fs/gfs2/ops_fstype.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/sched.h>
  10 #include <linux/slab.h>
  11 #include <linux/spinlock.h>
  12 #include <linux/completion.h>
  13 #include <linux/buffer_head.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/kthread.h>
  16 #include <linux/export.h>
  17 #include <linux/namei.h>
  18 #include <linux/mount.h>
  19 #include <linux/gfs2_ondisk.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/lockdep.h>
  22 #include <linux/module.h>
  23 #include <linux/backing-dev.h>
  24
  25 #include "gfs2.h"
  26 #include "incore.h"
  27 #include "bmap.h"
  28 #include "glock.h"
  29 #include "glops.h"
  30 #include "inode.h"
  31 #include "recovery.h"
  32 #include "rgrp.h"
  33 #include "super.h"
  34 #include "sys.h"
  35 #include "util.h"
  36 #include "log.h"
  37 #include "quota.h"
  38 #include "dir.h"
  39 #include "meta_io.h"
  40 #include "trace_gfs2.h"
  41 #include "lops.h"
  42
  43 #define DO 0
  44 #define UNDO 1
  45
  46 /**
  47  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  48  * @gt: tune
  49  *
  50  */
  51
  52 static void gfs2_tune_init(struct gfs2_tune *gt)
  53 {
  54         spin_lock_init(&gt->gt_spin);
  55
  56         gt->gt_quota_warn_period = 10;
  57         gt->gt_quota_scale_num = 1;
  58         gt->gt_quota_scale_den = 1;
  59         gt->gt_new_files_jdata = 0;
  60         gt->gt_max_readahead = BIT(18);
  61         gt->gt_complain_secs = 10;
  62 }
  63
  64 void free_sbd(struct gfs2_sbd *sdp)
  65 {
  66         if (sdp->sd_lkstats)
  67                 free_percpu(sdp->sd_lkstats);
  68         kfree(sdp);
  69 }
  70
  71 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  72 {
  73         struct gfs2_sbd *sdp;
  74         struct address_space *mapping;
  75
  76         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  77         if (!sdp)
  78                 return NULL;
  79
  80         sdp->sd_vfs = sb;
  81         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  82         if (!sdp->sd_lkstats)
  83                 goto fail;
  84         sb->s_fs_info = sdp;
  85
  86         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  87         gfs2_tune_init(&sdp->sd_tune);
  88
  89         init_waitqueue_head(&sdp->sd_glock_wait);
  90         init_waitqueue_head(&sdp->sd_async_glock_wait);
  91         atomic_set(&sdp->sd_glock_disposal, 0);
  92         init_completion(&sdp->sd_locking_init);
  93         init_completion(&sdp->sd_wdack);
  94         spin_lock_init(&sdp->sd_statfs_spin);
  95
  96         spin_lock_init(&sdp->sd_rindex_spin);
  97         sdp->sd_rindex_tree.rb_node = NULL;
  98
  99         INIT_LIST_HEAD(&sdp->sd_jindex_list);
 100         spin_lock_init(&sdp->sd_jindex_spin);
 101         mutex_init(&sdp->sd_jindex_mutex);
 102         init_completion(&sdp->sd_journal_ready);
 103
 104         INIT_LIST_HEAD(&sdp->sd_quota_list);
 105         mutex_init(&sdp->sd_quota_mutex);
 106         mutex_init(&sdp->sd_quota_sync_mutex);
 107         init_waitqueue_head(&sdp->sd_quota_wait);
 108         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 109         spin_lock_init(&sdp->sd_trunc_lock);
 110         spin_lock_init(&sdp->sd_bitmap_lock);
 111
 112         mapping = &sdp->sd_aspace;
 113
 114         address_space_init_once(mapping);
 115         mapping->a_ops = &gfs2_rgrp_aops;
 116         mapping->host = sb->s_bdev->bd_inode;
 117         mapping->flags = 0;
 118         mapping_set_gfp_mask(mapping, GFP_NOFS);
 119         mapping->private_data = NULL;
 120         mapping->writeback_index = 0;
 121
 122         spin_lock_init(&sdp->sd_log_lock);
 123         atomic_set(&sdp->sd_log_pinned, 0);
 124         INIT_LIST_HEAD(&sdp->sd_log_revokes);
 125         INIT_LIST_HEAD(&sdp->sd_log_ordered);
 126         spin_lock_init(&sdp->sd_ordered_lock);
 127
 128         init_waitqueue_head(&sdp->sd_log_waitq);
 129         init_waitqueue_head(&sdp->sd_logd_waitq);
 130         spin_lock_init(&sdp->sd_ail_lock);
 131         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 132         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 133
 134         init_rwsem(&sdp->sd_log_flush_lock);
 135         atomic_set(&sdp->sd_log_in_flight, 0);
 136         atomic_set(&sdp->sd_reserving_log, 0);
 137         init_waitqueue_head(&sdp->sd_reserving_log_wait);
 138         init_waitqueue_head(&sdp->sd_log_flush_wait);
 139         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 140         mutex_init(&sdp->sd_freeze_mutex);
 141
 142         return sdp;
 143
 144 fail:
 145         free_sbd(sdp);
 146         return NULL;
 147 }
 148
 149 /**
 150  * gfs2_check_sb - Check superblock
 151  * @sdp: the filesystem
 152  * @sb: The superblock
 153  * @silent: Don't print a message if the check fails
 154  *
 155  * Checks the version code of the FS is one that we understand how to
 156  * read and that the sizes of the various on-disk structures have not
 157  * changed.
