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 #include <linux/seq_file.h>
  44
  45 #define MLOG_MASK_PREFIX ML_SUPER
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 /* this should be the only file to include a version 1 header */
  51 #include "ocfs1_fs_compat.h"
  52
  53 #include "alloc.h"
  54 #include "dlmglue.h"
  55 #include "export.h"
  56 #include "extent_map.h"
  57 #include "heartbeat.h"
  58 #include "inode.h"
  59 #include "journal.h"
  60 #include "localalloc.h"
  61 #include "namei.h"
  62 #include "slot_map.h"
  63 #include "super.h"
  64 #include "sysfile.h"
  65 #include "uptodate.h"
  66 #include "ver.h"
  67
  68 #include "buffer_head_io.h"
  69
  70 static struct kmem_cache *ocfs2_inode_cachep = NULL;
  71
  72 /* OCFS2 needs to schedule several differnt types of work which
  73  * require cluster locking, disk I/O, recovery waits, etc. Since these
  74  * types of work tend to be heavy we avoid using the kernel events
  75  * workqueue and schedule on our own. */
  76 struct workqueue_struct *ocfs2_wq = NULL;
  77
  78 static struct dentry *ocfs2_debugfs_root = NULL;
  79
  80 MODULE_AUTHOR("Oracle");
  81 MODULE_LICENSE("GPL");
  82
  83 struct mount_options
  84 {
  85         unsigned long   commit_interval;
  86         unsigned long   mount_opt;
  87         unsigned int    atime_quantum;
  88         signed short    slot;
  89         unsigned int    localalloc_opt;
  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_check_volume(struct ocfs2_super *osb);
 112 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 113                                struct buffer_head *bh,
 114                                u32 sectsize);
 115 static int ocfs2_initialize_super(struct super_block *sb,
 116                                   struct buffer_head *bh,
 117                                   int sector_size);
 118 static int ocfs2_get_sector(struct super_block *sb,
 119                             struct buffer_head **bh,
 120                             int block,
 121                             int sect_size);
 122 static void ocfs2_write_super(struct super_block *sb);
 123 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 124 static void ocfs2_destroy_inode(struct inode *inode);
 125
 126 static const struct super_operations ocfs2_sops = {
 127         .statfs         = ocfs2_statfs,
 128         .alloc_inode    = ocfs2_alloc_inode,
 129         .destroy_inode  = ocfs2_destroy_inode,
 130         .drop_inode     = ocfs2_drop_inode,
 131         .clear_inode    = ocfs2_clear_inode,
 132         .delete_inode   = ocfs2_delete_inode,
 133         .sync_fs        = ocfs2_sync_fs,
 134         .write_super    = ocfs2_write_super,
 135         .put_super      = ocfs2_put_super,
 136         .remount_fs     = ocfs2_remount,
 137         .show_options   = ocfs2_show_options,
 138 };
 139
 140 enum {
 141         Opt_barrier,
 142         Opt_err_panic,
 143         Opt_err_ro,
 144         Opt_intr,
 145         Opt_nointr,
 146         Opt_hb_none,
 147         Opt_hb_local,
 148         Opt_data_ordered,
 149         Opt_data_writeback,
 150         Opt_atime_quantum,
 151         Opt_slot,
 152         Opt_commit,
 153         Opt_localalloc,
 154         Opt_localflocks,
 155         Opt_err,
 156 };
 157
 158 static match_table_t tokens = {
 159         {Opt_barrier, "barrier=%u"},
 160         {Opt_err_panic, "errors=panic"},
 161         {Opt_err_ro, "errors=remount-ro"},
 162         {Opt_intr, "intr"},
 163         {Opt_nointr, "nointr"},
 164         {Opt_hb_none, OCFS2_HB_NONE},
 165         {Opt_hb_local, OCFS2_HB_LOCAL},
 166         {Opt_data_ordered, "data=ordered"},
 167         {Opt_data_writeback, "data=writeback"},
 168         {Opt_atime_quantum, "atime_quantum=%u"},
 169         {Opt_slot, "preferred_slot=%u"},
 170         {Opt_commit, "commit=%u"},
 171         {Opt_localalloc, "localalloc=%d"},
 172         {Opt_localflocks, "localflocks"},
 173         {Opt_err, NULL}
 174 };
 175
 176 /*
 177  * write_super and sync_fs ripped right out of ext3.
 178  */
 179 static void ocfs2_write_super(struct super_block *sb)
 180 {
 181         if (mutex_trylock(&sb->s_lock) != 0)
 182                 BUG();
 183         sb->s_dirt = 0;
 184 }
 185
 186 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 187 {
 188         int status;
 189         tid_t target;
 190         struct ocfs2_super *osb = OCFS2_SB(sb);
 191
 192         sb->s_dirt = 0;
 193
 194         if (ocfs2_is_hard_readonly(osb))
 195                 return -EROFS;
 196
 197         if (wait) {
 198                 status = ocfs2_flush_truncate_log(osb);
 199                 if (status < 0)
 200                         mlog_errno(status);
 201         } else {
 202                 ocfs2_schedule_truncate_log_flush(osb, 0);
 203         }
 204
 205         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
 206                 if (wait)
 207                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 208                                         target);
 209         }
 210         return 0;
 211 }
 212
 213 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 214 {
 215         struct inode *new = NULL;
 216         int status = 0;
 217         int i;
 218
 219         mlog_entry_void();
 220
 221         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 222         if (IS_ERR(new)) {
 223                 status = PTR_ERR(new);
 224                 mlog_errno(status);
 225                 goto bail;
 226         }
 227         osb->root_inode = new;
 228
 229         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
 230         if (IS_ERR(new)) {
 231                 status = PTR_ERR(new);
 232                 mlog_errno(status);
 233                 goto bail;
 234         }
 235         osb->sys_root_inode = new;
 236
 237         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 238              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 239                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 240                 if (!new) {
 241                         ocfs2_release_system_inodes(osb);
 242                         status = -EINVAL;
 243                         mlog_errno(status);
 244                         /* FIXME: Should ERROR_RO_FS */
 245                         mlog(ML_ERROR, "Unable to load system inode %d, "
 246                              "possibly corrupt fs?", i);
 247                         goto bail;
 248                 }
 249                 // the array now has one ref, so drop this one
 250                 iput(new);
 251         }
 252
 253 bail:
 254         mlog_exit(status);
 255         return status;
 256 }
 257
 258 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 259 {
 260         struct inode *new = NULL;
 261         int status = 0;
 262         int i;
 263
 264         mlog_entry_void();
 265
 266         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 267              i < NUM_SYSTEM_INODES;
 268              i++) {
 269                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 270                 if (!new) {
 271                         ocfs2_release_system_inodes(osb);
 272                         status = -EINVAL;
 273                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 274                              status, i, osb->slot_num);
 275                         goto bail;
 276                 }
 277                 /* the array now has one ref, so drop this one */
 278                 iput(new);
 279         }
 280
 281 bail:
 282         mlog_exit(status);
 283         return status;
 284 }
 285
 286 static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
 287 {
 288         int i;
 289         struct inode *inode;
 290
 291         mlog_entry_void();
 292
 293         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
 294                 inode = osb->system_inodes[i];
 295                 if (inode) {
 296                         iput(inode);
 297                         osb->system_inodes[i] = NULL;
 298                 }
 299         }
 300
 301         inode = osb->sys_root_inode;
 302         if (inode) {
 303                 iput(inode);
 304                 osb->sys_root_inode = NULL;
 305         }
