fs/gfs2/ops_fstype.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/kthread.h>
  17 #include <linux/namei.h>
  18 #include <linux/mount.h>
  19 #include <linux/gfs2_ondisk.h>
  20 #include <linux/lm_interface.h>
  21
  22 #include "gfs2.h"
  23 #include "incore.h"
  24 #include "bmap.h"
  25 #include "daemon.h"
  26 #include "glock.h"
  27 #include "glops.h"
  28 #include "inode.h"
  29 #include "mount.h"
  30 #include "ops_fstype.h"
  31 #include "ops_dentry.h"
  32 #include "ops_super.h"
  33 #include "recovery.h"
  34 #include "rgrp.h"
  35 #include "super.h"
  36 #include "sys.h"
  37 #include "util.h"
  38 #include "log.h"
  39
  40 #define DO 0
  41 #define UNDO 1
  42
  43 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  44 {
  45         struct gfs2_sbd *sdp;
  46
  47         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  48         if (!sdp)
  49                 return NULL;
  50
  51         sb->s_fs_info = sdp;
  52         sdp->sd_vfs = sb;
  53
  54         gfs2_tune_init(&sdp->sd_tune);
  55
  56         INIT_LIST_HEAD(&sdp->sd_reclaim_list);
  57         spin_lock_init(&sdp->sd_reclaim_lock);
  58         init_waitqueue_head(&sdp->sd_reclaim_wq);
  59
  60         mutex_init(&sdp->sd_inum_mutex);
  61         spin_lock_init(&sdp->sd_statfs_spin);
  62
  63         spin_lock_init(&sdp->sd_rindex_spin);
  64         mutex_init(&sdp->sd_rindex_mutex);
  65         INIT_LIST_HEAD(&sdp->sd_rindex_list);
  66         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
  67         INIT_LIST_HEAD(&sdp->sd_rindex_recent_list);
  68
  69         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  70         spin_lock_init(&sdp->sd_jindex_spin);
  71         mutex_init(&sdp->sd_jindex_mutex);
  72
  73         INIT_LIST_HEAD(&sdp->sd_quota_list);
  74         spin_lock_init(&sdp->sd_quota_spin);
  75         mutex_init(&sdp->sd_quota_mutex);
  76
  77         spin_lock_init(&sdp->sd_log_lock);
  78
  79         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
  80         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
  81         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
  82         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
  83         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
  84
  85         mutex_init(&sdp->sd_log_reserve_mutex);
  86         INIT_LIST_HEAD(&sdp->sd_ail1_list);
  87         INIT_LIST_HEAD(&sdp->sd_ail2_list);
  88
  89         init_rwsem(&sdp->sd_log_flush_lock);
  90         atomic_set(&sdp->sd_log_in_flight, 0);
  91         init_waitqueue_head(&sdp->sd_log_flush_wait);
  92
  93         INIT_LIST_HEAD(&sdp->sd_revoke_list);
  94
  95         mutex_init(&sdp->sd_freeze_lock);
  96
  97         return sdp;
  98 }
  99
 100 static void init_vfs(struct super_block *sb, unsigned noatime)
 101 {
 102         struct gfs2_sbd *sdp = sb->s_fs_info;
 103
 104         sb->s_magic = GFS2_MAGIC;
 105         sb->s_op = &gfs2_super_ops;
 106         sb->s_export_op = &gfs2_export_ops;
 107         sb->s_time_gran = 1;
 108         sb->s_maxbytes = MAX_LFS_FILESIZE;
 109
 110         if (sb->s_flags & (MS_NOATIME | MS_NODIRATIME))
 111                 set_bit(noatime, &sdp->sd_flags);
 112
 113         /* Don't let the VFS update atimes.  GFS2 handles this itself. */
 114         sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
 115 }
 116
 117 static int init_names(struct gfs2_sbd *sdp, int silent)
 118 {
 119         char *proto, *table;
 120         int error = 0;
 121
 122         proto = sdp->sd_args.ar_lockproto;
 123         table = sdp->sd_args.ar_locktable;
 124
 125         /*  Try to autodetect  */
 126
 127         if (!proto[0] || !table[0]) {
 128                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 129                 if (error)
 130                         return error;
 131
 132                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 133                 if (error)
 134                         goto out;
 135
 136                 if (!proto[0])
 137                         proto = sdp->sd_sb.sb_lockproto;
 138                 if (!table[0])
 139                         table = sdp->sd_sb.sb_locktable;
 140         }
 141
 142         if (!table[0])
 143                 table = sdp->sd_vfs->s_id;
 144
 145         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 146         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 147
 148         table = sdp->sd_table_name;
 149         while ((table = strchr(table, '/')))
 150                 *table = '_';
 151
 152 out:
 153         return error;
 154 }
 155
 156 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 157                         int undo)
 158 {
 159         struct task_struct *p;
 160         int error = 0;
 161
 162         if (undo)
 163                 goto fail_trans;
 164
 165         for (sdp->sd_glockd_num = 0;
 166              sdp->sd_glockd_num < sdp->sd_args.ar_num_glockd;
 167              sdp->sd_glockd_num++) {
 168                 p = kthread_run(gfs2_glockd, sdp, "gfs2_glockd");
 169                 error = IS_ERR(p);
 170                 if (error) {
 171                         fs_err(sdp, "can't start glockd thread: %d\n", error);
 172                         goto fail;
 173                 }
