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 "meta_io.h"
  40 #include "trace_gfs2.h"
  41
  42 #define DO 0
  43 #define UNDO 1
  44
  45 /**
  46  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  47  * @gt: tune
  48  *
  49  */
  50
  51 static void gfs2_tune_init(struct gfs2_tune *gt)
  52 {
  53         spin_lock_init(&gt->gt_spin);
  54
  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         struct address_space *mapping;
  67
  68         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  69         if (!sdp)
  70                 return NULL;
  71
  72         sb->s_fs_info = sdp;
  73         sdp->sd_vfs = sb;
  74         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  75         if (!sdp->sd_lkstats) {
  76                 kfree(sdp);
  77                 return NULL;
  78         }
  79
  80         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  81         gfs2_tune_init(&sdp->sd_tune);
  82
  83         init_waitqueue_head(&sdp->sd_glock_wait);
  84         atomic_set(&sdp->sd_glock_disposal, 0);
  85         init_completion(&sdp->sd_locking_init);
  86         init_completion(&sdp->sd_wdack);
  87         spin_lock_init(&sdp->sd_statfs_spin);
  88
  89         spin_lock_init(&sdp->sd_rindex_spin);
  90         sdp->sd_rindex_tree.rb_node = NULL;
  91
  92         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  93         spin_lock_init(&sdp->sd_jindex_spin);
  94         mutex_init(&sdp->sd_jindex_mutex);
  95
  96         INIT_LIST_HEAD(&sdp->sd_quota_list);
  97         mutex_init(&sdp->sd_quota_mutex);
  98         mutex_init(&sdp->sd_quota_sync_mutex);
  99         init_waitqueue_head(&sdp->sd_quota_wait);
 100         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 101         spin_lock_init(&sdp->sd_trunc_lock);
 102
 103         mapping = &sdp->sd_aspace;
 104
 105         address_space_init_once(mapping);
 106         mapping->a_ops = &gfs2_meta_aops;
 107         mapping->host = sb->s_bdev->bd_inode;
 108         mapping->flags = 0;
 109         mapping_set_gfp_mask(mapping, GFP_NOFS);
 110         mapping->private_data = NULL;
 111         mapping->backing_dev_info = sb->s_bdi;
 112         mapping->writeback_index = 0;
 113
 114         spin_lock_init(&sdp->sd_log_lock);
 115         atomic_set(&sdp->sd_log_pinned, 0);
 116         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
 117         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
 118         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
 119         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 120         spin_lock_init(&sdp->sd_ordered_lock);
 121
 122         init_waitqueue_head(&sdp->sd_log_waitq);
 123         init_waitqueue_head(&sdp->sd_logd_waitq);
 124         spin_lock_init(&sdp->sd_ail_lock);
 125         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 126         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 127
 128         init_rwsem(&sdp->sd_log_flush_lock);
 129         atomic_set(&sdp->sd_log_in_flight, 0);
 130         init_waitqueue_head(&sdp->sd_log_flush_wait);
 131
 132         INIT_LIST_HEAD(&sdp->sd_revoke_list);
 133
 134         return sdp;
 135 }
 136
 137
 138 /**
 139  * gfs2_check_sb - Check superblock
 140  * @sdp: the filesystem
 141  * @sb: The superblock
 142  * @silent: Don't print a message if the check fails
 143  *
 144  * Checks the version code of the FS is one that we understand how to
 145  * read and that the sizes of the various on-disk structures have not
 146  * changed.
 147  */
 148
 149 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 150 {
 151         struct gfs2_sb_host *sb = &sdp->sd_sb;
 152
 153         if (sb->sb_magic != GFS2_MAGIC ||
 154             sb->sb_type != GFS2_METATYPE_SB) {
 155                 if (!silent)
 156                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 157                 return -EINVAL;
 158         }
 159
 160         /*  If format numbers match exactly, we're done.  */
 161
 162         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 163             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 164                 return 0;
 165
 166         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 167
 168         return -EINVAL;
 169 }
 170
 171 static void end_bio_io_page(struct bio *bio, int error)
 172 {
 173         struct page *page = bio->bi_private;
 174
 175         if (!error)
 176                 SetPageUptodate(page);
 177         else
 178                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 179         unlock_page(page);
 180 }
 181
 182 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 183 {
 184         struct gfs2_sb_host *sb = &sdp->sd_sb;
 185         struct super_block *s = sdp->sd_vfs;
 186         const struct gfs2_sb *str = buf;
 187
 188         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 189         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 190         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 191         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 192         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 193         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 194         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 195         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 196         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 197         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 198         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 199
 200         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 201         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 202         memcpy(s->s_uuid, str->sb_uuid, 16);
 203 }
 204
 205 /**
 206  * gfs2_read_super - Read the gfs2 super block from disk
 207  * @sdp: The GFS2 super block
 208  * @sector: The location of the super block
 209  * @error: The error code to return
 210  *
 211  * This uses the bio functions to read the super block from disk
 212  * because we want to be 100% sure that we never read cached data.
 213  * A super block is read twice only during each GFS2 mount and is
 214  * never written to by the filesystem. The first time its read no
 215  * locks are held, and the only details which are looked at are those
 216  * relating to the locking protocol. Once locking is up and working,
 217  * the sb is read again under the lock to establish the location of
 218  * the master directory (contains pointers to journals etc) and the
 219  * root directory.