 306
 307         inode = osb->root_inode;
 308         if (inode) {
 309                 iput(inode);
 310                 osb->root_inode = NULL;
 311         }
 312
 313         mlog_exit(0);
 314 }
 315
 316 /* We're allocating fs objects, use GFP_NOFS */
 317 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 318 {
 319         struct ocfs2_inode_info *oi;
 320
 321         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 322         if (!oi)
 323                 return NULL;
 324
 325         return &oi->vfs_inode;
 326 }
 327
 328 static void ocfs2_destroy_inode(struct inode *inode)
 329 {
 330         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 331 }
 332
 333 static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
 334                                                 unsigned int cbits)
 335 {
 336         unsigned int bytes = 1 << cbits;
 337         unsigned int trim = bytes;
 338         unsigned int bitshift = 32;
 339
 340         /*
 341          * i_size and all block offsets in ocfs2 are always 64 bits
 342          * wide. i_clusters is 32 bits, in cluster-sized units. So on
 343          * 64 bit platforms, cluster size will be the limiting factor.
 344          */
 345
 346 #if BITS_PER_LONG == 32
 347 # if defined(CONFIG_LBD)
 348         BUILD_BUG_ON(sizeof(sector_t) != 8);
 349         /*
 350          * We might be limited by page cache size.
 351          */
 352         if (bytes > PAGE_CACHE_SIZE) {
 353                 bytes = PAGE_CACHE_SIZE;
 354                 trim = 1;
 355                 /*
 356                  * Shift by 31 here so that we don't get larger than
 357                  * MAX_LFS_FILESIZE
 358                  */
 359                 bitshift = 31;
 360         }
 361 # else
 362         /*
 363          * We are limited by the size of sector_t. Use block size, as
 364          * that's what we expose to the VFS.
 365          */
 366         bytes = 1 << bbits;
 367         trim = 1;
 368         bitshift = 31;
 369 # endif
 370 #endif
 371
 372         /*
 373          * Trim by a whole cluster when we can actually approach the
 374          * on-disk limits. Otherwise we can overflow i_clusters when
 375          * an extent start is at the max offset.
 376          */
 377         return (((unsigned long long)bytes) << bitshift) - trim;
 378 }
 379
 380 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 381 {
 382         int incompat_features;
 383         int ret = 0;
 384         struct mount_options parsed_options;
 385         struct ocfs2_super *osb = OCFS2_SB(sb);
 386
 387         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 388                 ret = -EINVAL;
 389                 goto out;
 390         }
 391
 392         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
 393             (parsed_options.mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 394                 ret = -EINVAL;
 395                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 396                 goto out;
 397         }
 398
 399         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 400             (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
 401                 ret = -EINVAL;
 402                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 403                 goto out;
 404         }
 405
 406         /* We're going to/from readonly mode. */
 407         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 408                 /* Lock here so the check of HARD_RO and the potential
 409                  * setting of SOFT_RO is atomic. */
 410                 spin_lock(&osb->osb_lock);
 411                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 412                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 413                         ret = -EROFS;
 414                         goto unlock_osb;
 415                 }
 416
 417                 if (*flags & MS_RDONLY) {
 418                         mlog(0, "Going to ro mode.\n");
 419                         sb->s_flags |= MS_RDONLY;
 420                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 421                 } else {
 422                         mlog(0, "Making ro filesystem writeable.\n");
 423
 424                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 425                                 mlog(ML_ERROR, "Cannot remount RDWR "
 426                                      "filesystem due to previous errors.\n");
 427                                 ret = -EROFS;
 428                                 goto unlock_osb;
 429                         }
 430                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 431                         if (incompat_features) {
 432                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 433                                      "of unsupported optional features "
 434                                      "(%x).\n", incompat_features);
 435                                 ret = -EINVAL;
 436                                 goto unlock_osb;
 437                         }
 438                         sb->s_flags &= ~MS_RDONLY;
 439                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 440                 }
 441 unlock_osb:
 442                 spin_unlock(&osb->osb_lock);
 443         }
 444
 445         if (!ret) {
 446                 /* Only save off the new mount options in case of a successful
 447                  * remount. */
 448                 osb->s_mount_opt = parsed_options.mount_opt;
 449                 osb->s_atime_quantum = parsed_options.atime_quantum;
 450                 osb->preferred_slot = parsed_options.slot;
 451                 if (parsed_options.commit_interval)
 452                         osb->osb_commit_interval = parsed_options.commit_interval;
 453
 454                 if (!ocfs2_is_hard_readonly(osb))
 455                         ocfs2_set_journal_params(osb);
 456         }
 457 out:
 458         return ret;
 459 }
 460
 461 static int ocfs2_sb_probe(struct super_block *sb,
 462                           struct buffer_head **bh,
 463                           int *sector_size)
 464 {
 465         int status, tmpstat;
 466         struct ocfs1_vol_disk_hdr *hdr;
 467         struct ocfs2_dinode *di;
 468         int blksize;
 469
 470         *bh = NULL;
 471
 472         /* may be > 512 */
 473         *sector_size = bdev_hardsect_size(sb->s_bdev);
 474         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 475                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 476                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 477                 status = -EINVAL;
 478                 goto bail;
 479         }
 480
 481         /* Can this really happen? */
 482         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 483                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 484
 485         /* check block zero for old format */
 486         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 487         if (status < 0) {
 488                 mlog_errno(status);
 489                 goto bail;
 490         }
 491         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 492         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 493                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 494                      hdr->major_version, hdr->minor_version);
 495                 status = -EINVAL;
 496         }
 497         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 498                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 499                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 500                      hdr->signature);
 501                 status = -EINVAL;
 502         }
 503         brelse(*bh);
 504         *bh = NULL;
 505         if (status < 0) {
 506                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 507                      "upgraded before mounting with ocfs v2\n");
 508                 goto bail;
 509         }
 510
 511         /*
 512          * Now check at magic offset for 512, 1024, 2048, 4096
 513          * blocksizes.  4096 is the maximum blocksize because it is
 514          * the minimum clustersize.