 174                 sdp->sd_glockd_process[sdp->sd_glockd_num] = p;
 175         }
 176
 177         error = gfs2_glock_nq_num(sdp,
 178                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 179                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 180                                   mount_gh);
 181         if (error) {
 182                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 183                 goto fail;
 184         }
 185
 186         error = gfs2_glock_nq_num(sdp,
 187                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 188                                   LM_ST_SHARED,
 189                                   LM_FLAG_NOEXP | GL_EXACT,
 190                                   &sdp->sd_live_gh);
 191         if (error) {
 192                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 193                 goto fail_mount;
 194         }
 195
 196         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 197                                CREATE, &sdp->sd_rename_gl);
 198         if (error) {
 199                 fs_err(sdp, "can't create rename glock: %d\n", error);
 200                 goto fail_live;
 201         }
 202
 203         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 204                                CREATE, &sdp->sd_trans_gl);
 205         if (error) {
 206                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 207                 goto fail_rename;
 208         }
 209         set_bit(GLF_STICKY, &sdp->sd_trans_gl->gl_flags);
 210
 211         return 0;
 212
 213 fail_trans:
 214         gfs2_glock_put(sdp->sd_trans_gl);
 215 fail_rename:
 216         gfs2_glock_put(sdp->sd_rename_gl);
 217 fail_live:
 218         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 219 fail_mount:
 220         gfs2_glock_dq_uninit(mount_gh);
 221 fail:
 222         while (sdp->sd_glockd_num--)
 223                 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
 224
 225         return error;
 226 }
 227
 228 static inline struct inode *gfs2_lookup_root(struct super_block *sb,
 229                                              u64 no_addr)
 230 {
 231         return gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 232 }
 233
 234 static int init_sb(struct gfs2_sbd *sdp, int silent, int undo)
 235 {
 236         struct super_block *sb = sdp->sd_vfs;
 237         struct gfs2_holder sb_gh;
 238         u64 no_addr;
 239         struct inode *inode;
 240         int error = 0;
 241
 242         if (undo) {
 243                 if (sb->s_root) {
 244                         dput(sb->s_root);
 245                         sb->s_root = NULL;
 246                 }
 247                 return 0;
 248         }
 249
 250         error = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 251                                  LM_ST_SHARED, 0, &sb_gh);
 252         if (error) {
 253                 fs_err(sdp, "can't acquire superblock glock: %d\n", error);
 254                 return error;
 255         }
 256
 257         error = gfs2_read_sb(sdp, sb_gh.gh_gl, silent);
 258         if (error) {
 259                 fs_err(sdp, "can't read superblock: %d\n", error);
 260                 goto out;
 261         }
 262
 263         /* Set up the buffer cache and SB for real */
 264         if (sdp->sd_sb.sb_bsize < bdev_hardsect_size(sb->s_bdev)) {
 265                 error = -EINVAL;
 266                 fs_err(sdp, "FS block size (%u) is too small for device "
 267                        "block size (%u)\n",
 268                        sdp->sd_sb.sb_bsize, bdev_hardsect_size(sb->s_bdev));
 269                 goto out;
 270         }
 271         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 272                 error = -EINVAL;
 273                 fs_err(sdp, "FS block size (%u) is too big for machine "
 274                        "page size (%u)\n",
 275                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 276                 goto out;
 277         }
 278         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 279
 280         /* Get the root inode */
 281         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 282         if (sb->s_type == &gfs2meta_fs_type)
 283                 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 284         inode = gfs2_lookup_root(sb, no_addr);
 285         if (IS_ERR(inode)) {
 286                 error = PTR_ERR(inode);
 287                 fs_err(sdp, "can't read in root inode: %d\n", error);
 288                 goto out;
 289         }
 290
 291         sb->s_root = d_alloc_root(inode);
 292         if (!sb->s_root) {
 293                 fs_err(sdp, "can't get root dentry\n");
 294                 error = -ENOMEM;
 295                 iput(inode);
 296         } else
 297                 sb->s_root->d_op = &gfs2_dops;
 298
 299 out:
 300         gfs2_glock_dq_uninit(&sb_gh);
 301         return error;
 302 }
 303
 304 /**
 305  * map_journal_extents - create a reusable "extent" mapping from all logical
 306  * blocks to all physical blocks for the given journal.  This will save
 307  * us time when writing journal blocks.  Most journals will have only one
 308  * extent that maps all their logical blocks.  That's because gfs2.mkfs
 309  * arranges the journal blocks sequentially to maximize performance.