 158  */
 159
 160 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 161 {
 162         struct gfs2_sb_host *sb = &sdp->sd_sb;
 163
 164         if (sb->sb_magic != GFS2_MAGIC ||
 165             sb->sb_type != GFS2_METATYPE_SB) {
 166                 if (!silent)
 167                         pr_warn("not a GFS2 filesystem\n");
 168                 return -EINVAL;
 169         }
 170
 171         /*  If format numbers match exactly, we're done.  */
 172
 173         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 174             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 175                 return 0;
 176
 177         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 178
 179         return -EINVAL;
 180 }
 181
 182 static void end_bio_io_page(struct bio *bio)
 183 {
 184         struct page *page = bio->bi_private;
 185
 186         if (!bio->bi_status)
 187                 SetPageUptodate(page);
 188         else
 189                 pr_warn("error %d reading superblock\n", bio->bi_status);
 190         unlock_page(page);
 191 }
 192
 193 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 194 {
 195         struct gfs2_sb_host *sb = &sdp->sd_sb;
 196         struct super_block *s = sdp->sd_vfs;
 197         const struct gfs2_sb *str = buf;
 198
 199         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 200         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 201         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 202         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 203         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 204         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 205         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 206         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 207         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 208         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 209         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 210
 211         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 212         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 213         memcpy(&s->s_uuid, str->sb_uuid, 16);
 214 }
 215
 216 /**
 217  * gfs2_read_super - Read the gfs2 super block from disk
 218  * @sdp: The GFS2 super block
 219  * @sector: The location of the super block
 220  * @error: The error code to return
 221  *
 222  * This uses the bio functions to read the super block from disk
 223  * because we want to be 100% sure that we never read cached data.
 224  * A super block is read twice only during each GFS2 mount and is
 225  * never written to by the filesystem. The first time its read no
 226  * locks are held, and the only details which are looked at are those
 227  * relating to the locking protocol. Once locking is up and working,
 228  * the sb is read again under the lock to establish the location of
 229  * the master directory (contains pointers to journals etc) and the
 230  * root directory.