 515          */
 516         status = -EINVAL;
 517         for (blksize = *sector_size;
 518              blksize <= OCFS2_MAX_BLOCKSIZE;
 519              blksize <<= 1) {
 520                 tmpstat = ocfs2_get_sector(sb, bh,
 521                                            OCFS2_SUPER_BLOCK_BLKNO,
 522                                            blksize);
 523                 if (tmpstat < 0) {
 524                         status = tmpstat;
 525                         mlog_errno(status);
 526                         goto bail;
 527                 }
 528                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 529                 status = ocfs2_verify_volume(di, *bh, blksize);
 530                 if (status >= 0)
 531                         goto bail;
 532                 brelse(*bh);
 533                 *bh = NULL;
 534                 if (status != -EAGAIN)
 535                         break;
 536         }
 537
 538 bail:
 539         return status;
 540 }
 541
 542 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 543 {
 544         if (ocfs2_mount_local(osb)) {
 545                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 546                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 547                              "mounted device.\n");
 548                         return -EINVAL;
 549                 }
 550         }
 551
 552         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 553                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
 554                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 555                              "a read-write clustered device.\n");
 556                         return -EINVAL;
 557                 }
 558         }
 559
 560         return 0;
 561 }
 562
 563 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 564 {
 565         struct dentry *root;
 566         int status, sector_size;
 567         struct mount_options parsed_options;
 568         struct inode *inode = NULL;
 569         struct ocfs2_super *osb = NULL;
 570         struct buffer_head *bh = NULL;
 571         char nodestr[8];
 572
 573         mlog_entry("%p, %p, %i", sb, data, silent);
 574
 575         if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
 576                 status = -EINVAL;
 577                 goto read_super_error;
 578         }
 579
 580         /* probe for superblock */
 581         status = ocfs2_sb_probe(sb, &bh, &sector_size);
 582         if (status < 0) {
 583                 mlog(ML_ERROR, "superblock probe failed!\n");
 584                 goto read_super_error;
 585         }
 586
 587         status = ocfs2_initialize_super(sb, bh, sector_size);
 588         osb = OCFS2_SB(sb);
 589         if (status < 0) {
 590                 mlog_errno(status);
 591                 goto read_super_error;
 592         }
 593         brelse(bh);
 594         bh = NULL;
 595         osb->s_mount_opt = parsed_options.mount_opt;
 596         osb->s_atime_quantum = parsed_options.atime_quantum;
 597         osb->preferred_slot = parsed_options.slot;
 598         osb->osb_commit_interval = parsed_options.commit_interval;
 599         osb->local_alloc_size = parsed_options.localalloc_opt;
 600
 601         sb->s_magic = OCFS2_SUPER_MAGIC;
 602
 603         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
 604          * heartbeat=none */
 605         if (bdev_read_only(sb->s_bdev)) {
 606                 if (!(sb->s_flags & MS_RDONLY)) {
 607                         status = -EACCES;
 608                         mlog(ML_ERROR, "Readonly device detected but readonly "
 609                              "mount was not specified.\n");
 610                         goto read_super_error;
 611                 }
 612
 613                 /* You should not be able to start a local heartbeat
 614                  * on a readonly device. */
 615                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 616                         status = -EROFS;
 617                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
 618                              "device.\n");
 619                         goto read_super_error;
 620                 }
 621
 622                 status = ocfs2_check_journals_nolocks(osb);
 623                 if (status < 0) {
 624                         if (status == -EROFS)
 625                                 mlog(ML_ERROR, "Recovery required on readonly "
 626                                      "file system, but write access is "
 627                                      "unavailable.\n");
 628                         else
 629                                 mlog_errno(status);
 630                         goto read_super_error;
 631                 }
 632
 633                 ocfs2_set_ro_flag(osb, 1);
 634
 635                 printk(KERN_NOTICE "Readonly device detected. No cluster "
 636                        "services will be utilized for this mount. Recovery "
 637                        "will be skipped.\n");
 638         }
 639
 640         if (!ocfs2_is_hard_readonly(osb)) {
 641                 if (sb->s_flags & MS_RDONLY)
 642                         ocfs2_set_ro_flag(osb, 0);
 643         }
 644
 645         status = ocfs2_verify_heartbeat(osb);
 646         if (status < 0) {
 647                 mlog_errno(status);
 648                 goto read_super_error;
 649         }
 650
 651         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
 652                                                  ocfs2_debugfs_root);
 653         if (!osb->osb_debug_root) {
 654                 status = -EINVAL;
 655                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
 656                 goto read_super_error;
 657         }
 658
 659         status = ocfs2_mount_volume(sb);
 660         if (osb->root_inode)
 661                 inode = igrab(osb->root_inode);
 662
 663         if (status < 0)
 664                 goto read_super_error;
 665
 666         if (!inode) {
 667                 status = -EIO;
 668                 mlog_errno(status);
 669                 goto read_super_error;
 670         }
 671
 672         root = d_alloc_root(inode);
 673         if (!root) {
 674                 status = -ENOMEM;
 675                 mlog_errno(status);
 676                 goto read_super_error;
 677         }
 678
 679         sb->s_root = root;
 680
 681         ocfs2_complete_mount_recovery(osb);
 682
 683         if (ocfs2_mount_local(osb))
 684                 snprintf(nodestr, sizeof(nodestr), "local");
 685         else
 686                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
 687
 688         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
 689                "with %s data mode.\n",
 690                osb->dev_str, nodestr, osb->slot_num,
 691                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
 692                "ordered");
 693
 694         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
 695         wake_up(&osb->osb_mount_event);
 696
 697         mlog_exit(status);
 698         return status;
 699
 700 read_super_error:
 701         if (bh != NULL)
 702                 brelse(bh);
 703
 704         if (inode)
 705                 iput(inode);
 706
 707         if (osb) {
 708                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
 709                 wake_up(&osb->osb_mount_event);
 710                 ocfs2_dismount_volume(sb, 1);
 711         }
 712
 713         mlog_exit(status);
 714         return status;
 715 }
 716
 717 static int ocfs2_get_sb(struct file_system_type *fs_type,
 718                         int flags,
 719                         const char *dev_name,
 720                         void *data,
 721                         struct vfsmount *mnt)
 722 {
 723         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
 724                            mnt);
 725 }
 726