 310  * So the extent would map the first block for the entire file length.
 311  * However, gfs2_jadd can happen while file activity is happening, so
 312  * those journals may not be sequential.  Less likely is the case where
 313  * the users created their own journals by mounting the metafs and
 314  * laying it out.  But it's still possible.  These journals might have
 315  * several extents.
 316  *
 317  * TODO: This should be done in bigger chunks rather than one block at a time,
 318  *       but since it's only done at mount time, I'm not worried about the
 319  *       time it takes.
 320  */
 321 static int map_journal_extents(struct gfs2_sbd *sdp)
 322 {
 323         struct gfs2_jdesc *jd = sdp->sd_jdesc;
 324         unsigned int lb;
 325         u64 db, prev_db; /* logical block, disk block, prev disk block */
 326         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 327         struct gfs2_journal_extent *jext = NULL;
 328         struct buffer_head bh;
 329         int rc = 0;
 330
 331         prev_db = 0;
 332
 333         for (lb = 0; lb < ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift; lb++) {
 334                 bh.b_state = 0;
 335                 bh.b_blocknr = 0;
 336                 bh.b_size = 1 << ip->i_inode.i_blkbits;
 337                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
 338                 db = bh.b_blocknr;
 339                 if (rc || !db) {
 340                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
 341                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
 342                         break;
 343                 }
 344                 if (!prev_db || db != prev_db + 1) {
 345                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
 346                                        GFP_KERNEL);
 347                         if (!jext) {
 348                                 printk(KERN_INFO "GFS2 error: out of memory "
 349                                        "mapping journal extents.\n");
 350                                 rc = -ENOMEM;
 351                                 break;
 352                         }
 353                         jext->dblock = db;
 354                         jext->lblock = lb;
 355                         jext->blocks = 1;
 356                         list_add_tail(&jext->extent_list, &jd->extent_list);
 357                 } else {
 358                         jext->blocks++;
 359                 }
 360                 prev_db = db;
 361         }
 362         return rc;
 363 }
 364
 365 static void gfs2_lm_others_may_mount(struct gfs2_sbd *sdp)
 366 {
 367         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 368                 sdp->sd_lockstruct.ls_ops->lm_others_may_mount(
 369                                         sdp->sd_lockstruct.ls_lockspace);
 370 }
 371
 372 static int init_journal(struct gfs2_sbd *sdp, int undo)
 373 {
 374         struct gfs2_holder ji_gh;
 375         struct task_struct *p;
 376         struct gfs2_inode *ip;
 377         int jindex = 1;
 378         int error = 0;
 379
 380         if (undo) {
 381                 jindex = 0;
 382                 goto fail_recoverd;
 383         }
 384
 385         sdp->sd_jindex = gfs2_lookup_simple(sdp->sd_master_dir, "jindex");
 386         if (IS_ERR(sdp->sd_jindex)) {
 387                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 388                 return PTR_ERR(sdp->sd_jindex);
 389         }
 390         ip = GFS2_I(sdp->sd_jindex);
 391         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
 392
 393         /* Load in the journal index special file */
 394
 395         error = gfs2_jindex_hold(sdp, &ji_gh);
 396         if (error) {
 397                 fs_err(sdp, "can't read journal index: %d\n", error);
 398                 goto fail;
 399         }
 400
 401         error = -EINVAL;
 402         if (!gfs2_jindex_size(sdp)) {
 403                 fs_err(sdp, "no journals!\n");
 404                 goto fail_jindex;
 405         }
 406
 407         if (sdp->sd_args.ar_spectator) {
 408                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 409                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 410         } else {
 411                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 412                         fs_err(sdp, "can't mount journal #%u\n",
 413                                sdp->sd_lockstruct.ls_jid);
