SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ocfs2/super.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * super.c
   5  *
   6  * load/unload driver, mount/dismount volumes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2 of the License, or (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public
  21  * License along with this program; if not, write to the
  22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  23  * Boston, MA 021110-1307, USA.
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31 #include <linux/utsname.h>
  32 #include <linux/init.h>
  33 #include <linux/random.h>
  34 #include <linux/statfs.h>
  35 #include <linux/moduleparam.h>
  36 #include <linux/blkdev.h>
  37 #include <linux/socket.h>
  38 #include <linux/inet.h>
  39 #include <linux/parser.h>
  40 #include <linux/crc32.h>
  41 #include <linux/debugfs.h>
  42
  43 #include <cluster/nodemanager.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 #include "vote.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 static int ocfs2_parse_options(struct super_block *sb, char *options,
  85                                unsigned long *mount_opt, int is_remount);
  86 static void ocfs2_put_super(struct super_block *sb);
  87 static int ocfs2_mount_volume(struct super_block *sb);
  88 static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
  89 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
  90 static int ocfs2_initialize_mem_caches(void);
  91 static void ocfs2_free_mem_caches(void);
  92 static void ocfs2_delete_osb(struct ocfs2_super *osb);
  93
  94 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
  95
  96 static int ocfs2_sync_fs(struct super_block *sb, int wait);
  97
  98 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
  99 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
 100 static int ocfs2_release_system_inodes(struct ocfs2_super *osb);
 101 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb);
 102 static int ocfs2_check_volume(struct ocfs2_super *osb);
 103 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
 104                                struct buffer_head *bh,
 105                                u32 sectsize);
 106 static int ocfs2_initialize_super(struct super_block *sb,
 107                                   struct buffer_head *bh,
 108                                   int sector_size);
 109 static int ocfs2_get_sector(struct super_block *sb,
 110                             struct buffer_head **bh,
 111                             int block,
 112                             int sect_size);
 113 static void ocfs2_write_super(struct super_block *sb);
 114 static struct inode *ocfs2_alloc_inode(struct super_block *sb);
 115 static void ocfs2_destroy_inode(struct inode *inode);
 116
 117 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift);
 118
 119 static const struct super_operations ocfs2_sops = {
 120         .statfs         = ocfs2_statfs,
 121         .alloc_inode    = ocfs2_alloc_inode,
 122         .destroy_inode  = ocfs2_destroy_inode,
 123         .drop_inode     = ocfs2_drop_inode,
 124         .clear_inode    = ocfs2_clear_inode,
 125         .delete_inode   = ocfs2_delete_inode,
 126         .sync_fs        = ocfs2_sync_fs,
 127         .write_super    = ocfs2_write_super,
 128         .put_super      = ocfs2_put_super,
 129         .remount_fs     = ocfs2_remount,
 130 };
 131
 132 enum {
 133         Opt_barrier,
 134         Opt_err_panic,
 135         Opt_err_ro,
 136         Opt_intr,
 137         Opt_nointr,
 138         Opt_hb_none,
 139         Opt_hb_local,
 140         Opt_data_ordered,
 141         Opt_data_writeback,
 142         Opt_atime_quantum,
 143         Opt_err,
 144 };
 145
 146 static match_table_t tokens = {
 147         {Opt_barrier, "barrier=%u"},
 148         {Opt_err_panic, "errors=panic"},
 149         {Opt_err_ro, "errors=remount-ro"},
 150         {Opt_intr, "intr"},
 151         {Opt_nointr, "nointr"},
 152         {Opt_hb_none, OCFS2_HB_NONE},
 153         {Opt_hb_local, OCFS2_HB_LOCAL},
 154         {Opt_data_ordered, "data=ordered"},
 155         {Opt_data_writeback, "data=writeback"},
 156         {Opt_atime_quantum, "atime_quantum=%u"},
 157         {Opt_err, NULL}
 158 };
 159
 160 /*
 161  * write_super and sync_fs ripped right out of ext3.
 162  */
 163 static void ocfs2_write_super(struct super_block *sb)
 164 {
 165         if (mutex_trylock(&sb->s_lock) != 0)
 166                 BUG();
 167         sb->s_dirt = 0;
 168 }
 169
 170 static int ocfs2_sync_fs(struct super_block *sb, int wait)
 171 {
 172         int status = 0;
 173         tid_t target;
 174         struct ocfs2_super *osb = OCFS2_SB(sb);
 175
 176         sb->s_dirt = 0;
 177
 178         if (ocfs2_is_hard_readonly(osb))
 179                 return -EROFS;
 180
 181         if (wait) {
 182                 status = ocfs2_flush_truncate_log(osb);
 183                 if (status < 0)
 184                         mlog_errno(status);
 185         } else {
 186                 ocfs2_schedule_truncate_log_flush(osb, 0);
 187         }
 188
 189         if (journal_start_commit(OCFS2_SB(sb)->journal->j_journal, &target)) {
 190                 if (wait)
 191                         log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
 192                                         target);
 193         }
 194         return 0;
 195 }
 196
 197 static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 198 {
 199         struct inode *new = NULL;
 200         int status = 0;
 201         int i;
 202
 203         mlog_entry_void();
 204
 205         new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
 206         if (IS_ERR(new)) {
 207                 status = PTR_ERR(new);
 208                 mlog_errno(status);
 209                 goto bail;
 210         }
 211         osb->root_inode = new;
 212
 213         new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
 214         if (IS_ERR(new)) {
 215                 status = PTR_ERR(new);
 216                 mlog_errno(status);
 217                 goto bail;
 218         }
 219         osb->sys_root_inode = new;
 220
 221         for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
 222              i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
 223                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 224                 if (!new) {
 225                         ocfs2_release_system_inodes(osb);
 226                         status = -EINVAL;
 227                         mlog_errno(status);
 228                         /* FIXME: Should ERROR_RO_FS */
 229                         mlog(ML_ERROR, "Unable to load system inode %d, "
 230                              "possibly corrupt fs?", i);
 231                         goto bail;
 232                 }
 233                 // the array now has one ref, so drop this one
 234                 iput(new);
 235         }
 236
 237 bail:
 238         mlog_exit(status);
 239         return status;
 240 }
 241
 242 static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
 243 {
 244         struct inode *new = NULL;
 245         int status = 0;
 246         int i;
 247
 248         mlog_entry_void();
 249
 250         for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
 251              i < NUM_SYSTEM_INODES;
 252              i++) {
 253                 new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
 254                 if (!new) {
 255                         ocfs2_release_system_inodes(osb);
 256                         status = -EINVAL;
 257                         mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
 258                              status, i, osb->slot_num);
 259                         goto bail;
 260                 }
 261                 /* the array now has one ref, so drop this one */
 262                 iput(new);
 263         }
 264
 265 bail:
 266         mlog_exit(status);
 267         return status;
 268 }
 269
 270 static int ocfs2_release_system_inodes(struct ocfs2_super *osb)
 271 {
 272         int status = 0, i;
 273         struct inode *inode;
 274
 275         mlog_entry_void();
 276
 277         for (i = 0; i < NUM_SYSTEM_INODES; i++) {
 278                 inode = osb->system_inodes[i];
 279                 if (inode) {
 280                         iput(inode);
 281                         osb->system_inodes[i] = NULL;
 282                 }
 283         }
 284
 285         inode = osb->sys_root_inode;
 286         if (inode) {
 287                 iput(inode);
 288                 osb->sys_root_inode = NULL;
 289         }
 290
 291         inode = osb->root_inode;
 292         if (inode) {
 293                 iput(inode);
 294                 osb->root_inode = NULL;
 295         }
 296
 297         mlog_exit(status);
 298         return status;
 299 }
 300
 301 /* We're allocating fs objects, use GFP_NOFS */
 302 static struct inode *ocfs2_alloc_inode(struct super_block *sb)
 303 {
 304         struct ocfs2_inode_info *oi;
 305
 306         oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
 307         if (!oi)
 308                 return NULL;
 309
 310         return &oi->vfs_inode;
 311 }
 312
 313 static void ocfs2_destroy_inode(struct inode *inode)
 314 {
 315         kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
 316 }
 317
 318 /* From xfs_super.c:xfs_max_file_offset
 319  * Copyright (c) 2000-2004 Silicon Graphics, Inc.
