SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/utsname.h>
  32 #include <linux/init.h>
  33 #include <linux/random.h>
  34 #include <linux/statfs.h>
  35 #include <linux/moduleparam.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/socket.h>
  38 #include <linux/inet.h>
  39 #include <linux/parser.h>
  40 #include <linux/crc32.h>
  41 #include <linux/debugfs.h>
  42 #include <linux/mount.h>
  43
  44 #include <cluster/nodemanager.h>
  45
  46 #define MLOG_MASK_PREFIX ML_SUPER
  47 #include <cluster/masklog.h>
  48
  49 #include "ocfs2.h"
  50
  51 /* this should be the only file to include a version 1 header */
  52 #include "ocfs1_fs_compat.h"
  53
  54 #include "alloc.h"
  55 #include "dlmglue.h"
  56 #include "export.h"
  57 #include "extent_map.h"
  58 #include "heartbeat.h"
  59 #include "inode.h"
  60 #include "journal.h"
  61 #include "localalloc.h"
  62 #include "namei.h"
  63 #include "slot_map.h"
  64 #include "super.h"
  65 #include "sysfile.h"
  66 #include "uptodate.h"
  67 #include "ver.h"
  68
  69 #include "buffer_head_io.h"
  70
  71 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  72
  73 /* OCFS2 needs to schedule several differnt types of work which
  74  * require cluster locking, disk I/O, recovery waits, etc. Since these
  75  * types of work tend to be heavy we avoid using the kernel events
  76  * workqueue and schedule on our own. */
  77 struct workqueue_struct *ocfs2_wq = NULL;
  78
  79 static struct dentry *ocfs2_debugfs_root = NULL;
  80
  81 MODULE_AUTHOR("Oracle");
  82 MODULE_LICENSE("GPL");
  83
  84 struct mount_options
  85 {
  86         unsigned long   commit_interval;
  87         unsigned long   mount_opt;
  88         unsigned int    atime_quantum;
  89         signed short    slot;
  90         unsigned int    localalloc_opt;
  91 };
  92
  93 static int ocfs2_parse_options(struct super_block *sb, char *options,
  94                                struct mount_options *mopt,
  95                                int is_remount);
  96 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt);
  97 static void ocfs2_put_super(struct super_block *sb);
  98 static int ocfs2_mount_volume(struct super_block *sb);
  99 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
 100 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
 101 static int ocfs2_initialize_mem_caches(void);
 102 static void ocfs2_free_mem_caches(void);
 103 static void ocfs2_delete_osb(struct ocfs2_super *osb);
 104
 105 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
 106
 107 static int ocfs2_sync_fs(struct super_block *sb, int wait);
 108
 109 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
 110 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 111 static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
 112 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
 113 static int ocfs2_check_volume(struct ocfs2_super *osb);
 114 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 115                                struct buffer_head *bh,
 116                                u32 sectsize);
 117 static int ocfs2_initialize_super(struct super_block *sb,
 118                                   struct buffer_head *bh,
 119                                   int sector_size);
 120 static int ocfs2_get_sector(struct super_block *sb,
 121                             struct buffer_head **bh,
 122                             int block,
 123                             int sect_size);
 124 static void ocfs2_write_super(struct super_block *sb);
 125 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 126 static void ocfs2_destroy_inode(struct inode *inode);
 127
 128 static const struct super_operations ocfs2_sops = {
 129         .statfs         = ocfs2_statfs,
 130         .alloc_inode    = ocfs2_alloc_inode,
 131         .destroy_inode  = ocfs2_destroy_inode,
 132         .drop_inode     = ocfs2_drop_inode,
 133         .clear_inode    = ocfs2_clear_inode,
 134         .delete_inode   = ocfs2_delete_inode,
 135         .sync_fs        = ocfs2_sync_fs,
 136         .write_super    = ocfs2_write_super,
 137         .put_super      = ocfs2_put_super,
 138         .remount_fs     = ocfs2_remount,
 139         .show_options   = ocfs2_show_options,
 140 };
 141
 142 enum {
 143         Opt_barrier,
 144         Opt_err_panic,
 145         Opt_err_ro,
 146         Opt_intr,
 147         Opt_nointr,
 148         Opt_hb_none,
 149         Opt_hb_local,
 150         Opt_data_ordered,
 151         Opt_data_writeback,
 152         Opt_atime_quantum,
 153         Opt_slot,
 154         Opt_commit,
 155         Opt_localalloc,
 156         Opt_localflocks,
 157         Opt_err,
 158 };
 159
 160 static match_table_t tokens = {
 161         {Opt_barrier, "barrier=%u"},
 162         {Opt_err_panic, "errors=panic"},
 163         {Opt_err_ro, "errors=remount-ro"},
 164         {Opt_intr, "intr"},
 165         {Opt_nointr, "nointr"},
 166         {Opt_hb_none, OCFS2_HB_NONE},
 167         {Opt_hb_local, OCFS2_HB_LOCAL},
 168         {Opt_data_ordered, "data=ordered"},
 169         {Opt_data_writeback, "data=writeback"},
 170         {Opt_atime_quantum, "atime_quantum=%u"},
 171         {Opt_slot, "preferred_slot=%u"},
 172         {Opt_commit, "commit=%u"},
 173         {Opt_localalloc, "localalloc=%d"},
 174         {Opt_localflocks, "localflocks"},
 175         {Opt_err, NULL}
 176 };
 177
 178 /*
 179  * write_super and sync_fs ripped right out of ext3.