 414                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 415                                gfs2_jindex_size(sdp),
 416                                gfs2_jindex_size(sdp) - 1);
 417                         goto fail_jindex;
 418                 }
 419                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 420
 421                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 422                                           &gfs2_journal_glops,
 423                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 424                                           &sdp->sd_journal_gh);
 425                 if (error) {
 426                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 427                         goto fail_jindex;
 428                 }
 429
 430                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 431                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 432                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 433                                            &sdp->sd_jinode_gh);
 434                 if (error) {
 435                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 436                                error);
 437                         goto fail_journal_gh;
 438                 }
 439
 440                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 441                 if (error) {
 442                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 443                                sdp->sd_jdesc->jd_jid, error);
 444                         goto fail_jinode_gh;
 445                 }
 446                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 447
 448                 /* Map the extents for this journal's blocks */
 449                 map_journal_extents(sdp);
 450         }
 451
 452         if (sdp->sd_lockstruct.ls_first) {
 453                 unsigned int x;
 454                 for (x = 0; x < sdp->sd_journals; x++) {
 455                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
 456                         if (error) {
 457                                 fs_err(sdp, "error recovering journal %u: %d\n",
 458                                        x, error);
 459                                 goto fail_jinode_gh;
 460                         }
 461                 }
 462
 463                 gfs2_lm_others_may_mount(sdp);
 464         } else if (!sdp->sd_args.ar_spectator) {
 465                 error = gfs2_recover_journal(sdp->sd_jdesc);
 466                 if (error) {
 467                         fs_err(sdp, "error recovering my journal: %d\n", error);
 468                         goto fail_jinode_gh;
 469                 }
 470         }
 471
 472         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 473         gfs2_glock_dq_uninit(&ji_gh);
 474         jindex = 0;
 475
 476         p = kthread_run(gfs2_recoverd, sdp, "gfs2_recoverd");
 477         error = IS_ERR(p);
 478         if (error) {
 479                 fs_err(sdp, "can't start recoverd thread: %d\n", error);
 480                 goto fail_jinode_gh;
 481         }
 482         sdp->sd_recoverd_process = p;
 483
 484         return 0;
 485
 486 fail_recoverd:
 487         kthread_stop(sdp->sd_recoverd_process);
 488 fail_jinode_gh:
 489         if (!sdp->sd_args.ar_spectator)
 490                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 491 fail_journal_gh:
 492         if (!sdp->sd_args.ar_spectator)
 493                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 494 fail_jindex:
 495         gfs2_jindex_free(sdp);
 496         if (jindex)
 497                 gfs2_glock_dq_uninit(&ji_gh);
 498 fail:
 499         iput(sdp->sd_jindex);
 500         return error;
 501 }
 502
 503
 504 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 505 {
 506         int error = 0;
 507         struct gfs2_inode *ip;
 508         struct inode *inode;
 509
 510         if (undo)
 511                 goto fail_qinode;
 512
 513         inode = gfs2_lookup_root(sdp->sd_vfs, sdp->sd_sb.sb_master_dir.no_addr);
 514         if (IS_ERR(inode)) {
 515                 error = PTR_ERR(inode);
 516                 fs_err(sdp, "can't read in master directory: %d\n", error);
 517                 goto fail;
 518         }
 519         sdp->sd_master_dir = inode;
 520
 521         error = init_journal(sdp, undo);
 522         if (error)
 523                 goto fail_master;
 524
 525         /* Read in the master inode number inode */
