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