 727 static struct file_system_type ocfs2_fs_type = {
 728         .owner          = THIS_MODULE,
 729         .name           = "ocfs2",
 730         .get_sb         = ocfs2_get_sb, /* is this called when we mount
 731                                         * the fs? */
 732         .kill_sb        = kill_block_super, /* set to the generic one
 733                                              * right now, but do we
 734                                              * need to change that? */
 735         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 736         .next           = NULL
 737 };
 738
 739 static int ocfs2_parse_options(struct super_block *sb,
 740                                char *options,
 741                                struct mount_options *mopt,
 742                                int is_remount)
 743 {
 744         int status;
 745         char *p;
 746
 747         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 748                    options ? options : "(none)");
 749
 750         mopt->commit_interval = 0;
 751         mopt->mount_opt = 0;
 752         mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 753         mopt->slot = OCFS2_INVALID_SLOT;
 754         mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
 755
 756         if (!options) {
 757                 status = 1;
 758                 goto bail;
 759         }
 760
 761         while ((p = strsep(&options, ",")) != NULL) {
 762                 int token, option;
 763                 substring_t args[MAX_OPT_ARGS];
 764
 765                 if (!*p)
 766                         continue;
 767
 768                 token = match_token(p, tokens, args);
 769                 switch (token) {
 770                 case Opt_hb_local:
 771                         mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 772                         break;
 773                 case Opt_hb_none:
 774                         mopt->mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 775                         break;
 776                 case Opt_barrier:
 777                         if (match_int(&args[0], &option)) {
 778                                 status = 0;
 779                                 goto bail;
 780                         }
 781                         if (option)
 782                                 mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
 783                         else
 784                                 mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
 785                         break;
 786                 case Opt_intr:
 787                         mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
 788                         break;
 789                 case Opt_nointr:
 790                         mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
 791                         break;
 792                 case Opt_err_panic:
 793                         mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 794                         break;
 795                 case Opt_err_ro:
 796                         mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 797                         break;
 798                 case Opt_data_ordered:
 799                         mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 800                         break;
 801                 case Opt_data_writeback:
 802                         mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 803                         break;
 804                 case Opt_atime_quantum:
 805                         if (match_int(&args[0], &option)) {
 806                                 status = 0;
 807                                 goto bail;
 808                         }
 809                         if (option >= 0)
 810                                 mopt->atime_quantum = option;
 811                         break;
 812                 case Opt_slot:
 813                         option = 0;
 814                         if (match_int(&args[0], &option)) {
 815                                 status = 0;
 816                                 goto bail;
 817                         }
 818                         if (option)
 819                                 mopt->slot = (s16)option;
 820                         break;
 821                 case Opt_commit:
 822                         option = 0;
 823                         if (match_int(&args[0], &option)) {
 824                                 status = 0;
 825                                 goto bail;
 826                         }
 827                         if (option < 0)
 828                                 return 0;
 829                         if (option == 0)
 830                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 831                         mopt->commit_interval = HZ * option;
 832                         break;
 833                 case Opt_localalloc:
 834                         option = 0;
 835                         if (match_int(&args[0], &option)) {
 836                                 status = 0;
 837                                 goto bail;
 838                         }
 839                         if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
 840                                 mopt->localalloc_opt = option;
 841                         break;
 842                 case Opt_localflocks:
 843                         /*
 844                          * Changing this during remount could race
 845                          * flock() requests, or "unbalance" existing
 846                          * ones (e.g., a lock is taken in one mode but
 847                          * dropped in the other). If users care enough
 848                          * to flip locking modes during remount, we
 849                          * could add a "local" flag to individual
 850                          * flock structures for proper tracking of
 851                          * state.
 852                          */
 853                         if (!is_remount)
 854                                 mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
 855                         break;
 856                 default:
 857                         mlog(ML_ERROR,
 858                              "Unrecognized mount option \"%s\" "
 859                              "or missing value\n", p);
 860                         status = 0;
 861                         goto bail;
 862                 }
 863         }
 864
 865         status = 1;
 866
 867 bail:
 868         mlog_exit(status);
 869         return status;
 870 }
 871
 872 static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 873 {
 874         struct ocfs2_super *osb = OCFS2_SB(mnt->mnt_sb);
 875         unsigned long opts = osb->s_mount_opt;
 876
 877         if (opts & OCFS2_MOUNT_HB_LOCAL)
 878                 seq_printf(s, ",_netdev,heartbeat=local");
 879         else
 880                 seq_printf(s, ",heartbeat=none");
 881
 882         if (opts & OCFS2_MOUNT_NOINTR)
 883                 seq_printf(s, ",nointr");
 884
 885         if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
 886                 seq_printf(s, ",data=writeback");
 887         else
 888                 seq_printf(s, ",data=ordered");
 889
 890         if (opts & OCFS2_MOUNT_BARRIER)
 891                 seq_printf(s, ",barrier=1");
 892
 893         if (opts & OCFS2_MOUNT_ERRORS_PANIC)
 894                 seq_printf(s, ",errors=panic");
 895         else
 896                 seq_printf(s, ",errors=remount-ro");
 897
 898         if (osb->preferred_slot != OCFS2_INVALID_SLOT)
 899                 seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
 900
 901         if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
 902                 seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
 903
 904         if (osb->osb_commit_interval)
 905                 seq_printf(s, ",commit=%u",