 231  *
 232  * Returns: 0 on success or error
 233  */
 234
 235 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 236 {
 237         struct super_block *sb = sdp->sd_vfs;
 238         struct gfs2_sb *p;
 239         struct page *page;
 240         struct bio *bio;
 241
 242         page = alloc_page(GFP_NOFS);
 243         if (unlikely(!page))
 244                 return -ENOMEM;
 245
 246         ClearPageUptodate(page);
 247         ClearPageDirty(page);
 248         lock_page(page);
 249
 250         bio = bio_alloc(GFP_NOFS, 1);
 251         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 252         bio_set_dev(bio, sb->s_bdev);
 253         bio_add_page(bio, page, PAGE_SIZE, 0);
 254
 255         bio->bi_end_io = end_bio_io_page;
 256         bio->bi_private = page;
 257         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
 258         submit_bio(bio);
 259         wait_on_page_locked(page);
 260         bio_put(bio);
 261         if (!PageUptodate(page)) {
 262                 __free_page(page);
 263                 return -EIO;
 264         }
 265         p = kmap(page);
 266         gfs2_sb_in(sdp, p);
 267         kunmap(page);
 268         __free_page(page);
 269         return gfs2_check_sb(sdp, silent);
 270 }
 271
 272 /**
 273  * gfs2_read_sb - Read super block
 274  * @sdp: The GFS2 superblock
 275  * @silent: Don't print message if mount fails
 276  *
 277  */
 278
 279 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 280 {
 281         u32 hash_blocks, ind_blocks, leaf_blocks;
 282         u32 tmp_blocks;
 283         unsigned int x;
 284         int error;
 285
 286         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 287         if (error) {
 288                 if (!silent)
 289                         fs_err(sdp, "can't read superblock\n");
 290                 return error;
 291         }
 292
 293         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 294                                GFS2_BASIC_BLOCK_SHIFT;
 295         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 296         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 297                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 298         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 299                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 300         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 301         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 302         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 303         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 304         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 305                                 sizeof(struct gfs2_meta_header)) /
 306                                 sizeof(struct gfs2_quota_change);
 307         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 308                                      sizeof(struct gfs2_meta_header))
 309                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 310
 311         /* Compute maximum reservation required to add a entry to a directory */
 312
 313         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 314                              sdp->sd_jbsize);
 315
 316         ind_blocks = 0;
 317         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 318                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 319                 ind_blocks += tmp_blocks;
 320         }
 321
 322         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 323
 324         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 325
 326         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 327                                 sizeof(struct gfs2_dinode);
 328         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 329         for (x = 2;; x++) {
 330                 u64 space, d;
 331                 u32 m;
 332
 333                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 334                 d = space;
 335                 m = do_div(d, sdp->sd_inptrs);
 336
 337                 if (d != sdp->sd_heightsize[x - 1] || m)
 338                         break;
 339                 sdp->sd_heightsize[x] = space;
 340         }
 341         sdp->sd_max_height = x;
 342         sdp->sd_heightsize[x] = ~0;
 343         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 344
 345         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 346                                       sizeof(struct gfs2_leaf)) /
 347                                      GFS2_MIN_DIRENT_SIZE;
 348         return 0;
 349 }
 350
 351 static int init_names(struct gfs2_sbd *sdp, int silent)
 352 {
 353         char *proto, *table;
 354         int error = 0;
 355
 356         proto = sdp->sd_args.ar_lockproto;
 357         table = sdp->sd_args.ar_locktable;
 358
 359         /*  Try to autodetect  */
 360
 361         if (!proto[0] || !table[0]) {
 362                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 363                 if (error)
 364                         return error;
 365
 366                 if (!proto[0])
 367                         proto = sdp->sd_sb.sb_lockproto;
 368                 if (!table[0])
 369                         table = sdp->sd_sb.sb_locktable;
 370         }
 371
 372         if (!table[0])
 373                 table = sdp->sd_vfs->s_id;
 374
 375         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 376         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 377
 378         table = sdp->sd_table_name;
 379         while ((table = strchr(table, '/')))
 380                 *table = '_';
 381
 382         return error;
 383 }
 384
 385 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 386                         int undo)
 387 {
 388         int error = 0;
 389
 390         if (undo)
 391                 goto fail_trans;
 392
 393         error = gfs2_glock_nq_num(sdp,
 394                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 395                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 396                                   mount_gh);
 397         if (error) {
 398                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 399                 goto fail;
 400         }
 401
