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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/completion.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/kthread.h>
  19 #include <linux/export.h>
  20 #include <linux/namei.h>
  21 #include <linux/mount.h>
  22 #include <linux/gfs2_ondisk.h>
  23 #include <linux/quotaops.h>
  24 #include <linux/lockdep.h>
  25 #include <linux/module.h>
  26 #include <linux/backing-dev.h>
  27
  28 #include "gfs2.h"
  29 #include "incore.h"
  30 #include "bmap.h"
  31 #include "glock.h"
  32 #include "glops.h"
  33 #include "inode.h"
  34 #include "recovery.h"
  35 #include "rgrp.h"
  36 #include "super.h"
  37 #include "sys.h"
  38 #include "util.h"
  39 #include "log.h"
  40 #include "quota.h"
  41 #include "dir.h"
  42 #include "meta_io.h"
  43 #include "trace_gfs2.h"
  44
  45 #define DO 0
  46 #define UNDO 1
  47
  48 /**
  49  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  50  * @gt: tune
  51  *
  52  */
  53
  54 static void gfs2_tune_init(struct gfs2_tune *gt)
  55 {
  56         spin_lock_init(&gt->gt_spin);
  57
  58         gt->gt_quota_warn_period = 10;
  59         gt->gt_quota_scale_num = 1;
  60         gt->gt_quota_scale_den = 1;
  61         gt->gt_new_files_jdata = 0;
  62         gt->gt_max_readahead = BIT(18);
  63         gt->gt_complain_secs = 10;
  64 }
  65
  66 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  67 {
  68         struct gfs2_sbd *sdp;
  69         struct address_space *mapping;
  70
  71         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  72         if (!sdp)
  73                 return NULL;
  74
  75         sdp->sd_vfs = sb;
  76         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  77         if (!sdp->sd_lkstats) {
  78                 kfree(sdp);
  79                 return NULL;
  80         }
  81         sb->s_fs_info = sdp;
  82
  83         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  84         gfs2_tune_init(&sdp->sd_tune);
  85
  86         init_waitqueue_head(&sdp->sd_glock_wait);
  87         atomic_set(&sdp->sd_glock_disposal, 0);
  88         init_completion(&sdp->sd_locking_init);
  89         init_completion(&sdp->sd_wdack);
  90         spin_lock_init(&sdp->sd_statfs_spin);
  91
  92         spin_lock_init(&sdp->sd_rindex_spin);
  93         sdp->sd_rindex_tree.rb_node = NULL;
  94
  95         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  96         spin_lock_init(&sdp->sd_jindex_spin);
  97         mutex_init(&sdp->sd_jindex_mutex);
  98         init_completion(&sdp->sd_journal_ready);
  99
 100         INIT_LIST_HEAD(&sdp->sd_quota_list);
 101         mutex_init(&sdp->sd_quota_mutex);
 102         mutex_init(&sdp->sd_quota_sync_mutex);
 103         init_waitqueue_head(&sdp->sd_quota_wait);
 104         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 105         spin_lock_init(&sdp->sd_trunc_lock);
 106         spin_lock_init(&sdp->sd_bitmap_lock);
 107
 108         mapping = &sdp->sd_aspace;
 109
 110         address_space_init_once(mapping);
 111         mapping->a_ops = &gfs2_rgrp_aops;
 112         mapping->host = sb->s_bdev->bd_inode;
 113         mapping->flags = 0;
 114         mapping_set_gfp_mask(mapping, GFP_NOFS);
 115         mapping->private_data = NULL;
 116         mapping->writeback_index = 0;
 117
 118         spin_lock_init(&sdp->sd_log_lock);
 119         atomic_set(&sdp->sd_log_pinned, 0);
 120         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
 121         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 122         spin_lock_init(&sdp->sd_ordered_lock);
 123
 124         init_waitqueue_head(&sdp->sd_log_waitq);
 125         init_waitqueue_head(&sdp->sd_logd_waitq);
 126         spin_lock_init(&sdp->sd_ail_lock);
 127         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 128         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 129
 130         init_rwsem(&sdp->sd_log_flush_lock);
 131         atomic_set(&sdp->sd_log_in_flight, 0);
 132         atomic_set(&sdp->sd_reserving_log, 0);
 133         init_waitqueue_head(&sdp->sd_reserving_log_wait);
 134         init_waitqueue_head(&sdp->sd_log_flush_wait);
 135         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 136         mutex_init(&sdp->sd_freeze_mutex);
 137
 138         return sdp;
 139 }
 140
 141
 142 /**
 143  * gfs2_check_sb - Check superblock
 144  * @sdp: the filesystem
 145  * @sb: The superblock
 146  * @silent: Don't print a message if the check fails
 147  *
 148  * Checks the version code of the FS is one that we understand how to
 149  * read and that the sizes of the various on-disk structures have not
 150  * changed.