 526         sdp->sd_inum_inode = gfs2_lookup_simple(sdp->sd_master_dir, "inum");
 527         if (IS_ERR(sdp->sd_inum_inode)) {
 528                 error = PTR_ERR(sdp->sd_inum_inode);
 529                 fs_err(sdp, "can't read in inum inode: %d\n", error);
 530                 goto fail_journal;
 531         }
 532
 533
 534         /* Read in the master statfs inode */
 535         sdp->sd_statfs_inode = gfs2_lookup_simple(sdp->sd_master_dir, "statfs");
 536         if (IS_ERR(sdp->sd_statfs_inode)) {
 537                 error = PTR_ERR(sdp->sd_statfs_inode);
 538                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 539                 goto fail_inum;
 540         }
 541
 542         /* Read in the resource index inode */
 543         sdp->sd_rindex = gfs2_lookup_simple(sdp->sd_master_dir, "rindex");
 544         if (IS_ERR(sdp->sd_rindex)) {
 545                 error = PTR_ERR(sdp->sd_rindex);
 546                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 547                 goto fail_statfs;
 548         }
 549         ip = GFS2_I(sdp->sd_rindex);
 550         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
 551         sdp->sd_rindex_uptodate = 0;
 552
 553         /* Read in the quota inode */
 554         sdp->sd_quota_inode = gfs2_lookup_simple(sdp->sd_master_dir, "quota");
 555         if (IS_ERR(sdp->sd_quota_inode)) {
 556                 error = PTR_ERR(sdp->sd_quota_inode);
 557                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 558                 goto fail_rindex;
 559         }
 560         return 0;
 561
 562 fail_qinode:
 563         iput(sdp->sd_quota_inode);
 564 fail_rindex:
 565         gfs2_clear_rgrpd(sdp);
 566         iput(sdp->sd_rindex);
 567 fail_statfs:
 568         iput(sdp->sd_statfs_inode);
 569 fail_inum:
 570         iput(sdp->sd_inum_inode);
 571 fail_journal:
 572         init_journal(sdp, UNDO);
 573 fail_master:
 574         iput(sdp->sd_master_dir);
 575 fail:
 576         return error;
 577 }
 578
 579 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 580 {
 581         struct inode *pn = NULL;
 582         char buf[30];
 583         int error = 0;
 584         struct gfs2_inode *ip;
 585
 586         if (sdp->sd_args.ar_spectator)
 587                 return 0;
 588
 589         if (undo)
 590                 goto fail_qc_gh;
 591
 592         pn = gfs2_lookup_simple(sdp->sd_master_dir, "per_node");
 593         if (IS_ERR(pn)) {
 594                 error = PTR_ERR(pn);
 595                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 596                 return error;
 597         }
 598
 599         sprintf(buf, "inum_range%u", sdp->sd_jdesc->jd_jid);
 600         sdp->sd_ir_inode = gfs2_lookup_simple(pn, buf);
 601         if (IS_ERR(sdp->sd_ir_inode)) {
 602                 error = PTR_ERR(sdp->sd_ir_inode);
 603                 fs_err(sdp, "can't find local \"ir\" file: %d\n", error);
 604                 goto fail;
 605         }
 606
 607         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 608         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 609         if (IS_ERR(sdp->sd_sc_inode)) {
 610                 error = PTR_ERR(sdp->sd_sc_inode);
 611                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 612                 goto fail_ir_i;
 613         }
 614
 615         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 616         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 617         if (IS_ERR(sdp->sd_qc_inode)) {
 618                 error = PTR_ERR(sdp->sd_qc_inode);
 619                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 620                 goto fail_ut_i;
 621         }
 622
 623         iput(pn);
 624         pn = NULL;
 625
 626         ip = GFS2_I(sdp->sd_ir_inode);
 627         error = gfs2_glock_nq_init(ip->i_gl,
 628                                    LM_ST_EXCLUSIVE, 0,
 629                                    &sdp->sd_ir_gh);
 630         if (error) {
 631                 fs_err(sdp, "can't lock local \"ir\" file: %d\n", error);
 632                 goto fail_qc_i;
 633         }
 634
 635         ip = GFS2_I(sdp->sd_sc_inode);
 636         error = gfs2_glock_nq_init(ip->i_gl,
 637                                    LM_ST_EXCLUSIVE, 0,
 638                                    &sdp->sd_sc_gh);
 639         if (error) {
 640                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 641                 goto fail_ir_gh;
 642         }