 402         error = gfs2_glock_nq_num(sdp,
 403                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 404                                   LM_ST_SHARED,
 405                                   LM_FLAG_NOEXP | GL_EXACT,
 406                                   &sdp->sd_live_gh);
 407         if (error) {
 408                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 409                 goto fail_mount;
 410         }
 411
 412         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 413                                CREATE, &sdp->sd_rename_gl);
 414         if (error) {
 415                 fs_err(sdp, "can't create rename glock: %d\n", error);
 416                 goto fail_live;
 417         }
 418
 419         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 420                                CREATE, &sdp->sd_freeze_gl);
 421         if (error) {
 422                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 423                 goto fail_rename;
 424         }
 425
 426         return 0;
 427
 428 fail_trans:
 429         gfs2_glock_put(sdp->sd_freeze_gl);
 430 fail_rename:
 431         gfs2_glock_put(sdp->sd_rename_gl);
 432 fail_live:
 433         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 434 fail_mount:
 435         gfs2_glock_dq_uninit(mount_gh);
 436 fail:
 437         return error;
 438 }
 439
 440 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 441                             u64 no_addr, const char *name)
 442 {
 443         struct gfs2_sbd *sdp = sb->s_fs_info;
 444         struct dentry *dentry;
 445         struct inode *inode;
 446
 447         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 448                                   GFS2_BLKST_FREE /* ignore */);
 449         if (IS_ERR(inode)) {
 450                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 451                 return PTR_ERR(inode);
 452         }
 453         dentry = d_make_root(inode);
 454         if (!dentry) {
 455                 fs_err(sdp, "can't alloc %s dentry\n", name);
 456                 return -ENOMEM;
 457         }
 458         *dptr = dentry;
 459         return 0;
 460 }
 461
 462 static int init_sb(struct gfs2_sbd *sdp, int silent)
 463 {
 464         struct super_block *sb = sdp->sd_vfs;
 465         struct gfs2_holder sb_gh;
 466         u64 no_addr;
 467         int ret;
 468
 469         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 470                                 LM_ST_SHARED, 0, &sb_gh);
 471         if (ret) {
 472                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 473                 return ret;
 474         }
 475
 476         ret = gfs2_read_sb(sdp, silent);
 477         if (ret) {
 478                 fs_err(sdp, "can't read superblock: %d\n", ret);
 479                 goto out;
 480         }
 481
 482         /* Set up the buffer cache and SB for real */
 483         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 484                 ret = -EINVAL;
 485                 fs_err(sdp, "FS block size (%u) is too small for device "
 486                        "block size (%u)\n",
 487                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 488                 goto out;
 489         }
 490         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 491                 ret = -EINVAL;
 492                 fs_err(sdp, "FS block size (%u) is too big for machine "
 493                        "page size (%u)\n",
 494                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 495                 goto out;
 496         }
 497         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 498
 499         /* Get the root inode */
 500         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 501         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 502         if (ret)
 503                 goto out;
 504
 505         /* Get the master inode */
 506         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 507         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 508         if (ret) {
 509                 dput(sdp->sd_root_dir);
 510                 goto out;
 511         }
 512         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 513 out:
 514         gfs2_glock_dq_uninit(&sb_gh);
 515         return ret;
 516 }
 517
 518 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 519 {
 520         char *message = "FIRSTMOUNT=Done";
 521         char *envp[] = { message, NULL };
 522
 523         fs_info(sdp, "first mount done, others may mount\n");
 524
 525         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 526                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 527
 528         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 529 }
 530
 531 /**
 532  * gfs2_jindex_hold - Grab a lock on the jindex
 533  * @sdp: The GFS2 superblock
 534  * @ji_gh: the holder for the jindex glock
 535  *
 536  * Returns: errno
 537  */
 538
 539 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 540 {
 541         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 542         struct qstr name;
 543         char buf[20];
 544         struct gfs2_jdesc *jd;
 545         int error;
 546
 547         name.name = buf;
 548
 549         mutex_lock(&sdp->sd_jindex_mutex);
 550
 551         for (;;) {
 552                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 553                 if (error)
 554                         break;
 555
 556                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 557                 name.hash = gfs2_disk_hash(name.name, name.len);
 558
 559                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 560                 if (error == -ENOENT) {
 561                         error = 0;
 562                         break;
 563                 }
 564
 565                 gfs2_glock_dq_uninit(ji_gh);
 566
 567                 if (error)
 568                         break;
 569
 570                 error = -ENOMEM;
 571                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 572                 if (!jd)
 573                         break;
 574
 575                 INIT_LIST_HEAD(&jd->extent_list);
 576                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 577