 180  */
 181 static void ocfs2_write_super(struct super_block *sb)
 182 {
 183         if (mutex_trylock(&sb->s_lock) != 0)
 184                 BUG();
 185         sb->s_dirt = 0;
 186 }
 187
 188 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 189 {
 190         int status;
 191         tid_t target;
 192         struct ocfs2_super *osb = OCFS2_SB(sb);
 193
 194         sb->s_dirt = 0;
 195
 196         if (ocfs2_is_hard_readonly(osb))
 197                 return -EROFS;
 198
 199         if (wait) {
 200                 status = ocfs2_flush_truncate_log(osb);
 201                 if (status < 0)
 202                         mlog_errno(status);
 203         } else {
 204                 ocfs2_schedule_truncate_log_flush(osb, 0);
 205         }
 206
 207         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
 208                 if (wait)
 209                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 210                                         target);
 211         }
 212         return 0;
 213 }
 214
 215 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 216 {
 217         struct inode *new = NULL;
 218         int status = 0;
 219         int i;
 220
 221         mlog_entry_void();
 222
 223         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
 224         if (IS_ERR(new)) {
 225                 status = PTR_ERR(new);
 226                 mlog_errno(status);
 227                 goto bail;
 228         }
 229         osb->root_inode = new;
 230
 231         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
 232         if (IS_ERR(new)) {
 233                 status = PTR_ERR(new);
 234                 mlog_errno(status);
 235                 goto bail;
 236         }
 237         osb->sys_root_inode = new;
 238
 239         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 240              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 241                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 242                 if (!new) {
 243                         ocfs2_release_system_inodes(osb);
 244                         status = -EINVAL;
 245                         mlog_errno(status);
 246                         /* FIXME: Should ERROR_RO_FS */
 247                         mlog(ML_ERROR, "Unable to load system inode %d, "
 248                              "possibly corrupt fs?", i);
 249                         goto bail;
 250                 }
 251                 // the array now has one ref, so drop this one
 252                 iput(new);
 253         }
 254
 255 bail:
 256         mlog_exit(status);
 257         return status;
 258 }
 259
 260 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 261 {
 262         struct inode *new = NULL;
 263         int status = 0;
 264         int i;
 265
 266         mlog_entry_void();
 267
 268         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 269              i < NUM_SYSTEM_INODES;
 270              i++) {
 271                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 272                 if (!new) {
 273                         ocfs2_release_system_inodes(osb);
 274                         status = -EINVAL;
 275                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 276                              status, i, osb->slot_num);
 277                         goto bail;
 278                 }
 279                 /* the array now has one ref, so drop this one */
 280                 iput(new);
 281         }
 282
 283 bail:
 284         mlog_exit(status);
 285         return status;
 286 }
 287
 288 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 289 {
 290         int i;
 291         struct inode *inode;
 292
 293         mlog_entry_void();
 294
 295         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
 296                 inode = osb->system_inodes[i];
 297                 if (inode) {
 298                         iput(inode);
 299                         osb->system_inodes[i] = NULL;
 300                 }
 301         }
 302
 303         inode = osb->sys_root_inode;
 304         if (inode) {
 305                 iput(inode);
 306                 osb->sys_root_inode = NULL;
 307         }
 308
 309         inode = osb->root_inode;
 310         if (inode) {
 311                 iput(inode);
 312                 osb->root_inode = NULL;
 313         }
 314
 315         mlog_exit(0);
 316 }
 317
 318 /* We're allocating fs objects, use GFP_NOFS */
 319 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 320 {
 321         struct ocfs2_inode_info *oi;
 322
 323         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 324         if (!oi)
 325                 return NULL;
 326
 327         return &oi->vfs_inode;
 328 }
 329
 330 static void ocfs2_destroy_inode(struct inode *inode)
 331 {
 332         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 333 }
 334
 335 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 336                                                 unsigned int cbits)
 337 {
 338         unsigned int bytes = 1 << cbits;
 339         unsigned int trim = bytes;
 340         unsigned int bitshift = 32;
 341
 342         /*
 343          * i_size and all block offsets in ocfs2 are always 64 bits
 344          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 345          * 64 bit platforms, cluster size will be the limiting factor.
 346          */
 347
 348 #if BITS_PER_LONG == 32
 349 # if defined(CONFIG_LBD)
 350         BUILD_BUG_ON(sizeof(sector_t) != 8);
 351         /*
 352          * We might be limited by page cache size.
 353          */
 354         if (bytes > PAGE_CACHE_SIZE) {
 355                 bytes = PAGE_CACHE_SIZE;
 356                 trim = 1;
 357                 /*
 358                  * Shift by 31 here so that we don't get larger than
 359                  * MAX_LFS_FILESIZE
 360                  */
 361                 bitshift = 31;
 362         }
 363 # else
 364         /*
 365          * We are limited by the size of sector_t. Use block size, as
 366          * that's what we expose to the VFS.
 367          */
 368         bytes = 1 << bbits;
 369         trim = 1;
 370         bitshift = 31;
 371 # endif
 372 #endif
 373
 374         /*
 375          * Trim by a whole cluster when we can actually approach the
 376          * on-disk limits. Otherwise we can overflow i_clusters when
 377          * an extent start is at the max offset.
 378          */
 379         return (((unsigned long long)bytes) << bitshift) - trim;
 380 }
 381
 382 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 383 {
 384         int incompat_features;
 385         int ret = 0;
 386         struct mount_options parsed_options;
 387         struct ocfs2_super *osb = OCFS2_SB(sb);
 388
 389         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 390                 ret = -EINVAL;
 391                 goto out;
 392         }
 393
 394         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
 395             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 396                 ret = -EINVAL;
 397                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 398                 goto out;
 399         }
 400
 401         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 402             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 403                 ret = -EINVAL;
 404                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 405                 goto out;
 406         }
 407
 408         /* We're going to/from readonly mode. */
 409         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 410                 /* Lock here so the check of HARD_RO and the potential
 411                  * setting of SOFT_RO is atomic. */
 412                 spin_lock(&osb->osb_lock);
 413                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 414                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 415                         ret = -EROFS;
 416                         goto unlock_osb;
 417                 }
 418
 419                 if (*flags & MS_RDONLY) {
 420                         mlog(0, "Going to ro mode.\n");
 421                         sb->s_flags |= MS_RDONLY;
 422                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 423                 } else {
 424                         mlog(0, "Making ro filesystem writeable.\n");
 425
 426                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 427                                 mlog(ML_ERROR, "Cannot remount RDWR "
 428                                      "filesystem due to previous errors.\n");
 429                                 ret = -EROFS;
 430                                 goto unlock_osb;
 431                         }
 432                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 433                         if (incompat_features) {
 434                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 435                                      "of unsupported optional features "
 436                                      "(%x).\n", incompat_features);
 437                                 ret = -EINVAL;
 438                                 goto unlock_osb;
 439                         }
 440                         sb->s_flags &= ~MS_RDONLY;
 441                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 442                 }
 443 unlock_osb:
 444                 spin_unlock(&osb->osb_lock);
 445         }
 446
 447         if (!ret) {
 448                 /* Only save off the new mount options in case of a successful
 449                  * remount. */
 450                 osb->s_mount_opt = parsed_options.mount_opt;
 451                 osb->s_atime_quantum = parsed_options.atime_quantum;
 452                 osb->preferred_slot = parsed_options.slot;
 453                 if (parsed_options.commit_interval)
 454                         osb->osb_commit_interval = parsed_options.commit_interval;
 455
 456                 if (!ocfs2_is_hard_readonly(osb))
 457                         ocfs2_set_journal_params(osb);
 458         }
 459 out:
 460         return ret;
 461 }
 462
 463 static int ocfs2_sb_probe(struct super_block *sb,
 464                           struct buffer_head **bh,
 465                           int *sector_size)
 466 {
 467         int status, tmpstat;
 468         struct ocfs1_vol_disk_hdr *hdr;
 469         struct ocfs2_dinode *di;
 470         int blksize;
 471
 472         *bh = NULL;
 473
 474         /* may be > 512 */
 475         *sector_size = bdev_hardsect_size(sb->s_bdev);
 476         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 477                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 478                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 479                 status = -EINVAL;
 480                 goto bail;
 481         }
 482
 483         /* Can this really happen? */
 484         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 485                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 486
 487         /* check block zero for old format */
 488         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 489         if (status < 0) {
 490                 mlog_errno(status);
 491                 goto bail;
 492         }
 493         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 494         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 495                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 496                      hdr->major_version, hdr->minor_version);
 497                 status = -EINVAL;
 498         }
 499         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 500                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 501                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 502                      hdr->signature);
 503                 status = -EINVAL;
 504         }
 505         brelse(*bh);
 506         *bh = NULL;
 507         if (status < 0) {
 508                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 509                      "upgraded before mounting with ocfs v2\n");
 510                 goto bail;
 511         }
 512
 513         /*
 514          * Now check at magic offset for 512, 1024, 2048, 4096
 515          * blocksizes.  4096 is the maximum blocksize because it is
 516          * the minimum clustersize.
