fs/gfs2/ops_fstype.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/kthread.h>
  17 #include <linux/export.h>
  18 #include <linux/namei.h>
  19 #include <linux/mount.h>
  20 #include <linux/gfs2_ondisk.h>
  21 #include <linux/quotaops.h>
  22
  23 #include "gfs2.h"
  24 #include "incore.h"
  25 #include "bmap.h"
  26 #include "glock.h"
  27 #include "glops.h"
  28 #include "inode.h"
  29 #include "recovery.h"
  30 #include "rgrp.h"
  31 #include "super.h"
  32 #include "sys.h"
  33 #include "util.h"
  34 #include "log.h"
  35 #include "quota.h"
  36 #include "dir.h"
  37 #include "trace_gfs2.h"
  38
  39 #define DO 0
  40 #define UNDO 1
  41
  42 /**
  43  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  44  * @gt: tune
  45  *
  46  */
  47
  48 static void gfs2_tune_init(struct gfs2_tune *gt)
  49 {
  50         spin_lock_init(&gt->gt_spin);
  51
  52         gt->gt_quota_simul_sync = 64;
  53         gt->gt_quota_warn_period = 10;
  54         gt->gt_quota_scale_num = 1;
  55         gt->gt_quota_scale_den = 1;
  56         gt->gt_new_files_jdata = 0;
  57         gt->gt_max_readahead = 1 << 18;
  58         gt->gt_complain_secs = 10;
  59 }
  60
  61 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  62 {
  63         struct gfs2_sbd *sdp;
  64
  65         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  66         if (!sdp)
  67                 return NULL;
  68
  69         sb->s_fs_info = sdp;
  70         sdp->sd_vfs = sb;
  71         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  72         gfs2_tune_init(&sdp->sd_tune);
  73
  74         init_waitqueue_head(&sdp->sd_glock_wait);
  75         atomic_set(&sdp->sd_glock_disposal, 0);
  76         init_completion(&sdp->sd_locking_init);
  77         spin_lock_init(&sdp->sd_statfs_spin);
  78
  79         spin_lock_init(&sdp->sd_rindex_spin);
  80         mutex_init(&sdp->sd_rindex_mutex);
  81         sdp->sd_rindex_tree.rb_node = NULL;
  82
  83         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  84         spin_lock_init(&sdp->sd_jindex_spin);
  85         mutex_init(&sdp->sd_jindex_mutex);
  86
  87         INIT_LIST_HEAD(&sdp->sd_quota_list);
  88         mutex_init(&sdp->sd_quota_mutex);
  89         init_waitqueue_head(&sdp->sd_quota_wait);
  90         INIT_LIST_HEAD(&sdp->sd_trunc_list);
  91         spin_lock_init(&sdp->sd_trunc_lock);
  92
  93         spin_lock_init(&sdp->sd_log_lock);
  94         atomic_set(&sdp->sd_log_pinned, 0);
  95         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
  96         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
  97         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
  98         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
  99         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 100
 101         init_waitqueue_head(&sdp->sd_log_waitq);
 102         init_waitqueue_head(&sdp->sd_logd_waitq);
 103         spin_lock_init(&sdp->sd_ail_lock);
 104         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 105         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 106
 107         init_rwsem(&sdp->sd_log_flush_lock);
 108         atomic_set(&sdp->sd_log_in_flight, 0);
 109         init_waitqueue_head(&sdp->sd_log_flush_wait);
 110
 111         INIT_LIST_HEAD(&sdp->sd_revoke_list);
 112
 113         mutex_init(&sdp->sd_freeze_lock);
 114
 115         return sdp;
 116 }
 117
 118
 119 /**
 120  * gfs2_check_sb - Check superblock
 121  * @sdp: the filesystem
 122  * @sb: The superblock
 123  * @silent: Don't print a message if the check fails
 124  *
 125  * Checks the version code of the FS is one that we understand how to
 126  * read and that the sizes of the various on-disk structures have not
 127  * changed.
 128  */
 129
 130 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 131 {
 132         struct gfs2_sb_host *sb = &sdp->sd_sb;
 133
 134         if (sb->sb_magic != GFS2_MAGIC ||
 135             sb->sb_type != GFS2_METATYPE_SB) {
 136                 if (!silent)
 137                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 138                 return -EINVAL;
 139         }
 140
 141         /*  If format numbers match exactly, we're done.  */
 142
 143         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 144             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 145                 return 0;
 146
 147         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 148
 149         return -EINVAL;
 150 }
 151
 152 static void end_bio_io_page(struct bio *bio, int error)
 153 {
 154         struct page *page = bio->bi_private;
 155
 156         if (!error)
 157                 SetPageUptodate(page);
 158         else
 159                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 160         unlock_page(page);
 161 }
 162
 163 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 164 {
 165         struct gfs2_sb_host *sb = &sdp->sd_sb;
 166         struct super_block *s = sdp->sd_vfs;
 167         const struct gfs2_sb *str = buf;
 168
 169         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 170         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 171         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 172         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 173         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 174         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 175         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 176         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 177         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 178         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 179         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 180
 181         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 182         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 183         memcpy(s->s_uuid, str->sb_uuid, 16);
 184 }
 185
 186 /**
 187  * gfs2_read_super - Read the gfs2 super block from disk
 188  * @sdp: The GFS2 super block
 189  * @sector: The location of the super block
 190  * @error: The error code to return
 191  *
 192  * This uses the bio functions to read the super block from disk
 193  * because we want to be 100% sure that we never read cached data.