 151  */
 152
 153 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 154 {
 155         struct gfs2_sb_host *sb = &sdp->sd_sb;
 156
 157         if (sb->sb_magic != GFS2_MAGIC ||
 158             sb->sb_type != GFS2_METATYPE_SB) {
 159                 if (!silent)
 160                         pr_warn("not a GFS2 filesystem\n");
 161                 return -EINVAL;
 162         }
 163
 164         if (sb->sb_fs_format != GFS2_FORMAT_FS ||
 165             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 166                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 167                 return -EINVAL;
 168         }
 169
 170         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
 171             (sb->sb_bsize & (sb->sb_bsize - 1))) {
 172                 pr_warn("Invalid superblock size\n");
 173                 return -EINVAL;
 174         }
 175
 176         return 0;
 177 }
 178
 179 static void end_bio_io_page(struct bio *bio)
 180 {
 181         struct page *page = bio->bi_private;
 182
 183         if (!bio->bi_status)
 184                 SetPageUptodate(page);
 185         else
 186                 pr_warn("error %d reading superblock\n", bio->bi_status);
 187         unlock_page(page);
 188 }
 189
 190 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 191 {
 192         struct gfs2_sb_host *sb = &sdp->sd_sb;
 193         struct super_block *s = sdp->sd_vfs;
 194         const struct gfs2_sb *str = buf;
 195
 196         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 197         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 198         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 199         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 200         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 201         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 202         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 203         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 204         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 205         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 206         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 207
 208         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 209         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 210         memcpy(&s->s_uuid, str->sb_uuid, 16);
 211 }
 212
 213 /**
 214  * gfs2_read_super - Read the gfs2 super block from disk
 215  * @sdp: The GFS2 super block
 216  * @sector: The location of the super block
 217  * @error: The error code to return
 218  *
 219  * This uses the bio functions to read the super block from disk
 220  * because we want to be 100% sure that we never read cached data.
 221  * A super block is read twice only during each GFS2 mount and is
 222  * never written to by the filesystem. The first time its read no
 223  * locks are held, and the only details which are looked at are those
 224  * relating to the locking protocol. Once locking is up and working,
 225  * the sb is read again under the lock to establish the location of
 226  * the master directory (contains pointers to journals etc) and the
 227  * root directory.
 228  *
 229  * Returns: 0 on success or error
 230  */
 231
 232 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 233 {
 234         struct super_block *sb = sdp->sd_vfs;
 235         struct gfs2_sb *p;
 236         struct page *page;
 237         struct bio *bio;
 238
 239         page = alloc_page(GFP_NOFS);
 240         if (unlikely(!page))
 241                 return -ENOMEM;
 242
 243         ClearPageUptodate(page);
 244         ClearPageDirty(page);
 245         lock_page(page);
 246
 247         bio = bio_alloc(GFP_NOFS, 1);
 248         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 249         bio_set_dev(bio, sb->s_bdev);
 250         bio_add_page(bio, page, PAGE_SIZE, 0);
 251
 252         bio->bi_end_io = end_bio_io_page;
 253         bio->bi_private = page;
 254         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
 255         submit_bio(bio);
 256         wait_on_page_locked(page);
 257         bio_put(bio);
 258         if (!PageUptodate(page)) {
 259                 __free_page(page);
 260                 return -EIO;
 261         }
 262         p = kmap(page);
 263         gfs2_sb_in(sdp, p);
 264         kunmap(page);
 265         __free_page(page);
 266         return gfs2_check_sb(sdp, silent);
 267 }
 268
 269 /**
 270  * gfs2_read_sb - Read super block
 271  * @sdp: The GFS2 superblock
 272  * @silent: Don't print message if mount fails
 273  *
 274  */
 275
 276 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 277 {
 278         u32 hash_blocks, ind_blocks, leaf_blocks;
 279         u32 tmp_blocks;
 280         unsigned int x;
 281         int error;
 282
 283         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 284         if (error) {
 285                 if (!silent)
 286                         fs_err(sdp, "can't read superblock\n");
 287                 return error;
 288         }
 289
 290         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 291                                GFS2_BASIC_BLOCK_SHIFT;
 292         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 293         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 294                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 295         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 296                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 297         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 298         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 299         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 300         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 301         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 302                                 sizeof(struct gfs2_meta_header)) /
 303                                 sizeof(struct gfs2_quota_change);
 304         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 305                                      sizeof(struct gfs2_meta_header))
 306                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 307
 308         /* Compute maximum reservation required to add a entry to a directory */
 309
 310         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 311                              sdp->sd_jbsize);
 312
 313         ind_blocks = 0;
 314         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 315                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 316                 ind_blocks += tmp_blocks;
 317         }
 318
 319         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 320
 321         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 322
 323         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 324                                 sizeof(struct gfs2_dinode);
 325         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 326         for (x = 2;; x++) {
 327                 u64 space, d;
 328                 u32 m;
 329
 330                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 331                 d = space;
 332                 m = do_div(d, sdp->sd_inptrs);
 333
 334                 if (d != sdp->sd_heightsize[x - 1] || m)
 335                         break;
 336                 sdp->sd_heightsize[x] = space;
 337         }
 338         sdp->sd_max_height = x;