 320  */
 321 static unsigned long long ocfs2_max_file_offset(unsigned int blockshift)
 322 {
 323         unsigned int pagefactor = 1;
 324         unsigned int bitshift = BITS_PER_LONG - 1;
 325
 326         /* Figure out maximum filesize, on Linux this can depend on
 327          * the filesystem blocksize (on 32 bit platforms).
 328          * __block_prepare_write does this in an [unsigned] long...
 329          *      page->index << (PAGE_CACHE_SHIFT - bbits)
 330          * So, for page sized blocks (4K on 32 bit platforms),
 331          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
 332          *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
 333          * but for smaller blocksizes it is less (bbits = log2 bsize).
 334          * Note1: get_block_t takes a long (implicit cast from above)
 335          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
 336          * can optionally convert the [unsigned] long from above into
 337          * an [unsigned] long long.
 338          */
 339
 340 #if BITS_PER_LONG == 32
 341 # if defined(CONFIG_LBD)
 342         BUILD_BUG_ON(sizeof(sector_t) != 8);
 343         pagefactor = PAGE_CACHE_SIZE;
 344         bitshift = BITS_PER_LONG;
 345 # else
 346         pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
 347 # endif
 348 #endif
 349
 350         return (((unsigned long long)pagefactor) << bitshift) - 1;
 351 }
 352
 353 static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
 354 {
 355         int incompat_features;
 356         int ret = 0;
 357         unsigned long parsed_options;
 358         struct ocfs2_super *osb = OCFS2_SB(sb);
 359
 360         if (!ocfs2_parse_options(sb, data, &parsed_options, 1)) {
 361                 ret = -EINVAL;
 362                 goto out;
 363         }
 364
 365         if ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) !=
 366             (parsed_options & OCFS2_MOUNT_HB_LOCAL)) {
 367                 ret = -EINVAL;
 368                 mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
 369                 goto out;
 370         }
 371
 372         if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
 373             (parsed_options & OCFS2_MOUNT_DATA_WRITEBACK)) {
 374                 ret = -EINVAL;
 375                 mlog(ML_ERROR, "Cannot change data mode on remount\n");
 376                 goto out;
 377         }
 378
 379         /* We're going to/from readonly mode. */
 380         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
 381                 /* Lock here so the check of HARD_RO and the potential
 382                  * setting of SOFT_RO is atomic. */
 383                 spin_lock(&osb->osb_lock);
 384                 if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
 385                         mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
 386                         ret = -EROFS;
 387                         goto unlock_osb;
 388                 }
 389
 390                 if (*flags & MS_RDONLY) {
 391                         mlog(0, "Going to ro mode.\n");
 392                         sb->s_flags |= MS_RDONLY;
 393                         osb->osb_flags |= OCFS2_OSB_SOFT_RO;
 394                 } else {
 395                         mlog(0, "Making ro filesystem writeable.\n");
 396
 397                         if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
 398                                 mlog(ML_ERROR, "Cannot remount RDWR "
 399                                      "filesystem due to previous errors.\n");
 400                                 ret = -EROFS;
 401                                 goto unlock_osb;
 402                         }
 403                         incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
 404                         if (incompat_features) {
 405                                 mlog(ML_ERROR, "Cannot remount RDWR because "
 406                                      "of unsupported optional features "
 407                                      "(%x).\n", incompat_features);
 408                                 ret = -EINVAL;
 409                                 goto unlock_osb;
 410                         }
 411                         sb->s_flags &= ~MS_RDONLY;
 412                         osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
 413                 }
 414 unlock_osb:
 415                 spin_unlock(&osb->osb_lock);
 416         }
 417
 418         if (!ret) {
 419                 if (!ocfs2_is_hard_readonly(osb))
 420                         ocfs2_set_journal_params(osb);
 421
 422                 /* Only save off the new mount options in case of a successful
 423                  * remount. */
 424                 osb->s_mount_opt = parsed_options;
 425         }
 426 out:
 427         return ret;
 428 }
 429
 430 static int ocfs2_sb_probe(struct super_block *sb,
 431                           struct buffer_head **bh,
 432                           int *sector_size)
 433 {
 434         int status = 0, tmpstat;
 435         struct ocfs1_vol_disk_hdr *hdr;
 436         struct ocfs2_dinode *di;
 437         int blksize;
 438
 439         *bh = NULL;
 440
 441         /* may be > 512 */
 442         *sector_size = bdev_hardsect_size(sb->s_bdev);
 443         if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
 444                 mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
 445                      *sector_size, OCFS2_MAX_BLOCKSIZE);
 446                 status = -EINVAL;
 447                 goto bail;
 448         }
 449
 450         /* Can this really happen? */
 451         if (*sector_size < OCFS2_MIN_BLOCKSIZE)
 452                 *sector_size = OCFS2_MIN_BLOCKSIZE;
 453
 454         /* check block zero for old format */
 455         status = ocfs2_get_sector(sb, bh, 0, *sector_size);
 456         if (status < 0) {
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460         hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
 461         if (hdr->major_version == OCFS1_MAJOR_VERSION) {
 462                 mlog(ML_ERROR, "incompatible version: %u.%u\n",
 463                      hdr->major_version, hdr->minor_version);
 464                 status = -EINVAL;
 465         }
 466         if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
 467                    strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
 468                 mlog(ML_ERROR, "incompatible volume signature: %8s\n",
 469                      hdr->signature);
 470                 status = -EINVAL;
 471         }
 472         brelse(*bh);
 473         *bh = NULL;
 474         if (status < 0) {
 475                 mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
 476                      "upgraded before mounting with ocfs v2\n");
 477                 goto bail;
 478         }
 479
 480         /*
 481          * Now check at magic offset for 512, 1024, 2048, 4096
 482          * blocksizes.  4096 is the maximum blocksize because it is
 483          * the minimum clustersize.