 517          */
 518         status = -EINVAL;
 519         for (blksize = *sector_size;
 520              blksize <= OCFS2_MAX_BLOCKSIZE;
 521              blksize <<= 1) {
 522                 tmpstat = ocfs2_get_sector(sb, bh,
 523                                            OCFS2_SUPER_BLOCK_BLKNO,
 524                                            blksize);
 525                 if (tmpstat < 0) {
 526                         status = tmpstat;
 527                         mlog_errno(status);
 528                         goto bail;
 529                 }
 530                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 531                 status = ocfs2_verify_volume(di, *bh, blksize);
 532                 if (status >= 0)
 533                         goto bail;
 534                 brelse(*bh);
 535                 *bh = NULL;
 536                 if (status != -EAGAIN)
 537                         break;
 538         }
 539
 540 bail:
 541         return status;
 542 }
 543
 544 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 545 {
 546         if (ocfs2_mount_local(osb)) {
 547                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 548                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 549                              "mounted device.\n");
 550                         return -EINVAL;
 551                 }
 552         }
 553
 554         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 555                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
 556                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 557                              "a read-write clustered device.\n");
 558                         return -EINVAL;
 559                 }
 560         }
 561
 562         return 0;
 563 }
 564
 565 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 566 {
 567         struct dentry *root;
 568         int status, sector_size;
 569         struct mount_options parsed_options;
 570         struct inode *inode = NULL;
 571         struct ocfs2_super *osb = NULL;
 572         struct buffer_head *bh = NULL;
 573         char nodestr[8];
 574
 575         mlog_entry("%p, %p, %i", sb, data, silent);
 576
 577         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 578                 status = -EINVAL;
 579                 goto read_super_error;
 580         }
 581
 582         /* for now we only have one cluster/node, make sure we see it
 583          * in the heartbeat universe */
 584         if (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 585                 if (!o2hb_check_local_node_heartbeating()) {
 586                         status = -EINVAL;
 587                         goto read_super_error;
 588                 }
 589         }
 590
 591         /* probe for superblock */
 592         status = ocfs2_sb_probe(sb, &bh, &sector_size);
 593         if (status < 0) {
 594                 mlog(ML_ERROR, "superblock probe failed!\n");
 595                 goto read_super_error;
 596         }
 597
 598         status = ocfs2_initialize_super(sb, bh, sector_size);
 599         osb = OCFS2_SB(sb);
 600         if (status < 0) {
 601                 mlog_errno(status);
 602                 goto read_super_error;
 603         }
 604         brelse(bh);
 605         bh = NULL;
 606         osb->s_mount_opt = parsed_options.mount_opt;
 607         osb->s_atime_quantum = parsed_options.atime_quantum;
 608         osb->preferred_slot = parsed_options.slot;
 609         osb->osb_commit_interval = parsed_options.commit_interval;
 610         osb->local_alloc_size = parsed_options.localalloc_opt;
 611
 612         sb->s_magic = OCFS2_SUPER_MAGIC;
 613
 614         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
 615          * heartbeat=none */
 616         if (bdev_read_only(sb->s_bdev)) {
 617                 if (!(sb->s_flags & MS_RDONLY)) {
 618                         status = -EACCES;
 619                         mlog(ML_ERROR, "Readonly device detected but readonly "
 620                              "mount was not specified.\n");
 621                         goto read_super_error;
 622                 }
 623
 624                 /* You should not be able to start a local heartbeat
 625                  * on a readonly device. */
 626                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 627                         status = -EROFS;
 628                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
 629                              "device.\n");
 630                         goto read_super_error;
 631                 }
 632
 633                 status = ocfs2_check_journals_nolocks(osb);
 634                 if (status < 0) {
 635                         if (status == -EROFS)
 636                                 mlog(ML_ERROR, "Recovery required on readonly "
 637                                      "file system, but write access is "
 638                                      "unavailable.\n");
 639                         else
 640                                 mlog_errno(status);
 641                         goto read_super_error;
 642                 }
 643
 644                 ocfs2_set_ro_flag(osb, 1);
 645
 646                 printk(KERN_NOTICE "Readonly device detected. No cluster "
 647                        "services will be utilized for this mount. Recovery "
 648                        "will be skipped.\n");
 649         }
 650
 651         if (!ocfs2_is_hard_readonly(osb)) {
 652                 if (sb->s_flags & MS_RDONLY)
 653                         ocfs2_set_ro_flag(osb, 0);
 654         }
 655
 656         status = ocfs2_verify_heartbeat(osb);
 657         if (status < 0) {
 658                 mlog_errno(status);
 659                 goto read_super_error;
 660         }
 661
 662         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
 663                                                  ocfs2_debugfs_root);
 664         if (!osb->osb_debug_root) {
 665                 status = -EINVAL;
 666                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
 667                 goto read_super_error;
 668         }
 669
 670         status = ocfs2_mount_volume(sb);
 671         if (osb->root_inode)
 672                 inode = igrab(osb->root_inode);
 673
 674         if (status < 0)
 675                 goto read_super_error;
 676
 677         if (!inode) {
 678                 status = -EIO;
 679                 mlog_errno(status);
 680                 goto read_super_error;
 681         }
 682
 683         root = d_alloc_root(inode);
 684         if (!root) {
 685                 status = -ENOMEM;
 686                 mlog_errno(status);
 687                 goto read_super_error;
 688         }
 689
 690         sb->s_root = root;
 691
 692         ocfs2_complete_mount_recovery(osb);
 693
 694         if (ocfs2_mount_local(osb))
 695                 snprintf(nodestr, sizeof(nodestr), "local");
 696         else
 697                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
 698
 699         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
