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