 194  * A super block is read twice only during each GFS2 mount and is
 195  * never written to by the filesystem. The first time its read no
 196  * locks are held, and the only details which are looked at are those
 197  * relating to the locking protocol. Once locking is up and working,
 198  * the sb is read again under the lock to establish the location of
 199  * the master directory (contains pointers to journals etc) and the
 200  * root directory.
 201  *
 202  * Returns: 0 on success or error
 203  */
 204
 205 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 206 {
 207         struct super_block *sb = sdp->sd_vfs;
 208         struct gfs2_sb *p;
 209         struct page *page;
 210         struct bio *bio;
 211
 212         page = alloc_page(GFP_NOFS);
 213         if (unlikely(!page))
 214                 return -ENOBUFS;
 215
 216         ClearPageUptodate(page);
 217         ClearPageDirty(page);
 218         lock_page(page);
 219
 220         bio = bio_alloc(GFP_NOFS, 1);
 221         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 222         bio->bi_bdev = sb->s_bdev;
 223         bio_add_page(bio, page, PAGE_SIZE, 0);
 224
 225         bio->bi_end_io = end_bio_io_page;
 226         bio->bi_private = page;
 227         submit_bio(READ_SYNC | REQ_META | REQ_PRIO, bio);
 228         wait_on_page_locked(page);
 229         bio_put(bio);
 230         if (!PageUptodate(page)) {
 231                 __free_page(page);
 232                 return -EIO;
 233         }
 234         p = kmap(page);
 235         gfs2_sb_in(sdp, p);
 236         kunmap(page);
 237         __free_page(page);
 238         return gfs2_check_sb(sdp, silent);
 239 }
 240
 241 /**
 242  * gfs2_read_sb - Read super block
 243  * @sdp: The GFS2 superblock
 244  * @silent: Don't print message if mount fails
 245  *
 246  */
 247
 248 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 249 {
 250         u32 hash_blocks, ind_blocks, leaf_blocks;
 251         u32 tmp_blocks;
 252         unsigned int x;
 253         int error;
 254
 255         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 256         if (error) {
 257                 if (!silent)
 258                         fs_err(sdp, "can't read superblock\n");
 259                 return error;
 260         }
 261
 262         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 263                                GFS2_BASIC_BLOCK_SHIFT;
 264         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 265         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 266                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 267         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 268                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 269         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 270         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 271         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 272         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 273         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 274                                 sizeof(struct gfs2_meta_header)) /
 275                                 sizeof(struct gfs2_quota_change);
 276
 277         /* Compute maximum reservation required to add a entry to a directory */
 278
 279         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 280                              sdp->sd_jbsize);
 281
 282         ind_blocks = 0;
 283         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 284                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 285                 ind_blocks += tmp_blocks;
 286         }
 287
 288         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 289
 290         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 291
 292         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 293                                 sizeof(struct gfs2_dinode);
 294         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 295         for (x = 2;; x++) {
 296                 u64 space, d;
 297                 u32 m;
 298
 299                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 300                 d = space;
 301                 m = do_div(d, sdp->sd_inptrs);
 302
 303                 if (d != sdp->sd_heightsize[x - 1] || m)
 304                         break;
 305                 sdp->sd_heightsize[x] = space;
 306         }
 307         sdp->sd_max_height = x;
 308         sdp->sd_heightsize[x] = ~0;
 309         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 310
 311         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 312                                  sizeof(struct gfs2_dinode);
 313         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 314         for (x = 2;; x++) {
 315                 u64 space, d;
 316                 u32 m;
 317
 318                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 319                 d = space;
 320                 m = do_div(d, sdp->sd_inptrs);
 321
 322                 if (d != sdp->sd_jheightsize[x - 1] || m)
 323                         break;
 324                 sdp->sd_jheightsize[x] = space;
 325         }
 326         sdp->sd_max_jheight = x;
 327         sdp->sd_jheightsize[x] = ~0;
 328         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 329
 330         return 0;
 331 }
 332
 333 static int init_names(struct gfs2_sbd *sdp, int silent)
 334 {
 335         char *proto, *table;
 336         int error = 0;
 337
 338         proto = sdp->sd_args.ar_lockproto;
 339         table = sdp->sd_args.ar_locktable;
 340
 341         /*  Try to autodetect  */
 342
 343         if (!proto[0] || !table[0]) {
 344                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 345                 if (error)
 346                         return error;
 347
 348                 if (!proto[0])
 349                         proto = sdp->sd_sb.sb_lockproto;
 350                 if (!table[0])
 351                         table = sdp->sd_sb.sb_locktable;
 352         }
 353
 354         if (!table[0])
 355                 table = sdp->sd_vfs->s_id;
 356