 700                "with %s data mode.\n",
 701                osb->dev_str, nodestr, osb->slot_num,
 702                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
 703                "ordered");
 704
 705         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
 706         wake_up(&osb->osb_mount_event);
 707
 708         mlog_exit(status);
 709         return status;
 710
 711 read_super_error:
 712         if (bh != NULL)
 713                 brelse(bh);
 714
 715         if (inode)
 716                 iput(inode);
 717
 718         if (osb) {
 719                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
 720                 wake_up(&osb->osb_mount_event);
 721                 ocfs2_dismount_volume(sb, 1);
 722         }
 723
 724         mlog_exit(status);
 725         return status;
 726 }
 727
 728 static int ocfs2_get_sb(struct file_system_type *fs_type,
 729                         int flags,
 730                         const char *dev_name,
 731                         void *data,
 732                         struct vfsmount *mnt)
 733 {
 734         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
 735                            mnt);
 736 }
 737
 738 static struct file_system_type ocfs2_fs_type = {
 739         .owner          = THIS_MODULE,
 740         .name           = "ocfs2",
 741         .get_sb         = ocfs2_get_sb, /* is this called when we mount
 742                                         * the fs? */
 743         .kill_sb        = kill_block_super, /* set to the generic one
 744                                              * right now, but do we
 745                                              * need to change that? */
 746         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 747         .next           = NULL
 748 };
 749
 750 static int ocfs2_parse_options(struct super_block *sb,
 751                                char *options,
 752                                struct mount_options *mopt,
 753                                int is_remount)
 754 {
 755         int status;
 756         char *p;
 757
 758         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 759                    options ? options : "(none)");
 760
 761         mopt->commit_interval = 0;
 762         mopt->mount_opt = 0;
 763         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 764         mopt->slot = OCFS2_INVALID_SLOT;
 765         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
 766
 767         if (!options) {
 768                 status = 1;
 769                 goto bail;
 770         }
 771
 772         while ((p = strsep(&options, ",")) != NULL) {
 773                 int token, option;
 774                 substring_t args[MAX_OPT_ARGS];
 775
 776                 if (!*p)
 777                         continue;
 778
 779                 token = match_token(p, tokens, args);
 780                 switch (token) {
 781                 case Opt_hb_local:
 782                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 783                         break;
 784                 case Opt_hb_none:
 785                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 786                         break;
 787                 case Opt_barrier:
 788                         if (match_int(&args[0], &option)) {
 789                                 status = 0;
 790                                 goto bail;
 791                         }
 792                         if (option)
 793                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
 794                         else
 795                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
 796                         break;
 797                 case Opt_intr:
 798                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
 799                         break;
 800                 case Opt_nointr:
 801                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
 802                         break;
 803                 case Opt_err_panic:
 804                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 805                         break;
 806                 case Opt_err_ro:
 807                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 808                         break;
 809                 case Opt_data_ordered:
 810                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 811                         break;
 812                 case Opt_data_writeback:
 813                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 814                         break;
 815                 case Opt_atime_quantum:
 816                         if (match_int(&args[0], &option)) {
 817                                 status = 0;
 818                                 goto bail;
 819                         }
 820                         if (option >= 0)
 821                                 mopt->atime_quantum = option;
 822                         break;
 823                 case Opt_slot:
 824                         option = 0;
 825                         if (match_int(&args[0], &option)) {
 826                                 status = 0;
 827                                 goto bail;
 828                         }
 829                         if (option)
 830                                 mopt->slot = (s16)option;
 831                         break;
 832                 case Opt_commit:
 833                         option = 0;
 834                         if (match_int(&args[0], &option)) {
 835                                 status = 0;
 836                                 goto bail;
 837                         }
 838                         if (option < 0)
 839                                 return 0;
 840                         if (option == 0)
 841                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 842                         mopt->commit_interval = HZ * option;
 843                         break;
 844                 case Opt_localalloc:
 845                         option = 0;
 846                         if (match_int(&args[0], &option)) {
 847                                 status = 0;
 848                                 goto bail;
 849                         }
 850                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
 851                                 mopt->localalloc_opt = option;
 852                         break;
 853                 case Opt_localflocks:
 854                         /*
 855                          * Changing this during remount could race
 856                          * flock() requests, or "unbalance" existing
 857                          * ones (e.g., a lock is taken in one mode but
 858                          * dropped in the other). If users care enough
 859                          * to flip locking modes during remount, we
 860                          * could add a "local" flag to individual
 861                          * flock structures for proper tracking of
 862                          * state.