 643
 644         ip = GFS2_I(sdp->sd_qc_inode);
 645         error = gfs2_glock_nq_init(ip->i_gl,
 646                                    LM_ST_EXCLUSIVE, 0,
 647                                    &sdp->sd_qc_gh);
 648         if (error) {
 649                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 650                 goto fail_ut_gh;
 651         }
 652
 653         return 0;
 654
 655 fail_qc_gh:
 656         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 657 fail_ut_gh:
 658         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 659 fail_ir_gh:
 660         gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
 661 fail_qc_i:
 662         iput(sdp->sd_qc_inode);
 663 fail_ut_i:
 664         iput(sdp->sd_sc_inode);
 665 fail_ir_i:
 666         iput(sdp->sd_ir_inode);
 667 fail:
 668         if (pn)
 669                 iput(pn);
 670         return error;
 671 }
 672
 673 static int init_threads(struct gfs2_sbd *sdp, int undo)
 674 {
 675         struct task_struct *p;
 676         int error = 0;
 677
 678         if (undo)
 679                 goto fail_quotad;
 680
 681         sdp->sd_log_flush_time = jiffies;
 682         sdp->sd_jindex_refresh_time = jiffies;
 683
 684         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 685         error = IS_ERR(p);
 686         if (error) {
 687                 fs_err(sdp, "can't start logd thread: %d\n", error);
 688                 return error;
 689         }
 690         sdp->sd_logd_process = p;
 691
 692         sdp->sd_statfs_sync_time = jiffies;
 693         sdp->sd_quota_sync_time = jiffies;
 694
 695         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 696         error = IS_ERR(p);
 697         if (error) {
 698                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 699                 goto fail;
 700         }
 701         sdp->sd_quotad_process = p;
 702
 703         return 0;
 704
 705
 706 fail_quotad:
 707         kthread_stop(sdp->sd_quotad_process);
 708 fail:
 709         kthread_stop(sdp->sd_logd_process);
 710         return error;
 711 }
 712
 713 /**
 714  * gfs2_lm_mount - mount a locking protocol
 715  * @sdp: the filesystem
 716  * @args: mount arguements
 717  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 718  *
 719  * Returns: errno
 720  */
 721
 722 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 723 {
 724         char *proto = sdp->sd_proto_name;
 725         char *table = sdp->sd_table_name;
 726         int flags = LM_MFLAG_CONV_NODROP;
 727         int error;
 728
 729         if (sdp->sd_args.ar_spectator)
 730                 flags |= LM_MFLAG_SPECTATOR;
 731
 732         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 733
 734         error = gfs2_mount_lockproto(proto, table, sdp->sd_args.ar_hostdata,
 735                                      gfs2_glock_cb, sdp,
 736                                      GFS2_MIN_LVB_SIZE, flags,
 737                                      &sdp->sd_lockstruct, &sdp->sd_kobj);
 738         if (error) {
 739                 fs_info(sdp, "can't mount proto=%s, table=%s, hostdata=%s\n",
 740                         proto, table, sdp->sd_args.ar_hostdata);
 741                 goto out;
 742         }
 743
 744         if (gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lockspace) ||
 745             gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_ops) ||
 746             gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lvb_size >=
 747                                   GFS2_MIN_LVB_SIZE)) {
 748                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
 749                 goto out;
 750         }
 751
 752         if (sdp->sd_args.ar_spectator)
 753                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
 754         else
 755                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
 756                          sdp->sd_lockstruct.ls_jid);
 757
 758         fs_info(sdp, "Joined cluster. Now mounting FS...\n");
 759
 760         if ((sdp->sd_lockstruct.ls_flags & LM_LSFLAG_LOCAL) &&
 761             !sdp->sd_args.ar_ignore_local_fs) {
 762                 sdp->sd_args.ar_localflocks = 1;
 763                 sdp->sd_args.ar_localcaching = 1;
 764         }
 765
 766 out:
 767         return error;
 768 }
 769
 770 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
 771 {
 772         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 773                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
