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/export.h>
  18 #include <linux/namei.h>
  19 #include <linux/mount.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/quotaops.h>
  22 #include <linux/lockdep.h>
  23 #include <linux/module.h>
  24
  25 #include "gfs2.h"
  26 #include "incore.h"
  27 #include "bmap.h"
  28 #include "glock.h"
  29 #include "glops.h"
  30 #include "inode.h"
  31 #include "recovery.h"
  32 #include "rgrp.h"
  33 #include "super.h"
  34 #include "sys.h"
  35 #include "util.h"
  36 #include "log.h"
  37 #include "quota.h"
  38 #include "dir.h"
  39 #include "trace_gfs2.h"
  40
  41 #define DO 0
  42 #define UNDO 1
  43
  44 /**
  45  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  46  * @gt: tune
  47  *
  48  */
  49
  50 static void gfs2_tune_init(struct gfs2_tune *gt)
  51 {
  52         spin_lock_init(&gt->gt_spin);
  53
  54         gt->gt_quota_simul_sync = 64;
  55         gt->gt_quota_warn_period = 10;
  56         gt->gt_quota_scale_num = 1;
  57         gt->gt_quota_scale_den = 1;
  58         gt->gt_new_files_jdata = 0;
  59         gt->gt_max_readahead = 1 << 18;
  60         gt->gt_complain_secs = 10;
  61 }
  62
  63 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  64 {
  65         struct gfs2_sbd *sdp;
  66
  67         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  68         if (!sdp)
  69                 return NULL;
  70
  71         sb->s_fs_info = sdp;
  72         sdp->sd_vfs = sb;
  73         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  74         if (!sdp->sd_lkstats) {
  75                 kfree(sdp);
  76                 return NULL;
  77         }
  78
  79         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  80         gfs2_tune_init(&sdp->sd_tune);
  81
  82         init_waitqueue_head(&sdp->sd_glock_wait);
  83         atomic_set(&sdp->sd_glock_disposal, 0);
  84         init_completion(&sdp->sd_locking_init);
  85         init_completion(&sdp->sd_wdack);
  86         spin_lock_init(&sdp->sd_statfs_spin);
  87
  88         spin_lock_init(&sdp->sd_rindex_spin);
  89         sdp->sd_rindex_tree.rb_node = NULL;
  90
  91         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  92         spin_lock_init(&sdp->sd_jindex_spin);
  93         mutex_init(&sdp->sd_jindex_mutex);
  94
  95         INIT_LIST_HEAD(&sdp->sd_quota_list);
  96         mutex_init(&sdp->sd_quota_mutex);
  97         init_waitqueue_head(&sdp->sd_quota_wait);
  98         INIT_LIST_HEAD(&sdp->sd_trunc_list);
  99         spin_lock_init(&sdp->sd_trunc_lock);
 100
 101         spin_lock_init(&sdp->sd_log_lock);
 102         atomic_set(&sdp->sd_log_pinned, 0);
 103         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
 104         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
 105         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
 106         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 107         spin_lock_init(&sdp->sd_ordered_lock);
 108
 109         init_waitqueue_head(&sdp->sd_log_waitq);
 110         init_waitqueue_head(&sdp->sd_logd_waitq);
 111         spin_lock_init(&sdp->sd_ail_lock);
 112         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 113         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 114
 115         init_rwsem(&sdp->sd_log_flush_lock);
 116         atomic_set(&sdp->sd_log_in_flight, 0);
 117         init_waitqueue_head(&sdp->sd_log_flush_wait);
 118
 119         INIT_LIST_HEAD(&sdp->sd_revoke_list);
 120
 121         return sdp;
 122 }
 123
 124
 125 /**
 126  * gfs2_check_sb - Check superblock
 127  * @sdp: the filesystem
 128  * @sb: The superblock
 129  * @silent: Don't print a message if the check fails
 130  *
 131  * Checks the version code of the FS is one that we understand how to
 132  * read and that the sizes of the various on-disk structures have not
 133  * changed.
 134  */
 135
 136 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 137 {
 138         struct gfs2_sb_host *sb = &sdp->sd_sb;
 139
 140         if (sb->sb_magic != GFS2_MAGIC ||
 141             sb->sb_type != GFS2_METATYPE_SB) {
 142                 if (!silent)
 143                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 144                 return -EINVAL;
 145         }
 146
 147         /*  If format numbers match exactly, we're done.  */
 148
 149         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 150             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 151                 return 0;
 152
 153         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 154
 155         return -EINVAL;
 156 }
 157
 158 static void end_bio_io_page(struct bio *bio, int error)
 159 {
 160         struct page *page = bio->bi_private;
 161
 162         if (!error)
 163                 SetPageUptodate(page);
 164         else
 165                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 166         unlock_page(page);
 167 }
 168
 169 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 170 {
 171         struct gfs2_sb_host *sb = &sdp->sd_sb;
 172         struct super_block *s = sdp->sd_vfs;
 173         const struct gfs2_sb *str = buf;
 174
 175         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 176         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 177         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 178         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 179         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 180         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 181         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 182         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 183         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 184         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 185         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 186
 187         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 188         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 189         memcpy(s->s_uuid, str->sb_uuid, 16);
 190 }
 191
 192 /**
 193  * gfs2_read_super - Read the gfs2 super block from disk
 194  * @sdp: The GFS2 super block
 195  * @sector: The location of the super block
 196  * @error: The error code to return
 197  *
 198  * This uses the bio functions to read the super block from disk
 199  * because we want to be 100% sure that we never read cached data.