 220  *
 221  * Returns: 0 on success or error
 222  */
 223
 224 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 225 {
 226         struct super_block *sb = sdp->sd_vfs;
 227         struct gfs2_sb *p;
 228         struct page *page;
 229         struct bio *bio;
 230
 231         page = alloc_page(GFP_NOFS);
 232         if (unlikely(!page))
 233                 return -ENOBUFS;
 234
 235         ClearPageUptodate(page);
 236         ClearPageDirty(page);
 237         lock_page(page);
 238
 239         bio = bio_alloc(GFP_NOFS, 1);
 240         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 241         bio->bi_bdev = sb->s_bdev;
 242         bio_add_page(bio, page, PAGE_SIZE, 0);
 243
 244         bio->bi_end_io = end_bio_io_page;
 245         bio->bi_private = page;
 246         submit_bio(READ_SYNC | REQ_META, bio);
 247         wait_on_page_locked(page);
 248         bio_put(bio);
 249         if (!PageUptodate(page)) {
 250                 __free_page(page);
 251                 return -EIO;
 252         }
 253         p = kmap(page);
 254         gfs2_sb_in(sdp, p);
 255         kunmap(page);
 256         __free_page(page);
 257         return gfs2_check_sb(sdp, silent);
 258 }
 259
 260 /**
 261  * gfs2_read_sb - Read super block
 262  * @sdp: The GFS2 superblock
 263  * @silent: Don't print message if mount fails
 264  *
 265  */
 266
 267 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 268 {
 269         u32 hash_blocks, ind_blocks, leaf_blocks;
 270         u32 tmp_blocks;
 271         unsigned int x;
 272         int error;
 273
 274         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 275         if (error) {
 276                 if (!silent)
 277                         fs_err(sdp, "can't read superblock\n");
 278                 return error;
 279         }
 280
 281         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 282                                GFS2_BASIC_BLOCK_SHIFT;
 283         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 284         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 285                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 286         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 287                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 288         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 289         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 290         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 291         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 292         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 293                                 sizeof(struct gfs2_meta_header)) /
 294                                 sizeof(struct gfs2_quota_change);
 295         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 296                                      sizeof(struct gfs2_meta_header))
 297                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 298
 299         /* Compute maximum reservation required to add a entry to a directory */
 300
 301         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 302                              sdp->sd_jbsize);
 303
 304         ind_blocks = 0;
 305         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 306                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 307                 ind_blocks += tmp_blocks;
 308         }
 309
 310         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 311
 312         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 313
 314         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 315                                 sizeof(struct gfs2_dinode);
 316         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 317         for (x = 2;; x++) {
 318                 u64 space, d;
 319                 u32 m;
 320
 321                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 322                 d = space;
 323                 m = do_div(d, sdp->sd_inptrs);
 324
 325                 if (d != sdp->sd_heightsize[x - 1] || m)
 326                         break;
 327                 sdp->sd_heightsize[x] = space;
 328         }
 329         sdp->sd_max_height = x;
 330         sdp->sd_heightsize[x] = ~0;
 331         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 332
 333         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 334                                  sizeof(struct gfs2_dinode);
 335         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 336         for (x = 2;; x++) {
 337                 u64 space, d;
 338                 u32 m;
 339
 340                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 341                 d = space;
 342                 m = do_div(d, sdp->sd_inptrs);
 343
 344                 if (d != sdp->sd_jheightsize[x - 1] || m)
 345                         break;
 346                 sdp->sd_jheightsize[x] = space;
 347         }
 348         sdp->sd_max_jheight = x;
 349         sdp->sd_jheightsize[x] = ~0;
 350         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 351
 352         return 0;
 353 }
 354
 355 static int init_names(struct gfs2_sbd *sdp, int silent)
 356 {
 357         char *proto, *table;
 358         int error = 0;
 359
 360         proto = sdp->sd_args.ar_lockproto;
 361         table = sdp->sd_args.ar_locktable;
 362
 363         /*  Try to autodetect  */
 364
 365         if (!proto[0] || !table[0]) {
 366                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 367                 if (error)
 368                         return error;
 369
 370                 if (!proto[0])
 371                         proto = sdp->sd_sb.sb_lockproto;
 372                 if (!table[0])
 373                         table = sdp->sd_sb.sb_locktable;
 374         }
 375
 376         if (!table[0])
 377                 table = sdp->sd_vfs->s_id;
 378
 379         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 380         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 381
 382         table = sdp->sd_table_name;
 383         while ((table = strchr(table, '/')))
 384                 *table = '_';
 385
 386         return error;
 387 }
 388
 389 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 390                         int undo)
 391 {
 392         int error = 0;
 393
 394         if (undo)
 395                 goto fail_trans;
 396
 397         error = gfs2_glock_nq_num(sdp,
 398                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 399                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 400                                   mount_gh);
 401         if (error) {
 402                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 403                 goto fail;
 404         }
 405
 406         error = gfs2_glock_nq_num(sdp,
 407                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 408                                   LM_ST_SHARED,
 409                                   LM_FLAG_NOEXP | GL_EXACT,
 410                                   &sdp->sd_live_gh);
 411         if (error) {
 412                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 413                 goto fail_mount;
 414         }
 415