 357         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 358         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 359
 360         table = sdp->sd_table_name;
 361         while ((table = strchr(table, '/')))
 362                 *table = '_';
 363
 364         return error;
 365 }
 366
 367 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 368                         int undo)
 369 {
 370         int error = 0;
 371
 372         if (undo)
 373                 goto fail_trans;
 374
 375         error = gfs2_glock_nq_num(sdp,
 376                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 377                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 378                                   mount_gh);
 379         if (error) {
 380                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 381                 goto fail;
 382         }
 383
 384         error = gfs2_glock_nq_num(sdp,
 385                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 386                                   LM_ST_SHARED,
 387                                   LM_FLAG_NOEXP | GL_EXACT,
 388                                   &sdp->sd_live_gh);
 389         if (error) {
 390                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 391                 goto fail_mount;
 392         }
 393
 394         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 395                                CREATE, &sdp->sd_rename_gl);
 396         if (error) {
 397                 fs_err(sdp, "can't create rename glock: %d\n", error);
 398                 goto fail_live;
 399         }
 400
 401         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 402                                CREATE, &sdp->sd_trans_gl);
 403         if (error) {
 404                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 405                 goto fail_rename;
 406         }
 407
 408         return 0;
 409
 410 fail_trans:
 411         gfs2_glock_put(sdp->sd_trans_gl);
 412 fail_rename:
 413         gfs2_glock_put(sdp->sd_rename_gl);
 414 fail_live:
 415         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 416 fail_mount:
 417         gfs2_glock_dq_uninit(mount_gh);
 418 fail:
 419         return error;
 420 }
 421
 422 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 423                             u64 no_addr, const char *name)
 424 {
 425         struct gfs2_sbd *sdp = sb->s_fs_info;
 426         struct dentry *dentry;
 427         struct inode *inode;
 428
 429         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 430         if (IS_ERR(inode)) {
 431                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 432                 return PTR_ERR(inode);
 433         }
 434         dentry = d_alloc_root(inode);
 435         if (!dentry) {
 436                 fs_err(sdp, "can't alloc %s dentry\n", name);
 437                 iput(inode);
 438                 return -ENOMEM;
 439         }
 440         *dptr = dentry;
 441         return 0;
 442 }
 443
 444 static int init_sb(struct gfs2_sbd *sdp, int silent)
 445 {
 446         struct super_block *sb = sdp->sd_vfs;
 447         struct gfs2_holder sb_gh;
 448         u64 no_addr;
 449         int ret;
 450
 451         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 452                                 LM_ST_SHARED, 0, &sb_gh);
 453         if (ret) {
 454                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 455                 return ret;
 456         }
 457
 458         ret = gfs2_read_sb(sdp, silent);
 459         if (ret) {
 460                 fs_err(sdp, "can't read superblock: %d\n", ret);
 461                 goto out;
 462         }
 463
 464         /* Set up the buffer cache and SB for real */
 465         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 466                 ret = -EINVAL;
 467                 fs_err(sdp, "FS block size (%u) is too small for device "
 468                        "block size (%u)\n",
 469                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 470                 goto out;
 471         }
 472         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 473                 ret = -EINVAL;
 474                 fs_err(sdp, "FS block size (%u) is too big for machine "
 475                        "page size (%u)\n",
 476                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 477                 goto out;
 478         }
 479         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 480
 481         /* Get the root inode */
 482         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 483         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 484         if (ret)
 485                 goto out;
 486
 487         /* Get the master inode */
 488         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 489         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 490         if (ret) {
 491                 dput(sdp->sd_root_dir);
 492                 goto out;
 493         }
 494         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 495 out:
 496         gfs2_glock_dq_uninit(&sb_gh);
 497         return ret;
 498 }
 499
 500 /**
 501  * map_journal_extents - create a reusable "extent" mapping from all logical
 502  * blocks to all physical blocks for the given journal.  This will save
 503  * us time when writing journal blocks.  Most journals will have only one
 504  * extent that maps all their logical blocks.  That's because gfs2.mkfs
 505  * arranges the journal blocks sequentially to maximize performance.
 506  * So the extent would map the first block for the entire file length.
 507  * However, gfs2_jadd can happen while file activity is happening, so
 508  * those journals may not be sequential.  Less likely is the case where
 509  * the users created their own journals by mounting the metafs and
 510  * laying it out.  But it's still possible.  These journals might have
 511  * several extents.
 512  *
 513  * TODO: This should be done in bigger chunks rather than one block at a time,
 514  *       but since it's only done at mount time, I'm not worried about the
 515  *       time it takes.
