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