 416         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 417                                CREATE, &sdp->sd_rename_gl);
 418         if (error) {
 419                 fs_err(sdp, "can't create rename glock: %d\n", error);
 420                 goto fail_live;
 421         }
 422
 423         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 424                                CREATE, &sdp->sd_trans_gl);
 425         if (error) {
 426                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 427                 goto fail_rename;
 428         }
 429
 430         return 0;
 431
 432 fail_trans:
 433         gfs2_glock_put(sdp->sd_trans_gl);
 434 fail_rename:
 435         gfs2_glock_put(sdp->sd_rename_gl);
 436 fail_live:
 437         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 438 fail_mount:
 439         gfs2_glock_dq_uninit(mount_gh);
 440 fail:
 441         return error;
 442 }
 443
 444 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 445                             u64 no_addr, const char *name)
 446 {
 447         struct gfs2_sbd *sdp = sb->s_fs_info;
 448         struct dentry *dentry;
 449         struct inode *inode;
 450
 451         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 452         if (IS_ERR(inode)) {
 453                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 454                 return PTR_ERR(inode);
 455         }
 456         dentry = d_make_root(inode);
 457         if (!dentry) {
 458                 fs_err(sdp, "can't alloc %s dentry\n", name);
 459                 return -ENOMEM;
 460         }
 461         *dptr = dentry;
 462         return 0;
 463 }
 464
 465 static int init_sb(struct gfs2_sbd *sdp, int silent)
 466 {
 467         struct super_block *sb = sdp->sd_vfs;
 468         struct gfs2_holder sb_gh;
 469         u64 no_addr;
 470         int ret;
 471
 472         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 473                                 LM_ST_SHARED, 0, &sb_gh);
 474         if (ret) {
 475                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 476                 return ret;
 477         }
 478
 479         ret = gfs2_read_sb(sdp, silent);
 480         if (ret) {
 481                 fs_err(sdp, "can't read superblock: %d\n", ret);
 482                 goto out;
 483         }
 484
 485         /* Set up the buffer cache and SB for real */
 486         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 487                 ret = -EINVAL;
 488                 fs_err(sdp, "FS block size (%u) is too small for device "
 489                        "block size (%u)\n",
 490                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 491                 goto out;
 492         }
 493         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 494                 ret = -EINVAL;
 495                 fs_err(sdp, "FS block size (%u) is too big for machine "
 496                        "page size (%u)\n",
 497                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 498                 goto out;
 499         }
 500         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 501
 502         /* Get the root inode */
 503         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 504         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 505         if (ret)
 506                 goto out;
 507
 508         /* Get the master inode */
 509         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 510         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 511         if (ret) {
 512                 dput(sdp->sd_root_dir);
 513                 goto out;
 514         }
 515         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 516 out:
 517         gfs2_glock_dq_uninit(&sb_gh);
 518         return ret;
 519 }
 520
 521 /**
 522  * map_journal_extents - create a reusable "extent" mapping from all logical
 523  * blocks to all physical blocks for the given journal.  This will save
 524  * us time when writing journal blocks.  Most journals will have only one
 525  * extent that maps all their logical blocks.  That's because gfs2.mkfs
 526  * arranges the journal blocks sequentially to maximize performance.
 527  * So the extent would map the first block for the entire file length.
 528  * However, gfs2_jadd can happen while file activity is happening, so
 529  * those journals may not be sequential.  Less likely is the case where
 530  * the users created their own journals by mounting the metafs and
 531  * laying it out.  But it's still possible.  These journals might have
 532  * several extents.
 533  *
 534  * TODO: This should be done in bigger chunks rather than one block at a time,
 535  *       but since it's only done at mount time, I'm not worried about the
 536  *       time it takes.
 537  */
 538 static int map_journal_extents(struct gfs2_sbd *sdp)
 539 {
 540         struct gfs2_jdesc *jd = sdp->sd_jdesc;
 541         unsigned int lb;
 542         u64 db, prev_db; /* logical block, disk block, prev disk block */
 543         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 544         struct gfs2_journal_extent *jext = NULL;
 545         struct buffer_head bh;
 546         int rc = 0;
 547
 548         prev_db = 0;
 549
 550         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
 551                 bh.b_state = 0;
 552                 bh.b_blocknr = 0;
 553                 bh.b_size = 1 << ip->i_inode.i_blkbits;
 554                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
 555                 db = bh.b_blocknr;
 556                 if (rc || !db) {
 557                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
 558                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
 559                         break;
 560                 }
 561                 if (!prev_db || db != prev_db + 1) {
 562                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
 563                                        GFP_KERNEL);
 564                         if (!jext) {
 565                                 printk(KERN_INFO "GFS2 error: out of memory "
 566                                        "mapping journal extents.\n");
 567                                 rc = -ENOMEM;
 568                                 break;
 569                         }
 570                         jext->dblock = db;
 571                         jext->lblock = lb;
 572                         jext->blocks = 1;
 573                         list_add_tail(&jext->extent_list, &jd->extent_list);
 574                 } else {
 575                         jext->blocks++;
 576                 }
 577                 prev_db = db;
 578         }
 579         return rc;
 580 }
 581
 582 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 583 {
 584         char *message = "FIRSTMOUNT=Done";
 585         char *envp[] = { message, NULL };
 586
 587         fs_info(sdp, "first mount done, others may mount\n");
 588
 589         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 590                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 591
 592         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 593 }
 594
 595 /**
 596  * gfs2_jindex_hold - Grab a lock on the jindex
 597  * @sdp: The GFS2 superblock
 598  * @ji_gh: the holder for the jindex glock