 516  */
 517 static int map_journal_extents(struct gfs2_sbd *sdp)
 518 {
 519         struct gfs2_jdesc *jd = sdp->sd_jdesc;
 520         unsigned int lb;
 521         u64 db, prev_db; /* logical block, disk block, prev disk block */
 522         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 523         struct gfs2_journal_extent *jext = NULL;
 524         struct buffer_head bh;
 525         int rc = 0;
 526
 527         prev_db = 0;
 528
 529         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
 530                 bh.b_state = 0;
 531                 bh.b_blocknr = 0;
 532                 bh.b_size = 1 << ip->i_inode.i_blkbits;
 533                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
 534                 db = bh.b_blocknr;
 535                 if (rc || !db) {
 536                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
 537                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
 538                         break;
 539                 }
 540                 if (!prev_db || db != prev_db + 1) {
 541                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
 542                                        GFP_KERNEL);
 543                         if (!jext) {
 544                                 printk(KERN_INFO "GFS2 error: out of memory "
 545                                        "mapping journal extents.\n");
 546                                 rc = -ENOMEM;
 547                                 break;
 548                         }
 549                         jext->dblock = db;
 550                         jext->lblock = lb;
 551                         jext->blocks = 1;
 552                         list_add_tail(&jext->extent_list, &jd->extent_list);
 553                 } else {
 554                         jext->blocks++;
 555                 }
 556                 prev_db = db;
 557         }
 558         return rc;
 559 }
 560
 561 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 562 {
 563         char *message = "FIRSTMOUNT=Done";
 564         char *envp[] = { message, NULL };
 565         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 566         ls->ls_first_done = 1;
 567         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 568 }
 569
 570 /**
 571  * gfs2_jindex_hold - Grab a lock on the jindex
 572  * @sdp: The GFS2 superblock
 573  * @ji_gh: the holder for the jindex glock
 574  *
 575  * Returns: errno
 576  */
 577
 578 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 579 {
 580         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 581         struct qstr name;
 582         char buf[20];
 583         struct gfs2_jdesc *jd;
 584         int error;
 585
 586         name.name = buf;
 587
 588         mutex_lock(&sdp->sd_jindex_mutex);
 589
 590         for (;;) {
 591                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 592                 if (error)
 593                         break;
 594
 595                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 596                 name.hash = gfs2_disk_hash(name.name, name.len);
 597
 598                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 599                 if (error == -ENOENT) {
 600                         error = 0;
 601                         break;
 602                 }
 603
 604                 gfs2_glock_dq_uninit(ji_gh);
 605
 606                 if (error)
 607                         break;
 608
 609                 error = -ENOMEM;
 610                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 611                 if (!jd)
 612                         break;
 613
 614                 INIT_LIST_HEAD(&jd->extent_list);
 615                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 616                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 617                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 618                         if (!jd->jd_inode)
 619                                 error = -ENOENT;
 620                         else
 621                                 error = PTR_ERR(jd->jd_inode);
 622                         kfree(jd);
 623                         break;
 624                 }
 625
 626                 spin_lock(&sdp->sd_jindex_spin);
 627                 jd->jd_jid = sdp->sd_journals++;
 628                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 629                 spin_unlock(&sdp->sd_jindex_spin);
 630         }
 631
 632         mutex_unlock(&sdp->sd_jindex_mutex);
 633
 634         return error;
 635 }
 636
 637 static int init_journal(struct gfs2_sbd *sdp, int undo)
 638 {
 639         struct inode *master = sdp->sd_master_dir->d_inode;
 640         struct gfs2_holder ji_gh;
 641         struct gfs2_inode *ip;
 642         int jindex = 1;
 643         int error = 0;
 644
 645         if (undo) {
 646                 jindex = 0;
 647                 goto fail_jinode_gh;
 648         }
 649
 650         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 651         if (IS_ERR(sdp->sd_jindex)) {
 652                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 653                 return PTR_ERR(sdp->sd_jindex);
 654         }
 655
 656         /* Load in the journal index special file */
 657
 658         error = gfs2_jindex_hold(sdp, &ji_gh);
 659         if (error) {
 660                 fs_err(sdp, "can't read journal index: %d\n", error);
 661                 goto fail;
 662         }
 663
 664         error = -EUSERS;
 665         if (!gfs2_jindex_size(sdp)) {
 666                 fs_err(sdp, "no journals!\n");
 667                 goto fail_jindex;
 668         }
 669
 670         if (sdp->sd_args.ar_spectator) {
 671                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 672                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 673                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 674                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 675         } else {
 676                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 677                         fs_err(sdp, "can't mount journal #%u\n",
 678                                sdp->sd_lockstruct.ls_jid);
 679                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 680                                gfs2_jindex_size(sdp),
 681                                gfs2_jindex_size(sdp) - 1);
 682                         goto fail_jindex;
 683                 }
 684                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 685
 686                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 687                                           &gfs2_journal_glops,