 200  * A super block is read twice only during each GFS2 mount and is
 201  * never written to by the filesystem. The first time its read no
 202  * locks are held, and the only details which are looked at are those
 203  * relating to the locking protocol. Once locking is up and working,
 204  * the sb is read again under the lock to establish the location of
 205  * the master directory (contains pointers to journals etc) and the
 206  * root directory.
 207  *
 208  * Returns: 0 on success or error
 209  */
 210
 211 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 212 {
 213         struct super_block *sb = sdp->sd_vfs;
 214         struct gfs2_sb *p;
 215         struct page *page;
 216         struct bio *bio;
 217
 218         page = alloc_page(GFP_NOFS);
 219         if (unlikely(!page))
 220                 return -ENOBUFS;
 221
 222         ClearPageUptodate(page);
 223         ClearPageDirty(page);
 224         lock_page(page);
 225
 226         bio = bio_alloc(GFP_NOFS, 1);
 227         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 228         bio->bi_bdev = sb->s_bdev;
 229         bio_add_page(bio, page, PAGE_SIZE, 0);
 230
 231         bio->bi_end_io = end_bio_io_page;
 232         bio->bi_private = page;
 233         submit_bio(READ_SYNC | REQ_META, bio);
 234         wait_on_page_locked(page);
 235         bio_put(bio);
 236         if (!PageUptodate(page)) {
 237                 __free_page(page);
 238                 return -EIO;
 239         }
 240         p = kmap(page);
 241         gfs2_sb_in(sdp, p);
 242         kunmap(page);
 243         __free_page(page);
 244         return gfs2_check_sb(sdp, silent);
 245 }
 246
 247 /**
 248  * gfs2_read_sb - Read super block
 249  * @sdp: The GFS2 superblock
 250  * @silent: Don't print message if mount fails
 251  *
 252  */
 253
 254 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 255 {
 256         u32 hash_blocks, ind_blocks, leaf_blocks;
 257         u32 tmp_blocks;
 258         unsigned int x;
 259         int error;
 260
 261         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 262         if (error) {
 263                 if (!silent)
 264                         fs_err(sdp, "can't read superblock\n");
 265                 return error;
 266         }
 267
 268         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 269                                GFS2_BASIC_BLOCK_SHIFT;
 270         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 271         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 272                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 273         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 274                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 275         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 276         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 277         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 278         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 279         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 280                                 sizeof(struct gfs2_meta_header)) /
 281                                 sizeof(struct gfs2_quota_change);
 282         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 283                                      sizeof(struct gfs2_meta_header))
 284                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 285
 286         /* Compute maximum reservation required to add a entry to a directory */
 287
 288         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 289                              sdp->sd_jbsize);
 290
 291         ind_blocks = 0;
 292         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 293                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 294                 ind_blocks += tmp_blocks;
 295         }
 296
 297         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 298
 299         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 300
 301         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 302                                 sizeof(struct gfs2_dinode);
 303         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 304         for (x = 2;; x++) {
 305                 u64 space, d;
 306                 u32 m;
 307
 308                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 309                 d = space;
 310                 m = do_div(d, sdp->sd_inptrs);
 311
 312                 if (d != sdp->sd_heightsize[x - 1] || m)
 313                         break;
 314                 sdp->sd_heightsize[x] = space;
 315         }
 316         sdp->sd_max_height = x;
 317         sdp->sd_heightsize[x] = ~0;
 318         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 319
 320         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 321                                  sizeof(struct gfs2_dinode);
 322         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 323         for (x = 2;; x++) {
 324                 u64 space, d;
 325                 u32 m;
 326
 327                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 328                 d = space;
 329                 m = do_div(d, sdp->sd_inptrs);
 330
 331                 if (d != sdp->sd_jheightsize[x - 1] || m)
 332                         break;
 333                 sdp->sd_jheightsize[x] = space;
 334         }
 335         sdp->sd_max_jheight = x;
 336         sdp->sd_jheightsize[x] = ~0;
 337         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 338
 339         return 0;
 340 }
 341
 342 static int init_names(struct gfs2_sbd *sdp, int silent)
 343 {
 344         char *proto, *table;
 345         int error = 0;
 346
 347         proto = sdp->sd_args.ar_lockproto;
 348         table = sdp->sd_args.ar_locktable;
 349
 350         /*  Try to autodetect  */
 351
 352         if (!proto[0] || !table[0]) {
 353                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 354                 if (error)
 355                         return error;
 356
 357                 if (!proto[0])
 358                         proto = sdp->sd_sb.sb_lockproto;
 359                 if (!table[0])
 360                         table = sdp->sd_sb.sb_locktable;