 599  *
 600  * Returns: errno
 601  */
 602
 603 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 604 {
 605         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 606         struct qstr name;
 607         char buf[20];
 608         struct gfs2_jdesc *jd;
 609         int error;
 610
 611         name.name = buf;
 612
 613         mutex_lock(&sdp->sd_jindex_mutex);
 614
 615         for (;;) {
 616                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 617                 if (error)
 618                         break;
 619
 620                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 621                 name.hash = gfs2_disk_hash(name.name, name.len);
 622
 623                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 624                 if (error == -ENOENT) {
 625                         error = 0;
 626                         break;
 627                 }
 628
 629                 gfs2_glock_dq_uninit(ji_gh);
 630
 631                 if (error)
 632                         break;
 633
 634                 error = -ENOMEM;
 635                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 636                 if (!jd)
 637                         break;
 638
 639                 INIT_LIST_HEAD(&jd->extent_list);
 640                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 641                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 642                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 643                         if (!jd->jd_inode)
 644                                 error = -ENOENT;
 645                         else
 646                                 error = PTR_ERR(jd->jd_inode);
 647                         kfree(jd);
 648                         break;
 649                 }
 650
 651                 spin_lock(&sdp->sd_jindex_spin);
 652                 jd->jd_jid = sdp->sd_journals++;
 653                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 654                 spin_unlock(&sdp->sd_jindex_spin);
 655         }
 656
 657         mutex_unlock(&sdp->sd_jindex_mutex);
 658
 659         return error;
 660 }
 661
 662 /**
 663  * check_journal_clean - Make sure a journal is clean for a spectator mount
 664  * @sdp: The GFS2 superblock
 665  * @jd: The journal descriptor
 666  *
 667  * Returns: 0 if the journal is clean or locked, else an error
 668  */
 669 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 670 {
 671         int error;
 672         struct gfs2_holder j_gh;
 673         struct gfs2_log_header_host head;
 674         struct gfs2_inode *ip;
 675
 676         ip = GFS2_I(jd->jd_inode);
 677         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 678                                    GL_EXACT | GL_NOCACHE, &j_gh);
 679         if (error) {
 680                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 681                 return -EPERM;
 682         }
 683         error = gfs2_jdesc_check(jd);
 684         if (error) {
 685                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 686                 goto out_unlock;
 687         }
 688         error = gfs2_find_jhead(jd, &head);
 689         if (error) {
 690                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 691                 goto out_unlock;
 692         }
 693         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 694                 error = -EPERM;
 695                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 696                        "must not be a spectator.\n", jd->jd_jid);
 697         }
 698
 699 out_unlock:
 700         gfs2_glock_dq_uninit(&j_gh);
 701         return error;
 702 }
 703
 704 static int init_journal(struct gfs2_sbd *sdp, int undo)
 705 {
 706         struct inode *master = sdp->sd_master_dir->d_inode;
 707         struct gfs2_holder ji_gh;
 708         struct gfs2_inode *ip;
 709         int jindex = 1;
 710         int error = 0;
 711
 712         if (undo) {
 713                 jindex = 0;
 714                 goto fail_jinode_gh;
 715         }
 716
 717         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 718         if (IS_ERR(sdp->sd_jindex)) {
 719                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 720                 return PTR_ERR(sdp->sd_jindex);
 721         }
 722
 723         /* Load in the journal index special file */
 724
 725         error = gfs2_jindex_hold(sdp, &ji_gh);
 726         if (error) {
 727                 fs_err(sdp, "can't read journal index: %d\n", error);
 728                 goto fail;
 729         }
 730
 731         error = -EUSERS;
 732         if (!gfs2_jindex_size(sdp)) {
 733                 fs_err(sdp, "no journals!\n");
 734                 goto fail_jindex;
 735         }
 736
 737         if (sdp->sd_args.ar_spectator) {
 738                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 739                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 740                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 741                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 742         } else {
 743                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 744                         fs_err(sdp, "can't mount journal #%u\n",
 745                                sdp->sd_lockstruct.ls_jid);
 746                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 747                                gfs2_jindex_size(sdp),
 748                                gfs2_jindex_size(sdp) - 1);
 749                         goto fail_jindex;
 750                 }
 751                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 752
 753                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 754                                           &gfs2_journal_glops,
 755                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 756                                           &sdp->sd_journal_gh);
 757                 if (error) {
 758                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 759                         goto fail_jindex;
 760                 }
 761
 762                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 763                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 764                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 765                                            &sdp->sd_jinode_gh);
 766                 if (error) {
 767                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 768                                error);
 769                         goto fail_journal_gh;
 770                 }
 771
 772                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 773                 if (error) {
 774                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 775                                sdp->sd_jdesc->jd_jid, error);
 776                         goto fail_jinode_gh;
 777                 }
 778                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 779                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 780                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 781
 782                 /* Map the extents for this journal's blocks */
 783                 map_journal_extents(sdp);