 339         sdp->sd_heightsize[x] = ~0;
 340         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 341
 342         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 343                                  sizeof(struct gfs2_dinode);
 344         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 345         for (x = 2;; x++) {
 346                 u64 space, d;
 347                 u32 m;
 348
 349                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 350                 d = space;
 351                 m = do_div(d, sdp->sd_inptrs);
 352
 353                 if (d != sdp->sd_jheightsize[x - 1] || m)
 354                         break;
 355                 sdp->sd_jheightsize[x] = space;
 356         }
 357         sdp->sd_max_jheight = x;
 358         sdp->sd_jheightsize[x] = ~0;
 359         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 360
 361         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 362                                       sizeof(struct gfs2_leaf)) /
 363                                      GFS2_MIN_DIRENT_SIZE;
 364         return 0;
 365 }
 366
 367 static int init_names(struct gfs2_sbd *sdp, int silent)
 368 {
 369         char *proto, *table;
 370         int error = 0;
 371
 372         proto = sdp->sd_args.ar_lockproto;
 373         table = sdp->sd_args.ar_locktable;
 374
 375         /*  Try to autodetect  */
 376
 377         if (!proto[0] || !table[0]) {
 378                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 379                 if (error)
 380                         return error;
 381
 382                 if (!proto[0])
 383                         proto = sdp->sd_sb.sb_lockproto;
 384                 if (!table[0])
 385                         table = sdp->sd_sb.sb_locktable;
 386         }
 387
 388         if (!table[0])
 389                 table = sdp->sd_vfs->s_id;
 390
 391         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 392         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 393
 394         table = sdp->sd_table_name;
 395         while ((table = strchr(table, '/')))
 396                 *table = '_';
 397
 398         return error;
 399 }
 400
 401 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 402                         int undo)
 403 {
 404         int error = 0;
 405
 406         if (undo)
 407                 goto fail_trans;
 408
 409         error = gfs2_glock_nq_num(sdp,
 410                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 411                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 412                                   mount_gh);
 413         if (error) {
 414                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 415                 goto fail;
 416         }
 417
 418         error = gfs2_glock_nq_num(sdp,
 419                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 420                                   LM_ST_SHARED,
 421                                   LM_FLAG_NOEXP | GL_EXACT,
 422                                   &sdp->sd_live_gh);
 423         if (error) {
 424                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 425                 goto fail_mount;
 426         }
 427
 428         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 429                                CREATE, &sdp->sd_rename_gl);
 430         if (error) {
 431                 fs_err(sdp, "can't create rename glock: %d\n", error);
 432                 goto fail_live;
 433         }
 434
 435         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 436                                CREATE, &sdp->sd_freeze_gl);
 437         if (error) {
 438                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 439                 goto fail_rename;
 440         }
 441
 442         return 0;
 443
 444 fail_trans:
 445         gfs2_glock_put(sdp->sd_freeze_gl);
 446 fail_rename:
 447         gfs2_glock_put(sdp->sd_rename_gl);
 448 fail_live:
 449         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 450 fail_mount:
 451         gfs2_glock_dq_uninit(mount_gh);
 452 fail:
 453         return error;
 454 }
 455
 456 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 457                             u64 no_addr, const char *name)
 458 {
 459         struct gfs2_sbd *sdp = sb->s_fs_info;
 460         struct dentry *dentry;
 461         struct inode *inode;
 462
 463         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 464                                   GFS2_BLKST_FREE /* ignore */);
 465         if (IS_ERR(inode)) {
 466                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 467                 return PTR_ERR(inode);
 468         }
 469         dentry = d_make_root(inode);
 470         if (!dentry) {
 471                 fs_err(sdp, "can't alloc %s dentry\n", name);
 472                 return -ENOMEM;
 473         }
 474         *dptr = dentry;
 475         return 0;
 476 }
 477
 478 static int init_sb(struct gfs2_sbd *sdp, int silent)
 479 {
 480         struct super_block *sb = sdp->sd_vfs;
 481         struct gfs2_holder sb_gh;
 482         u64 no_addr;
 483         int ret;
 484
 485         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 486                                 LM_ST_SHARED, 0, &sb_gh);
 487         if (ret) {
 488                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 489                 return ret;
 490         }
 491
 492         ret = gfs2_read_sb(sdp, silent);
 493         if (ret) {
 494                 fs_err(sdp, "can't read superblock: %d\n", ret);
 495                 goto out;
 496         }
 497
 498         /* Set up the buffer cache and SB for real */
 499         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 500                 ret = -EINVAL;
 501                 fs_err(sdp, "FS block size (%u) is too small for device "
 502                        "block size (%u)\n",
 503                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 504                 goto out;
 505         }
 506         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 507                 ret = -EINVAL;
 508                 fs_err(sdp, "FS block size (%u) is too big for machine "
 509                        "page size (%u)\n",
 510                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 511                 goto out;
 512         }
 513         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 514
 515         /* Get the root inode */
 516         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 517         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 518         if (ret)
 519                 goto out;
 520
 521         /* Get the master inode */
 522         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 523         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 524         if (ret) {
 525                 dput(sdp->sd_root_dir);
 526                 goto out;
 527         }
 528         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 529 out:
 530         gfs2_glock_dq_uninit(&sb_gh);
 531         return ret;
 532 }
 533
 534 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 535 {
 536         char *message = "FIRSTMOUNT=Done";
 537         char *envp[] = { message, NULL };
 538
 539         fs_info(sdp, "first mount done, others may mount\n");
 540
 541         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 542                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 543