 361         }
 362
 363         if (!table[0])
 364                 table = sdp->sd_vfs->s_id;
 365
 366         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 367         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 368
 369         table = sdp->sd_table_name;
 370         while ((table = strchr(table, '/')))
 371                 *table = '_';
 372
 373         return error;
 374 }
 375
 376 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 377                         int undo)
 378 {
 379         int error = 0;
 380
 381         if (undo)
 382                 goto fail_trans;
 383
 384         error = gfs2_glock_nq_num(sdp,
 385                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 386                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 387                                   mount_gh);
 388         if (error) {
 389                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 390                 goto fail;
 391         }
 392
 393         error = gfs2_glock_nq_num(sdp,
 394                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 395                                   LM_ST_SHARED,
 396                                   LM_FLAG_NOEXP | GL_EXACT,
 397                                   &sdp->sd_live_gh);
 398         if (error) {
 399                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 400                 goto fail_mount;
 401         }
 402
 403         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 404                                CREATE, &sdp->sd_rename_gl);
 405         if (error) {
 406                 fs_err(sdp, "can't create rename glock: %d\n", error);
 407                 goto fail_live;
 408         }
 409
 410         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 411                                CREATE, &sdp->sd_trans_gl);
 412         if (error) {
 413                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 414                 goto fail_rename;
 415         }
 416
 417         return 0;
 418
 419 fail_trans:
 420         gfs2_glock_put(sdp->sd_trans_gl);
 421 fail_rename:
 422         gfs2_glock_put(sdp->sd_rename_gl);
 423 fail_live:
 424         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 425 fail_mount:
 426         gfs2_glock_dq_uninit(mount_gh);
 427 fail:
 428         return error;
 429 }
 430
 431 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 432                             u64 no_addr, const char *name)
 433 {
 434         struct gfs2_sbd *sdp = sb->s_fs_info;
 435         struct dentry *dentry;
 436         struct inode *inode;
 437
 438         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 439         if (IS_ERR(inode)) {
 440                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 441                 return PTR_ERR(inode);
 442         }
 443         dentry = d_make_root(inode);
 444         if (!dentry) {
 445                 fs_err(sdp, "can't alloc %s dentry\n", name);
 446                 return -ENOMEM;
 447         }
 448         *dptr = dentry;
 449         return 0;
 450 }
 451
 452 static int init_sb(struct gfs2_sbd *sdp, int silent)
 453 {
 454         struct super_block *sb = sdp->sd_vfs;
 455         struct gfs2_holder sb_gh;
 456         u64 no_addr;
 457         int ret;
 458
 459         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 460                                 LM_ST_SHARED, 0, &sb_gh);
 461         if (ret) {
 462                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 463                 return ret;
 464         }
 465
 466         ret = gfs2_read_sb(sdp, silent);
 467         if (ret) {
 468                 fs_err(sdp, "can't read superblock: %d\n", ret);
 469                 goto out;
 470         }
 471
 472         /* Set up the buffer cache and SB for real */
 473         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 474                 ret = -EINVAL;
 475                 fs_err(sdp, "FS block size (%u) is too small for device "
 476                        "block size (%u)\n",
 477                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 478                 goto out;
 479         }
 480         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 481                 ret = -EINVAL;
 482                 fs_err(sdp, "FS block size (%u) is too big for machine "
 483                        "page size (%u)\n",
 484                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 485                 goto out;
 486         }
 487         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 488
 489         /* Get the root inode */
 490         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 491         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 492         if (ret)
 493                 goto out;
 494
 495         /* Get the master inode */
 496         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 497         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 498         if (ret) {
 499                 dput(sdp->sd_root_dir);
 500                 goto out;
 501         }
 502         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 503 out:
 504         gfs2_glock_dq_uninit(&sb_gh);
 505         return ret;
 506 }
 507
 508 /**
 509  * map_journal_extents - create a reusable "extent" mapping from all logical
 510  * blocks to all physical blocks for the given journal.  This will save
 511  * us time when writing journal blocks.  Most journals will have only one
 512  * extent that maps all their logical blocks.  That's because gfs2.mkfs
 513  * arranges the journal blocks sequentially to maximize performance.
 514  * So the extent would map the first block for the entire file length.
 515  * However, gfs2_jadd can happen while file activity is happening, so
 516  * those journals may not be sequential.  Less likely is the case where
 517  * the users created their own journals by mounting the metafs and
 518  * laying it out.  But it's still possible.  These journals might have
 519  * several extents.
 520  *
 521  * TODO: This should be done in bigger chunks rather than one block at a time,
 522  *       but since it's only done at mount time, I'm not worried about the
 523  *       time it takes.