 784         }
 785         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 786
 787         if (sdp->sd_lockstruct.ls_first) {
 788                 unsigned int x;
 789                 for (x = 0; x < sdp->sd_journals; x++) {
 790                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 791
 792                         if (sdp->sd_args.ar_spectator) {
 793                                 error = check_journal_clean(sdp, jd);
 794                                 if (error)
 795                                         goto fail_jinode_gh;
 796                                 continue;
 797                         }
 798                         error = gfs2_recover_journal(jd, true);
 799                         if (error) {
 800                                 fs_err(sdp, "error recovering journal %u: %d\n",
 801                                        x, error);
 802                                 goto fail_jinode_gh;
 803                         }
 804                 }
 805
 806                 gfs2_others_may_mount(sdp);
 807         } else if (!sdp->sd_args.ar_spectator) {
 808                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 809                 if (error) {
 810                         fs_err(sdp, "error recovering my journal: %d\n", error);
 811                         goto fail_jinode_gh;
 812                 }
 813         }
 814
 815         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 816         gfs2_glock_dq_uninit(&ji_gh);
 817         jindex = 0;
 818
 819         return 0;
 820
 821 fail_jinode_gh:
 822         if (!sdp->sd_args.ar_spectator)
 823                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 824 fail_journal_gh:
 825         if (!sdp->sd_args.ar_spectator)
 826                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 827 fail_jindex:
 828         gfs2_jindex_free(sdp);
 829         if (jindex)
 830                 gfs2_glock_dq_uninit(&ji_gh);
 831 fail:
 832         iput(sdp->sd_jindex);
 833         return error;
 834 }
 835
 836 static struct lock_class_key gfs2_quota_imutex_key;
 837
 838 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 839 {
 840         int error = 0;
 841         struct inode *master = sdp->sd_master_dir->d_inode;
 842
 843         if (undo)
 844                 goto fail_qinode;
 845
 846         error = init_journal(sdp, undo);
 847         if (error)
 848                 goto fail;
 849
 850         /* Read in the master statfs inode */
 851         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 852         if (IS_ERR(sdp->sd_statfs_inode)) {
 853                 error = PTR_ERR(sdp->sd_statfs_inode);
 854                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 855                 goto fail_journal;
 856         }
 857
 858         /* Read in the resource index inode */
 859         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 860         if (IS_ERR(sdp->sd_rindex)) {
 861                 error = PTR_ERR(sdp->sd_rindex);
 862                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 863                 goto fail_statfs;
 864         }
 865         sdp->sd_rindex_uptodate = 0;
 866
 867         /* Read in the quota inode */
 868         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 869         if (IS_ERR(sdp->sd_quota_inode)) {
 870                 error = PTR_ERR(sdp->sd_quota_inode);
 871                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 872                 goto fail_rindex;
 873         }
 874         /*
 875          * i_mutex on quota files is special. Since this inode is hidden system
 876          * file, we are safe to define locking ourselves.
 877          */
 878         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
 879                           &gfs2_quota_imutex_key);
 880
 881         error = gfs2_rindex_update(sdp);
 882         if (error)
 883                 goto fail_qinode;
 884
 885         return 0;
 886
 887 fail_qinode:
 888         iput(sdp->sd_quota_inode);
 889 fail_rindex:
 890         gfs2_clear_rgrpd(sdp);
 891         iput(sdp->sd_rindex);
 892 fail_statfs:
 893         iput(sdp->sd_statfs_inode);
 894 fail_journal:
 895         init_journal(sdp, UNDO);
 896 fail:
 897         return error;
 898 }
 899
 900 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 901 {
 902         struct inode *pn = NULL;
 903         char buf[30];
 904         int error = 0;
 905         struct gfs2_inode *ip;
 906         struct inode *master = sdp->sd_master_dir->d_inode;
 907
 908         if (sdp->sd_args.ar_spectator)
 909                 return 0;
 910
 911         if (undo)
 912                 goto fail_qc_gh;
 913
 914         pn = gfs2_lookup_simple(master, "per_node");
 915         if (IS_ERR(pn)) {
 916                 error = PTR_ERR(pn);
 917                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 918                 return error;
 919         }
 920
 921         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 922         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 923         if (IS_ERR(sdp->sd_sc_inode)) {
 924                 error = PTR_ERR(sdp->sd_sc_inode);
 925                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 926                 goto fail;
 927         }
 928
 929         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 930         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 931         if (IS_ERR(sdp->sd_qc_inode)) {
 932                 error = PTR_ERR(sdp->sd_qc_inode);
 933                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 934                 goto fail_ut_i;
 935         }
 936
 937         iput(pn);
 938         pn = NULL;
 939
 940         ip = GFS2_I(sdp->sd_sc_inode);
 941         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 942                                    &sdp->sd_sc_gh);
 943         if (error) {
 944                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 945                 goto fail_qc_i;
 946         }
 947
 948         ip = GFS2_I(sdp->sd_qc_inode);
 949         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 950                                    &sdp->sd_qc_gh);
 951         if (error) {
 952                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 953                 goto fail_ut_gh;
 954         }
 955
 956         return 0;
 957
 958 fail_qc_gh:
 959         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 960 fail_ut_gh:
 961         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 962 fail_qc_i:
 963         iput(sdp->sd_qc_inode);
 964 fail_ut_i:
 965         iput(sdp->sd_sc_inode);
 966 fail:
 967         if (pn)
 968                 iput(pn);
 969         return error;
 970 }
 971
 972 static const match_table_t nolock_tokens = {
 973         { Opt_jid, "jid=%d\n", },
 974         { Opt_err, NULL },
 975 };
 976
 977 static const struct lm_lockops nolock_ops = {
 978         .lm_proto_name = "lock_nolock",
 979         .lm_put_lock = gfs2_glock_free,
 980         .lm_tokens = &nolock_tokens,
 981 };
 982
 983 /**
 984  * gfs2_lm_mount - mount a locking protocol
 985  * @sdp: the filesystem
 986  * @args: mount arguments
 987  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 988  *