 906                            (unsigned) (osb->osb_commit_interval / HZ));
 907
 908         if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
 909                 seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
 910
 911         if (opts & OCFS2_MOUNT_LOCALFLOCKS)
 912                 seq_printf(s, ",localflocks,");
 913
 914         return 0;
 915 }
 916
 917 static int __init ocfs2_init(void)
 918 {
 919         int status;
 920
 921         mlog_entry_void();
 922
 923         ocfs2_print_version();
 924
 925         status = init_ocfs2_uptodate_cache();
 926         if (status < 0) {
 927                 mlog_errno(status);
 928                 goto leave;
 929         }
 930
 931         status = ocfs2_initialize_mem_caches();
 932         if (status < 0) {
 933                 mlog_errno(status);
 934                 goto leave;
 935         }
 936
 937         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
 938         if (!ocfs2_wq) {
 939                 status = -ENOMEM;
 940                 goto leave;
 941         }
 942
 943         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
 944         if (!ocfs2_debugfs_root) {
 945                 status = -EFAULT;
 946                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
 947         }
 948
 949         ocfs2_set_locking_protocol();
 950
 951 leave:
 952         if (status < 0) {
 953                 ocfs2_free_mem_caches();
 954                 exit_ocfs2_uptodate_cache();
 955         }
 956
 957         mlog_exit(status);
 958
 959         if (status >= 0) {
 960                 return register_filesystem(&ocfs2_fs_type);
 961         } else
 962                 return -1;
 963 }
 964
 965 static void __exit ocfs2_exit(void)
 966 {
 967         mlog_entry_void();
 968
 969         if (ocfs2_wq) {
 970                 flush_workqueue(ocfs2_wq);
 971                 destroy_workqueue(ocfs2_wq);
 972         }
 973
 974         debugfs_remove(ocfs2_debugfs_root);
 975
 976         ocfs2_free_mem_caches();
 977
 978         unregister_filesystem(&ocfs2_fs_type);
 979
 980         exit_ocfs2_uptodate_cache();
 981
 982         mlog_exit_void();
 983 }
 984
 985 static void ocfs2_put_super(struct super_block *sb)
 986 {
 987         mlog_entry("(0x%p)\n", sb);
 988
 989         ocfs2_sync_blockdev(sb);
 990         ocfs2_dismount_volume(sb, 0);
 991
 992         mlog_exit_void();
 993 }
 994
 995 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 996 {
 997         struct ocfs2_super *osb;
 998         u32 numbits, freebits;
 999         int status;
1000         struct ocfs2_dinode *bm_lock;
1001         struct buffer_head *bh = NULL;
1002         struct inode *inode = NULL;
1003
1004         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
1005
1006         osb = OCFS2_SB(dentry->d_sb);
1007
1008         inode = ocfs2_get_system_file_inode(osb,
1009                                             GLOBAL_BITMAP_SYSTEM_INODE,
1010                                             OCFS2_INVALID_SLOT);
1011         if (!inode) {
1012                 mlog(ML_ERROR, "failed to get bitmap inode\n");
1013                 status = -EIO;
1014                 goto bail;
1015         }
1016
1017         status = ocfs2_inode_lock(inode, &bh, 0);
1018         if (status < 0) {
1019                 mlog_errno(status);
1020                 goto bail;
1021         }
1022
1023         bm_lock = (struct ocfs2_dinode *) bh->b_data;
1024
1025         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
1026         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
1027
1028         buf->f_type = OCFS2_SUPER_MAGIC;
1029         buf->f_bsize = dentry->d_sb->s_blocksize;
1030         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
1031         buf->f_blocks = ((sector_t) numbits) *
1032                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1033         buf->f_bfree = ((sector_t) freebits) *
1034                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
1035         buf->f_bavail = buf->f_bfree;
1036         buf->f_files = numbits;
1037         buf->f_ffree = freebits;
1038
1039         brelse(bh);
1040
1041         ocfs2_inode_unlock(inode, 0);
1042         status = 0;
1043 bail:
1044         if (inode)
1045                 iput(inode);
1046
1047         mlog_exit(status);
1048
1049         return status;
1050 }
1051
1052 static void ocfs2_inode_init_once(struct kmem_cache *cachep, void *data)
1053 {
1054         struct ocfs2_inode_info *oi = data;
1055
1056         oi->ip_flags = 0;
1057         oi->ip_open_count = 0;
1058         spin_lock_init(&oi->ip_lock);
1059         ocfs2_extent_map_init(&oi->vfs_inode);
1060         INIT_LIST_HEAD(&oi->ip_io_markers);
1061         oi->ip_created_trans = 0;
1062         oi->ip_last_trans = 0;
1063         oi->ip_dir_start_lookup = 0;
1064
1065         init_rwsem(&oi->ip_alloc_sem);
1066         mutex_init(&oi->ip_io_mutex);
1067
1068         oi->ip_blkno = 0ULL;
1069         oi->ip_clusters = 0;
1070
1071         ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
1072         ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
1073         ocfs2_lock_res_init_once(&oi->ip_open_lockres);
1074
1075         ocfs2_metadata_cache_init(&oi->vfs_inode);
1076
1077         inode_init_once(&oi->vfs_inode);
1078 }
1079
1080 static int ocfs2_initialize_mem_caches(void)
1081 {
1082         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
1083                                        sizeof(struct ocfs2_inode_info),
1084                                        0,
1085                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
1086                                                 SLAB_MEM_SPREAD),
1087                                        ocfs2_inode_init_once);
1088         if (!ocfs2_inode_cachep)
1089                 return -ENOMEM;
1090
1091         return 0;
1092 }
1093
1094 static void ocfs2_free_mem_caches(void)
1095 {
1096         if (ocfs2_inode_cachep)
1097                 kmem_cache_destroy(ocfs2_inode_cachep);
1098
1099         ocfs2_inode_cachep = NULL;
1100 }
1101
1102 static int ocfs2_get_sector(struct super_block *sb,
1103                             struct buffer_head **bh,
1104                             int block,
1105                             int sect_size)
1106 {
1107         if (!sb_set_blocksize(sb, sect_size)) {
1108                 mlog(ML_ERROR, "unable to set blocksize\n");
1109                 return -EIO;
1110         }
1111
1112         *bh = sb_getblk(sb, block);
1113         if (!*bh) {
1114                 mlog_errno(-EIO);
1115                 return -EIO;
1116         }
1117         lock_buffer(*bh);
1118         if (!buffer_dirty(*bh))
1119                 clear_buffer_uptodate(*bh);
1120         unlock_buffer(*bh);
1121         ll_rw_block(READ, 1, bh);
1122         wait_on_buffer(*bh);
1123         return 0;
1124 }
1125
1126 static int ocfs2_mount_volume(struct super_block *sb)
1127 {
1128         int status = 0;
1129         int unlock_super = 0;
1130         struct ocfs2_super *osb = OCFS2_SB(sb);
1131
1132         mlog_entry_void();
1133
1134         if (ocfs2_is_hard_readonly(osb))
1135                 goto leave;
1136
1137         status = ocfs2_dlm_init(osb);
1138         if (status < 0) {
1139                 mlog_errno(status);
1140                 goto leave;
1141         }
1142
1143         status = ocfs2_super_lock(osb, 1);
1144         if (status < 0) {
1145                 mlog_errno(status);
1146                 goto leave;
1147         }
1148         unlock_super = 1;
1149
1150         /* This will load up the node map and add ourselves to it. */
1151         status = ocfs2_find_slot(osb);
1152         if (status < 0) {
1153                 mlog_errno(status);
1154                 goto leave;
1155         }
1156
1157         /* load all node-local system inodes */
1158         status = ocfs2_init_local_system_inodes(osb);