 544         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 545 }
 546
 547 /**
 548  * gfs2_jindex_hold - Grab a lock on the jindex
 549  * @sdp: The GFS2 superblock
 550  * @ji_gh: the holder for the jindex glock
 551  *
 552  * Returns: errno
 553  */
 554
 555 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 556 {
 557         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 558         struct qstr name;
 559         char buf[20];
 560         struct gfs2_jdesc *jd;
 561         int error;
 562
 563         name.name = buf;
 564
 565         mutex_lock(&sdp->sd_jindex_mutex);
 566
 567         for (;;) {
 568                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 569                 if (error)
 570                         break;
 571
 572                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 573                 name.hash = gfs2_disk_hash(name.name, name.len);
 574
 575                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 576                 if (error == -ENOENT) {
 577                         error = 0;
 578                         break;
 579                 }
 580
 581                 gfs2_glock_dq_uninit(ji_gh);
 582
 583                 if (error)
 584                         break;
 585
 586                 error = -ENOMEM;
 587                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 588                 if (!jd)
 589                         break;
 590
 591                 INIT_LIST_HEAD(&jd->extent_list);
 592                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 593
 594                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 595                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 596                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 597                         if (!jd->jd_inode)
 598                                 error = -ENOENT;
 599                         else
 600                                 error = PTR_ERR(jd->jd_inode);
 601                         kfree(jd);
 602                         break;
 603                 }
 604
 605                 spin_lock(&sdp->sd_jindex_spin);
 606                 jd->jd_jid = sdp->sd_journals++;
 607                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 608                 spin_unlock(&sdp->sd_jindex_spin);
 609         }
 610
 611         mutex_unlock(&sdp->sd_jindex_mutex);
 612
 613         return error;
 614 }
 615
 616 /**
 617  * check_journal_clean - Make sure a journal is clean for a spectator mount
 618  * @sdp: The GFS2 superblock
 619  * @jd: The journal descriptor
 620  *
 621  * Returns: 0 if the journal is clean or locked, else an error
 622  */
 623 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 624 {
 625         int error;
 626         struct gfs2_holder j_gh;
 627         struct gfs2_log_header_host head;
 628         struct gfs2_inode *ip;
 629
 630         ip = GFS2_I(jd->jd_inode);
 631         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 632                                    GL_EXACT | GL_NOCACHE, &j_gh);
 633         if (error) {
 634                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 635                 return -EPERM;
 636         }
 637         error = gfs2_jdesc_check(jd);
 638         if (error) {
 639                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 640                 goto out_unlock;
 641         }
 642         error = gfs2_find_jhead(jd, &head);
 643         if (error) {
 644                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 645                 goto out_unlock;
 646         }
 647         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 648                 error = -EPERM;
 649                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 650                        "must not be a spectator.\n", jd->jd_jid);
 651         }
 652
 653 out_unlock:
 654         gfs2_glock_dq_uninit(&j_gh);
 655         return error;
 656 }
 657
 658 static int init_journal(struct gfs2_sbd *sdp, int undo)
 659 {
 660         struct inode *master = d_inode(sdp->sd_master_dir);
 661         struct gfs2_holder ji_gh;
 662         struct gfs2_inode *ip;
 663         int jindex = 1;
 664         int error = 0;
 665
 666         if (undo) {
 667                 jindex = 0;
 668                 goto fail_jinode_gh;
 669         }
 670
 671         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 672         if (IS_ERR(sdp->sd_jindex)) {
 673                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 674                 return PTR_ERR(sdp->sd_jindex);
 675         }
 676
 677         /* Load in the journal index special file */
 678
 679         error = gfs2_jindex_hold(sdp, &ji_gh);
 680         if (error) {
 681                 fs_err(sdp, "can't read journal index: %d\n", error);
 682                 goto fail;
 683         }
 684
 685         error = -EUSERS;
 686         if (!gfs2_jindex_size(sdp)) {
 687                 fs_err(sdp, "no journals!\n");
 688                 goto fail_jindex;
 689         }
 690
 691         atomic_set(&sdp->sd_log_blks_needed, 0);
 692         if (sdp->sd_args.ar_spectator) {
 693                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 694                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 695                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 696                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 697         } else {
 698                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 699                         fs_err(sdp, "can't mount journal #%u\n",
 700                                sdp->sd_lockstruct.ls_jid);
 701                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 702                                gfs2_jindex_size(sdp),
 703                                gfs2_jindex_size(sdp) - 1);
 704                         goto fail_jindex;
 705                 }
 706                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 707
 708                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 709                                           &gfs2_journal_glops,
 710                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 711                                           &sdp->sd_journal_gh);
 712                 if (error) {
 713                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 714                         goto fail_jindex;
 715                 }
 716
 717                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 718                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 719                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 720                                            &sdp->sd_jinode_gh);
 721                 if (error) {
 722                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 723                                error);
 724                         goto fail_journal_gh;
 725                 }
 726
 727                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 728                 if (error) {
 729                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 730                                sdp->sd_jdesc->jd_jid, error);
 731                         goto fail_jinode_gh;
 732                 }
