfs2_blockcheck_stats *stats)
2378 {
2379         int status = -EAGAIN;
2380
2381         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
2382                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
2383                 /* We have to do a raw check of the feature here */
2384                 if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
2385                     OCFS2_FEATURE_INCOMPAT_META_ECC) {
2386                         status = ocfs2_block_check_validate(bh->b_data,
2387                                                             bh->b_size,
2388                                                             &di->i_check,
2389                                                             stats);
2390                         if (status)
2391                                 goto out;
2392                 }
2393                 status = -EINVAL;
2394                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
2395                         mlog(ML_ERROR, "found superblock with incorrect block "
2396                              "size: found %u, should be %u\n",
2397                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
2398                                blksz);
2399                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
2400                            OCFS2_MAJOR_REV_LEVEL ||
2401                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
2402                            OCFS2_MINOR_REV_LEVEL) {
2403                         mlog(ML_ERROR, "found superblock with bad version: "
2404                              "found %u.%u, should be %u.%u\n",
2405                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
2406                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
2407                              OCFS2_MAJOR_REV_LEVEL,
2408                              OCFS2_MINOR_REV_LEVEL);
2409                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
2410                         mlog(ML_ERROR, "bad block number on superblock: "
2411                              "found %llu, should be %llu\n",
2412                              (unsigned long long)le64_to_cpu(di->i_blkno),
2413                              (unsigned long long)bh->b_blocknr);
2414                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
2415                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
2416                         mlog(ML_ERROR, "bad cluster size found: %u\n",
2417                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
2418                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
2419                         mlog(ML_ERROR, "bad root_blkno: 0\n");
2420                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
2421                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
2422                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
2423                         mlog(ML_ERROR,
2424                              "Superblock slots found greater than file system "
2425                              "maximum: found %u, max %u\n",
2426                              le16_to_cpu(di->id2.i_super.s_max_slots),
2427                              OCFS2_MAX_SLOTS);
2428                 } else {
2429                         /* found it! */
2430                         status = 0;
2431                 }
2432         }
2433
2434 out:
2435         if (status && status != -EAGAIN)
2436                 mlog_errno(status);
2437         return status;
2438 }
2439
2440 static int ocfs2_check_volume(struct ocfs2_super *osb)
2441 {
2442         int status;
2443         int dirty;
2444         int local;
2445         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
2446                                                   * recover
2447                                                   * ourselves. */
2448
2449         /* Init our journal object. */
2450         status = ocfs2_journal_init(osb->journal, &dirty);
2451         if (status < 0) {
2452                 mlog(ML_ERROR, "Could not initialize journal!\n");
2453                 goto finally;
2454         }
2455
2456         /* Now that journal has been initialized, check to make sure
2457            entire volume is addressable. */
2458         status = ocfs2_journal_addressable(osb);
2459         if (status)
2460                 goto finally;
2461
2462         /* If the journal was unmounted cleanly then we don't want to
2463          * recover anything. Otherwise, journal_load will do that
2464          * dirty work for us :) */
2465         if (!dirty) {
2466                 status = ocfs2_journal_wipe(osb->journal, 0);
2467                 if (status < 0) {
2468                         mlog_errno(status);
2469                         goto finally;
2470                 }
2471         } else {
2472                 printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
2473                        "unmounted cleanly, recovering it.\n", osb->dev_str);
2474         }
2475
2476         local = ocfs2_mount_local(osb);
2477
2478         /* will play back anything left in the journal. */
2479         status = ocfs2_journal_load(osb->journal, local, dirty);
2480         if (status < 0) {
2481                 mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
2482                 goto finally;
2483         }
2484
2485         if (dirty) {
2486                 /* recover my local alloc if we didn't unmount cleanly. */
2487                 status = ocfs2_begin_local_alloc_recovery(osb,
2488                                                           osb->slot_num,
2489                                                           &local_alloc);
2490                 if (status < 0) {
2491                         mlog_errno(status);
2492                         goto finally;
2493                 }
2494                 /* we complete the recovery process after we've marked
2495                  * ourselves as mounted. */
2496         }
2497
2498         status = ocfs2_load_local_alloc(osb);
2499         if (status < 0) {
2500                 mlog_errno(status);
2501                 goto finally;
2502         }
2503
2504         if (dirty) {
2505                 /* Recovery will be completed after we've mounted the
2506                  * rest of the volume. */
2507                 osb->dirty = 1;
2508                 osb->local_alloc_copy = local_alloc;
2509                 local_alloc = NULL;
2510         }
2511
2512         /* go through each journal, trylock it and if you get the
2513          * lock, and it's marked as dirty, set the bit in the recover
2514          * map and launch a recovery thread for it. */
2515         status = ocfs2_mark_dead_nodes(osb);
2516         if (status < 0) {
2517                 mlog_errno(status);
2518                 goto finally;
2519         }
2520
2521         status = ocfs2_compute_replay_slots(osb);
2522         if (status < 0)
2523                 mlog_errno(status);
2524
2525 finally:
2526         if (local_alloc)
2527                 kfree(local_alloc);
2528
2529         if (status)
2530                 mlog_errno(status);
2531         return status;
2532 }
2533
2534 /*
2535  * The routine gets called from dismount or close whenever a dismount on
2536  * volume is requested and the osb open count becomes 1.