 688                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 689                                           &sdp->sd_journal_gh);
 690                 if (error) {
 691                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 692                         goto fail_jindex;
 693                 }
 694
 695                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 696                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 697                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 698                                            &sdp->sd_jinode_gh);
 699                 if (error) {
 700                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 701                                error);
 702                         goto fail_journal_gh;
 703                 }
 704
 705                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 706                 if (error) {
 707                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 708                                sdp->sd_jdesc->jd_jid, error);
 709                         goto fail_jinode_gh;
 710                 }
 711                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 712                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 713                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 714
 715                 /* Map the extents for this journal's blocks */
 716                 map_journal_extents(sdp);
 717         }
 718         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 719
 720         if (sdp->sd_lockstruct.ls_first) {
 721                 unsigned int x;
 722                 for (x = 0; x < sdp->sd_journals; x++) {
 723                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
 724                                                      true);
 725                         if (error) {
 726                                 fs_err(sdp, "error recovering journal %u: %d\n",
 727                                        x, error);
 728                                 goto fail_jinode_gh;
 729                         }
 730                 }
 731
 732                 gfs2_others_may_mount(sdp);
 733         } else if (!sdp->sd_args.ar_spectator) {
 734                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 735                 if (error) {
 736                         fs_err(sdp, "error recovering my journal: %d\n", error);
 737                         goto fail_jinode_gh;
 738                 }
 739         }
 740
 741         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 742         gfs2_glock_dq_uninit(&ji_gh);
 743         jindex = 0;
 744
 745         return 0;
 746
 747 fail_jinode_gh:
 748         if (!sdp->sd_args.ar_spectator)
 749                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 750 fail_journal_gh:
 751         if (!sdp->sd_args.ar_spectator)
 752                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 753 fail_jindex:
 754         gfs2_jindex_free(sdp);
 755         if (jindex)
 756                 gfs2_glock_dq_uninit(&ji_gh);
 757 fail:
 758         iput(sdp->sd_jindex);
 759         return error;
 760 }
 761
 762
 763 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 764 {
 765         int error = 0;
 766         struct inode *master = sdp->sd_master_dir->d_inode;
 767
 768         if (undo)
 769                 goto fail_qinode;
 770
 771         error = init_journal(sdp, undo);
 772         if (error)
 773                 goto fail;
 774
 775         /* Read in the master statfs inode */
 776         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 777         if (IS_ERR(sdp->sd_statfs_inode)) {
 778                 error = PTR_ERR(sdp->sd_statfs_inode);
 779                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 780                 goto fail_journal;
 781         }
 782
 783         /* Read in the resource index inode */
 784         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 785         if (IS_ERR(sdp->sd_rindex)) {
 786                 error = PTR_ERR(sdp->sd_rindex);
 787                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 788                 goto fail_statfs;
 789         }
 790         sdp->sd_rindex_uptodate = 0;
 791
 792         /* Read in the quota inode */
 793         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 794         if (IS_ERR(sdp->sd_quota_inode)) {
 795                 error = PTR_ERR(sdp->sd_quota_inode);
 796                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 797                 goto fail_rindex;
 798         }
 799         return 0;
 800
 801 fail_qinode:
 802         iput(sdp->sd_quota_inode);
 803 fail_rindex:
 804         gfs2_clear_rgrpd(sdp);
 805         iput(sdp->sd_rindex);
 806 fail_statfs:
 807         iput(sdp->sd_statfs_inode);
 808 fail_journal:
 809         init_journal(sdp, UNDO);
 810 fail:
 811         return error;
 812 }
 813
 814 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 815 {
 816         struct inode *pn = NULL;
 817         char buf[30];
 818         int error = 0;
 819         struct gfs2_inode *ip;
 820         struct inode *master = sdp->sd_master_dir->d_inode;
 821
 822         if (sdp->sd_args.ar_spectator)
 823                 return 0;
 824
 825         if (undo)
 826                 goto fail_qc_gh;
 827
 828         pn = gfs2_lookup_simple(master, "per_node");
 829         if (IS_ERR(pn)) {
 830                 error = PTR_ERR(pn);
 831                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 832                 return error;
 833         }
 834
 835         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 836         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 837         if (IS_ERR(sdp->sd_sc_inode)) {
 838                 error = PTR_ERR(sdp->sd_sc_inode);
 839                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 840                 goto fail;
 841         }
 842
 843         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 844         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 845         if (IS_ERR(sdp->sd_qc_inode)) {
 846                 error = PTR_ERR(sdp->sd_qc_inode);
 847                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 848                 goto fail_ut_i;
 849         }
 850
 851         iput(pn);
 852         pn = NULL;
 853
 854         ip = GFS2_I(sdp->sd_sc_inode);
 855         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 856                                    &sdp->sd_sc_gh);
 857         if (error) {
 858                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 859                 goto fail_qc_i;