 989  * Returns: errno
 990  */
 991
 992 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 993 {
 994         const struct lm_lockops *lm;
 995         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 996         struct gfs2_args *args = &sdp->sd_args;
 997         const char *proto = sdp->sd_proto_name;
 998         const char *table = sdp->sd_table_name;
 999         char *o, *options;
1000         int ret;
1001
1002         if (!strcmp("lock_nolock", proto)) {
1003                 lm = &nolock_ops;
1004                 sdp->sd_args.ar_localflocks = 1;
1005 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1006         } else if (!strcmp("lock_dlm", proto)) {
1007                 lm = &gfs2_dlm_ops;
1008 #endif
1009         } else {
1010                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
1011                 return -ENOENT;
1012         }
1013
1014         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1015
1016         ls->ls_ops = lm;
1017         ls->ls_first = 1;
1018
1019         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1020                 substring_t tmp[MAX_OPT_ARGS];
1021                 int token, option;
1022
1023                 if (!o || !*o)
1024                         continue;
1025
1026                 token = match_token(o, *lm->lm_tokens, tmp);
1027                 switch (token) {
1028                 case Opt_jid:
1029                         ret = match_int(&tmp[0], &option);
1030                         if (ret || option < 0)
1031                                 goto hostdata_error;
1032                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1033                                 ls->ls_jid = option;
1034                         break;
1035                 case Opt_id:
1036                 case Opt_nodir:
1037                         /* Obsolete, but left for backward compat purposes */
1038                         break;
1039                 case Opt_first:
1040                         ret = match_int(&tmp[0], &option);
1041                         if (ret || (option != 0 && option != 1))
1042                                 goto hostdata_error;
1043                         ls->ls_first = option;
1044                         break;
1045                 case Opt_err:
1046                 default:
1047 hostdata_error:
1048                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1049                         return -EINVAL;
1050                 }
1051         }
1052
1053         if (lm->lm_mount == NULL) {
1054                 fs_info(sdp, "Now mounting FS...\n");
1055                 complete_all(&sdp->sd_locking_init);
1056                 return 0;
1057         }
1058         ret = lm->lm_mount(sdp, table);
1059         if (ret == 0)
1060                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1061         complete_all(&sdp->sd_locking_init);
1062         return ret;
1063 }
1064
1065 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1066 {
1067         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1068         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1069             lm->lm_unmount)
1070                 lm->lm_unmount(sdp);
1071 }
1072
1073 static int gfs2_journalid_wait(void *word)
1074 {
1075         if (signal_pending(current))
1076                 return -EINTR;
1077         schedule();
1078         return 0;
1079 }
1080
1081 static int wait_on_journal(struct gfs2_sbd *sdp)
1082 {
1083         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1084                 return 0;
1085
1086         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1087 }
1088
1089 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1090 {
1091         struct super_block *sb = sdp->sd_vfs;
1092         char ro[20];
1093         char spectator[20];
1094         char *envp[] = { ro, spectator, NULL };
1095         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1096         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1097         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1098 }
1099
1100 /**
1101  * fill_super - Read in superblock
1102  * @sb: The VFS superblock
1103  * @data: Mount options
1104  * @silent: Don't complain if it's not a GFS2 filesystem
1105  *
1106  * Returns: errno
1107  */
1108
1109 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1110 {
1111         struct gfs2_sbd *sdp;
1112         struct gfs2_holder mount_gh;
1113         int error;
1114
1115         sdp = init_sbd(sb);
1116         if (!sdp) {
1117                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1118                 return -ENOMEM;
1119         }
1120         sdp->sd_args = *args;
1121
1122         if (sdp->sd_args.ar_spectator) {
1123                 sb->s_flags |= MS_RDONLY;
1124                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1125         }
1126         if (sdp->sd_args.ar_posix_acl)
1127                 sb->s_flags |= MS_POSIXACL;
1128         if (sdp->sd_args.ar_nobarrier)
1129                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1130
1131         sb->s_flags |= MS_NOSEC;
1132         sb->s_magic = GFS2_MAGIC;
1133         sb->s_op = &gfs2_super_ops;
1134         sb->s_d_op = &gfs2_dops;
1135         sb->s_export_op = &gfs2_export_ops;
1136         sb->s_xattr = gfs2_xattr_handlers;
1137         sb->s_qcop = &gfs2_quotactl_ops;
1138         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1139         sb->s_time_gran = 1;
1140         sb->s_maxbytes = MAX_LFS_FILESIZE;
1141
1142         /* Set up the buffer cache and fill in some fake block size values
1143            to allow us to read-in the on-disk superblock. */
1144         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1145         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1146         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1147                                GFS2_BASIC_BLOCK_SHIFT;
1148         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1149
1150         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1151         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1152         if (sdp->sd_args.ar_statfs_quantum) {
1153                 sdp->sd_tune.gt_statfs_slow = 0;
1154                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1155         } else {
1156                 sdp->sd_tune.gt_statfs_slow = 1;
1157                 sdp->sd_tune.gt_statfs_quantum = 30;
1158         }
1159
1160         error = init_names(sdp, silent);
1161         if (error) {
1162                 /* In this case, we haven't initialized sysfs, so we have to
1163                    manually free the sdp. */
1164                 free_percpu(sdp->sd_lkstats);
1165                 kfree(sdp);
1166                 sb->s_fs_info = NULL;
1167                 return error;
1168         }
1169
1170         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1171
1172         error = gfs2_sys_fs_add(sdp);
1173         /*
1174          * If we hit an error here, gfs2_sys_fs_add will have called function
1175          * kobject_put which causes the sysfs usage count to go to zero, which
1176          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1177          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1178          * kobject_put to free sdp.