 733                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 734                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 735                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 736
 737                 /* Map the extents for this journal's blocks */
 738                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 739         }
 740         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 741
 742         if (sdp->sd_lockstruct.ls_first) {
 743                 unsigned int x;
 744                 for (x = 0; x < sdp->sd_journals; x++) {
 745                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 746
 747                         if (sdp->sd_args.ar_spectator) {
 748                                 error = check_journal_clean(sdp, jd);
 749                                 if (error)
 750                                         goto fail_jinode_gh;
 751                                 continue;
 752                         }
 753                         error = gfs2_recover_journal(jd, true);
 754                         if (error) {
 755                                 fs_err(sdp, "error recovering journal %u: %d\n",
 756                                        x, error);
 757                                 goto fail_jinode_gh;
 758                         }
 759                 }
 760
 761                 gfs2_others_may_mount(sdp);
 762         } else if (!sdp->sd_args.ar_spectator) {
 763                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 764                 if (error) {
 765                         fs_err(sdp, "error recovering my journal: %d\n", error);
 766                         goto fail_jinode_gh;
 767                 }
 768         }
 769
 770         sdp->sd_log_idle = 1;
 771         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 772         gfs2_glock_dq_uninit(&ji_gh);
 773         jindex = 0;
 774         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 775         return 0;
 776
 777 fail_jinode_gh:
 778         if (!sdp->sd_args.ar_spectator)
 779                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 780 fail_journal_gh:
 781         if (!sdp->sd_args.ar_spectator)
 782                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 783 fail_jindex:
 784         gfs2_jindex_free(sdp);
 785         if (jindex)
 786                 gfs2_glock_dq_uninit(&ji_gh);
 787 fail:
 788         iput(sdp->sd_jindex);
 789         return error;
 790 }
 791
 792 static struct lock_class_key gfs2_quota_imutex_key;
 793
 794 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 795 {
 796         int error = 0;
 797         struct inode *master = d_inode(sdp->sd_master_dir);
 798
 799         if (undo)
 800                 goto fail_qinode;
 801
 802         error = init_journal(sdp, undo);
 803         complete_all(&sdp->sd_journal_ready);
 804         if (error)
 805                 goto fail;
 806
 807         /* Read in the master statfs inode */
 808         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 809         if (IS_ERR(sdp->sd_statfs_inode)) {
 810                 error = PTR_ERR(sdp->sd_statfs_inode);
 811                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 812                 goto fail_journal;
 813         }
 814
 815         /* Read in the resource index inode */
 816         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 817         if (IS_ERR(sdp->sd_rindex)) {
 818                 error = PTR_ERR(sdp->sd_rindex);
 819                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 820                 goto fail_statfs;
 821         }
 822         sdp->sd_rindex_uptodate = 0;
 823
 824         /* Read in the quota inode */
 825         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 826         if (IS_ERR(sdp->sd_quota_inode)) {
 827                 error = PTR_ERR(sdp->sd_quota_inode);
 828                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 829                 goto fail_rindex;
 830         }
 831         /*
 832          * i_mutex on quota files is special. Since this inode is hidden system
 833          * file, we are safe to define locking ourselves.
 834          */
 835         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 836                           &gfs2_quota_imutex_key);
 837
 838         error = gfs2_rindex_update(sdp);
 839         if (error)
 840                 goto fail_qinode;
 841
 842         return 0;
 843
 844 fail_qinode:
 845         iput(sdp->sd_quota_inode);
 846 fail_rindex:
 847         gfs2_clear_rgrpd(sdp);
 848         iput(sdp->sd_rindex);
 849 fail_statfs:
 850         iput(sdp->sd_statfs_inode);
 851 fail_journal:
 852         init_journal(sdp, UNDO);
 853 fail:
 854         return error;
 855 }
 856
 857 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 858 {
 859         struct inode *pn = NULL;
 860         char buf[30];
 861         int error = 0;
 862         struct gfs2_inode *ip;
 863         struct inode *master = d_inode(sdp->sd_master_dir);
 864
 865         if (sdp->sd_args.ar_spectator)
 866                 return 0;
 867
 868         if (undo)
 869                 goto fail_qc_gh;
 870
 871         pn = gfs2_lookup_simple(master, "per_node");
 872         if (IS_ERR(pn)) {
 873                 error = PTR_ERR(pn);
 874                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 875                 return error;
 876         }
 877
 878         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 879         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 880         if (IS_ERR(sdp->sd_sc_inode)) {
 881                 error = PTR_ERR(sdp->sd_sc_inode);
 882                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 883                 goto fail;
 884         }
 885
 886         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 887         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 888         if (IS_ERR(sdp->sd_qc_inode)) {
 889                 error = PTR_ERR(sdp->sd_qc_inode);
 890                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 891                 goto fail_ut_i;
 892         }
 893
 894         iput(pn);
 895         pn = NULL;
 896
 897         ip = GFS2_I(sdp->sd_sc_inode);
 898         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 899                                    &sdp->sd_sc_gh);
 900         if (error) {
 901                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 902                 goto fail_qc_i;
 903         }
 904
 905         ip = GFS2_I(sdp->sd_qc_inode);
 906         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 907                                    &sdp->sd_qc_gh);
 908         if (error) {
 909                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 910                 goto fail_ut_gh;
 911         }
 912
 913         return 0;
 914
 915 fail_qc_gh:
 916         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 917 fail_ut_gh:
 918         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 919 fail_qc_i:
 920         iput(sdp->sd_qc_inode);
 921 fail_ut_i:
 922         iput(sdp->sd_sc_inode);
 923 fail:
 924         iput(pn);
 925         return error;
 926 }
 927
 928 static const match_table_t nolock_tokens = {
 929         { Opt_jid, "jid=%d", },
 930         { Opt_err, NULL },
 931 };
 932
 933 static const struct lm_lockops nolock_ops = {
 934         .lm_proto_name = "lock_nolock",
 935         .lm_put_lock = gfs2_glock_free,