 860         }
 861
 862         ip = GFS2_I(sdp->sd_qc_inode);
 863         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 864                                    &sdp->sd_qc_gh);
 865         if (error) {
 866                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 867                 goto fail_ut_gh;
 868         }
 869
 870         return 0;
 871
 872 fail_qc_gh:
 873         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 874 fail_ut_gh:
 875         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 876 fail_qc_i:
 877         iput(sdp->sd_qc_inode);
 878 fail_ut_i:
 879         iput(sdp->sd_sc_inode);
 880 fail:
 881         if (pn)
 882                 iput(pn);
 883         return error;
 884 }
 885
 886 static int init_threads(struct gfs2_sbd *sdp, int undo)
 887 {
 888         struct task_struct *p;
 889         int error = 0;
 890
 891         if (undo)
 892                 goto fail_quotad;
 893
 894         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 895         error = IS_ERR(p);
 896         if (error) {
 897                 fs_err(sdp, "can't start logd thread: %d\n", error);
 898                 return error;
 899         }
 900         sdp->sd_logd_process = p;
 901
 902         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 903         error = IS_ERR(p);
 904         if (error) {
 905                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 906                 goto fail;
 907         }
 908         sdp->sd_quotad_process = p;
 909
 910         return 0;
 911
 912
 913 fail_quotad:
 914         kthread_stop(sdp->sd_quotad_process);
 915 fail:
 916         kthread_stop(sdp->sd_logd_process);
 917         return error;
 918 }
 919
 920 static const match_table_t nolock_tokens = {
 921         { Opt_jid, "jid=%d\n", },
 922         { Opt_err, NULL },
 923 };
 924
 925 static const struct lm_lockops nolock_ops = {
 926         .lm_proto_name = "lock_nolock",
 927         .lm_put_lock = gfs2_glock_free,
 928         .lm_tokens = &nolock_tokens,
 929 };
 930
 931 /**
 932  * gfs2_lm_mount - mount a locking protocol
 933  * @sdp: the filesystem
 934  * @args: mount arguments
 935  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 936  *
 937  * Returns: errno
 938  */
 939
 940 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 941 {
 942         const struct lm_lockops *lm;
 943         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 944         struct gfs2_args *args = &sdp->sd_args;
 945         const char *proto = sdp->sd_proto_name;
 946         const char *table = sdp->sd_table_name;
 947         const char *fsname;
 948         char *o, *options;
 949         int ret;
 950
 951         if (!strcmp("lock_nolock", proto)) {
 952                 lm = &nolock_ops;
 953                 sdp->sd_args.ar_localflocks = 1;
 954 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 955         } else if (!strcmp("lock_dlm", proto)) {
 956                 lm = &gfs2_dlm_ops;
 957 #endif
 958         } else {
 959                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
 960                 return -ENOENT;
 961         }
 962
 963         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 964
 965         ls->ls_ops = lm;
 966         ls->ls_first = 1;
 967
 968         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 969                 substring_t tmp[MAX_OPT_ARGS];
 970                 int token, option;
 971
 972                 if (!o || !*o)
 973                         continue;
 974
 975                 token = match_token(o, *lm->lm_tokens, tmp);
 976                 switch (token) {
 977                 case Opt_jid:
 978                         ret = match_int(&tmp[0], &option);
 979                         if (ret || option < 0)
 980                                 goto hostdata_error;
 981                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 982                                 ls->ls_jid = option;
 983                         break;
 984                 case Opt_id:
 985                         /* Obsolete, but left for backward compat purposes */
 986                         break;
 987                 case Opt_first:
 988                         ret = match_int(&tmp[0], &option);
 989                         if (ret || (option != 0 && option != 1))
 990                                 goto hostdata_error;
 991                         ls->ls_first = option;
 992                         break;
 993                 case Opt_nodir:
 994                         ret = match_int(&tmp[0], &option);
 995                         if (ret || (option != 0 && option != 1))
 996                                 goto hostdata_error;
 997                         ls->ls_nodir = option;
 998                         break;
 999                 case Opt_err:
1000                 default:
1001 hostdata_error:
1002                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1003                         return -EINVAL;
1004                 }
1005         }
1006
1007         if (sdp->sd_args.ar_spectator)
1008                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1009         else
1010                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1011                          sdp->sd_lockstruct.ls_jid);
1012
1013         fsname = strchr(table, ':');
1014         if (fsname)
1015                 fsname++;
1016         if (lm->lm_mount == NULL) {
1017                 fs_info(sdp, "Now mounting FS...\n");
1018                 complete_all(&sdp->sd_locking_init);
1019                 return 0;
1020         }
1021         ret = lm->lm_mount(sdp, fsname);
1022         if (ret == 0)
1023                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1024         complete_all(&sdp->sd_locking_init);
1025         return ret;
1026 }
1027
1028 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1029 {
1030         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1031         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1032             lm->lm_unmount)
1033                 lm->lm_unmount(sdp);
1034 }
1035
1036 static int gfs2_journalid_wait(void *word)
1037 {
1038         if (signal_pending(current))
1039                 return -EINTR;
1040         schedule();
1041         return 0;
1042 }
1043
1044 static int wait_on_journal(struct gfs2_sbd *sdp)
1045 {
1046         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1047                 return 0;
1048
1049         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1050 }
1051
1052 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1053 {
1054         struct super_block *sb = sdp->sd_vfs;