 484          */
 485         status = -EINVAL;
 486         for (blksize = *sector_size;
 487              blksize <= OCFS2_MAX_BLOCKSIZE;
 488              blksize <<= 1) {
 489                 tmpstat = ocfs2_get_sector(sb, bh,
 490                                            OCFS2_SUPER_BLOCK_BLKNO,
 491                                            blksize);
 492                 if (tmpstat < 0) {
 493                         status = tmpstat;
 494                         mlog_errno(status);
 495                         goto bail;
 496                 }
 497                 di = (struct ocfs2_dinode *) (*bh)->b_data;
 498                 status = ocfs2_verify_volume(di, *bh, blksize);
 499                 if (status >= 0)
 500                         goto bail;
 501                 brelse(*bh);
 502                 *bh = NULL;
 503                 if (status != -EAGAIN)
 504                         break;
 505         }
 506
 507 bail:
 508         return status;
 509 }
 510
 511 static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
 512 {
 513         if (ocfs2_mount_local(osb)) {
 514                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 515                         mlog(ML_ERROR, "Cannot heartbeat on a locally "
 516                              "mounted device.\n");
 517                         return -EINVAL;
 518                 }
 519         }
 520
 521         if (!(osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL)) {
 522                 if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb)) {
 523                         mlog(ML_ERROR, "Heartbeat has to be started to mount "
 524                              "a read-write clustered device.\n");
 525                         return -EINVAL;
 526                 }
 527         }
 528
 529         return 0;
 530 }
 531
 532 static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 533 {
 534         struct dentry *root;
 535         int status, sector_size;
 536         unsigned long parsed_opt;
 537         struct inode *inode = NULL;
 538         struct ocfs2_super *osb = NULL;
 539         struct buffer_head *bh = NULL;
 540         char nodestr[8];
 541
 542         mlog_entry("%p, %p, %i", sb, data, silent);
 543
 544         if (!ocfs2_parse_options(sb, data, &parsed_opt, 0)) {
 545                 status = -EINVAL;
 546                 goto read_super_error;
 547         }
 548
 549         /* for now we only have one cluster/node, make sure we see it
 550          * in the heartbeat universe */
 551         if (parsed_opt & OCFS2_MOUNT_HB_LOCAL) {
 552                 if (!o2hb_check_local_node_heartbeating()) {
 553                         status = -EINVAL;
 554                         goto read_super_error;
 555                 }
 556         }
 557
 558         /* probe for superblock */
 559         status = ocfs2_sb_probe(sb, &bh, &sector_size);
 560         if (status < 0) {
 561                 mlog(ML_ERROR, "superblock probe failed!\n");
 562                 goto read_super_error;
 563         }
 564
 565         status = ocfs2_initialize_super(sb, bh, sector_size);
 566         osb = OCFS2_SB(sb);
 567         if (status < 0) {
 568                 mlog_errno(status);
 569                 goto read_super_error;
 570         }
 571         brelse(bh);
 572         bh = NULL;
 573         osb->s_mount_opt = parsed_opt;
 574
 575         sb->s_magic = OCFS2_SUPER_MAGIC;
 576
 577         /* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
 578          * heartbeat=none */
 579         if (bdev_read_only(sb->s_bdev)) {
 580                 if (!(sb->s_flags & MS_RDONLY)) {
 581                         status = -EACCES;
 582                         mlog(ML_ERROR, "Readonly device detected but readonly "
 583                              "mount was not specified.\n");
 584                         goto read_super_error;
 585                 }
 586
 587                 /* You should not be able to start a local heartbeat
 588                  * on a readonly device. */
 589                 if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
 590                         status = -EROFS;
 591                         mlog(ML_ERROR, "Local heartbeat specified on readonly "
 592                              "device.\n");
 593                         goto read_super_error;
 594                 }
 595
 596                 status = ocfs2_check_journals_nolocks(osb);
 597                 if (status < 0) {
 598                         if (status == -EROFS)
 599                                 mlog(ML_ERROR, "Recovery required on readonly "
 600                                      "file system, but write access is "
 601                                      "unavailable.\n");
 602                         else
 603                                 mlog_errno(status);
 604                         goto read_super_error;
 605                 }
 606
 607                 ocfs2_set_ro_flag(osb, 1);
 608
 609                 printk(KERN_NOTICE "Readonly device detected. No cluster "
 610                        "services will be utilized for this mount. Recovery "
 611                        "will be skipped.\n");
 612         }
 613
 614         if (!ocfs2_is_hard_readonly(osb)) {
 615                 if (sb->s_flags & MS_RDONLY)
 616                         ocfs2_set_ro_flag(osb, 0);
 617         }
 618
 619         status = ocfs2_verify_heartbeat(osb);
 620         if (status < 0) {
 621                 mlog_errno(status);
 622                 goto read_super_error;
 623         }
 624
 625         osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
 626                                                  ocfs2_debugfs_root);
 627         if (!osb->osb_debug_root) {
 628                 status = -EINVAL;
 629                 mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
 630                 goto read_super_error;
 631         }
 632
 633         status = ocfs2_mount_volume(sb);
 634         if (osb->root_inode)
 635                 inode = igrab(osb->root_inode);
 636
 637         if (status < 0)
 638                 goto read_super_error;
 639
 640         if (!inode) {
 641                 status = -EIO;
 642                 mlog_errno(status);
 643                 goto read_super_error;
 644         }
 645
 646         root = d_alloc_root(inode);
 647         if (!root) {
 648                 status = -ENOMEM;
 649                 mlog_errno(status);
 650                 goto read_super_error;
 651         }
 652
 653         sb->s_root = root;
 654
 655         ocfs2_complete_mount_recovery(osb);
 656
 657         if (ocfs2_mount_local(osb))
 658                 snprintf(nodestr, sizeof(nodestr), "local");
 659         else
 660                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
 661
 662         printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
 663                "with %s data mode.\n",
 664                osb->dev_str, nodestr, osb->slot_num,
 665                osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
 666                "ordered");
 667
 668         atomic_set(&osb->vol_state, VOLUME_MOUNTED);
 669         wake_up(&osb->osb_mount_event);
 670
 671         mlog_exit(status);
 672         return status;
 673
 674 read_super_error:
 675         if (bh != NULL)
 676                 brelse(bh);
 677
 678         if (inode)
 679                 iput(inode);
 680
 681         if (osb) {
 682                 atomic_set(&osb->vol_state, VOLUME_DISABLED);
 683                 wake_up(&osb->osb_mount_event);
 684                 ocfs2_dismount_volume(sb, 1);
 685         }
 686
 687         mlog_exit(status);
 688         return status;
 689 }
 690
 691 static int ocfs2_get_sb(struct file_system_type *fs_type,
 692                         int flags,
 693                         const char *dev_name,
 694                         void *data,
 695                         struct vfsmount *mnt)
 696 {
 697         return get_sb_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super,
 698                            mnt);