 936         .lm_tokens = &nolock_tokens,
 937 };
 938
 939 /**
 940  * gfs2_lm_mount - mount a locking protocol
 941  * @sdp: the filesystem
 942  * @args: mount arguments
 943  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 944  *
 945  * Returns: errno
 946  */
 947
 948 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 949 {
 950         const struct lm_lockops *lm;
 951         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 952         struct gfs2_args *args = &sdp->sd_args;
 953         const char *proto = sdp->sd_proto_name;
 954         const char *table = sdp->sd_table_name;
 955         char *o, *options;
 956         int ret;
 957
 958         if (!strcmp("lock_nolock", proto)) {
 959                 lm = &nolock_ops;
 960                 sdp->sd_args.ar_localflocks = 1;
 961 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 962         } else if (!strcmp("lock_dlm", proto)) {
 963                 lm = &gfs2_dlm_ops;
 964 #endif
 965         } else {
 966                 pr_info("can't find protocol %s\n", proto);
 967                 return -ENOENT;
 968         }
 969
 970         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 971
 972         ls->ls_ops = lm;
 973         ls->ls_first = 1;
 974
 975         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 976                 substring_t tmp[MAX_OPT_ARGS];
 977                 int token, option;
 978
 979                 if (!o || !*o)
 980                         continue;
 981
 982                 token = match_token(o, *lm->lm_tokens, tmp);
 983                 switch (token) {
 984                 case Opt_jid:
 985                         ret = match_int(&tmp[0], &option);
 986                         if (ret || option < 0)
 987                                 goto hostdata_error;
 988                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 989                                 ls->ls_jid = option;
 990                         break;
 991                 case Opt_id:
 992                 case Opt_nodir:
 993                         /* Obsolete, but left for backward compat purposes */
 994                         break;
 995                 case Opt_first:
 996                         ret = match_int(&tmp[0], &option);
 997                         if (ret || (option != 0 && option != 1))
 998                                 goto hostdata_error;
 999                         ls->ls_first = option;
1000                         break;
1001                 case Opt_err:
1002                 default:
1003 hostdata_error:
1004                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1005                         return -EINVAL;
1006                 }
1007         }
1008
1009         if (lm->lm_mount == NULL) {
1010                 fs_info(sdp, "Now mounting FS...\n");
1011                 complete_all(&sdp->sd_locking_init);
1012                 return 0;
1013         }
1014         ret = lm->lm_mount(sdp, table);
1015         if (ret == 0)
1016                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1017         complete_all(&sdp->sd_locking_init);
1018         return ret;
1019 }
1020
1021 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1022 {
1023         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1024         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1025             lm->lm_unmount)
1026                 lm->lm_unmount(sdp);
1027 }
1028
1029 static int wait_on_journal(struct gfs2_sbd *sdp)
1030 {
1031         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1032                 return 0;
1033
1034         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1035                 ? -EINTR : 0;
1036 }
1037
1038 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1039 {
1040         struct super_block *sb = sdp->sd_vfs;
1041         char ro[20];
1042         char spectator[20];
1043         char *envp[] = { ro, spectator, NULL };
1044         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1045         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1046         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1047 }
1048
1049 /**
1050  * fill_super - Read in superblock
1051  * @sb: The VFS superblock
1052  * @data: Mount options
1053  * @silent: Don't complain if it's not a GFS2 filesystem
1054  *
1055  * Returns: errno
1056  */
1057
1058 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1059 {
1060         struct gfs2_sbd *sdp;
1061         struct gfs2_holder mount_gh;
1062         int error;
1063
1064         sdp = init_sbd(sb);
1065         if (!sdp) {
1066                 pr_warn("can't alloc struct gfs2_sbd\n");
1067                 return -ENOMEM;
1068         }
1069         sdp->sd_args = *args;
1070
1071         if (sdp->sd_args.ar_spectator) {
1072                 sb->s_flags |= MS_RDONLY;
1073                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1074         }
1075         if (sdp->sd_args.ar_posix_acl)
1076                 sb->s_flags |= MS_POSIXACL;
1077         if (sdp->sd_args.ar_nobarrier)
1078                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1079
1080         sb->s_flags |= MS_NOSEC;
1081         sb->s_magic = GFS2_MAGIC;
1082         sb->s_op = &gfs2_super_ops;
1083         sb->s_d_op = &gfs2_dops;
1084         sb->s_export_op = &gfs2_export_ops;
1085         sb->s_xattr = gfs2_xattr_handlers;
1086         sb->s_qcop = &gfs2_quotactl_ops;
1087         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1088         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1089         sb->s_time_gran = 1;
1090         sb->s_maxbytes = MAX_LFS_FILESIZE;
1091
1092         /* Set up the buffer cache and fill in some fake block size values
1093            to allow us to read-in the on-disk superblock. */
1094         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1095         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1096         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1097                                GFS2_BASIC_BLOCK_SHIFT;
1098         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1099
1100         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1101         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1102         if (sdp->sd_args.ar_statfs_quantum) {
1103                 sdp->sd_tune.gt_statfs_slow = 0;
1104                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1105         } else {
1106                 sdp->sd_tune.gt_statfs_slow = 1;
1107                 sdp->sd_tune.gt_statfs_quantum = 30;
1108         }
1109
1110         error = init_names(sdp, silent);
1111         if (error) {
1112                 /* In this case, we haven't initialized sysfs, so we have to
1113                    manually free the sdp. */
1114                 free_percpu(sdp->sd_lkstats);
1115                 kfree(sdp);
1116                 sb->s_fs_info = NULL;
1117                 return error;
1118         }
1119
1120         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1121
1122         error = gfs2_sys_fs_add(sdp);
1123         /*
1124          * If we hit an error here, gfs2_sys_fs_add will have called function
1125          * kobject_put which causes the sysfs usage count to go to zero, which
1126          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1127          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1128          * kobject_put to free sdp.