 524  */
 525 static int map_journal_extents(struct gfs2_sbd *sdp)
 526 {
 527         struct gfs2_jdesc *jd = sdp->sd_jdesc;
 528         unsigned int lb;
 529         u64 db, prev_db; /* logical block, disk block, prev disk block */
 530         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 531         struct gfs2_journal_extent *jext = NULL;
 532         struct buffer_head bh;
 533         int rc = 0;
 534
 535         prev_db = 0;
 536
 537         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
 538                 bh.b_state = 0;
 539                 bh.b_blocknr = 0;
 540                 bh.b_size = 1 << ip->i_inode.i_blkbits;
 541                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
 542                 db = bh.b_blocknr;
 543                 if (rc || !db) {
 544                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
 545                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
 546                         break;
 547                 }
 548                 if (!prev_db || db != prev_db + 1) {
 549                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
 550                                        GFP_KERNEL);
 551                         if (!jext) {
 552                                 printk(KERN_INFO "GFS2 error: out of memory "
 553                                        "mapping journal extents.\n");
 554                                 rc = -ENOMEM;
 555                                 break;
 556                         }
 557                         jext->dblock = db;
 558                         jext->lblock = lb;
 559                         jext->blocks = 1;
 560                         list_add_tail(&jext->extent_list, &jd->extent_list);
 561                 } else {
 562                         jext->blocks++;
 563                 }
 564                 prev_db = db;
 565         }
 566         return rc;
 567 }
 568
 569 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 570 {
 571         char *message = "FIRSTMOUNT=Done";
 572         char *envp[] = { message, NULL };
 573
 574         fs_info(sdp, "first mount done, others may mount\n");
 575
 576         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 577                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 578
 579         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 580 }
 581
 582 /**
 583  * gfs2_jindex_hold - Grab a lock on the jindex
 584  * @sdp: The GFS2 superblock
 585  * @ji_gh: the holder for the jindex glock
 586  *
 587  * Returns: errno
 588  */
 589
 590 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 591 {
 592         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 593         struct qstr name;
 594         char buf[20];
 595         struct gfs2_jdesc *jd;
 596         int error;
 597
 598         name.name = buf;
 599
 600         mutex_lock(&sdp->sd_jindex_mutex);
 601
 602         for (;;) {
 603                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 604                 if (error)
 605                         break;
 606
 607                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 608                 name.hash = gfs2_disk_hash(name.name, name.len);
 609
 610                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 611                 if (error == -ENOENT) {
 612                         error = 0;
 613                         break;
 614                 }
 615
 616                 gfs2_glock_dq_uninit(ji_gh);
 617
 618                 if (error)
 619                         break;
 620
 621                 error = -ENOMEM;
 622                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 623                 if (!jd)
 624                         break;
 625
 626                 INIT_LIST_HEAD(&jd->extent_list);
 627                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 628                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 629                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 630                         if (!jd->jd_inode)
 631                                 error = -ENOENT;
 632                         else
 633                                 error = PTR_ERR(jd->jd_inode);
 634                         kfree(jd);
 635                         break;
 636                 }
 637
 638                 spin_lock(&sdp->sd_jindex_spin);
 639                 jd->jd_jid = sdp->sd_journals++;
 640                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 641                 spin_unlock(&sdp->sd_jindex_spin);
 642         }
 643
 644         mutex_unlock(&sdp->sd_jindex_mutex);
 645
 646         return error;
 647 }
 648
 649 static int init_journal(struct gfs2_sbd *sdp, int undo)
 650 {
 651         struct inode *master = sdp->sd_master_dir->d_inode;
 652         struct gfs2_holder ji_gh;
 653         struct gfs2_inode *ip;
 654         int jindex = 1;
 655         int error = 0;
 656
 657         if (undo) {
 658                 jindex = 0;
 659                 goto fail_jinode_gh;
 660         }
 661
 662         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 663         if (IS_ERR(sdp->sd_jindex)) {
 664                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 665                 return PTR_ERR(sdp->sd_jindex);
 666         }
 667
 668         /* Load in the journal index special file */
 669
 670         error = gfs2_jindex_hold(sdp, &ji_gh);
 671         if (error) {
 672                 fs_err(sdp, "can't read journal index: %d\n", error);
 673                 goto fail;
 674         }
 675
 676         error = -EUSERS;
 677         if (!gfs2_jindex_size(sdp)) {
 678                 fs_err(sdp, "no journals!\n");
 679                 goto fail_jindex;
 680         }
 681
 682         if (sdp->sd_args.ar_spectator) {
 683                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 684                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 685                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 686                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 687         } else {
 688                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 689                         fs_err(sdp, "can't mount journal #%u\n",
 690                                sdp->sd_lockstruct.ls_jid);
 691                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 692                                gfs2_jindex_size(sdp),
 693                                gfs2_jindex_size(sdp) - 1);
 694                         goto fail_jindex;
 695                 }
 696                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 697
 698                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 699                                           &gfs2_journal_glops,
 700                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 701                                           &sdp->sd_journal_gh);
 702                 if (error) {
 703                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 704                         goto fail_jindex;
 705                 }
 706
 707                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 708                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 709                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 710                                            &sdp->sd_jinode_gh);
 711                 if (error) {
 712                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 713                                error);
 714                         goto fail_journal_gh;
 715                 }
 716
 717                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 718                 if (error) {
 719                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 720                                sdp->sd_jdesc->jd_jid, error);
