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