1129          */
1130         if (error)
1131                 return error;
1132
1133         gfs2_create_debugfs_file(sdp);
1134
1135         error = gfs2_lm_mount(sdp, silent);
1136         if (error)
1137                 goto fail_debug;
1138
1139         error = init_locking(sdp, &mount_gh, DO);
1140         if (error)
1141                 goto fail_lm;
1142
1143         error = init_sb(sdp, silent);
1144         if (error)
1145                 goto fail_locking;
1146
1147         error = wait_on_journal(sdp);
1148         if (error)
1149                 goto fail_sb;
1150
1151         /*
1152          * If user space has failed to join the cluster or some similar
1153          * failure has occurred, then the journal id will contain a
1154          * negative (error) number. This will then be returned to the
1155          * caller (of the mount syscall). We do this even for spectator
1156          * mounts (which just write a jid of 0 to indicate "ok" even though
1157          * the jid is unused in the spectator case)
1158          */
1159         if (sdp->sd_lockstruct.ls_jid < 0) {
1160                 error = sdp->sd_lockstruct.ls_jid;
1161                 sdp->sd_lockstruct.ls_jid = 0;
1162                 goto fail_sb;
1163         }
1164
1165         if (sdp->sd_args.ar_spectator)
1166                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1167                          sdp->sd_table_name);
1168         else
1169                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1170                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1171
1172         error = init_inodes(sdp, DO);
1173         if (error)
1174                 goto fail_sb;
1175
1176         error = init_per_node(sdp, DO);
1177         if (error)
1178                 goto fail_inodes;
1179
1180         error = gfs2_statfs_init(sdp);
1181         if (error) {
1182                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1183                 goto fail_per_node;
1184         }
1185
1186         if (sb_rdonly(sb)) {
1187                 struct gfs2_holder freeze_gh;
1188
1189                 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1190                                            GL_EXACT, &freeze_gh);
1191                 if (error) {
1192                         fs_err(sdp, "can't make FS RO: %d\n", error);
1193                         goto fail_per_node;
1194                 }
1195                 gfs2_glock_dq_uninit(&freeze_gh);
1196         } else {
1197                 error = gfs2_make_fs_rw(sdp);
1198                 if (error) {
1199                         fs_err(sdp, "can't make FS RW: %d\n", error);
1200                         goto fail_per_node;
1201                 }
1202         }
1203
1204         gfs2_glock_dq_uninit(&mount_gh);
1205         gfs2_online_uevent(sdp);
1206         return 0;
1207
1208 fail_per_node:
1209         init_per_node(sdp, UNDO);
1210 fail_inodes:
1211         init_inodes(sdp, UNDO);
1212 fail_sb:
1213         if (sdp->sd_root_dir)
1214                 dput(sdp->sd_root_dir);
1215         if (sdp->sd_master_dir)
1216                 dput(sdp->sd_master_dir);
1217         if (sb->s_root)
1218                 dput(sb->s_root);
1219         sb->s_root = NULL;
1220 fail_locking:
1221         init_locking(sdp, &mount_gh, UNDO);
1222 fail_lm:
1223         complete_all(&sdp->sd_journal_ready);
1224         gfs2_gl_hash_clear(sdp);
1225         gfs2_lm_unmount(sdp);
1226 fail_debug:
1227         gfs2_delete_debugfs_file(sdp);
1228         free_percpu(sdp->sd_lkstats);
1229         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1230          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1231         gfs2_sys_fs_del(sdp);
1232         sb->s_fs_info = NULL;
1233         return error;
1234 }
1235
1236 static int set_gfs2_super(struct super_block *s, void *data)
1237 {
1238         s->s_bdev = data;
1239         s->s_dev = s->s_bdev->bd_dev;
1240         s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1241         return 0;
1242 }
1243
1244 static int test_gfs2_super(struct super_block *s, void *ptr)
1245 {
1246         struct block_device *bdev = ptr;
1247         return (bdev == s->s_bdev);
1248 }
1249
1250 /**
1251  * gfs2_mount - Get the GFS2 superblock
1252  * @fs_type: The GFS2 filesystem type
1253  * @flags: Mount flags
1254  * @dev_name: The name of the device
1255  * @data: The mount arguments
1256  *
1257  * Q. Why not use get_sb_bdev() ?