 721                         goto fail_jinode_gh;
 722                 }
 723                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 724                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 725                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 726
 727                 /* Map the extents for this journal's blocks */
 728                 map_journal_extents(sdp);
 729         }
 730         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 731
 732         if (sdp->sd_lockstruct.ls_first) {
 733                 unsigned int x;
 734                 for (x = 0; x < sdp->sd_journals; x++) {
 735                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
 736                                                      true);
 737                         if (error) {
 738                                 fs_err(sdp, "error recovering journal %u: %d\n",
 739                                        x, error);
 740                                 goto fail_jinode_gh;
 741                         }
 742                 }
 743
 744                 gfs2_others_may_mount(sdp);
 745         } else if (!sdp->sd_args.ar_spectator) {
 746                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 747                 if (error) {
 748                         fs_err(sdp, "error recovering my journal: %d\n", error);
 749                         goto fail_jinode_gh;
 750                 }
 751         }
 752
 753         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 754         gfs2_glock_dq_uninit(&ji_gh);
 755         jindex = 0;
 756
 757         return 0;
 758
 759 fail_jinode_gh:
 760         if (!sdp->sd_args.ar_spectator)
 761                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 762 fail_journal_gh:
 763         if (!sdp->sd_args.ar_spectator)
 764                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 765 fail_jindex:
 766         gfs2_jindex_free(sdp);
 767         if (jindex)
 768                 gfs2_glock_dq_uninit(&ji_gh);
 769 fail:
 770         iput(sdp->sd_jindex);
 771         return error;
 772 }
 773
 774 static struct lock_class_key gfs2_quota_imutex_key;
 775
 776 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 777 {
 778         int error = 0;
 779         struct inode *master = sdp->sd_master_dir->d_inode;
 780
 781         if (undo)
 782                 goto fail_qinode;
 783
 784         error = init_journal(sdp, undo);
 785         if (error)
 786                 goto fail;
 787
 788         /* Read in the master statfs inode */
 789         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 790         if (IS_ERR(sdp->sd_statfs_inode)) {
 791                 error = PTR_ERR(sdp->sd_statfs_inode);
 792                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 793                 goto fail_journal;
 794         }
 795
 796         /* Read in the resource index inode */
 797         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 798         if (IS_ERR(sdp->sd_rindex)) {
 799                 error = PTR_ERR(sdp->sd_rindex);
 800                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 801                 goto fail_statfs;
 802         }
 803         sdp->sd_rindex_uptodate = 0;
 804
 805         /* Read in the quota inode */
 806         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 807         if (IS_ERR(sdp->sd_quota_inode)) {
 808                 error = PTR_ERR(sdp->sd_quota_inode);
 809                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 810                 goto fail_rindex;
 811         }
 812         /*
 813          * i_mutex on quota files is special. Since this inode is hidden system
 814          * file, we are safe to define locking ourselves.
 815          */
 816         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
 817                           &gfs2_quota_imutex_key);
 818
 819         error = gfs2_rindex_update(sdp);
 820         if (error)
 821                 goto fail_qinode;
 822
 823         return 0;
 824
 825 fail_qinode:
 826         iput(sdp->sd_quota_inode);
 827 fail_rindex:
 828         gfs2_clear_rgrpd(sdp);
 829         iput(sdp->sd_rindex);
 830 fail_statfs:
 831         iput(sdp->sd_statfs_inode);
 832 fail_journal:
 833         init_journal(sdp, UNDO);
 834 fail:
 835         return error;
 836 }
 837
 838 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 839 {
 840         struct inode *pn = NULL;
 841         char buf[30];
 842         int error = 0;
 843         struct gfs2_inode *ip;
 844         struct inode *master = sdp->sd_master_dir->d_inode;
 845
 846         if (sdp->sd_args.ar_spectator)
 847                 return 0;
 848
 849         if (undo)
 850                 goto fail_qc_gh;
 851
 852         pn = gfs2_lookup_simple(master, "per_node");
 853         if (IS_ERR(pn)) {
 854                 error = PTR_ERR(pn);
 855                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 856                 return error;
 857         }
 858
 859         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 860         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 861         if (IS_ERR(sdp->sd_sc_inode)) {
 862                 error = PTR_ERR(sdp->sd_sc_inode);
 863                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 864                 goto fail;
 865         }
 866
 867         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 868         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 869         if (IS_ERR(sdp->sd_qc_inode)) {
 870                 error = PTR_ERR(sdp->sd_qc_inode);
 871                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 872                 goto fail_ut_i;
 873         }
 874
 875         iput(pn);
 876         pn = NULL;
 877
 878         ip = GFS2_I(sdp->sd_sc_inode);
 879         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 880                                    &sdp->sd_sc_gh);
 881         if (error) {
 882                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 883                 goto fail_qc_i;
 884         }
 885
 886         ip = GFS2_I(sdp->sd_qc_inode);
 887         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 888                                    &sdp->sd_qc_gh);
 889         if (error) {
 890                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 891                 goto fail_ut_gh;
 892         }
 893
 894         return 0;
 895
 896 fail_qc_gh:
 897         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 898 fail_ut_gh:
 899         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 900 fail_qc_i:
 901         iput(sdp->sd_qc_inode);
 902 fail_ut_i:
 903         iput(sdp->sd_sc_inode);
 904 fail:
 905         if (pn)
 906                 iput(pn);
 907         return error;
 908 }
 909
 910 static int init_threads(struct gfs2_sbd *sdp, int undo)
 911 {
 912         struct task_struct *p;
 913         int error = 0;
 914
 915         if (undo)
 916                 goto fail_quotad;
 917
 918         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 919         error = IS_ERR(p);
 920         if (error) {
 921                 fs_err(sdp, "can't start logd thread: %d\n", error);
 922                 return error;
 923         }
 924         sdp->sd_logd_process = p;
 925
 926         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 927         error = IS_ERR(p);
 928         if (error) {
 929                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 930                 goto fail;
 931         }
 932         sdp->sd_quotad_process = p;
 933
 934         return 0;
 935
 936
 937 fail_quotad:
 938         kthread_stop(sdp->sd_quotad_process);
 939 fail:
 940         kthread_stop(sdp->sd_logd_process);