 699 }
 700
 701 static struct file_system_type ocfs2_fs_type = {
 702         .owner          = THIS_MODULE,
 703         .name           = "ocfs2",
 704         .get_sb         = ocfs2_get_sb, /* is this called when we mount
 705                                         * the fs? */
 706         .kill_sb        = kill_block_super, /* set to the generic one
 707                                              * right now, but do we
 708                                              * need to change that? */
 709         .fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 710         .next           = NULL
 711 };
 712
 713 static int ocfs2_parse_options(struct super_block *sb,
 714                                char *options,
 715                                unsigned long *mount_opt,
 716                                int is_remount)
 717 {
 718         int status;
 719         char *p;
 720
 721         mlog_entry("remount: %d, options: \"%s\"\n", is_remount,
 722                    options ? options : "(none)");
 723
 724         *mount_opt = 0;
 725
 726         if (!options) {
 727                 status = 1;
 728                 goto bail;
 729         }
 730
 731         while ((p = strsep(&options, ",")) != NULL) {
 732                 int token, option;
 733                 substring_t args[MAX_OPT_ARGS];
 734                 struct ocfs2_super * osb = OCFS2_SB(sb);
 735
 736                 if (!*p)
 737                         continue;
 738
 739                 token = match_token(p, tokens, args);
 740                 switch (token) {
 741                 case Opt_hb_local:
 742                         *mount_opt |= OCFS2_MOUNT_HB_LOCAL;
 743                         break;
 744                 case Opt_hb_none:
 745                         *mount_opt &= ~OCFS2_MOUNT_HB_LOCAL;
 746                         break;
 747                 case Opt_barrier:
 748                         if (match_int(&args[0], &option)) {
 749                                 status = 0;
 750                                 goto bail;
 751                         }
 752                         if (option)
 753                                 *mount_opt |= OCFS2_MOUNT_BARRIER;
 754                         else
 755                                 *mount_opt &= ~OCFS2_MOUNT_BARRIER;
 756                         break;
 757                 case Opt_intr:
 758                         *mount_opt &= ~OCFS2_MOUNT_NOINTR;
 759                         break;
 760                 case Opt_nointr:
 761                         *mount_opt |= OCFS2_MOUNT_NOINTR;
 762                         break;
 763                 case Opt_err_panic:
 764                         *mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
 765                         break;
 766                 case Opt_err_ro:
 767                         *mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
 768                         break;
 769                 case Opt_data_ordered:
 770                         *mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
 771                         break;
 772                 case Opt_data_writeback:
 773                         *mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
 774                         break;
 775                 case Opt_atime_quantum:
 776                         if (match_int(&args[0], &option)) {
 777                                 status = 0;
 778                                 goto bail;
 779                         }
 780                         if (option >= 0)
 781                                 osb->s_atime_quantum = option;
 782                         else
 783                                 osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 784                         break;
 785                 default:
 786                         mlog(ML_ERROR,
 787                              "Unrecognized mount option \"%s\" "
 788                              "or missing value\n", p);
 789                         status = 0;
 790                         goto bail;
 791                 }
 792         }
 793
 794         status = 1;
 795
 796 bail:
 797         mlog_exit(status);
 798         return status;
 799 }
 800
 801 static int __init ocfs2_init(void)
 802 {
 803         int status;
 804
 805         mlog_entry_void();
 806
 807         ocfs2_print_version();
 808
 809         if (init_ocfs2_extent_maps())
 810                 return -ENOMEM;
 811
 812         status = init_ocfs2_uptodate_cache();
 813         if (status < 0) {
 814                 mlog_errno(status);
 815                 goto leave;
 816         }
 817
 818         status = ocfs2_initialize_mem_caches();
 819         if (status < 0) {
 820                 mlog_errno(status);
 821                 goto leave;
 822         }
 823
 824         ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
 825         if (!ocfs2_wq) {
 826                 status = -ENOMEM;
 827                 goto leave;
 828         }
 829
 830         ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
 831         if (!ocfs2_debugfs_root) {
 832                 status = -EFAULT;
 833                 mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
 834         }
 835
 836 leave:
 837         if (status < 0) {
 838                 ocfs2_free_mem_caches();
 839                 exit_ocfs2_uptodate_cache();
 840                 exit_ocfs2_extent_maps();
 841         }
 842
 843         mlog_exit(status);
 844
 845         if (status >= 0) {
 846                 return register_filesystem(&ocfs2_fs_type);
 847         } else
 848                 return -1;
 849 }
 850
 851 static void __exit ocfs2_exit(void)
 852 {
 853         mlog_entry_void();
 854
 855         if (ocfs2_wq) {
 856                 flush_workqueue(ocfs2_wq);
 857                 destroy_workqueue(ocfs2_wq);
 858         }
 859
 860         debugfs_remove(ocfs2_debugfs_root);
 861
 862         ocfs2_free_mem_caches();
 863
 864         unregister_filesystem(&ocfs2_fs_type);
 865
 866         exit_ocfs2_extent_maps();
 867
 868         exit_ocfs2_uptodate_cache();
 869
 870         mlog_exit_void();
 871 }
 872
 873 static void ocfs2_put_super(struct super_block *sb)
 874 {
 875         mlog_entry("(0x%p)\n", sb);
 876
 877         ocfs2_sync_blockdev(sb);
 878         ocfs2_dismount_volume(sb, 0);
 879
 880         mlog_exit_void();
 881 }
 882
 883 static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
 884 {
 885         struct ocfs2_super *osb;
 886         u32 numbits, freebits;
 887         int status;
 888         struct ocfs2_dinode *bm_lock;
 889         struct buffer_head *bh = NULL;
 890         struct inode *inode = NULL;
 891
 892         mlog_entry("(%p, %p)\n", dentry->d_sb, buf);
 893
 894         osb = OCFS2_SB(dentry->d_sb);
 895
 896         inode = ocfs2_get_system_file_inode(osb,
 897                                             GLOBAL_BITMAP_SYSTEM_INODE,
 898                                             OCFS2_INVALID_SLOT);
 899         if (!inode) {
 900                 mlog(ML_ERROR, "failed to get bitmap inode\n");
 901                 status = -EIO;
 902                 goto bail;
 903         }
 904
 905         status = ocfs2_meta_lock(inode, &bh, 0);
 906         if (status < 0) {
 907                 mlog_errno(status);
 908                 goto bail;
 909         }
 910
 911         bm_lock = (struct ocfs2_dinode *) bh->b_data;
 912
 913         numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
 914         freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
 915
 916         buf->f_type = OCFS2_SUPER_MAGIC;
 917         buf->f_bsize = dentry->d_sb->s_blocksize;
 918         buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
 919         buf->f_blocks = ((sector_t) numbits) *
 920                         (osb->s_clustersize >> osb->sb->s_blocksize_bits);
 921         buf->f_bfree = ((sector_t) freebits) *
 922                        (osb->s_clustersize >> osb->sb->s_blocksize_bits);
 923         buf->f_bavail = buf->f_bfree;
 924         buf->f_files = numbits;