 578                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 579                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 580                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 581                         if (!jd->jd_inode)
 582                                 error = -ENOENT;
 583                         else
 584                                 error = PTR_ERR(jd->jd_inode);
 585                         kfree(jd);
 586                         break;
 587                 }
 588
 589                 spin_lock(&sdp->sd_jindex_spin);
 590                 jd->jd_jid = sdp->sd_journals++;
 591                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 592                 spin_unlock(&sdp->sd_jindex_spin);
 593         }
 594
 595         mutex_unlock(&sdp->sd_jindex_mutex);
 596
 597         return error;
 598 }
 599
 600 /**
 601  * check_journal_clean - Make sure a journal is clean for a spectator mount
 602  * @sdp: The GFS2 superblock
 603  * @jd: The journal descriptor
 604  *
 605  * Returns: 0 if the journal is clean or locked, else an error
 606  */
 607 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 608 {
 609         int error;
 610         struct gfs2_holder j_gh;
 611         struct gfs2_log_header_host head;
 612         struct gfs2_inode *ip;
 613
 614         ip = GFS2_I(jd->jd_inode);
 615         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 616                                    GL_EXACT | GL_NOCACHE, &j_gh);
 617         if (error) {
 618                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 619                 return -EPERM;
 620         }
 621         error = gfs2_jdesc_check(jd);
 622         if (error) {
 623                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 624                 goto out_unlock;
 625         }
 626         error = gfs2_find_jhead(jd, &head, false);
 627         if (error) {
 628                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 629                 goto out_unlock;
 630         }
 631         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 632                 error = -EPERM;
 633                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 634                        "must not be a spectator.\n", jd->jd_jid);
 635         }
 636
 637 out_unlock:
 638         gfs2_glock_dq_uninit(&j_gh);
 639         return error;
 640 }
 641
 642 static int init_journal(struct gfs2_sbd *sdp, int undo)
 643 {
 644         struct inode *master = d_inode(sdp->sd_master_dir);
 645         struct gfs2_holder ji_gh;
 646         struct gfs2_inode *ip;
 647         int jindex = 1;
 648         int error = 0;
 649
 650         if (undo) {
 651                 jindex = 0;
 652                 goto fail_jinode_gh;
 653         }
 654
 655         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 656         if (IS_ERR(sdp->sd_jindex)) {
 657                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 658                 return PTR_ERR(sdp->sd_jindex);
 659         }
 660
 661         /* Load in the journal index special file */
 662
 663         error = gfs2_jindex_hold(sdp, &ji_gh);
 664         if (error) {
 665                 fs_err(sdp, "can't read journal index: %d\n", error);
 666                 goto fail;
 667         }
 668
 669         error = -EUSERS;
 670         if (!gfs2_jindex_size(sdp)) {
 671                 fs_err(sdp, "no journals!\n");
 672                 goto fail_jindex;
 673         }
 674
 675         atomic_set(&sdp->sd_log_blks_needed, 0);
 676         if (sdp->sd_args.ar_spectator) {
 677                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 678                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 679                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 680                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 681         } else {
 682                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 683                         fs_err(sdp, "can't mount journal #%u\n",
 684                                sdp->sd_lockstruct.ls_jid);
 685                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 686                                gfs2_jindex_size(sdp),
 687                                gfs2_jindex_size(sdp) - 1);
 688                         goto fail_jindex;
 689                 }
 690                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 691
 692                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 693                                           &gfs2_journal_glops,
 694                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 695                                           &sdp->sd_journal_gh);
 696                 if (error) {
 697                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 698                         goto fail_jindex;
 699                 }
 700
 701                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 702                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 703                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 704                                            &sdp->sd_jinode_gh);
 705                 if (error) {
 706                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 707                                error);
 708                         goto fail_journal_gh;
 709                 }
 710
 711                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 712                 if (error) {
 713                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 714                                sdp->sd_jdesc->jd_jid, error);
 715                         goto fail_jinode_gh;
 716                 }
 717                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 718                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 719                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 720
 721                 /* Map the extents for this journal's blocks */
 722                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 723         }
 724         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 725
 726         if (sdp->sd_lockstruct.ls_first) {
 727                 unsigned int x;