 941         return error;
 942 }
 943
 944 static const match_table_t nolock_tokens = {
 945         { Opt_jid, "jid=%d\n", },
 946         { Opt_err, NULL },
 947 };
 948
 949 static const struct lm_lockops nolock_ops = {
 950         .lm_proto_name = "lock_nolock",
 951         .lm_put_lock = gfs2_glock_free,
 952         .lm_tokens = &nolock_tokens,
 953 };
 954
 955 /**
 956  * gfs2_lm_mount - mount a locking protocol
 957  * @sdp: the filesystem
 958  * @args: mount arguments
 959  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 960  *
 961  * Returns: errno
 962  */
 963
 964 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 965 {
 966         const struct lm_lockops *lm;
 967         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 968         struct gfs2_args *args = &sdp->sd_args;
 969         const char *proto = sdp->sd_proto_name;
 970         const char *table = sdp->sd_table_name;
 971         char *o, *options;
 972         int ret;
 973
 974         if (!strcmp("lock_nolock", proto)) {
 975                 lm = &nolock_ops;
 976                 sdp->sd_args.ar_localflocks = 1;
 977 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 978         } else if (!strcmp("lock_dlm", proto)) {
 979                 lm = &gfs2_dlm_ops;
 980 #endif
 981         } else {
 982                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
 983                 return -ENOENT;
 984         }
 985
 986         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 987
 988         ls->ls_ops = lm;
 989         ls->ls_first = 1;
 990
 991         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 992                 substring_t tmp[MAX_OPT_ARGS];
 993                 int token, option;
 994
 995                 if (!o || !*o)
 996                         continue;
 997
 998                 token = match_token(o, *lm->lm_tokens, tmp);
 999                 switch (token) {
1000                 case Opt_jid:
1001                         ret = match_int(&tmp[0], &option);
1002                         if (ret || option < 0)
1003                                 goto hostdata_error;
1004                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1005                                 ls->ls_jid = option;
1006                         break;
1007                 case Opt_id:
1008                 case Opt_nodir:
1009                         /* Obsolete, but left for backward compat purposes */
1010                         break;
1011                 case Opt_first:
1012                         ret = match_int(&tmp[0], &option);
1013                         if (ret || (option != 0 && option != 1))
1014                                 goto hostdata_error;
1015                         ls->ls_first = option;
1016                         break;
1017                 case Opt_err:
1018                 default:
1019 hostdata_error:
1020                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1021                         return -EINVAL;
1022                 }
1023         }
1024
1025         if (lm->lm_mount == NULL) {
1026                 fs_info(sdp, "Now mounting FS...\n");
1027                 complete_all(&sdp->sd_locking_init);
1028                 return 0;
1029         }
1030         ret = lm->lm_mount(sdp, table);
1031         if (ret == 0)
1032                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1033         complete_all(&sdp->sd_locking_init);
1034         return ret;
1035 }
1036
1037 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1038 {
1039         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1040         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1041             lm->lm_unmount)
1042                 lm->lm_unmount(sdp);
1043 }
1044
1045 static int gfs2_journalid_wait(void *word)
1046 {
1047         if (signal_pending(current))
1048                 return -EINTR;
1049         schedule();
1050         return 0;
1051 }
1052
1053 static int wait_on_journal(struct gfs2_sbd *sdp)
1054 {
1055         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1056                 return 0;
1057
1058         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1059 }
1060
1061 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1062 {
1063         struct super_block *sb = sdp->sd_vfs;
1064         char ro[20];
1065         char spectator[20];
1066         char *envp[] = { ro, spectator, NULL };
1067         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1068         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1069         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1070 }
1071
1072 /**
1073  * fill_super - Read in superblock
1074  * @sb: The VFS superblock
1075  * @data: Mount options
1076  * @silent: Don't complain if it's not a GFS2 filesystem
1077  *
1078  * Returns: errno
1079  */
1080
1081 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1082 {
1083         struct gfs2_sbd *sdp;
1084         struct gfs2_holder mount_gh;
1085         int error;
1086
1087         sdp = init_sbd(sb);
1088         if (!sdp) {
1089                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1090                 return -ENOMEM;
1091         }
1092         sdp->sd_args = *args;
1093
1094         if (sdp->sd_args.ar_spectator) {
1095                 sb->s_flags |= MS_RDONLY;
1096                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1097         }
1098         if (sdp->sd_args.ar_posix_acl)
1099                 sb->s_flags |= MS_POSIXACL;
1100         if (sdp->sd_args.ar_nobarrier)
1101                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1102
1103         sb->s_flags |= MS_NOSEC;
1104         sb->s_magic = GFS2_MAGIC;
1105         sb->s_op = &gfs2_super_ops;
1106         sb->s_d_op = &gfs2_dops;
1107         sb->s_export_op = &gfs2_export_ops;
1108         sb->s_xattr = gfs2_xattr_handlers;
1109         sb->s_qcop = &gfs2_quotactl_ops;
1110         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1111         sb->s_time_gran = 1;
1112         sb->s_maxbytes = MAX_LFS_FILESIZE;
1113
1114         /* Set up the buffer cache and fill in some fake block size values
1115            to allow us to read-in the on-disk superblock. */
1116         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1117         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1118         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1119                                GFS2_BASIC_BLOCK_SHIFT;
1120         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1121
1122         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1123         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1124         if (sdp->sd_args.ar_statfs_quantum) {
1125                 sdp->sd_tune.gt_statfs_slow = 0;
1126                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1127         } else {
1128                 sdp->sd_tune.gt_statfs_slow = 1;
1129                 sdp->sd_tune.gt_statfs_quantum = 30;
1130         }
1131
1132         error = init_names(sdp, silent);
1133         if (error) {
1134                 /* In this case, we haven't initialized sysfs, so we have to
1135                    manually free the sdp. */
1136                 free_percpu(sdp->sd_lkstats);
1137                 kfree(sdp);
1138                 sb->s_fs_info = NULL;
1139                 return error;
1140         }
1141
1142         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1143
1144         error = gfs2_sys_fs_add(sdp);
1145         /*
1146          * If we hit an error here, gfs2_sys_fs_add will have called function
1147          * kobject_put which causes the sysfs usage count to go to zero, which
1148          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1149          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1150          * kobject_put to free sdp.