 925         buf->f_ffree = freebits;
 926
 927         brelse(bh);
 928
 929         ocfs2_meta_unlock(inode, 0);
 930         status = 0;
 931 bail:
 932         if (inode)
 933                 iput(inode);
 934
 935         mlog_exit(status);
 936
 937         return status;
 938 }
 939
 940 static void ocfs2_inode_init_once(void *data,
 941                                   struct kmem_cache *cachep,
 942                                   unsigned long flags)
 943 {
 944         struct ocfs2_inode_info *oi = data;
 945
 946         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 947             SLAB_CTOR_CONSTRUCTOR) {
 948                 oi->ip_flags = 0;
 949                 oi->ip_open_count = 0;
 950                 spin_lock_init(&oi->ip_lock);
 951                 ocfs2_extent_map_init(&oi->vfs_inode);
 952                 INIT_LIST_HEAD(&oi->ip_io_markers);
 953                 oi->ip_created_trans = 0;
 954                 oi->ip_last_trans = 0;
 955                 oi->ip_dir_start_lookup = 0;
 956
 957                 init_rwsem(&oi->ip_alloc_sem);
 958                 mutex_init(&oi->ip_io_mutex);
 959
 960                 oi->ip_blkno = 0ULL;
 961                 oi->ip_clusters = 0;
 962
 963                 ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
 964                 ocfs2_lock_res_init_once(&oi->ip_meta_lockres);
 965                 ocfs2_lock_res_init_once(&oi->ip_data_lockres);
 966                 ocfs2_lock_res_init_once(&oi->ip_open_lockres);
 967
 968                 ocfs2_metadata_cache_init(&oi->vfs_inode);
 969
 970                 inode_init_once(&oi->vfs_inode);
 971         }
 972 }
 973
 974 static int ocfs2_initialize_mem_caches(void)
 975 {
 976         ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
 977                                        sizeof(struct ocfs2_inode_info),
 978                                        0,
 979                                        (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 980                                                 SLAB_MEM_SPREAD),
 981                                        ocfs2_inode_init_once, NULL);
 982         if (!ocfs2_inode_cachep)
 983                 return -ENOMEM;
 984
 985         return 0;
 986 }
 987
 988 static void ocfs2_free_mem_caches(void)
 989 {
 990         if (ocfs2_inode_cachep)
 991                 kmem_cache_destroy(ocfs2_inode_cachep);
 992
 993         ocfs2_inode_cachep = NULL;
 994 }
 995
 996 static int ocfs2_get_sector(struct super_block *sb,
 997                             struct buffer_head **bh,
 998                             int block,
 999                             int sect_size)
1000 {
1001         if (!sb_set_blocksize(sb, sect_size)) {
1002                 mlog(ML_ERROR, "unable to set blocksize\n");
1003                 return -EIO;
1004         }
1005
1006         *bh = sb_getblk(sb, block);
1007         if (!*bh) {
1008                 mlog_errno(-EIO);
1009                 return -EIO;
1010         }
1011         lock_buffer(*bh);
1012         if (!buffer_dirty(*bh))
1013                 clear_buffer_uptodate(*bh);
1014         unlock_buffer(*bh);
1015         ll_rw_block(READ, 1, bh);
1016         wait_on_buffer(*bh);
1017         return 0;
1018 }
1019
1020 /* ocfs2 1.0 only allows one cluster and node identity per kernel image. */
1021 static int ocfs2_fill_local_node_info(struct ocfs2_super *osb)
1022 {
1023         int status;
1024
1025         /* XXX hold a ref on the node while mounte?  easy enough, if
1026          * desirable. */
1027         if (ocfs2_mount_local(osb))
1028                 osb->node_num = 0;
1029         else
1030                 osb->node_num = o2nm_this_node();
1031
1032         if (osb->node_num == O2NM_MAX_NODES) {
1033                 mlog(ML_ERROR, "could not find this host's node number\n");
1034                 status = -ENOENT;
1035                 goto bail;
1036         }
1037
1038         mlog(0, "I am node %d\n", osb->node_num);
1039
1040         status = 0;
1041 bail:
1042         return status;
1043 }
1044
1045 static int ocfs2_mount_volume(struct super_block *sb)
1046 {
1047         int status = 0;
1048         int unlock_super = 0;
1049         struct ocfs2_super *osb = OCFS2_SB(sb);
1050
1051         mlog_entry_void();
1052
1053         if (ocfs2_is_hard_readonly(osb))
1054                 goto leave;
1055
1056         status = ocfs2_fill_local_node_info(osb);
1057         if (status < 0) {
1058                 mlog_errno(status);
1059                 goto leave;
1060         }
1061
1062         status = ocfs2_register_hb_callbacks(osb);
1063         if (status < 0) {
1064                 mlog_errno(status);
1065                 goto leave;
1066         }
1067
1068         status = ocfs2_dlm_init(osb);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto leave;
1072         }
1073
1074         /* requires vote_thread to be running. */
1075         status = ocfs2_register_net_handlers(osb);
1076         if (status < 0) {
1077                 mlog_errno(status);
1078                 goto leave;
1079         }
1080
1081         status = ocfs2_super_lock(osb, 1);
1082         if (status < 0) {
1083                 mlog_errno(status);
1084                 goto leave;
1085         }
1086         unlock_super = 1;
1087
1088         /* This will load up the node map and add ourselves to it. */
1089         status = ocfs2_find_slot(osb);
1090         if (status < 0) {
1091                 mlog_errno(status);
1092                 goto leave;
1093         }
1094
1095         ocfs2_populate_mounted_map(osb);
1096
1097         /* load all node-local system inodes */
1098         status = ocfs2_init_local_system_inodes(osb);
1099         if (status < 0) {
1100                 mlog_errno(status);
1101                 goto leave;
1102         }
1103
1104         status = ocfs2_check_volume(osb);
1105         if (status < 0) {
1106                 mlog_errno(status);
1107                 goto leave;
1108         }
1109
1110         status = ocfs2_truncate_log_init(osb);
1111         if (status < 0) {
1112                 mlog_errno(status);
1113                 goto leave;
1114         }
1115
1116         if (ocfs2_mount_local(osb))
1117                 goto leave;
1118
1119         /* This should be sent *after* we recovered our journal as it
1120          * will cause other nodes to unmark us as needing
1121          * recovery. However, we need to send it *before* dropping the
1122          * super block lock as otherwise their recovery threads might
1123          * try to clean us up while we're live! */
1124         status = ocfs2_request_mount_vote(osb);
1125         if (status < 0)
1126                 mlog_errno(status);
1127
1128 leave:
1129         if (unlock_super)
1130                 ocfs2_super_unlock(osb, 1);
1131
1132         mlog_exit(status);
1133         return status;
1134 }
1135
1136 /* we can't grab the goofy sem lock from inside wait_event, so we use
1137  * memory barriers to make sure that we'll see the null task before
1138  * being woken up */
1139 static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
1140 {
1141         mb();
1142         return osb->recovery_thread_task != NULL;
1143 }
1144
1145 static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
1146 {
1147         int tmp;
1148         struct ocfs2_super *osb = NULL;
1149         char nodestr[8];
1150
1151         mlog_entry("(0x%p)\n", sb);
1152
1153         BUG_ON(!sb);
1154         osb = OCFS2_SB(sb);
1155         BUG_ON(!osb);
1156
1157         ocfs2_shutdown_local_alloc(osb);
1158
1159         ocfs2_truncate_log_shutdown(osb);
1160
1161         /* disable any new recovery threads and wait for any currently
1162          * running ones to exit. Do this before setting the vol_state. */
1163         mutex_lock(&osb->recovery_lock);
1164         osb->disable_recovery = 1;
1165         mutex_unlock(&osb->recovery_lock);
1166         wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
1167
1168         /* At this point, we know that no more recovery threads can be