1159         if (status < 0) {
1160                 mlog_errno(status);
1161                 goto leave;
1162         }
1163
1164         status = ocfs2_check_volume(osb);
1165         if (status < 0) {
1166                 mlog_errno(status);
1167                 goto leave;
1168         }
1169
1170         status = ocfs2_truncate_log_init(osb);
1171         if (status < 0) {
1172                 mlog_errno(status);
1173                 goto leave;
1174         }
1175
1176         if (ocfs2_mount_local(osb))
1177                 goto leave;
1178
1179 leave:
1180         if (unlock_super)
1181                 ocfs2_super_unlock(osb, 1);
1182
1183         mlog_exit(status);
1184         return status;
1185 }
1186
1187 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1188 {
1189         int tmp, hangup_needed = 0;
1190         struct ocfs2_super *osb = NULL;
1191         char nodestr[8];
1192
1193         mlog_entry("(0x%p)\n", sb);
1194
1195         BUG_ON(!sb);
1196         osb = OCFS2_SB(sb);
1197         BUG_ON(!osb);
1198
1199         ocfs2_shutdown_local_alloc(osb);
1200
1201         ocfs2_truncate_log_shutdown(osb);
1202
1203         /* This will disable recovery and flush any recovery work. */
1204         ocfs2_recovery_exit(osb);
1205
1206         ocfs2_journal_shutdown(osb);
1207
1208         ocfs2_sync_blockdev(sb);
1209
1210         /* No cluster connection means we've failed during mount, so skip
1211          * all the steps which depended on that to complete. */
1212         if (osb->cconn) {
1213                 tmp = ocfs2_super_lock(osb, 1);
1214                 if (tmp < 0) {
1215                         mlog_errno(tmp);
1216                         return;
1217                 }
1218         }
1219
1220         if (osb->slot_num != OCFS2_INVALID_SLOT)
1221                 ocfs2_put_slot(osb);
1222
1223         if (osb->cconn)
1224                 ocfs2_super_unlock(osb, 1);
1225
1226         ocfs2_release_system_inodes(osb);
1227
1228         /*
1229          * If we're dismounting due to mount error, mount.ocfs2 will clean
1230          * up heartbeat.  If we're a local mount, there is no heartbeat.
1231          * If we failed before we got a uuid_str yet, we can't stop
1232          * heartbeat.  Otherwise, do it.
1233          */
1234         if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str)
1235                 hangup_needed = 1;
1236
1237         if (osb->cconn)
1238                 ocfs2_dlm_shutdown(osb, hangup_needed);
1239
1240         debugfs_remove(osb->osb_debug_root);
1241
1242         if (hangup_needed)
1243                 ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
1244
1245         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1246
1247         if (ocfs2_mount_local(osb))
1248                 snprintf(nodestr, sizeof(nodestr), "local");
1249         else
1250                 snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
1251
1252         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1253                osb->dev_str, nodestr);
1254
1255         ocfs2_delete_osb(osb);
1256         kfree(osb);
1257         sb->s_dev = 0;
1258         sb->s_fs_info = NULL;
1259 }
1260
1261 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1262                                 unsigned uuid_bytes)
1263 {
1264         int i, ret;
1265         char *ptr;
1266
1267         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1268
1269         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1270         if (osb->uuid_str == NULL)
1271                 return -ENOMEM;
1272
1273         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1274                 /* print with null */
1275                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1276                 if (ret != 2) /* drop super cleans up */
1277                         return -EINVAL;
1278                 /* then only advance past the last char */
1279                 ptr += 2;
1280         }
1281
1282         return 0;
1283 }
1284
1285 static int ocfs2_initialize_super(struct super_block *sb,
1286                                   struct buffer_head *bh,
1287                                   int sector_size)
1288 {
1289         int status;
1290         int i, cbits, bbits;
1291         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1292         struct inode *inode = NULL;
1293         struct ocfs2_journal *journal;
1294         __le32 uuid_net_key;
1295         struct ocfs2_super *osb;
1296
1297         mlog_entry_void();
1298
1299         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1300         if (!osb) {
1301                 status = -ENOMEM;
1302                 mlog_errno(status);
1303                 goto bail;
1304         }
1305
1306         sb->s_fs_info = osb;
1307         sb->s_op = &ocfs2_sops;
1308         sb->s_export_op = &ocfs2_export_ops;
1309         sb->s_time_gran = 1;
1310         sb->s_flags |= MS_NOATIME;
1311         /* this is needed to support O_LARGEFILE */
1312         cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1313         bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
1314         sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
1315
1316         osb->sb = sb;
1317         /* Save off for ocfs2_rw_direct */
1318         osb->s_sectsize_bits = blksize_bits(sector_size);
1319         BUG_ON(!osb->s_sectsize_bits);
1320
1321         spin_lock_init(&osb->dc_task_lock);
1322         init_waitqueue_head(&osb->dc_event);
1323         osb->dc_work_sequence = 0;
1324         osb->dc_wake_sequence = 0;
1325         INIT_LIST_HEAD(&osb->blocked_lock_list);
1326         osb->blocked_lock_count = 0;
1327         spin_lock_init(&osb->osb_lock);
1328
1329         atomic_set(&osb->alloc_stats.moves, 0);
1330         atomic_set(&osb->alloc_stats.local_data, 0);
1331         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1332         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1333         atomic_set(&osb->alloc_stats.bg_extends, 0);
1334
1335         ocfs2_init_node_maps(osb);
1336
1337         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1338                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1339
1340         status = ocfs2_recovery_init(osb);
1341         if (status) {
1342                 mlog(ML_ERROR, "Unable to initialize recovery state\n");
1343                 mlog_errno(status);
1344                 goto bail;
1345         }
1346
1347         init_waitqueue_head(&osb->checkpoint_event);
1348         atomic_set(&osb->needs_checkpoint, 0);
1349
1350         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1351
1352         osb->slot_num = OCFS2_INVALID_SLOT;
1353
1354         osb->local_alloc_state = OCFS2_LA_UNUSED;
1355         osb->local_alloc_bh = NULL;
1356
1357         init_waitqueue_head(&osb->osb_mount_event);
1358
1359         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1360         if (!osb->vol_label) {
1361                 mlog(ML_ERROR, "unable to alloc vol label\n");
1362                 status = -ENOMEM;
1363                 goto bail;
1364         }
1365
1366         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1367         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1368                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1369                      osb->max_slots);
1370                 status = -EINVAL;
1371                 goto bail;
1372         }
1373         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1374
1375         init_waitqueue_head(&osb->osb_wipe_event);
1376         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1377                                         sizeof(*osb->osb_orphan_wipes),
1378                                         GFP_KERNEL);
1379         if (!osb->osb_orphan_wipes) {
1380                 status = -ENOMEM;
1381                 mlog_errno(status);
1382                 goto bail;
1383         }