 863                          */
 864                         if (!is_remount)
 865                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
 866                         break;
 867                 default:
 868                         mlog(ML_ERROR,
 869                              "Unrecognized mount option \"%s\" "
 870                              "or missing value\n", p);
 871                         status = 0;
 872                         goto bail;
 873                 }
 874         }
 875
 876         status = 1;
 877
 878 bail:
 879         mlog_exit(status);
 880         return status;
 881 }
 882
 883 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 884 {
 885         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
 886         unsigned long opts = osb->s_mount_opt;
 887
 888         if (opts & OCFS2_MOUNT_HB_LOCAL)
 889                 seq_printf(s, ",_netdev,heartbeat=local");
 890         else
 891                 seq_printf(s, ",heartbeat=none");
 892
 893         if (opts & OCFS2_MOUNT_NOINTR)
 894                 seq_printf(s, ",nointr");
 895
 896         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
 897                 seq_printf(s, ",data=writeback");
 898         else
 899                 seq_printf(s, ",data=ordered");
 900
 901         if (opts & OCFS2_MOUNT_BARRIER)
 902                 seq_printf(s, ",barrier=1");
 903
 904         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
 905                 seq_printf(s, ",errors=panic");
 906         else
 907                 seq_printf(s, ",errors=remount-ro");
 908
 909         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
 910                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
 911
 912         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
 913                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
 914
 915         if (osb->osb_commit_interval)
 916                 seq_printf(s, ",commit=%u",
 917                            (unsigned) (osb->osb_commit_interval / HZ));
 918
 919         if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
 920                 seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
 921
 922         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
 923                 seq_printf(s, ",localflocks,");
 924
 925         return 0;
 926 }
 927
 928 static int __init ocfs2_init(void)
 929 {
 930         int status;
 931
 932         mlog_entry_void();
 933
 934         ocfs2_print_version();
 935
 936         status = init_ocfs2_uptodate_cache();
 937         if (status < 0) {
 938                 mlog_errno(status);
 939                 goto leave;
 940         }
 941
 942         status = ocfs2_initialize_mem_caches();
 943         if (status < 0) {
 944                 mlog_errno(status);
 945                 goto leave;
 946         }
 947
 948         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
 949         if (!ocfs2_wq) {
 950                 status = -ENOMEM;
 951                 goto leave;
 952         }
 953
 954         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
 955         if (!ocfs2_debugfs_root) {
 956                 status = -EFAULT;
 957                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
 958         }
 959
 960 leave:
 961         if (status < 0) {
 962                 ocfs2_free_mem_caches();
 963                 exit_ocfs2_uptodate_cache();
 964         }
 965
 966         mlog_exit(status);
 967
 968         if (status >= 0) {
 969                 return register_filesystem(&ocfs2_fs_type);
 970         } else
 971                 return -1;
 972 }
 973
 974 static void __exit ocfs2_exit(void)
 975 {
 976         mlog_entry_void();
 977
 978         if (ocfs2_wq) {
 979                 flush_workqueue(ocfs2_wq);
 980                 destroy_workqueue(ocfs2_wq);
 981         }
 982
 983         debugfs_remove(ocfs2_debugfs_root);
 984
 985         ocfs2_free_mem_caches();
 986
 987         unregister_filesystem(&ocfs2_fs_type);
 988
 989         exit_ocfs2_uptodate_cache();
 990
 991         mlog_exit_void();
 992 }
 993
 994 static void ocfs2_put_super(struct super_block *sb)
 995 {
 996         mlog_entry("(0x%p)\n", sb);
 997
 998         ocfs2_sync_blockdev(sb);
 999         ocfs2_dismount_volume(sb, 0);
1000
1001         mlog_exit_void();
1002 }
1003
1004 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
1005 {
1006         struct ocfs2_super *osb;
1007         u32 numbits, freebits;
1008         int status;
1009         struct ocfs2_dinode *bm_lock;
1010         struct buffer_head *bh = NULL;
1011         struct inode *inode = NULL;
1012
1013         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1014
1015         osb = OCFS2_SB(dentry->d_sb);
1016
1017         inode = ocfs2_get_system_file_inode(osb,
1018                                             GLOBAL_BITMAP_SYSTEM_INODE,
1019                                             OCFS2_INVALID_SLOT);
1020         if (!inode) {
1021                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1022                 status = -EIO;
1023                 goto bail;
1024         }
1025
1026         status = ocfs2_inode_lock(inode, &bh, 0);
1027         if (status < 0) {
1028                 mlog_errno(status);
1029                 goto bail;
1030         }
1031
1032         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1033
1034         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1035         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1036
1037         buf->f_type = OCFS2_SUPER_MAGIC;
1038         buf->f_bsize = dentry->d_sb->s_blocksize;
1039         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1040         buf->f_blocks = ((sector_t) numbits) *
1041                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1042         buf->f_bfree = ((sector_t) freebits) *
1043                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1044         buf->f_bavail = buf->f_bfree;
1045         buf->f_files = numbits;
1046         buf->f_ffree = freebits;
1047
1048         brelse(bh);
1049
1050         ocfs2_inode_unlock(inode, 0);
1051         status = 0;
1052 bail:
1053         if (inode)
1054                 iput(inode);
1055
1056         mlog_exit(status);
1057
1058         return status;
1059 }
1060
1061 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1062 {
1063         struct ocfs2_inode_info *oi = data;
1064
1065         oi->ip_flags = 0;
1066         oi->ip_open_count = 0;
1067         spin_lock_init(&oi->ip_lock);
1068         ocfs2_extent_map_init(&oi->vfs_inode);
1069         INIT_LIST_HEAD(&oi->ip_io_markers);
1070         oi->ip_created_trans = 0;
1071         oi->ip_last_trans = 0;
1072         oi->ip_dir_start_lookup = 0;
1073
1074         init_rwsem(&oi->ip_alloc_sem);
1075         mutex_init(&oi->ip_io_mutex);
1076
1077         oi->ip_blkno = 0ULL;
1078         oi->ip_clusters = 0;
1079
1080         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1081         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1082         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1083
1084         ocfs2_metadata_cache_init(&oi->vfs_inode);
1085
1086         inode_init_once(&oi->vfs_inode);
1087 }
1088
1089 static int ocfs2_initialize_mem_caches(void)
1090 {
1091         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1092                                        sizeof(struct ocfs2_inode_info),
1093                                        0,
1094                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1095                                                 SLAB_MEM_SPREAD),
1096                                        ocfs2_inode_init_once);
1097         if (!ocfs2_inode_cachep)
1098                 return -ENOMEM;
1099
1100         return 0;
1101 }
1102
1103 static void ocfs2_free_mem_caches(void)
1104 {
1105         if (ocfs2_inode_cachep)
1106                 kmem_cache_destroy(ocfs2_inode_cachep);
1107
1108         ocfs2_inode_cachep = NULL;
1109 }
1110
1111 static int ocfs2_get_sector(struct super_block *sb,
1112                             struct buffer_head **bh,
1113                             int block,
1114                             int sect_size)
1115 {
1116         if (!sb_set_blocksize(sb, sect_size)) {
1117                 mlog(ML_ERROR, "unable to set blocksize\n");
1118                 return -EIO;
1119         }
1120
1121         *bh = sb_getblk(sb, block);
1122         if (!*bh) {
1123                 mlog_errno(-EIO);
1124                 return -EIO;
1125         }
1126         lock_buffer(*bh);
1127         if (!buffer_dirty(*bh))
1128                 clear_buffer_uptodate(*bh);
1129         unlock_buffer(*bh);