1169          * launched, so wait for any recovery completion work to
1170          * complete. */
1171         flush_workqueue(ocfs2_wq);
1172
1173         ocfs2_journal_shutdown(osb);
1174
1175         ocfs2_sync_blockdev(sb);
1176
1177         /* No dlm means we've failed during mount, so skip all the
1178          * steps which depended on that to complete. */
1179         if (osb->dlm) {
1180                 tmp = ocfs2_super_lock(osb, 1);
1181                 if (tmp < 0) {
1182                         mlog_errno(tmp);
1183                         return;
1184                 }
1185
1186                 tmp = ocfs2_request_umount_vote(osb);
1187                 if (tmp < 0)
1188                         mlog_errno(tmp);
1189
1190                 if (osb->slot_num != OCFS2_INVALID_SLOT)
1191                         ocfs2_put_slot(osb);
1192
1193                 ocfs2_super_unlock(osb, 1);
1194         }
1195
1196         ocfs2_release_system_inodes(osb);
1197
1198         if (osb->dlm) {
1199                 ocfs2_unregister_net_handlers(osb);
1200
1201                 ocfs2_dlm_shutdown(osb);
1202         }
1203
1204         ocfs2_clear_hb_callbacks(osb);
1205
1206         debugfs_remove(osb->osb_debug_root);
1207
1208         if (!mnt_err)
1209                 ocfs2_stop_heartbeat(osb);
1210
1211         atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
1212
1213         if (ocfs2_mount_local(osb))
1214                 snprintf(nodestr, sizeof(nodestr), "local");
1215         else
1216                 snprintf(nodestr, sizeof(nodestr), "%d", osb->node_num);
1217
1218         printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
1219                osb->dev_str, nodestr);
1220
1221         ocfs2_delete_osb(osb);
1222         kfree(osb);
1223         sb->s_dev = 0;
1224         sb->s_fs_info = NULL;
1225 }
1226
1227 static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
1228                                 unsigned uuid_bytes)
1229 {
1230         int i, ret;
1231         char *ptr;
1232
1233         BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
1234
1235         osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
1236         if (osb->uuid_str == NULL)
1237                 return -ENOMEM;
1238
1239         for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
1240                 /* print with null */
1241                 ret = snprintf(ptr, 3, "%02X", uuid[i]);
1242                 if (ret != 2) /* drop super cleans up */
1243                         return -EINVAL;
1244                 /* then only advance past the last char */
1245                 ptr += 2;
1246         }
1247
1248         return 0;
1249 }
1250
1251 static int ocfs2_initialize_super(struct super_block *sb,
1252                                   struct buffer_head *bh,
1253                                   int sector_size)
1254 {
1255         int status = 0;
1256         int i;
1257         struct ocfs2_dinode *di = NULL;
1258         struct inode *inode = NULL;
1259         struct buffer_head *bitmap_bh = NULL;
1260         struct ocfs2_journal *journal;
1261         __le32 uuid_net_key;
1262         struct ocfs2_super *osb;
1263
1264         mlog_entry_void();
1265
1266         osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
1267         if (!osb) {
1268                 status = -ENOMEM;
1269                 mlog_errno(status);
1270                 goto bail;
1271         }
1272
1273         sb->s_fs_info = osb;
1274         sb->s_op = &ocfs2_sops;
1275         sb->s_export_op = &ocfs2_export_ops;
1276         sb->s_flags |= MS_NOATIME;
1277         /* this is needed to support O_LARGEFILE */
1278         sb->s_maxbytes = ocfs2_max_file_offset(sb->s_blocksize_bits);
1279
1280         osb->sb = sb;
1281         /* Save off for ocfs2_rw_direct */
1282         osb->s_sectsize_bits = blksize_bits(sector_size);
1283         BUG_ON(!osb->s_sectsize_bits);
1284
1285         osb->net_response_ids = 0;
1286         spin_lock_init(&osb->net_response_lock);
1287         INIT_LIST_HEAD(&osb->net_response_list);
1288
1289         INIT_LIST_HEAD(&osb->osb_net_handlers);
1290         init_waitqueue_head(&osb->recovery_event);
1291         spin_lock_init(&osb->vote_task_lock);
1292         init_waitqueue_head(&osb->vote_event);
1293         osb->vote_work_sequence = 0;
1294         osb->vote_wake_sequence = 0;
1295         INIT_LIST_HEAD(&osb->blocked_lock_list);
1296         osb->blocked_lock_count = 0;
1297         INIT_LIST_HEAD(&osb->vote_list);
1298         spin_lock_init(&osb->osb_lock);
1299
1300         atomic_set(&osb->alloc_stats.moves, 0);
1301         atomic_set(&osb->alloc_stats.local_data, 0);
1302         atomic_set(&osb->alloc_stats.bitmap_data, 0);
1303         atomic_set(&osb->alloc_stats.bg_allocs, 0);
1304         atomic_set(&osb->alloc_stats.bg_extends, 0);
1305
1306         ocfs2_init_node_maps(osb);
1307
1308         snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
1309                  MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
1310
1311         mutex_init(&osb->recovery_lock);
1312
1313         osb->disable_recovery = 0;
1314         osb->recovery_thread_task = NULL;
1315
1316         init_waitqueue_head(&osb->checkpoint_event);
1317         atomic_set(&osb->needs_checkpoint, 0);
1318
1319         osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
1320
1321         osb->node_num = O2NM_INVALID_NODE_NUM;
1322         osb->slot_num = OCFS2_INVALID_SLOT;
1323
1324         osb->local_alloc_state = OCFS2_LA_UNUSED;
1325         osb->local_alloc_bh = NULL;
1326
1327         ocfs2_setup_hb_callbacks(osb);
1328
1329         init_waitqueue_head(&osb->osb_mount_event);
1330
1331         osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
1332         if (!osb->vol_label) {
1333                 mlog(ML_ERROR, "unable to alloc vol label\n");
1334                 status = -ENOMEM;
1335                 goto bail;
1336         }
1337
1338         di = (struct ocfs2_dinode *)bh->b_data;
1339
1340         osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
1341         if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
1342                 mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
1343                      osb->max_slots);
1344                 status = -EINVAL;
1345                 goto bail;
1346         }
1347         mlog(0, "max_slots for this device: %u\n", osb->max_slots);
1348
1349         init_waitqueue_head(&osb->osb_wipe_event);
1350         osb->osb_orphan_wipes = kcalloc(osb->max_slots,
1351                                         sizeof(*osb->osb_orphan_wipes),
1352                                         GFP_KERNEL);
1353         if (!osb->osb_orphan_wipes) {
1354                 status = -ENOMEM;
1355                 mlog_errno(status);
1356                 goto bail;
1357         }
1358
1359         osb->s_feature_compat =
1360                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
1361         osb->s_feature_ro_compat =
1362                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
1363         osb->s_feature_incompat =
1364                 le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
1365
1366         if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
1367                 mlog(ML_ERROR, "couldn't mount because of unsupported "
1368                      "optional features (%x).\n", i);
1369                 status = -EINVAL;
1370                 goto bail;
1371         }
1372         if (!(osb->sb->s_flags & MS_RDONLY) &&
1373             (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
1374                 mlog(ML_ERROR, "couldn't mount RDWR because of "
1375                      "unsupported optional features (%x).\n", i);
1376                 status = -EINVAL;
1377                 goto bail;
1378         }
1379
1380         get_random_bytes(&osb->s_next_generation, sizeof(u32));
1381
1382         /* FIXME
1383          * This should be done in ocfs2_journal_init(), but unknown
1384          * ordering issues will cause the filesystem to crash.