1258  * A. We need to select one of two root directories to mount, independent
1259  *    of whether this is the initial, or subsequent, mount of this sb
1260  *
1261  * Returns: 0 or -ve on error
1262  */
1263
1264 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1265                        const char *dev_name, void *data)
1266 {
1267         struct block_device *bdev;
1268         struct super_block *s;
1269         fmode_t mode = FMODE_READ | FMODE_EXCL;
1270         int error;
1271         struct gfs2_args args;
1272         struct gfs2_sbd *sdp;
1273
1274         if (!(flags & MS_RDONLY))
1275                 mode |= FMODE_WRITE;
1276
1277         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1278         if (IS_ERR(bdev))
1279                 return ERR_CAST(bdev);
1280
1281         /*
1282          * once the super is inserted into the list by sget, s_umount
1283          * will protect the lockfs code from trying to start a snapshot
1284          * while we are mounting
1285          */
1286         mutex_lock(&bdev->bd_fsfreeze_mutex);
1287         if (bdev->bd_fsfreeze_count > 0) {
1288                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1289                 error = -EBUSY;
1290                 goto error_bdev;
1291         }
1292         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1293         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1294         error = PTR_ERR(s);
1295         if (IS_ERR(s))
1296                 goto error_bdev;
1297
1298         if (s->s_root) {
1299                 /*
1300                  * s_umount nests inside bd_mutex during
1301                  * __invalidate_device().  blkdev_put() acquires
1302                  * bd_mutex and can't be called under s_umount.  Drop
1303                  * s_umount temporarily.  This is safe as we're
1304                  * holding an active reference.
1305                  */
1306                 up_write(&s->s_umount);
1307                 blkdev_put(bdev, mode);
1308                 down_write(&s->s_umount);
1309         } else {
1310                 /* s_mode must be set before deactivate_locked_super calls */
1311                 s->s_mode = mode;
1312         }
1313
1314         memset(&args, 0, sizeof(args));
1315         args.ar_quota = GFS2_QUOTA_DEFAULT;
1316         args.ar_data = GFS2_DATA_DEFAULT;
1317         args.ar_commit = 30;
1318         args.ar_statfs_quantum = 30;
1319         args.ar_quota_quantum = 60;
1320         args.ar_errors = GFS2_ERRORS_DEFAULT;
1321
1322         error = gfs2_mount_args(&args, data);
1323         if (error) {
1324                 pr_warn("can't parse mount arguments\n");
1325                 goto error_super;
1326         }
1327
1328         if (s->s_root) {
1329                 error = -EBUSY;
1330                 if ((flags ^ s->s_flags) & MS_RDONLY)
1331                         goto error_super;
1332         } else {
1333                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1334                 sb_set_blocksize(s, block_size(bdev));
1335                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1336                 if (error)
1337                         goto error_super;
1338                 s->s_flags |= MS_ACTIVE;
1339                 bdev->bd_super = s;
1340         }
1341
1342         sdp = s->s_fs_info;
1343         if (args.ar_meta)
1344                 return dget(sdp->sd_master_dir);
1345         else
1346                 return dget(sdp->sd_root_dir);
1347
1348 error_super:
1349         deactivate_locked_super(s);
1350         return ERR_PTR(error);
1351 error_bdev:
1352         blkdev_put(bdev, mode);
1353         return ERR_PTR(error);
1354 }
1355
1356 static int set_meta_super(struct super_block *s, void *ptr)
1357 {
1358         return -EINVAL;
1359 }
1360
1361 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1362                         int flags, const char *dev_name, void *data)
1363 {
1364         struct super_block *s;
1365         struct gfs2_sbd *sdp;
1366         struct path path;
1367         int error;
1368
1369         if (!dev_name || !*dev_name)
1370                 return ERR_PTR(-EINVAL);
1371
1372         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1373         if (error) {
1374                 pr_warn("path_lookup on %s returned error %d\n",
1375                         dev_name, error);
1376                 return ERR_PTR(error);
1377         }
1378         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1379                  path.dentry->d_sb->s_bdev);
1380         path_put(&path);
1381         if (IS_ERR(s)) {
1382                 pr_warn("gfs2 mount does not exist\n");
1383                 return ERR_CAST(s);
1384         }
1385         if ((flags ^ s->s_flags) & MS_RDONLY) {
1386                 deactivate_locked_super(s);
1387                 return ERR_PTR(-EBUSY);
1388         }
1389         sdp = s->s_fs_info;
1390         return dget(sdp->sd_master_dir);
1391 }
1392
1393 static void gfs2_kill_sb(struct super_block *sb)
1394 {
1395         struct gfs2_sbd *sdp = sb->s_fs_info;
1396
1397         if (sdp == NULL) {
1398                 kill_block_super(sb);
1399                 return;
1400         }
1401
1402         gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1403         dput(sdp->sd_root_dir);
1404         dput(sdp->sd_master_dir);
1405         sdp->sd_root_dir = NULL;
1406         sdp->sd_master_dir = NULL;
1407         shrink_dcache_sb(sb);
1408         free_percpu(sdp->sd_lkstats);
1409         kill_block_super(sb);
1410 }
1411
1412 struct file_system_type gfs2_fs_type = {
1413         .name = "gfs2",
1414         .fs_flags = FS_REQUIRES_DEV,
1415         .mount = gfs2_mount,
1416         .kill_sb = gfs2_kill_sb,
1417         .owner = THIS_MODULE,
1418 };
1419 MODULE_ALIAS_FS("gfs2");
1420
1421 struct file_system_type gfs2meta_fs_type = {
1422         .name = "gfs2meta",
1423         .fs_flags = FS_REQUIRES_DEV,
1424         .mount = gfs2_mount_meta,
1425         .owner = THIS_MODULE,
1426 };
1427 MODULE_ALIAS_FS("gfs2meta");