1130         ll_rw_block(READ, 1, bh);
1131         wait_on_buffer(*bh);
1132         return 0;
1133 }
1134
1135 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1136 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1137 {
1138         int status;
1139
1140         /* XXX hold a ref on the node while mounte?  easy enough, if
1141          * desirable. */
1142         if (ocfs2_mount_local(osb))
1143                 osb->node_num = 0;
1144         else
1145                 osb->node_num = o2nm_this_node();
1146
1147         if (osb->node_num == O2NM_MAX_NODES) {
1148                 mlog(ML_ERROR, "could not find this host's node number\n");
1149                 status = -ENOENT;
1150                 goto bail;
1151         }
1152
1153         mlog(0, "I am node %d\n", osb->node_num);
1154
1155         status = 0;
1156 bail:
1157         return status;
1158 }
1159
1160 static int ocfs2_mount_volume(struct super_block *sb)
1161 {
1162         int status = 0;
1163         int unlock_super = 0;
1164         struct ocfs2_super *osb = OCFS2_SB(sb);
1165
1166         mlog_entry_void();
1167
1168         if (ocfs2_is_hard_readonly(osb))
1169                 goto leave;
1170
1171         status = ocfs2_fill_local_node_info(osb);
1172         if (status < 0) {
1173                 mlog_errno(status);
1174                 goto leave;
1175         }
1176
1177         status = ocfs2_dlm_init(osb);
1178         if (status < 0) {
1179                 mlog_errno(status);
1180                 goto leave;
1181         }
1182
1183         status = ocfs2_super_lock(osb, 1);
1184         if (status < 0) {
1185                 mlog_errno(status);
1186                 goto leave;
1187         }
1188         unlock_super = 1;
1189
1190         /* This will load up the node map and add ourselves to it. */
1191         status = ocfs2_find_slot(osb);
1192         if (status < 0) {
1193                 mlog_errno(status);
1194                 goto leave;
1195         }
1196
1197         /* load all node-local system inodes */
1198         status = ocfs2_init_local_system_inodes(osb);
1199         if (status < 0) {
1200                 mlog_errno(status);
1201                 goto leave;
1202         }
1203
1204         status = ocfs2_check_volume(osb);
1205         if (status < 0) {
1206                 mlog_errno(status);
1207                 goto leave;
1208         }
1209
1210         status = ocfs2_truncate_log_init(osb);
1211         if (status < 0) {
1212                 mlog_errno(status);
1213                 goto leave;
1214         }
1215
1216         if (ocfs2_mount_local(osb))
1217                 goto leave;
1218
1219 leave:
1220         if (unlock_super)
1221                 ocfs2_super_unlock(osb, 1);
1222
1223         mlog_exit(status);
1224         return status;
1225 }
1226
1227 /* we can't grab the goofy sem lock from inside wait_event, so we use
1228  * memory barriers to make sure that we'll see the null task before
1229  * being woken up */
1230 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1231 {
1232         mb();
1233         return osb->recovery_thread_task != NULL;
1234 }
1235
1236 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1237 {
1238         int tmp;
1239         struct ocfs2_super *osb = NULL;
1240         char nodestr[8];
1241
1242         mlog_entry("(0x%p)\n", sb);
1243
1244         BUG_ON(!sb);
1245         osb = OCFS2_SB(sb);
1246         BUG_ON(!osb);
1247
1248         ocfs2_shutdown_local_alloc(osb);
1249
1250         ocfs2_truncate_log_shutdown(osb);
1251
1252         /* disable any new recovery threads and wait for any currently
1253          * running ones to exit. Do this before setting the vol_state. */
1254         mutex_lock(&osb->recovery_lock);
1255         osb->disable_recovery = 1;
1256         mutex_unlock(&osb->recovery_lock);
1257         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1258
1259         /* At this point, we know that no more recovery threads can be
1260          * launched, so wait for any recovery completion work to
1261          * complete. */
1262         flush_workqueue(ocfs2_wq);
1263
1264         ocfs2_journal_shutdown(osb);
1265
1266         ocfs2_sync_blockdev(sb);
1267
1268         /* No dlm means we've failed during mount, so skip all the
1269          * steps which depended on that to complete. */
1270         if (osb->dlm) {
1271                 tmp = ocfs2_super_lock(osb, 1);
1272                 if (tmp < 0) {
1273                         mlog_errno(tmp);
1274                         return;
1275                 }
1276         }
1277
1278         if (osb->slot_num != OCFS2_INVALID_SLOT)
1279                 ocfs2_put_slot(osb);
1280
1281         if (osb->dlm)
1282                 ocfs2_super_unlock(osb, 1);
1283
1284         ocfs2_release_system_inodes(osb);
1285
1286         if (osb->dlm)
1287                 ocfs2_dlm_shutdown(osb);
1288
1289         debugfs_remove(osb->osb_debug_root);
1290
1291         if (!mnt_err)
1292                 ocfs2_stop_heartbeat(osb);
1293
1294         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1295
1296         if (ocfs2_mount_local(osb))
1297                 snprintf(nodestr, sizeof(nodestr), "local");
1298         else
1299                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1300
1301         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1302                osb->dev_str, nodestr);
1303
1304         ocfs2_delete_osb(osb);
1305         kfree(osb);
1306         sb->s_dev = 0;
1307         sb->s_fs_info = NULL;
1308 }
1309
1310 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1311                                 unsigned uuid_bytes)
1312 {
1313         int i, ret;
1314         char *ptr;
1315
1316         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1317
1318         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1319         if (osb->uuid_str == NULL)
1320                 return -ENOMEM;
1321
1322         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1323                 /* print with null */
1324                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1325                 if (ret != 2) /* drop super cleans up */
1326                         return -EINVAL;
1327                 /* then only advance past the last char */
1328                 ptr += 2;
1329         }
1330
1331         return 0;
1332 }
1333
1334 static int ocfs2_initialize_super(struct super_block *sb,
1335                                   struct buffer_head *bh,
1336                                   int sector_size)
1337 {
1338         int status;
1339         int i, cbits, bbits;
1340         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1341         struct inode *inode = NULL;
1342         struct ocfs2_journal *journal;
1343         __le32 uuid_net_key;
1344         struct ocfs2_super *osb;
1345
1346         mlog_entry_void();
1347
1348         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1349         if (!osb) {
1350                 status = -ENOMEM;
1351                 mlog_errno(status);
1352                 goto bail;
1353         }
1354
1355         sb->s_fs_info = osb;
1356         sb->s_op = &ocfs2_sops;
1357         sb->s_export_op = &ocfs2_export_ops;
1358         sb->s_time_gran = 1;
1359         sb->s_flags |= MS_NOATIME;
1360         /* this is needed to support O_LARGEFILE */
1361         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1362         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1363         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1364
1365         osb->sb = sb;
1366         /* Save off for ocfs2_rw_direct */
1367         osb->s_sectsize_bits = blksize_bits(sector_size);
1368         BUG_ON(!osb->s_sectsize_bits);
1369
1370         init_waitqueue_head(&osb->recovery_event);
1371         spin_lock_init(&osb->dc_task_lock);
1372         init_waitqueue_head(&osb->dc_event);
1373         osb->dc_work_sequence = 0;
1374         osb->dc_wake_sequence = 0;
1375         INIT_LIST_HEAD(&osb->blocked_lock_list);
1376         osb->blocked_lock_count = 0;
1377         spin_lock_init(&osb->osb_lock);
1378
1379         atomic_set(&osb->alloc_stats.moves, 0);
1380         atomic_set(&osb->alloc_stats.local_data, 0);
1381         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1382         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1383         atomic_set(&osb->alloc_stats.bg_extends, 0);
1384
1385         ocfs2_init_node_maps(osb);
1386
1387         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1388                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1389
1390         mutex_init(&osb->recovery_lock);
1391
1392         osb->disable_recovery = 0;
1393         osb->recovery_thread_task = NULL;
1394