1179          */
1180         if (error)
1181                 return error;
1182
1183         gfs2_create_debugfs_file(sdp);
1184
1185         error = gfs2_lm_mount(sdp, silent);
1186         if (error)
1187                 goto fail_debug;
1188
1189         error = init_locking(sdp, &mount_gh, DO);
1190         if (error)
1191                 goto fail_lm;
1192
1193         error = init_sb(sdp, silent);
1194         if (error)
1195                 goto fail_locking;
1196
1197         error = wait_on_journal(sdp);
1198         if (error)
1199                 goto fail_sb;
1200
1201         /*
1202          * If user space has failed to join the cluster or some similar
1203          * failure has occurred, then the journal id will contain a
1204          * negative (error) number. This will then be returned to the
1205          * caller (of the mount syscall). We do this even for spectator
1206          * mounts (which just write a jid of 0 to indicate "ok" even though
1207          * the jid is unused in the spectator case)
1208          */
1209         if (sdp->sd_lockstruct.ls_jid < 0) {
1210                 error = sdp->sd_lockstruct.ls_jid;
1211                 sdp->sd_lockstruct.ls_jid = 0;
1212                 goto fail_sb;
1213         }
1214
1215         if (sdp->sd_args.ar_spectator)
1216                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1217                          sdp->sd_table_name);
1218         else
1219                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1220                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1221
1222         error = init_inodes(sdp, DO);
1223         if (error)
1224                 goto fail_sb;
1225
1226         error = init_per_node(sdp, DO);
1227         if (error)
1228                 goto fail_inodes;
1229
1230         error = gfs2_statfs_init(sdp);
1231         if (error) {
1232                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1233                 goto fail_per_node;
1234         }
1235
1236         if (!(sb->s_flags & MS_RDONLY)) {
1237                 error = gfs2_make_fs_rw(sdp);
1238                 if (error) {
1239                         fs_err(sdp, "can't make FS RW: %d\n", error);
1240                         goto fail_per_node;
1241                 }
1242         }
1243
1244         gfs2_glock_dq_uninit(&mount_gh);
1245         gfs2_online_uevent(sdp);
1246         return 0;
1247
1248 fail_per_node:
1249         init_per_node(sdp, UNDO);
1250 fail_inodes:
1251         init_inodes(sdp, UNDO);
1252 fail_sb:
1253         if (sdp->sd_root_dir)
1254                 dput(sdp->sd_root_dir);
1255         if (sdp->sd_master_dir)
1256                 dput(sdp->sd_master_dir);
1257         if (sb->s_root)
1258                 dput(sb->s_root);
1259         sb->s_root = NULL;
1260 fail_locking:
1261         init_locking(sdp, &mount_gh, UNDO);
1262 fail_lm:
1263         gfs2_gl_hash_clear(sdp);
1264         gfs2_lm_unmount(sdp);
1265 fail_debug:
1266         gfs2_delete_debugfs_file(sdp);
1267         free_percpu(sdp->sd_lkstats);
1268         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1269          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1270         gfs2_sys_fs_del(sdp);
1271         sb->s_fs_info = NULL;
1272         return error;
1273 }
1274
1275 static int set_gfs2_super(struct super_block *s, void *data)
1276 {
1277         s->s_bdev = data;
1278         s->s_dev = s->s_bdev->bd_dev;
1279
1280         /*
1281          * We set the bdi here to the queue backing, file systems can
1282          * overwrite this in ->fill_super()
1283          */
1284         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1285         return 0;
1286 }
1287
1288 static int test_gfs2_super(struct super_block *s, void *ptr)
1289 {
1290         struct block_device *bdev = ptr;
1291         return (bdev == s->s_bdev);
1292 }
1293
1294 /**
1295  * gfs2_mount - Get the GFS2 superblock
1296  * @fs_type: The GFS2 filesystem type
1297  * @flags: Mount flags
1298  * @dev_name: The name of the device
1299  * @data: The mount arguments
1300  *
1301  * Q. Why not use get_sb_bdev() ?