1151          */
1152         if (error)
1153                 return error;
1154
1155         gfs2_create_debugfs_file(sdp);
1156
1157         error = gfs2_lm_mount(sdp, silent);
1158         if (error)
1159                 goto fail_debug;
1160
1161         error = init_locking(sdp, &mount_gh, DO);
1162         if (error)
1163                 goto fail_lm;
1164
1165         error = init_sb(sdp, silent);
1166         if (error)
1167                 goto fail_locking;
1168
1169         error = wait_on_journal(sdp);
1170         if (error)
1171                 goto fail_sb;
1172
1173         /*
1174          * If user space has failed to join the cluster or some similar
1175          * failure has occurred, then the journal id will contain a
1176          * negative (error) number. This will then be returned to the
1177          * caller (of the mount syscall). We do this even for spectator
1178          * mounts (which just write a jid of 0 to indicate "ok" even though
1179          * the jid is unused in the spectator case)
1180          */
1181         if (sdp->sd_lockstruct.ls_jid < 0) {
1182                 error = sdp->sd_lockstruct.ls_jid;
1183                 sdp->sd_lockstruct.ls_jid = 0;
1184                 goto fail_sb;
1185         }
1186
1187         if (sdp->sd_args.ar_spectator)
1188                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1189                          sdp->sd_table_name);
1190         else
1191                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1192                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1193
1194         error = init_inodes(sdp, DO);
1195         if (error)
1196                 goto fail_sb;
1197
1198         error = init_per_node(sdp, DO);
1199         if (error)
1200                 goto fail_inodes;
1201
1202         error = gfs2_statfs_init(sdp);
1203         if (error) {
1204                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1205                 goto fail_per_node;
1206         }
1207
1208         error = init_threads(sdp, DO);
1209         if (error)
1210                 goto fail_per_node;
1211
1212         if (!(sb->s_flags & MS_RDONLY)) {
1213                 error = gfs2_make_fs_rw(sdp);
1214                 if (error) {
1215                         fs_err(sdp, "can't make FS RW: %d\n", error);
1216                         goto fail_threads;
1217                 }
1218         }
1219
1220         gfs2_glock_dq_uninit(&mount_gh);
1221         gfs2_online_uevent(sdp);
1222         return 0;
1223
1224 fail_threads:
1225         init_threads(sdp, UNDO);
1226 fail_per_node:
1227         init_per_node(sdp, UNDO);
1228 fail_inodes:
1229         init_inodes(sdp, UNDO);
1230 fail_sb:
1231         if (sdp->sd_root_dir)
1232                 dput(sdp->sd_root_dir);
1233         if (sdp->sd_master_dir)
1234                 dput(sdp->sd_master_dir);
1235         if (sb->s_root)
1236                 dput(sb->s_root);
1237         sb->s_root = NULL;
1238 fail_locking:
1239         init_locking(sdp, &mount_gh, UNDO);
1240 fail_lm:
1241         gfs2_gl_hash_clear(sdp);
1242         gfs2_lm_unmount(sdp);
1243 fail_debug:
1244         gfs2_delete_debugfs_file(sdp);
1245         free_percpu(sdp->sd_lkstats);
1246         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1247          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1248         gfs2_sys_fs_del(sdp);
1249         sb->s_fs_info = NULL;
1250         return error;
1251 }
1252
1253 static int set_gfs2_super(struct super_block *s, void *data)
1254 {
1255         s->s_bdev = data;
1256         s->s_dev = s->s_bdev->bd_dev;
1257
1258         /*
1259          * We set the bdi here to the queue backing, file systems can
1260          * overwrite this in ->fill_super()
1261          */
1262         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1263         return 0;
1264 }
1265
1266 static int test_gfs2_super(struct super_block *s, void *ptr)
1267 {
1268         struct block_device *bdev = ptr;
1269         return (bdev == s->s_bdev);
1270 }
1271
1272 /**
1273  * gfs2_mount - Get the GFS2 superblock
1274  * @fs_type: The GFS2 filesystem type
1275  * @flags: Mount flags
1276  * @dev_name: The name of the device
1277  * @data: The mount arguments
1278  *
1279  * Q. Why not use get_sb_bdev() ?