1395         init_waitqueue_head(&osb->checkpoint_event);
1396         atomic_set(&osb->needs_checkpoint, 0);
1397
1398         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1399
1400         osb->node_num = O2NM_INVALID_NODE_NUM;
1401         osb->slot_num = OCFS2_INVALID_SLOT;
1402
1403         osb->local_alloc_state = OCFS2_LA_UNUSED;
1404         osb->local_alloc_bh = NULL;
1405
1406         ocfs2_setup_hb_callbacks(osb);
1407
1408         init_waitqueue_head(&osb->osb_mount_event);
1409
1410         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1411         if (!osb->vol_label) {
1412                 mlog(ML_ERROR, "unable to alloc vol label\n");
1413                 status = -ENOMEM;
1414                 goto bail;
1415         }
1416
1417         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1418         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1419                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1420                      osb->max_slots);
1421                 status = -EINVAL;
1422                 goto bail;
1423         }
1424         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1425
1426         init_waitqueue_head(&osb->osb_wipe_event);
1427         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1428                                         sizeof(*osb->osb_orphan_wipes),
1429                                         GFP_KERNEL);
1430         if (!osb->osb_orphan_wipes) {
1431                 status = -ENOMEM;
1432                 mlog_errno(status);
1433                 goto bail;
1434         }
1435
1436         osb->s_feature_compat =
1437                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1438         osb->s_feature_ro_compat =
1439                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1440         osb->s_feature_incompat =
1441                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1442
1443         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1444                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1445                      "optional features (%x).\n", i);
1446                 status = -EINVAL;
1447                 goto bail;
1448         }
1449         if (!(osb->sb->s_flags & MS_RDONLY) &&
1450             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1451                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1452                      "unsupported optional features (%x).\n", i);
1453                 status = -EINVAL;
1454                 goto bail;
1455         }
1456
1457         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1458
1459         /* FIXME
1460          * This should be done in ocfs2_journal_init(), but unknown
1461          * ordering issues will cause the filesystem to crash.
1462          * If anyone wants to figure out what part of the code
1463          * refers to osb->journal before ocfs2_journal_init() is run,
1464          * be my guest.
1465          */
1466         /* initialize our journal structure */
1467
1468         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1469         if (!journal) {
1470                 mlog(ML_ERROR, "unable to alloc journal\n");
1471                 status = -ENOMEM;
1472                 goto bail;
1473         }
1474         osb->journal = journal;
1475         journal->j_osb = osb;
1476
1477         atomic_set(&journal->j_num_trans, 0);
1478         init_rwsem(&journal->j_trans_barrier);
1479         init_waitqueue_head(&journal->j_checkpointed);
1480         spin_lock_init(&journal->j_lock);
1481         journal->j_trans_id = (unsigned long) 1;
1482         INIT_LIST_HEAD(&journal->j_la_cleanups);
1483         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1484         journal->j_state = OCFS2_JOURNAL_FREE;
1485
1486         /* get some pseudo constants for clustersize bits */
1487         osb->s_clustersize_bits =
1488                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1489         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1490         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1491
1492         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1493             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1494                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1495                      osb->s_clustersize);
1496                 status = -EINVAL;
1497                 goto bail;
1498         }
1499
1500         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1501             > (u32)~0UL) {
1502                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1503                      "what jbd can address in 32 bits.\n");
1504                 status = -EINVAL;
1505                 goto bail;
1506         }
1507
1508         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1509                                  sizeof(di->id2.i_super.s_uuid))) {
1510                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1511                 status = -ENOMEM;
1512                 goto bail;
1513         }
1514
1515         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1516
1517         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1518         osb->vol_label[63] = '\0';
1519         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1520         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1521         osb->first_cluster_group_blkno =
1522                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1523         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1524         mlog(0, "vol_label: %s\n", osb->vol_label);
1525         mlog(0, "uuid: %s\n", osb->uuid_str);
1526         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1527              (unsigned long long)osb->root_blkno,
1528              (unsigned long long)osb->system_dir_blkno);
1529
1530         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1531         if (!osb->osb_dlm_debug) {
1532                 status = -ENOMEM;
1533                 mlog_errno(status);
1534                 goto bail;
1535         }
1536
1537         atomic_set(&osb->vol_state, VOLUME_INIT);
1538
1539         /* load root, system_dir, and all global system inodes */
1540         status = ocfs2_init_global_system_inodes(osb);
1541         if (status < 0) {
1542                 mlog_errno(status);
1543                 goto bail;
1544         }
1545
1546         /*
1547          * global bitmap
1548          */
1549         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1550                                             OCFS2_INVALID_SLOT);
1551         if (!inode) {
1552                 status = -EINVAL;
1553                 mlog_errno(status);
1554                 goto bail;
1555         }
1556
1557         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1558         iput(inode);
1559
1560         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1561
1562         status = ocfs2_init_slot_info(osb);
1563         if (status < 0) {
1564                 mlog_errno(status);
1565                 goto bail;
1566         }
1567
1568 bail:
1569         mlog_exit(status);
1570         return status;
1571 }
1572
1573 /*
1574  * will return: -EAGAIN if it is ok to keep searching for superblocks
1575  *              -EINVAL if there is a bad superblock
1576  *              0 on success
1577  */
1578 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1579                                struct buffer_head *bh,
1580                                u32 blksz)
1581 {
1582         int status = -EAGAIN;
1583
1584         mlog_entry_void();
1585
1586         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1587                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1588                 status = -EINVAL;
1589                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1590                         mlog(ML_ERROR, "found superblock with incorrect block "
1591                              "size: found %u, should be %u\n",
1592                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1593                                blksz);
1594                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1595                            OCFS2_MAJOR_REV_LEVEL ||
1596                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1597                            OCFS2_MINOR_REV_LEVEL) {
1598                         mlog(ML_ERROR, "found superblock with bad version: "
1599                              "found %u.%u, should be %u.%u\n",
1600                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1601                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1602                              OCFS2_MAJOR_REV_LEVEL,