 774 }
 775
 776 /**
 777  * fill_super - Read in superblock
 778  * @sb: The VFS superblock
 779  * @data: Mount options
 780  * @silent: Don't complain if it's not a GFS2 filesystem
 781  *
 782  * Returns: errno
 783  */
 784
 785 static int fill_super(struct super_block *sb, void *data, int silent)
 786 {
 787         struct gfs2_sbd *sdp;
 788         struct gfs2_holder mount_gh;
 789         int error;
 790
 791         sdp = init_sbd(sb);
 792         if (!sdp) {
 793                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
 794                 return -ENOMEM;
 795         }
 796
 797         error = gfs2_mount_args(sdp, (char *)data, 0);
 798         if (error) {
 799                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
 800                 goto fail;
 801         }
 802
 803         init_vfs(sb, SDF_NOATIME);
 804
 805         /* Set up the buffer cache and fill in some fake block size values
 806            to allow us to read-in the on-disk superblock. */
 807         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
 808         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
 809         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 810                                GFS2_BASIC_BLOCK_SHIFT;
 811         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 812
 813         error = init_names(sdp, silent);
 814         if (error)
 815                 goto fail;
 816
 817         gfs2_create_debugfs_file(sdp);
 818
 819         error = gfs2_sys_fs_add(sdp);
 820         if (error)
 821                 goto fail;
 822
 823         error = gfs2_lm_mount(sdp, silent);
 824         if (error)
 825                 goto fail_sys;
 826
 827         error = init_locking(sdp, &mount_gh, DO);
 828         if (error)
 829                 goto fail_lm;
 830
 831         error = init_sb(sdp, silent, DO);
 832         if (error)
 833                 goto fail_locking;
 834
 835         error = init_inodes(sdp, DO);
 836         if (error)
 837                 goto fail_sb;
 838
 839         error = init_per_node(sdp, DO);
 840         if (error)
 841                 goto fail_inodes;
 842
 843         error = gfs2_statfs_init(sdp);
 844         if (error) {
 845                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
 846                 goto fail_per_node;
 847         }
 848
 849         error = init_threads(sdp, DO);
 850         if (error)
 851                 goto fail_per_node;
 852
 853         if (!(sb->s_flags & MS_RDONLY)) {
 854                 error = gfs2_make_fs_rw(sdp);
 855                 if (error) {
 856                         fs_err(sdp, "can't make FS RW: %d\n", error);
 857                         goto fail_threads;
 858                 }
 859         }
 860
 861         gfs2_glock_dq_uninit(&mount_gh);
 862
 863         return 0;
 864
 865 fail_threads:
 866         init_threads(sdp, UNDO);
 867 fail_per_node:
 868         init_per_node(sdp, UNDO);
 869 fail_inodes:
 870         init_inodes(sdp, UNDO);
 871 fail_sb:
 872         init_sb(sdp, 0, UNDO);
 873 fail_locking:
 874         init_locking(sdp, &mount_gh, UNDO);
 875 fail_lm:
 876         gfs2_gl_hash_clear(sdp, WAIT);
 877         gfs2_lm_unmount(sdp);
 878         while (invalidate_inodes(sb))
 879                 yield();
 880 fail_sys:
 881         gfs2_sys_fs_del(sdp);
 882 fail:
 883         gfs2_delete_debugfs_file(sdp);
 884         kfree(sdp);
 885         sb->s_fs_info = NULL;
 886         return error;
 887 }
 888
 889 static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
 890                 const char *dev_name, void *data, struct vfsmount *mnt)
 891 {
 892         struct super_block *sb;
 893         struct gfs2_sbd *sdp;
 894         int error = get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
 895         if (error)
 896                 goto out;
 897         sb = mnt->mnt_sb;
 898         sdp = sb->s_fs_info;
 899         sdp->sd_gfs2mnt = mnt;
 900 out:
 901         return error;
 902 }
 903
 904 static int fill_super_meta(struct super_block *sb, struct super_block *new,
 905                            void *data, int silent)
 906 {
 907         struct gfs2_sbd *sdp = sb->s_fs_info;
 908         struct inode *inode;
 909         int error = 0;
 910
 911         new->s_fs_info = sdp;
 912         sdp->sd_vfs_meta = sb;
 913
 914         init_vfs(new, SDF_NOATIME);
 915
 916         /* Get the master inode */
 917         inode = igrab(sdp->sd_master_dir);