1302  * A. We need to select one of two root directories to mount, independent
1303  *    of whether this is the initial, or subsequent, mount of this sb
1304  *
1305  * Returns: 0 or -ve on error
1306  */
1307
1308 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1309                        const char *dev_name, void *data)
1310 {
1311         struct block_device *bdev;
1312         struct super_block *s;
1313         fmode_t mode = FMODE_READ | FMODE_EXCL;
1314         int error;
1315         struct gfs2_args args;
1316         struct gfs2_sbd *sdp;
1317
1318         if (!(flags & MS_RDONLY))
1319                 mode |= FMODE_WRITE;
1320
1321         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1322         if (IS_ERR(bdev))
1323                 return ERR_CAST(bdev);
1324
1325         /*
1326          * once the super is inserted into the list by sget, s_umount
1327          * will protect the lockfs code from trying to start a snapshot
1328          * while we are mounting
1329          */
1330         mutex_lock(&bdev->bd_fsfreeze_mutex);
1331         if (bdev->bd_fsfreeze_count > 0) {
1332                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1333                 error = -EBUSY;
1334                 goto error_bdev;
1335         }
1336         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1337         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1338         error = PTR_ERR(s);
1339         if (IS_ERR(s))
1340                 goto error_bdev;
1341
1342         if (s->s_root) {
1343                 /*
1344                  * s_umount nests inside bd_mutex during
1345                  * __invalidate_device().  blkdev_put() acquires
1346                  * bd_mutex and can't be called under s_umount.  Drop
1347                  * s_umount temporarily.  This is safe as we're
1348                  * holding an active reference.
1349                  */
1350                 up_write(&s->s_umount);
1351                 blkdev_put(bdev, mode);
1352                 down_write(&s->s_umount);
1353         }
1354
1355         memset(&args, 0, sizeof(args));
1356         args.ar_quota = GFS2_QUOTA_DEFAULT;
1357         args.ar_data = GFS2_DATA_DEFAULT;
1358         args.ar_commit = 30;
1359         args.ar_statfs_quantum = 30;
1360         args.ar_quota_quantum = 60;
1361         args.ar_errors = GFS2_ERRORS_DEFAULT;
1362
1363         error = gfs2_mount_args(&args, data);
1364         if (error) {
1365                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1366                 goto error_super;
1367         }
1368
1369         if (s->s_root) {
1370                 error = -EBUSY;
1371                 if ((flags ^ s->s_flags) & MS_RDONLY)
1372                         goto error_super;
1373         } else {
1374                 char b[BDEVNAME_SIZE];
1375
1376                 s->s_mode = mode;
1377                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1378                 sb_set_blocksize(s, block_size(bdev));
1379                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1380                 if (error)
1381                         goto error_super;
1382                 s->s_flags |= MS_ACTIVE;
1383                 bdev->bd_super = s;
1384         }
1385
1386         sdp = s->s_fs_info;
1387         if (args.ar_meta)
1388                 return dget(sdp->sd_master_dir);
1389         else
1390                 return dget(sdp->sd_root_dir);
1391
1392 error_super:
1393         deactivate_locked_super(s);
1394         return ERR_PTR(error);
1395 error_bdev:
1396         blkdev_put(bdev, mode);
1397         return ERR_PTR(error);
1398 }
1399
1400 static int set_meta_super(struct super_block *s, void *ptr)
1401 {
1402         return -EINVAL;
1403 }
1404
1405 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1406                         int flags, const char *dev_name, void *data)
1407 {
1408         struct super_block *s;
1409         struct gfs2_sbd *sdp;
1410         struct path path;
1411         int error;
1412
1413         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1414         if (error) {
1415                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1416                        dev_name, error);
1417                 return ERR_PTR(error);
1418         }
1419         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1420                  path.dentry->d_inode->i_sb->s_bdev);
1421         path_put(&path);
1422         if (IS_ERR(s)) {
1423                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1424                 return ERR_CAST(s);
1425         }
1426         if ((flags ^ s->s_flags) & MS_RDONLY) {
1427                 deactivate_locked_super(s);
1428                 return ERR_PTR(-EBUSY);
1429         }
1430         sdp = s->s_fs_info;
1431         return dget(sdp->sd_master_dir);
1432 }
1433
1434 static void gfs2_kill_sb(struct super_block *sb)
1435 {
1436         struct gfs2_sbd *sdp = sb->s_fs_info;
1437
1438         if (sdp == NULL) {
1439                 kill_block_super(sb);
1440                 return;
1441         }
1442
1443         gfs2_meta_syncfs(sdp);
1444         dput(sdp->sd_root_dir);
1445         dput(sdp->sd_master_dir);
1446         sdp->sd_root_dir = NULL;
1447         sdp->sd_master_dir = NULL;
1448         shrink_dcache_sb(sb);
1449         gfs2_delete_debugfs_file(sdp);
1450         free_percpu(sdp->sd_lkstats);
1451         kill_block_super(sb);
1452 }
1453
1454 struct file_system_type gfs2_fs_type = {
1455         .name = "gfs2",
1456         .fs_flags = FS_REQUIRES_DEV,
1457         .mount = gfs2_mount,
1458         .kill_sb = gfs2_kill_sb,
1459         .owner = THIS_MODULE,
1460 };
1461 MODULE_ALIAS_FS("gfs2");
1462
1463 struct file_system_type gfs2meta_fs_type = {
1464         .name = "gfs2meta",
1465         .fs_flags = FS_REQUIRES_DEV,
1466         .mount = gfs2_mount_meta,
1467         .owner = THIS_MODULE,
1468 };
1469 MODULE_ALIAS_FS("gfs2meta");