1603                              OCFS2_MINOR_REV_LEVEL);
1604                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1605                         mlog(ML_ERROR, "bad block number on superblock: "
1606                              "found %llu, should be %llu\n",
1607                              (unsigned long long)le64_to_cpu(di->i_blkno),
1608                              (unsigned long long)bh->b_blocknr);
1609                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1610                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1611                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1612                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1613                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1614                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1615                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1616                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1617                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1618                         mlog(ML_ERROR,
1619                              "Superblock slots found greater than file system "
1620                              "maximum: found %u, max %u\n",
1621                              le16_to_cpu(di->id2.i_super.s_max_slots),
1622                              OCFS2_MAX_SLOTS);
1623                 } else {
1624                         /* found it! */
1625                         status = 0;
1626                 }
1627         }
1628
1629         mlog_exit(status);
1630         return status;
1631 }
1632
1633 static int ocfs2_check_volume(struct ocfs2_super *osb)
1634 {
1635         int status;
1636         int dirty;
1637         int local;
1638         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1639                                                   * recover
1640                                                   * ourselves. */
1641
1642         mlog_entry_void();
1643
1644         /* Init our journal object. */
1645         status = ocfs2_journal_init(osb->journal, &dirty);
1646         if (status < 0) {
1647                 mlog(ML_ERROR, "Could not initialize journal!\n");
1648                 goto finally;
1649         }
1650
1651         /* If the journal was unmounted cleanly then we don't want to
1652          * recover anything. Otherwise, journal_load will do that
1653          * dirty work for us :) */
1654         if (!dirty) {
1655                 status = ocfs2_journal_wipe(osb->journal, 0);
1656                 if (status < 0) {
1657                         mlog_errno(status);
1658                         goto finally;
1659                 }
1660         } else {
1661                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1662                      "recovering volume.\n");
1663         }
1664
1665         local = ocfs2_mount_local(osb);
1666
1667         /* will play back anything left in the journal. */
1668         ocfs2_journal_load(osb->journal, local);
1669
1670         if (dirty) {
1671                 /* recover my local alloc if we didn't unmount cleanly. */
1672                 status = ocfs2_begin_local_alloc_recovery(osb,
1673                                                           osb->slot_num,
1674                                                           &local_alloc);
1675                 if (status < 0) {
1676                         mlog_errno(status);
1677                         goto finally;
1678                 }
1679                 /* we complete the recovery process after we've marked
1680                  * ourselves as mounted. */
1681         }
1682
1683         mlog(0, "Journal loaded.\n");
1684
1685         status = ocfs2_load_local_alloc(osb);
1686         if (status < 0) {
1687                 mlog_errno(status);
1688                 goto finally;
1689         }
1690
1691         if (dirty) {
1692                 /* Recovery will be completed after we've mounted the
1693                  * rest of the volume. */
1694                 osb->dirty = 1;
1695                 osb->local_alloc_copy = local_alloc;
1696                 local_alloc = NULL;
1697         }
1698
1699         /* go through each journal, trylock it and if you get the
1700          * lock, and it's marked as dirty, set the bit in the recover
1701          * map and launch a recovery thread for it. */
1702         status = ocfs2_mark_dead_nodes(osb);
1703         if (status < 0)
1704                 mlog_errno(status);
1705
1706 finally:
1707         if (local_alloc)
1708                 kfree(local_alloc);
1709
1710         mlog_exit(status);
1711         return status;
1712 }
1713
1714 /*
1715  * The routine gets called from dismount or close whenever a dismount on
1716  * volume is requested and the osb open count becomes 1.
1717  * It will remove the osb from the global list and also free up all the
1718  * initialized resources and fileobject.
1719  */
1720 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1721 {
1722         mlog_entry_void();
1723
1724         /* This function assumes that the caller has the main osb resource */
1725
1726         if (osb->slot_info)
1727                 ocfs2_free_slot_info(osb->slot_info);
1728
1729         kfree(osb->osb_orphan_wipes);
1730         /* FIXME
1731          * This belongs in journal shutdown, but because we have to
1732          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1733          * we free it here.
1734          */
1735         kfree(osb->journal);
1736         if (osb->local_alloc_copy)
1737                 kfree(osb->local_alloc_copy);
1738         kfree(osb->uuid_str);
1739         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1740         memset(osb, 0, sizeof(struct ocfs2_super));
1741
1742         mlog_exit_void();
1743 }
1744
1745 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1746  * panic(). We do not support continue-on-error operation. */
1747 static void ocfs2_handle_error(struct super_block *sb)
1748 {
1749         struct ocfs2_super *osb = OCFS2_SB(sb);
1750
1751         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1752                 panic("OCFS2: (device %s): panic forced after error\n",
1753                       sb->s_id);
1754
1755         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1756
1757         if (sb->s_flags & MS_RDONLY &&
1758             (ocfs2_is_soft_readonly(osb) ||
1759              ocfs2_is_hard_readonly(osb)))
1760                 return;
1761
1762         printk(KERN_CRIT "File system is now read-only due to the potential "
1763                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1764                "system is unmounted.\n");
1765         sb->s_flags |= MS_RDONLY;
1766         ocfs2_set_ro_flag(osb, 0);
1767 }
1768
1769 static char error_buf[1024];
1770
1771 void __ocfs2_error(struct super_block *sb,
1772                    const char *function,
1773                    const char *fmt, ...)
1774 {
1775         va_list args;
1776
1777         va_start(args, fmt);
1778         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1779         va_end(args);
1780
1781         /* Not using mlog here because we want to show the actual
1782          * function the error came from. */
1783         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1784                sb->s_id, function, error_buf);
1785
1786         ocfs2_handle_error(sb);
1787 }
1788
1789 /* Handle critical errors. This is intentionally more drastic than
1790  * ocfs2_handle_error, so we only use for things like journal errors,
1791  * etc. */
1792 void __ocfs2_abort(struct super_block* sb,
1793                    const char *function,
1794                    const char *fmt, ...)
1795 {
1796         va_list args;
1797
1798         va_start(args, fmt);
1799         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1800         va_end(args);
1801
1802         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1803                sb->s_id, function, error_buf);
1804
1805         /* We don't have the cluster support yet to go straight to
1806          * hard readonly in here. Until then, we want to keep
1807          * ocfs2_abort() so that we can at least mark critical
1808          * errors.
1809          *
1810          * TODO: This should abort the journal and alert other nodes
1811          * that our slot needs recovery. */
1812
1813         /* Force a panic(). This stinks, but it's better than letting
1814          * things continue without having a proper hard readonly
1815          * here. */
1816         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1817         ocfs2_handle_error(sb);
1818 }
1819
1820 module_init(ocfs2_init);
1821 module_exit(ocfs2_exit);