fs/ocfs2/super.c

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