2537  * It will remove the osb from the global list and also free up all the
2538  * initialized resources and fileobject.
2539  */
2540 static void ocfs2_delete_osb(struct ocfs2_super *osb)
2541 {
2542         /* This function assumes that the caller has the main osb resource */
2543
2544         ocfs2_free_slot_info(osb);
2545
2546         kfree(osb->osb_orphan_wipes);
2547         kfree(osb->slot_recovery_generations);
2548         /* FIXME
2549          * This belongs in journal shutdown, but because we have to
2550          * allocate osb->journal at the start of ocfs2_initialize_osb(),
2551          * we free it here.
2552          */
2553         kfree(osb->journal);
2554         if (osb->local_alloc_copy)
2555                 kfree(osb->local_alloc_copy);
2556         kfree(osb->uuid_str);
2557         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
2558         memset(osb, 0, sizeof(struct ocfs2_super));
2559 }
2560
2561 /* Put OCFS2 into a readonly state, or (if the user specifies it),
2562  * panic(). We do not support continue-on-error operation. */
2563 static void ocfs2_handle_error(struct super_block *sb)
2564 {
2565         struct ocfs2_super *osb = OCFS2_SB(sb);
2566
2567         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
2568                 panic("OCFS2: (device %s): panic forced after error\n",
2569                       sb->s_id);
2570
2571         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
2572
2573         if (sb->s_flags & MS_RDONLY &&
2574             (ocfs2_is_soft_readonly(osb) ||
2575              ocfs2_is_hard_readonly(osb)))
2576                 return;
2577
2578         printk(KERN_CRIT "File system is now read-only due to the potential "
2579                "of on-disk corruption. Please run fsck.ocfs2 once the file "
2580                "system is unmounted.\n");
2581         sb->s_flags |= MS_RDONLY;
2582         ocfs2_set_ro_flag(osb, 0);
2583 }
2584
2585 static char error_buf[1024];
2586
2587 void __ocfs2_error(struct super_block *sb,
2588                    const char *function,
2589                    const char *fmt, ...)
2590 {
2591         va_list args;
2592
2593         va_start(args, fmt);
2594         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2595         va_end(args);
2596
2597         /* Not using mlog here because we want to show the actual
2598          * function the error came from. */
2599         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
2600                sb->s_id, function, error_buf);
2601
2602         ocfs2_handle_error(sb);
2603 }
2604
2605 /* Handle critical errors. This is intentionally more drastic than
2606  * ocfs2_handle_error, so we only use for things like journal errors,
2607  * etc. */
2608 void __ocfs2_abort(struct super_block* sb,
2609                    const char *function,
2610                    const char *fmt, ...)
2611 {
2612         va_list args;
2613
2614         va_start(args, fmt);
2615         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
2616         va_end(args);
2617
2618         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
2619                sb->s_id, function, error_buf);
2620
2621         /* We don't have the cluster support yet to go straight to
2622          * hard readonly in here. Until then, we want to keep
2623          * ocfs2_abort() so that we can at least mark critical
2624          * errors.
2625          *
2626          * TODO: This should abort the journal and alert other nodes
2627          * that our slot needs recovery. */
2628
2629         /* Force a panic(). This stinks, but it's better than letting
2630          * things continue without having a proper hard readonly
2631          * here. */
2632         if (!ocfs2_mount_local(OCFS2_SB(sb)))
2633                 OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
2634         ocfs2_handle_error(sb);
2635 }
2636
2637 /*
2638  * Void signal blockers, because in-kernel sigprocmask() only fails
2639  * when SIG_* is wrong.
2640  */
2641 void ocfs2_block_signals(sigset_t *oldset)
2642 {
2643         int rc;
2644         sigset_t blocked;
2645
2646         sigfillset(&blocked);
2647         rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
2648         BUG_ON(rc);
2649 }
2650
2651 void ocfs2_unblock_signals(sigset_t *oldset)
2652 {
2653         int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
2654         BUG_ON(rc);
2655 }
2656
2657 module_init(ocfs2_init);
2658 module_exit(ocfs2_exit);