1385          * If anyone wants to figure out what part of the code
1386          * refers to osb->journal before ocfs2_journal_init() is run,
1387          * be my guest.
1388          */
1389         /* initialize our journal structure */
1390
1391         journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
1392         if (!journal) {
1393                 mlog(ML_ERROR, "unable to alloc journal\n");
1394                 status = -ENOMEM;
1395                 goto bail;
1396         }
1397         osb->journal = journal;
1398         journal->j_osb = osb;
1399
1400         atomic_set(&journal->j_num_trans, 0);
1401         init_rwsem(&journal->j_trans_barrier);
1402         init_waitqueue_head(&journal->j_checkpointed);
1403         spin_lock_init(&journal->j_lock);
1404         journal->j_trans_id = (unsigned long) 1;
1405         INIT_LIST_HEAD(&journal->j_la_cleanups);
1406         INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
1407         journal->j_state = OCFS2_JOURNAL_FREE;
1408
1409         /* get some pseudo constants for clustersize bits */
1410         osb->s_clustersize_bits =
1411                 le32_to_cpu(di->id2.i_super.s_clustersize_bits);
1412         osb->s_clustersize = 1 << osb->s_clustersize_bits;
1413         mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
1414
1415         if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
1416             osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
1417                 mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
1418                      osb->s_clustersize);
1419                 status = -EINVAL;
1420                 goto bail;
1421         }
1422
1423         if (ocfs2_clusters_to_blocks(osb->sb, le32_to_cpu(di->i_clusters) - 1)
1424             > (u32)~0UL) {
1425                 mlog(ML_ERROR, "Volume might try to write to blocks beyond "
1426                      "what jbd can address in 32 bits.\n");
1427                 status = -EINVAL;
1428                 goto bail;
1429         }
1430
1431         if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
1432                                  sizeof(di->id2.i_super.s_uuid))) {
1433                 mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
1434                 status = -ENOMEM;
1435                 goto bail;
1436         }
1437
1438         memcpy(&uuid_net_key, di->id2.i_super.s_uuid, sizeof(uuid_net_key));
1439         osb->net_key = le32_to_cpu(uuid_net_key);
1440
1441         strncpy(osb->vol_label, di->id2.i_super.s_label, 63);
1442         osb->vol_label[63] = '\0';
1443         osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
1444         osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
1445         osb->first_cluster_group_blkno =
1446                 le64_to_cpu(di->id2.i_super.s_first_cluster_group);
1447         osb->fs_generation = le32_to_cpu(di->i_fs_generation);
1448         mlog(0, "vol_label: %s\n", osb->vol_label);
1449         mlog(0, "uuid: %s\n", osb->uuid_str);
1450         mlog(0, "root_blkno=%llu, system_dir_blkno=%llu\n",
1451              (unsigned long long)osb->root_blkno,
1452              (unsigned long long)osb->system_dir_blkno);
1453
1454         osb->osb_dlm_debug = ocfs2_new_dlm_debug();
1455         if (!osb->osb_dlm_debug) {
1456                 status = -ENOMEM;
1457                 mlog_errno(status);
1458                 goto bail;
1459         }
1460
1461         atomic_set(&osb->vol_state, VOLUME_INIT);
1462
1463         /* load root, system_dir, and all global system inodes */
1464         status = ocfs2_init_global_system_inodes(osb);
1465         if (status < 0) {
1466                 mlog_errno(status);
1467                 goto bail;
1468         }
1469
1470         /*
1471          * global bitmap
1472          */
1473         inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
1474                                             OCFS2_INVALID_SLOT);
1475         if (!inode) {
1476                 status = -EINVAL;
1477                 mlog_errno(status);
1478                 goto bail;
1479         }
1480
1481         osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
1482
1483         /* We don't have a cluster lock on the bitmap here because
1484          * we're only interested in static information and the extra
1485          * complexity at mount time isn't worht it. Don't pass the
1486          * inode in to the read function though as we don't want it to
1487          * be put in the cache. */
1488         status = ocfs2_read_block(osb, osb->bitmap_blkno, &bitmap_bh, 0,
1489                                   NULL);
1490         iput(inode);
1491         if (status < 0) {
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495
1496         di = (struct ocfs2_dinode *) bitmap_bh->b_data;
1497         osb->bitmap_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
1498         brelse(bitmap_bh);
1499         mlog(0, "cluster bitmap inode: %llu, clusters per group: %u\n",
1500              (unsigned long long)osb->bitmap_blkno, osb->bitmap_cpg);
1501
1502         status = ocfs2_init_slot_info(osb);
1503         if (status < 0) {
1504                 mlog_errno(status);
1505                 goto bail;
1506         }
1507
1508 bail:
1509         mlog_exit(status);
1510         return status;
1511 }
1512
1513 /*
1514  * will return: -EAGAIN if it is ok to keep searching for superblocks
1515  *              -EINVAL if there is a bad superblock
1516  *              0 on success
1517  */
1518 static int ocfs2_verify_volume(struct ocfs2_dinode *di,
1519                                struct buffer_head *bh,
1520                                u32 blksz)
1521 {
1522         int status = -EAGAIN;
1523
1524         mlog_entry_void();
1525
1526         if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
1527                    strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
1528                 status = -EINVAL;
1529                 if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
1530                         mlog(ML_ERROR, "found superblock with incorrect block "
1531                              "size: found %u, should be %u\n",
1532                              1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
1533                                blksz);
1534                 } else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
1535                            OCFS2_MAJOR_REV_LEVEL ||
1536                            le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
1537                            OCFS2_MINOR_REV_LEVEL) {
1538                         mlog(ML_ERROR, "found superblock with bad version: "
1539                              "found %u.%u, should be %u.%u\n",
1540                              le16_to_cpu(di->id2.i_super.s_major_rev_level),
1541                              le16_to_cpu(di->id2.i_super.s_minor_rev_level),
1542                              OCFS2_MAJOR_REV_LEVEL,
1543                              OCFS2_MINOR_REV_LEVEL);
1544                 } else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
1545                         mlog(ML_ERROR, "bad block number on superblock: "
1546                              "found %llu, should be %llu\n",
1547                              (unsigned long long)di->i_blkno,
1548                              (unsigned long long)bh->b_blocknr);
1549                 } else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
1550                             le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
1551                         mlog(ML_ERROR, "bad cluster size found: %u\n",
1552                              1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
1553                 } else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
1554                         mlog(ML_ERROR, "bad root_blkno: 0\n");
1555                 } else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
