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