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