1055         char ro[20];
1056         char spectator[20];
1057         char *envp[] = { ro, spectator, NULL };
1058         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1059         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1060         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1061 }
1062
1063 /**
1064  * fill_super - Read in superblock
1065  * @sb: The VFS superblock
1066  * @data: Mount options
1067  * @silent: Don't complain if it's not a GFS2 filesystem
1068  *
1069  * Returns: errno
1070  */
1071
1072 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1073 {
1074         struct gfs2_sbd *sdp;
1075         struct gfs2_holder mount_gh;
1076         int error;
1077
1078         sdp = init_sbd(sb);
1079         if (!sdp) {
1080                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1081                 return -ENOMEM;
1082         }
1083         sdp->sd_args = *args;
1084
1085         if (sdp->sd_args.ar_spectator) {
1086                 sb->s_flags |= MS_RDONLY;
1087                 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1088         }
1089         if (sdp->sd_args.ar_posix_acl)
1090                 sb->s_flags |= MS_POSIXACL;
1091         if (sdp->sd_args.ar_nobarrier)
1092                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1093
1094         sb->s_flags |= MS_NOSEC;
1095         sb->s_magic = GFS2_MAGIC;
1096         sb->s_op = &gfs2_super_ops;
1097         sb->s_d_op = &gfs2_dops;
1098         sb->s_export_op = &gfs2_export_ops;
1099         sb->s_xattr = gfs2_xattr_handlers;
1100         sb->s_qcop = &gfs2_quotactl_ops;
1101         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1102         sb->s_time_gran = 1;
1103         sb->s_maxbytes = MAX_LFS_FILESIZE;
1104
1105         /* Set up the buffer cache and fill in some fake block size values
1106            to allow us to read-in the on-disk superblock. */
1107         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1108         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1109         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1110                                GFS2_BASIC_BLOCK_SHIFT;
1111         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1112
1113         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1114         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1115         if (sdp->sd_args.ar_statfs_quantum) {
1116                 sdp->sd_tune.gt_statfs_slow = 0;
1117                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1118         } else {
1119                 sdp->sd_tune.gt_statfs_slow = 1;
1120                 sdp->sd_tune.gt_statfs_quantum = 30;
1121         }
1122
1123         error = init_names(sdp, silent);
1124         if (error)
1125                 goto fail;
1126
1127         gfs2_create_debugfs_file(sdp);
1128
1129         error = gfs2_sys_fs_add(sdp);
1130         if (error)
1131                 goto fail;
1132
1133         error = gfs2_lm_mount(sdp, silent);
1134         if (error)
1135                 goto fail_sys;
1136
1137         error = init_locking(sdp, &mount_gh, DO);
1138         if (error)
1139                 goto fail_lm;
1140
1141         error = init_sb(sdp, silent);
1142         if (error)
1143                 goto fail_locking;
1144
1145         error = wait_on_journal(sdp);
1146         if (error)
1147                 goto fail_sb;
1148
1149         /*
1150          * If user space has failed to join the cluster or some similar
1151          * failure has occurred, then the journal id will contain a
1152          * negative (error) number. This will then be returned to the
1153          * caller (of the mount syscall). We do this even for spectator
1154          * mounts (which just write a jid of 0 to indicate "ok" even though
1155          * the jid is unused in the spectator case)
1156          */
1157         if (sdp->sd_lockstruct.ls_jid < 0) {
1158                 error = sdp->sd_lockstruct.ls_jid;
1159                 sdp->sd_lockstruct.ls_jid = 0;
1160                 goto fail_sb;
1161         }
1162
1163         error = init_inodes(sdp, DO);
1164         if (error)
1165                 goto fail_sb;
1166
1167         error = init_per_node(sdp, DO);
1168         if (error)
1169                 goto fail_inodes;
1170
1171         error = gfs2_statfs_init(sdp);
1172         if (error) {
1173                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1174                 goto fail_per_node;
1175         }
1176
1177         error = init_threads(sdp, DO);
1178         if (error)
1179                 goto fail_per_node;
1180
1181         if (!(sb->s_flags & MS_RDONLY)) {
1182                 error = gfs2_make_fs_rw(sdp);
1183                 if (error) {
1184                         fs_err(sdp, "can't make FS RW: %d\n", error);
1185                         goto fail_threads;
1186                 }
1187         }
1188
1189         gfs2_glock_dq_uninit(&mount_gh);
1190         gfs2_online_uevent(sdp);
1191         return 0;
1192
1193 fail_threads:
1194         init_threads(sdp, UNDO);
1195 fail_per_node:
1196         init_per_node(sdp, UNDO);
1197 fail_inodes:
1198         init_inodes(sdp, UNDO);
1199 fail_sb:
1200         if (sdp->sd_root_dir)
1201                 dput(sdp->sd_root_dir);
1202         if (sdp->sd_master_dir)
1203                 dput(sdp->sd_master_dir);
1204         if (sb->s_root)
1205                 dput(sb->s_root);
1206         sb->s_root = NULL;
1207 fail_locking:
1208         init_locking(sdp, &mount_gh, UNDO);
1209 fail_lm:
1210         gfs2_gl_hash_clear(sdp);
1211         gfs2_lm_unmount(sdp);
1212 fail_sys:
1213         gfs2_sys_fs_del(sdp);
1214 fail:
1215         gfs2_delete_debugfs_file(sdp);
1216         kfree(sdp);
1217         sb->s_fs_info = NULL;
1218         return error;
1219 }
1220
1221 static int set_gfs2_super(struct super_block *s, void *data)
1222 {
1223         s->s_bdev = data;
1224         s->s_dev = s->s_bdev->bd_dev;
1225
1226         /*
1227          * We set the bdi here to the queue backing, file systems can
1228          * overwrite this in ->fill_super()
1229          */
1230         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1231         return 0;
1232 }
1233
1234 static int test_gfs2_super(struct super_block *s, void *ptr)
1235 {
1236         struct block_device *bdev = ptr;
1237         return (bdev == s->s_bdev);
1238 }
1239
1240 /**
1241  * gfs2_mount - Get the GFS2 superblock
1242  * @fs_type: The GFS2 filesystem type
1243  * @flags: Mount flags
1244  * @dev_name: The name of the device
1245  * @data: The mount arguments
1246  *
1247  * Q. Why not use get_sb_bdev() ?