1384
1385         osb->s_feature_compat =
1386                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1387         osb->s_feature_ro_compat =
1388                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1389         osb->s_feature_incompat =
1390                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1391
1392         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1393                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1394                      "optional features (%x).\n", i);
1395                 status = -EINVAL;
1396                 goto bail;
1397         }
1398         if (!(osb->sb->s_flags & MS_RDONLY) &&
1399             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1400                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1401                      "unsupported optional features (%x).\n", i);
1402                 status = -EINVAL;
1403                 goto bail;
1404         }
1405
1406         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1407
1408         /* FIXME
1409          * This should be done in ocfs2_journal_init(), but unknown
1410          * ordering issues will cause the filesystem to crash.
1411          * If anyone wants to figure out what part of the code
1412          * refers to osb->journal before ocfs2_journal_init() is run,
1413          * be my guest.
1414          */
1415         /* initialize our journal structure */
1416
1417         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1418         if (!journal) {
1419                 mlog(ML_ERROR, "unable to alloc journal\n");
1420                 status = -ENOMEM;
1421                 goto bail;
1422         }
1423         osb->journal = journal;
1424         journal->j_osb = osb;
1425
1426         atomic_set(&journal->j_num_trans, 0);
1427         init_rwsem(&journal->j_trans_barrier);
1428         init_waitqueue_head(&journal->j_checkpointed);
1429         spin_lock_init(&journal->j_lock);
1430         journal->j_trans_id = (unsigned long) 1;
1431         INIT_LIST_HEAD(&journal->j_la_cleanups);
1432         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1433         journal->j_state = OCFS2_JOURNAL_FREE;
1434
1435         /* get some pseudo constants for clustersize bits */
1436         osb->s_clustersize_bits =
1437                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1438         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1439         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1440
1441         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1442             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1443                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1444                      osb->s_clustersize);
1445                 status = -EINVAL;
1446                 goto bail;
1447         }
1448
1449         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1450             > (u32)~0UL) {
1451                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1452                      "what jbd can address in 32 bits.\n");
1453                 status = -EINVAL;
1454                 goto bail;
1455         }
1456
1457         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1458                                  sizeof(di->id2.i_super.s_uuid))) {
1459                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1460                 status = -ENOMEM;
1461                 goto bail;
1462         }
1463
1464         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1465
1466         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1467         osb->vol_label[63] = '\0';
1468         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1469         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1470         osb->first_cluster_group_blkno =
1471                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1472         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1473         mlog(0, "vol_label: %s\n", osb->vol_label);
1474         mlog(0, "uuid: %s\n", osb->uuid_str);
1475         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1476              (unsigned long long)osb->root_blkno,
1477              (unsigned long long)osb->system_dir_blkno);
1478
1479         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1480         if (!osb->osb_dlm_debug) {
1481                 status = -ENOMEM;
1482                 mlog_errno(status);
1483                 goto bail;
1484         }
1485
1486         atomic_set(&osb->vol_state, VOLUME_INIT);
1487
1488         /* load root, system_dir, and all global system inodes */
1489         status = ocfs2_init_global_system_inodes(osb);
1490         if (status < 0) {
1491                 mlog_errno(status);
1492                 goto bail;
1493         }
1494
1495         /*
1496          * global bitmap
1497          */
1498         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1499                                             OCFS2_INVALID_SLOT);
1500         if (!inode) {
1501                 status = -EINVAL;
1502                 mlog_errno(status);
1503                 goto bail;
1504         }
1505
1506         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1507         iput(inode);
1508
1509         osb->bitmap_cpg = ocfs2_group_bitmap_size(sb) * 8;
1510
1511         status = ocfs2_init_slot_info(osb);
1512         if (status < 0) {
1513                 mlog_errno(status);
1514                 goto bail;
1515         }
1516
1517 bail:
1518         mlog_exit(status);
1519         return status;
1520 }
1521
1522 /*
1523  * will return: -EAGAIN if it is ok to keep searching for superblocks
1524  *              -EINVAL if there is a bad superblock
1525  *              0 on success
1526  */
1527 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1528                                struct buffer_head *bh,
1529                                u32 blksz)
1530 {
1531         int status = -EAGAIN;
1532
1533         mlog_entry_void();
1534
1535         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1536                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1537                 status = -EINVAL;
1538                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1539                         mlog(ML_ERROR, "found superblock with incorrect block "
1540                              "size: found %u, should be %u\n",
1541                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1542                                blksz);
1543                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1544                            OCFS2_MAJOR_REV_LEVEL ||
1545                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1546                            OCFS2_MINOR_REV_LEVEL) {
1547                         mlog(ML_ERROR, "found superblock with bad version: "
1548                              "found %u.%u, should be %u.%u\n",
1549                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1550                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1551                              OCFS2_MAJOR_REV_LEVEL,
1552                              OCFS2_MINOR_REV_LEVEL);
1553                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1554                         mlog(ML_ERROR, "bad block number on superblock: "
1555                              "found %llu, should be %llu\n",
1556                              (unsigned long long)le64_to_cpu(di->i_blkno),
1557                              (unsigned long long)bh->b_blocknr);