 918
 919         new->s_root = d_alloc_root(inode);
 920         if (!new->s_root) {
 921                 fs_err(sdp, "can't get root dentry\n");
 922                 error = -ENOMEM;
 923                 iput(inode);
 924         } else
 925                 new->s_root->d_op = &gfs2_dops;
 926
 927         return error;
 928 }
 929
 930 static int set_bdev_super(struct super_block *s, void *data)
 931 {
 932         s->s_bdev = data;
 933         s->s_dev = s->s_bdev->bd_dev;
 934         return 0;
 935 }
 936
 937 static int test_bdev_super(struct super_block *s, void *data)
 938 {
 939         return s->s_bdev == data;
 940 }
 941
 942 static struct super_block* get_gfs2_sb(const char *dev_name)
 943 {
 944         struct kstat stat;
 945         struct nameidata nd;
 946         struct super_block *sb = NULL, *s;
 947         int error;
 948
 949         error = path_lookup(dev_name, LOOKUP_FOLLOW, &nd);
 950         if (error) {
 951                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error\n",
 952                        dev_name);
 953                 goto out;
 954         }
 955         error = vfs_getattr(nd.path.mnt, nd.path.dentry, &stat);
 956
 957         list_for_each_entry(s, &gfs2_fs_type.fs_supers, s_instances) {
 958                 if ((S_ISBLK(stat.mode) && s->s_dev == stat.rdev) ||
 959                     (S_ISDIR(stat.mode) &&
 960                      s == nd.path.dentry->d_inode->i_sb)) {
 961                         sb = s;
 962                         goto free_nd;
 963                 }
 964         }
 965
 966         printk(KERN_WARNING "GFS2: Unrecognized block device or "
 967                "mount point %s\n", dev_name);
 968
 969 free_nd:
 970         path_put(&nd.path);
 971 out:
 972         return sb;
 973 }
 974
 975 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
 976                             const char *dev_name, void *data, struct vfsmount *mnt)
 977 {
 978         int error = 0;
 979         struct super_block *sb = NULL, *new;
 980         struct gfs2_sbd *sdp;
 981
 982         sb = get_gfs2_sb(dev_name);
 983         if (!sb) {
 984                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
 985                 error = -ENOENT;
 986                 goto error;
 987         }
 988         sdp = sb->s_fs_info;
 989         if (sdp->sd_vfs_meta) {
 990                 printk(KERN_WARNING "GFS2: gfs2meta mount already exists\n");
 991                 error = -EBUSY;
 992                 goto error;
 993         }
 994         down(&sb->s_bdev->bd_mount_sem);
 995         new = sget(fs_type, test_bdev_super, set_bdev_super, sb->s_bdev);
 996         up(&sb->s_bdev->bd_mount_sem);
 997         if (IS_ERR(new)) {
 998                 error = PTR_ERR(new);
 999                 goto error;
1000         }
1001         new->s_flags = flags;
1002         strlcpy(new->s_id, sb->s_id, sizeof(new->s_id));
1003         sb_set_blocksize(new, sb->s_blocksize);
1004         error = fill_super_meta(sb, new, data, flags & MS_SILENT ? 1 : 0);
1005         if (error) {
1006                 up_write(&new->s_umount);
1007                 deactivate_super(new);
1008                 goto error;
1009         }
1010
1011         new->s_flags |= MS_ACTIVE;
1012
1013         /* Grab a reference to the gfs2 mount point */
1014         atomic_inc(&sdp->sd_gfs2mnt->mnt_count);
1015         return simple_set_mnt(mnt, new);
1016 error:
1017         return error;
1018 }
1019
1020 static void gfs2_kill_sb(struct super_block *sb)
1021 {
1022         if (sb->s_fs_info) {
1023                 gfs2_delete_debugfs_file(sb->s_fs_info);
1024                 gfs2_meta_syncfs(sb->s_fs_info);
1025         }
1026         kill_block_super(sb);
1027 }
1028
1029 static void gfs2_kill_sb_meta(struct super_block *sb)
1030 {
1031         struct gfs2_sbd *sdp = sb->s_fs_info;
1032         generic_shutdown_super(sb);
1033         sdp->sd_vfs_meta = NULL;
1034         atomic_dec(&sdp->sd_gfs2mnt->mnt_count);
1035 }
1036
1037 struct file_system_type gfs2_fs_type = {
1038         .name = "gfs2",
1039         .fs_flags = FS_REQUIRES_DEV,
1040         .get_sb = gfs2_get_sb,
1041         .kill_sb = gfs2_kill_sb,
1042         .owner = THIS_MODULE,
1043 };
1044
1045 struct file_system_type gfs2meta_fs_type = {
1046         .name = "gfs2meta",
1047         .fs_flags = FS_REQUIRES_DEV,
1048         .get_sb = gfs2_get_sb_meta,
1049         .kill_sb = gfs2_kill_sb_meta,
1050         .owner = THIS_MODULE,
1051 };
1052