1556                         mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
1557                 } else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
1558                         mlog(ML_ERROR,
1559                              "Superblock slots found greater than file system "
1560                              "maximum: found %u, max %u\n",
1561                              le16_to_cpu(di->id2.i_super.s_max_slots),
1562                              OCFS2_MAX_SLOTS);
1563                 } else {
1564                         /* found it! */
1565                         status = 0;
1566                 }
1567         }
1568
1569         mlog_exit(status);
1570         return status;
1571 }
1572
1573 static int ocfs2_check_volume(struct ocfs2_super *osb)
1574 {
1575         int status = 0;
1576         int dirty;
1577         int local;
1578         struct ocfs2_dinode *local_alloc = NULL; /* only used if we
1579                                                   * recover
1580                                                   * ourselves. */
1581
1582         mlog_entry_void();
1583
1584         /* Init our journal object. */
1585         status = ocfs2_journal_init(osb->journal, &dirty);
1586         if (status < 0) {
1587                 mlog(ML_ERROR, "Could not initialize journal!\n");
1588                 goto finally;
1589         }
1590
1591         /* If the journal was unmounted cleanly then we don't want to
1592          * recover anything. Otherwise, journal_load will do that
1593          * dirty work for us :) */
1594         if (!dirty) {
1595                 status = ocfs2_journal_wipe(osb->journal, 0);
1596                 if (status < 0) {
1597                         mlog_errno(status);
1598                         goto finally;
1599                 }
1600         } else {
1601                 mlog(ML_NOTICE, "File system was not unmounted cleanly, "
1602                      "recovering volume.\n");
1603         }
1604
1605         local = ocfs2_mount_local(osb);
1606
1607         /* will play back anything left in the journal. */
1608         ocfs2_journal_load(osb->journal, local);
1609
1610         if (dirty) {
1611                 /* recover my local alloc if we didn't unmount cleanly. */
1612                 status = ocfs2_begin_local_alloc_recovery(osb,
1613                                                           osb->slot_num,
1614                                                           &local_alloc);
1615                 if (status < 0) {
1616                         mlog_errno(status);
1617                         goto finally;
1618                 }
1619                 /* we complete the recovery process after we've marked
1620                  * ourselves as mounted. */
1621         }
1622
1623         mlog(0, "Journal loaded.\n");
1624
1625         status = ocfs2_load_local_alloc(osb);
1626         if (status < 0) {
1627                 mlog_errno(status);
1628                 goto finally;
1629         }
1630
1631         if (dirty) {
1632                 /* Recovery will be completed after we've mounted the
1633                  * rest of the volume. */
1634                 osb->dirty = 1;
1635                 osb->local_alloc_copy = local_alloc;
1636                 local_alloc = NULL;
1637         }
1638
1639         /* go through each journal, trylock it and if you get the
1640          * lock, and it's marked as dirty, set the bit in the recover
1641          * map and launch a recovery thread for it. */
1642         status = ocfs2_mark_dead_nodes(osb);
1643         if (status < 0)
1644                 mlog_errno(status);
1645
1646 finally:
1647         if (local_alloc)
1648                 kfree(local_alloc);
1649
1650         mlog_exit(status);
1651         return status;
1652 }
1653
1654 /*
1655  * The routine gets called from dismount or close whenever a dismount on
1656  * volume is requested and the osb open count becomes 1.
1657  * It will remove the osb from the global list and also free up all the
1658  * initialized resources and fileobject.
1659  */
1660 static void ocfs2_delete_osb(struct ocfs2_super *osb)
1661 {
1662         mlog_entry_void();
1663
1664         /* This function assumes that the caller has the main osb resource */
1665
1666         if (osb->slot_info)
1667                 ocfs2_free_slot_info(osb->slot_info);
1668
1669         kfree(osb->osb_orphan_wipes);
1670         /* FIXME
1671          * This belongs in journal shutdown, but because we have to
1672          * allocate osb->journal at the start of ocfs2_initalize_osb(),
1673          * we free it here.
1674          */
1675         kfree(osb->journal);
1676         if (osb->local_alloc_copy)
1677                 kfree(osb->local_alloc_copy);
1678         kfree(osb->uuid_str);
1679         ocfs2_put_dlm_debug(osb->osb_dlm_debug);
1680         memset(osb, 0, sizeof(struct ocfs2_super));
1681
1682         mlog_exit_void();
1683 }
1684
1685 /* Put OCFS2 into a readonly state, or (if the user specifies it),
1686  * panic(). We do not support continue-on-error operation. */
1687 static void ocfs2_handle_error(struct super_block *sb)
1688 {
1689         struct ocfs2_super *osb = OCFS2_SB(sb);
1690
1691         if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
1692                 panic("OCFS2: (device %s): panic forced after error\n",
1693                       sb->s_id);
1694
1695         ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
1696
1697         if (sb->s_flags & MS_RDONLY &&
1698             (ocfs2_is_soft_readonly(osb) ||
1699              ocfs2_is_hard_readonly(osb)))
1700                 return;
1701
1702         printk(KERN_CRIT "File system is now read-only due to the potential "
1703                "of on-disk corruption. Please run fsck.ocfs2 once the file "
1704                "system is unmounted.\n");
1705         sb->s_flags |= MS_RDONLY;
1706         ocfs2_set_ro_flag(osb, 0);
1707 }
1708
1709 static char error_buf[1024];
1710
1711 void __ocfs2_error(struct super_block *sb,
1712                    const char *function,
1713                    const char *fmt, ...)
1714 {
1715         va_list args;
1716
1717         va_start(args, fmt);
1718         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1719         va_end(args);
1720
1721         /* Not using mlog here because we want to show the actual
1722          * function the error came from. */
1723         printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %s\n",
1724                sb->s_id, function, error_buf);
1725
1726         ocfs2_handle_error(sb);
1727 }
1728
1729 /* Handle critical errors. This is intentionally more drastic than
1730  * ocfs2_handle_error, so we only use for things like journal errors,
1731  * etc. */
1732 void __ocfs2_abort(struct super_block* sb,
1733                    const char *function,
1734                    const char *fmt, ...)
1735 {
1736         va_list args;
1737
1738         va_start(args, fmt);
1739         vsnprintf(error_buf, sizeof(error_buf), fmt, args);
1740         va_end(args);
1741
1742         printk(KERN_CRIT "OCFS2: abort (device %s): %s: %s\n",
1743                sb->s_id, function, error_buf);
1744
1745         /* We don't have the cluster support yet to go straight to
1746          * hard readonly in here. Until then, we want to keep
1747          * ocfs2_abort() so that we can at least mark critical
1748          * errors.
1749          *
1750          * TODO: This should abort the journal and alert other nodes
1751          * that our slot needs recovery. */
1752
1753         /* Force a panic(). This stinks, but it's better than letting
1754          * things continue without having a proper hard readonly
1755          * here. */
1756         OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
1757         ocfs2_handle_error(sb);
1758 }
1759
1760 module_init(ocfs2_init);
1761 module_exit(ocfs2_exit);