1248  * A. We need to select one of two root directories to mount, independent
1249  *    of whether this is the initial, or subsequent, mount of this sb
1250  *
1251  * Returns: 0 or -ve on error
1252  */
1253
1254 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1255                        const char *dev_name, void *data)
1256 {
1257         struct block_device *bdev;
1258         struct super_block *s;
1259         fmode_t mode = FMODE_READ | FMODE_EXCL;
1260         int error;
1261         struct gfs2_args args;
1262         struct gfs2_sbd *sdp;
1263
1264         if (!(flags & MS_RDONLY))
1265                 mode |= FMODE_WRITE;
1266
1267         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1268         if (IS_ERR(bdev))
1269                 return ERR_CAST(bdev);
1270
1271         /*
1272          * once the super is inserted into the list by sget, s_umount
1273          * will protect the lockfs code from trying to start a snapshot
1274          * while we are mounting
1275          */
1276         mutex_lock(&bdev->bd_fsfreeze_mutex);
1277         if (bdev->bd_fsfreeze_count > 0) {
1278                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1279                 error = -EBUSY;
1280                 goto error_bdev;
1281         }
1282         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1283         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1284         error = PTR_ERR(s);
1285         if (IS_ERR(s))
1286                 goto error_bdev;
1287
1288         if (s->s_root)
1289                 blkdev_put(bdev, mode);
1290
1291         memset(&args, 0, sizeof(args));
1292         args.ar_quota = GFS2_QUOTA_DEFAULT;
1293         args.ar_data = GFS2_DATA_DEFAULT;
1294         args.ar_commit = 30;
1295         args.ar_statfs_quantum = 30;
1296         args.ar_quota_quantum = 60;
1297         args.ar_errors = GFS2_ERRORS_DEFAULT;
1298
1299         error = gfs2_mount_args(&args, data);
1300         if (error) {
1301                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1302                 goto error_super;
1303         }
1304
1305         if (s->s_root) {
1306                 error = -EBUSY;
1307                 if ((flags ^ s->s_flags) & MS_RDONLY)
1308                         goto error_super;
1309         } else {
1310                 char b[BDEVNAME_SIZE];
1311
1312                 s->s_flags = flags;
1313                 s->s_mode = mode;
1314                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1315                 sb_set_blocksize(s, block_size(bdev));
1316                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1317                 if (error)
1318                         goto error_super;
1319                 s->s_flags |= MS_ACTIVE;
1320                 bdev->bd_super = s;
1321         }
1322
1323         sdp = s->s_fs_info;
1324         if (args.ar_meta)
1325                 return dget(sdp->sd_master_dir);
1326         else
1327                 return dget(sdp->sd_root_dir);
1328
1329 error_super:
1330         deactivate_locked_super(s);
1331         return ERR_PTR(error);
1332 error_bdev:
1333         blkdev_put(bdev, mode);
1334         return ERR_PTR(error);
1335 }
1336
1337 static int set_meta_super(struct super_block *s, void *ptr)
1338 {
1339         return -EINVAL;
1340 }
1341
1342 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1343                         int flags, const char *dev_name, void *data)
1344 {
1345         struct super_block *s;
1346         struct gfs2_sbd *sdp;
1347         struct path path;
1348         int error;
1349
1350         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1351         if (error) {
1352                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1353                        dev_name, error);
1354                 return ERR_PTR(error);
1355         }
1356         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1357                  path.dentry->d_inode->i_sb->s_bdev);
1358         path_put(&path);
1359         if (IS_ERR(s)) {
1360                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1361                 return ERR_CAST(s);
1362         }
1363         if ((flags ^ s->s_flags) & MS_RDONLY) {
1364                 deactivate_locked_super(s);
1365                 return ERR_PTR(-EBUSY);
1366         }
1367         sdp = s->s_fs_info;
1368         return dget(sdp->sd_master_dir);
1369 }
1370
1371 static void gfs2_kill_sb(struct super_block *sb)
1372 {
1373         struct gfs2_sbd *sdp = sb->s_fs_info;
1374
1375         if (sdp == NULL) {
1376                 kill_block_super(sb);
1377                 return;
1378         }
1379
1380         gfs2_meta_syncfs(sdp);
1381         dput(sdp->sd_root_dir);
1382         dput(sdp->sd_master_dir);
1383         sdp->sd_root_dir = NULL;
1384         sdp->sd_master_dir = NULL;
1385         shrink_dcache_sb(sb);
1386         kill_block_super(sb);
1387         gfs2_delete_debugfs_file(sdp);
1388         kfree(sdp);
1389 }
1390
1391 struct file_system_type gfs2_fs_type = {
1392         .name = "gfs2",
1393         .fs_flags = FS_REQUIRES_DEV,
1394         .mount = gfs2_mount,
1395         .kill_sb = gfs2_kill_sb,
1396         .owner = THIS_MODULE,
1397 };
1398
1399 struct file_system_type gfs2meta_fs_type = {
1400         .name = "gfs2meta",
1401         .fs_flags = FS_REQUIRES_DEV,
1402         .mount = gfs2_mount_meta,
1403         .owner = THIS_MODULE,
1404 };
1405