1280  * A. We need to select one of two root directories to mount, independent
1281  *    of whether this is the initial, or subsequent, mount of this sb
1282  *
1283  * Returns: 0 or -ve on error
1284  */
1285
1286 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1287                        const char *dev_name, void *data)
1288 {
1289         struct block_device *bdev;
1290         struct super_block *s;
1291         fmode_t mode = FMODE_READ | FMODE_EXCL;
1292         int error;
1293         struct gfs2_args args;
1294         struct gfs2_sbd *sdp;
1295
1296         if (!(flags & MS_RDONLY))
1297                 mode |= FMODE_WRITE;
1298
1299         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1300         if (IS_ERR(bdev))
1301                 return ERR_CAST(bdev);
1302
1303         /*
1304          * once the super is inserted into the list by sget, s_umount
1305          * will protect the lockfs code from trying to start a snapshot
1306          * while we are mounting
1307          */
1308         mutex_lock(&bdev->bd_fsfreeze_mutex);
1309         if (bdev->bd_fsfreeze_count > 0) {
1310                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1311                 error = -EBUSY;
1312                 goto error_bdev;
1313         }
1314         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1315         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1316         error = PTR_ERR(s);
1317         if (IS_ERR(s))
1318                 goto error_bdev;
1319
1320         if (s->s_root) {
1321                 /*
1322                  * s_umount nests inside bd_mutex during
1323                  * __invalidate_device().  blkdev_put() acquires
1324                  * bd_mutex and can't be called under s_umount.  Drop
1325                  * s_umount temporarily.  This is safe as we're
1326                  * holding an active reference.
1327                  */
1328                 up_write(&s->s_umount);
1329                 blkdev_put(bdev, mode);
1330                 down_write(&s->s_umount);
1331         }
1332
1333         memset(&args, 0, sizeof(args));
1334         args.ar_quota = GFS2_QUOTA_DEFAULT;
1335         args.ar_data = GFS2_DATA_DEFAULT;
1336         args.ar_commit = 30;
1337         args.ar_statfs_quantum = 30;
1338         args.ar_quota_quantum = 60;
1339         args.ar_errors = GFS2_ERRORS_DEFAULT;
1340
1341         error = gfs2_mount_args(&args, data);
1342         if (error) {
1343                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1344                 goto error_super;
1345         }
1346
1347         if (s->s_root) {
1348                 error = -EBUSY;
1349                 if ((flags ^ s->s_flags) & MS_RDONLY)
1350                         goto error_super;
1351         } else {
1352                 char b[BDEVNAME_SIZE];
1353
1354                 s->s_mode = mode;
1355                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1356                 sb_set_blocksize(s, block_size(bdev));
1357                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1358                 if (error)
1359                         goto error_super;
1360                 s->s_flags |= MS_ACTIVE;
1361                 bdev->bd_super = s;
1362         }
1363
1364         sdp = s->s_fs_info;
1365         if (args.ar_meta)
1366                 return dget(sdp->sd_master_dir);
1367         else
1368                 return dget(sdp->sd_root_dir);
1369
1370 error_super:
1371         deactivate_locked_super(s);
1372         return ERR_PTR(error);
1373 error_bdev:
1374         blkdev_put(bdev, mode);
1375         return ERR_PTR(error);
1376 }
1377
1378 static int set_meta_super(struct super_block *s, void *ptr)
1379 {
1380         return -EINVAL;
1381 }
1382
1383 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1384                         int flags, const char *dev_name, void *data)
1385 {
1386         struct super_block *s;
1387         struct gfs2_sbd *sdp;
1388         struct path path;
1389         int error;
1390
1391         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1392         if (error) {
1393                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1394                        dev_name, error);
1395                 return ERR_PTR(error);
1396         }
1397         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1398                  path.dentry->d_inode->i_sb->s_bdev);
1399         path_put(&path);
1400         if (IS_ERR(s)) {
1401                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1402                 return ERR_CAST(s);
1403         }
1404         if ((flags ^ s->s_flags) & MS_RDONLY) {
1405                 deactivate_locked_super(s);
1406                 return ERR_PTR(-EBUSY);
1407         }
1408         sdp = s->s_fs_info;
1409         return dget(sdp->sd_master_dir);
1410 }
1411
1412 static void gfs2_kill_sb(struct super_block *sb)
1413 {
1414         struct gfs2_sbd *sdp = sb->s_fs_info;
1415
1416         if (sdp == NULL) {
1417                 kill_block_super(sb);
1418                 return;
1419         }
1420
1421         gfs2_meta_syncfs(sdp);
1422         dput(sdp->sd_root_dir);
1423         dput(sdp->sd_master_dir);
1424         sdp->sd_root_dir = NULL;
1425         sdp->sd_master_dir = NULL;
1426         shrink_dcache_sb(sb);
1427         gfs2_delete_debugfs_file(sdp);
1428         free_percpu(sdp->sd_lkstats);
1429         kill_block_super(sb);
1430 }
1431
1432 struct file_system_type gfs2_fs_type = {
1433         .name = "gfs2",
1434         .fs_flags = FS_REQUIRES_DEV,
1435         .mount = gfs2_mount,
1436         .kill_sb = gfs2_kill_sb,
1437         .owner = THIS_MODULE,
1438 };
1439 MODULE_ALIAS_FS("gfs2");
1440
1441 struct file_system_type gfs2meta_fs_type = {
1442         .name = "gfs2meta",
1443         .fs_flags = FS_REQUIRES_DEV,
1444         .mount = gfs2_mount_meta,
1445         .owner = THIS_MODULE,
1446 };
1447 MODULE_ALIAS_FS("gfs2meta");