1558                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1559                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1560                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1561                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1562                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1563                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1564                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1565                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1566                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1567                         mlog(ML_ERROR,
1568                              "Superblock slots found greater than file system "
1569                              "maximum: found %u, max %u\n",
1570                              le16_to_cpu(di->id2.i_super.s_max_slots),
1571                              OCFS2_MAX_SLOTS);
1572                 } else {
1573                         /* found it! */
1574                         status = 0;
1575                 }
1576         }
1577
1578         mlog_exit(status);
1579         return status;
1580 }
1581
1582 static int ocfs2_check_volume(struct ocfs2_super *osb)
1583 {
1584         int status;
1585         int dirty;
1586         int local;
1587         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1588                                                   * recover
1589                                                   * ourselves. */
1590
1591         mlog_entry_void();
1592
1593         /* Init our journal object. */
1594         status = ocfs2_journal_init(osb->journal, &dirty);
1595         if (status < 0) {
1596                 mlog(ML_ERROR, "Could not initialize journal!\n");
1597                 goto finally;
1598         }
1599
1600         /* If the journal was unmounted cleanly then we don't want to
1601          * recover anything. Otherwise, journal_load will do that
1602          * dirty work for us :) */
1603         if (!dirty) {
1604                 status = ocfs2_journal_wipe(osb->journal, 0);
1605                 if (status < 0) {
1606                         mlog_errno(status);
1607                         goto finally;
1608                 }
1609         } else {
1610                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1611                      "recovering volume.\n");
1612         }
1613
1614         local = ocfs2_mount_local(osb);
1615
1616         /* will play back anything left in the journal. */
1617         ocfs2_journal_load(osb->journal, local);
1618
1619         if (dirty) {
1620                 /* recover my local alloc if we didn't unmount cleanly. */
1621                 status = ocfs2_begin_local_alloc_recovery(osb,
1622                                                           osb->slot_num,
1623                                                           &local_alloc);
1624                 if (status < 0) {
1625                         mlog_errno(status);
1626                         goto finally;
1627                 }
1628                 /* we complete the recovery process after we've marked
1629                  * ourselves as mounted. */
1630         }
1631
1632         mlog(0, "Journal loaded.\n");
1633
1634         status = ocfs2_load_local_alloc(osb);
1635         if (status < 0) {
1636                 mlog_errno(status);
1637                 goto finally;
1638         }
1639
1640         if (dirty) {
1641                 /* Recovery will be completed after we've mounted the
1642                  * rest of the volume. */
1643                 osb->dirty = 1;
1644                 osb->local_alloc_copy = local_alloc;
1645                 local_alloc = NULL;
1646         }
1647
1648         /* go through each journal, trylock it and if you get the
1649          * lock, and it's marked as dirty, set the bit in the recover
1650          * map and launch a recovery thread for it. */
1651         status = ocfs2_mark_dead_nodes(osb);
1652         if (status < 0)
1653                 mlog_errno(status);
1654
1655 finally:
1656         if (local_alloc)
1657                 kfree(local_alloc);
1658
1659         mlog_exit(status);
1660         return status;
1661 }
1662
1663 /*
1664  * The routine gets called from dismount or close whenever a dismount on
1665  * volume is requested and the osb open count becomes 1.
1666  * It will remove the osb from the global list and also free up all the
1667  * initialized resources and fileobject.
1668  */
1669 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1670 {
1671         mlog_entry_void();
1672
1673         /* This function assumes that the caller has the main osb resource */
1674
1675         ocfs2_free_slot_info(osb);
1676
1677         kfree(osb->osb_orphan_wipes);
1678         /* FIXME
1679          * This belongs in journal shutdown, but because we have to
1680          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1681          * we free it here.
1682          */
1683         kfree(osb->journal);
1684         if (osb->local_alloc_copy)
1685                 kfree(osb->local_alloc_copy);
1686         kfree(osb->uuid_str);
1687         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1688         memset(osb, 0, sizeof(struct ocfs2_super));
1689
1690         mlog_exit_void();
1691 }
1692
1693 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1694  * panic(). We do not support continue-on-error operation. */
1695 static void ocfs2_handle_error(struct super_block *sb)
1696 {
1697         struct ocfs2_super *osb = OCFS2_SB(sb);
1698
1699         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1700                 panic("OCFS2: (device %s): panic forced after error\n",
1701                       sb->s_id);
1702
1703         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1704
1705         if (sb->s_flags & MS_RDONLY &&
1706             (ocfs2_is_soft_readonly(osb) ||
1707              ocfs2_is_hard_readonly(osb)))
1708                 return;
1709
1710         printk(KERN_CRIT "File system is now read-only due to the potential "
1711                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1712                "system is unmounted.\n");
1713         sb->s_flags |= MS_RDONLY;
1714         ocfs2_set_ro_flag(osb, 0);
1715 }
1716
1717 static char error_buf[1024];
1718
1719 void __ocfs2_error(struct super_block *sb,
1720                    const char *function,
1721                    const char *fmt, ...)
1722 {
1723         va_list args;
1724
1725         va_start(args, fmt);
1726         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1727         va_end(args);
1728
1729         /* Not using mlog here because we want to show the actual
1730          * function the error came from. */
1731         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1732                sb->s_id, function, error_buf);
1733
1734         ocfs2_handle_error(sb);
1735 }
1736
1737 /* Handle critical errors. This is intentionally more drastic than
1738  * ocfs2_handle_error, so we only use for things like journal errors,
1739  * etc. */
1740 void __ocfs2_abort(struct super_block* sb,
1741                    const char *function,
1742                    const char *fmt, ...)
1743 {
1744         va_list args;
1745
1746         va_start(args, fmt);
1747         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1748         va_end(args);
1749
1750         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1751                sb->s_id, function, error_buf);
1752
1753         /* We don't have the cluster support yet to go straight to
1754          * hard readonly in here. Until then, we want to keep
1755          * ocfs2_abort() so that we can at least mark critical
1756          * errors.
1757          *
1758          * TODO: This should abort the journal and alert other nodes
1759          * that our slot needs recovery. */
1760
1761         /* Force a panic(). This stinks, but it's better than letting
1762          * things continue without having a proper hard readonly
1763          * here. */
1764         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1765         ocfs2_handle_error(sb);
1766 }
1767
1768 module_init(ocfs2_init);
1769 module_exit(ocfs2_exit);