 728                 for (x = 0; x < sdp->sd_journals; x++) {
 729                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 730
 731                         if (sdp->sd_args.ar_spectator) {
 732                                 error = check_journal_clean(sdp, jd);
 733                                 if (error)
 734                                         goto fail_jinode_gh;
 735                                 continue;
 736                         }
 737                         error = gfs2_recover_journal(jd, true);
 738                         if (error) {
 739                                 fs_err(sdp, "error recovering journal %u: %d\n",
 740                                        x, error);
 741                                 goto fail_jinode_gh;
 742                         }
 743                 }
 744
 745                 gfs2_others_may_mount(sdp);
 746         } else if (!sdp->sd_args.ar_spectator) {
 747                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 748                 if (error) {
 749                         fs_err(sdp, "error recovering my journal: %d\n", error);
 750                         goto fail_jinode_gh;
 751                 }
 752         }
 753
 754         sdp->sd_log_idle = 1;
 755         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 756         gfs2_glock_dq_uninit(&ji_gh);
 757         jindex = 0;
 758         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 759         return 0;
 760
 761 fail_jinode_gh:
 762         if (!sdp->sd_args.ar_spectator)
 763                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 764 fail_journal_gh:
 765         if (!sdp->sd_args.ar_spectator)
 766                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 767 fail_jindex:
 768         gfs2_jindex_free(sdp);
 769         if (jindex)
 770                 gfs2_glock_dq_uninit(&ji_gh);
 771 fail:
 772         iput(sdp->sd_jindex);
 773         return error;
 774 }
 775
 776 static struct lock_class_key gfs2_quota_imutex_key;
 777
 778 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 779 {
 780         int error = 0;
 781         struct inode *master = d_inode(sdp->sd_master_dir);
 782
 783         if (undo)
 784                 goto fail_qinode;
 785
 786         error = init_journal(sdp, undo);
 787         complete_all(&sdp->sd_journal_ready);
 788         if (error)
 789                 goto fail;
 790
 791         /* Read in the master statfs inode */
 792         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 793         if (IS_ERR(sdp->sd_statfs_inode)) {
 794                 error = PTR_ERR(sdp->sd_statfs_inode);
 795                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 796                 goto fail_journal;
 797         }
 798
 799         /* Read in the resource index inode */
 800         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 801         if (IS_ERR(sdp->sd_rindex)) {
 802                 error = PTR_ERR(sdp->sd_rindex);
 803                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 804                 goto fail_statfs;
 805         }
 806         sdp->sd_rindex_uptodate = 0;
 807
 808         /* Read in the quota inode */
 809         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 810         if (IS_ERR(sdp->sd_quota_inode)) {
 811                 error = PTR_ERR(sdp->sd_quota_inode);
 812                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 813                 goto fail_rindex;
 814         }
 815         /*
 816          * i_rwsem on quota files is special. Since this inode is hidden system
 817          * file, we are safe to define locking ourselves.
 818          */
 819         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 820                           &gfs2_quota_imutex_key);
 821
 822         error = gfs2_rindex_update(sdp);
 823         if (error)
 824                 goto fail_qinode;
 825
 826         return 0;
 827
 828 fail_qinode:
 829         iput(sdp->sd_quota_inode);
 830 fail_rindex:
 831         gfs2_clear_rgrpd(sdp);
 832         iput(sdp->sd_rindex);
 833 fail_statfs:
 834         iput(sdp->sd_statfs_inode);
 835 fail_journal:
 836         init_journal(sdp, UNDO);
 837 fail:
 838         return error;
 839 }
 840
 841 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 842 {
 843         struct inode *pn = NULL;
 844         char buf[30];
 845         int error = 0;
 846         struct gfs2_inode *ip;
 847         struct inode *master = d_inode(sdp->sd_master_dir);
 848
 849         if (sdp->sd_args.ar_spectator)
 850                 return 0;
 851
 852         if (undo)
 853                 goto fail_qc_gh;
 854
 855         pn = gfs2_lookup_simple(master, "per_node");
 856         if (IS_ERR(pn)) {
 857                 error = PTR_ERR(pn);
 858                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 859                 return error;
 860         }
 861
 862         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 863         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 864         if (IS_ERR(sdp->sd_sc_inode)) {
 865                 error = PTR_ERR(sdp->sd_sc_inode);
 866                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 867                 goto fail;
 868         }
 869
 870         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 871         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 872         if (IS_ERR(sdp->sd_qc_inode)) {
 873                 error = PTR_ERR(sdp->sd_qc_inode);
 874                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 875                 goto fail_ut_i;
 876         }
 877
 878         iput(pn);
 879         pn = NULL;
 880
 881         ip = GFS2_I(sdp->sd_sc_inode);
 882         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 883                                    &sdp->sd_sc_gh);
 884         if (error) {
 885                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 886                 goto fail_qc_i;
 887         }
 888
 889         ip = GFS2_I(sdp->sd_qc_inode);
 890         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 891                                    &sdp->sd_qc_gh);
 892         if (error) {
 893                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 894                 goto fail_ut_gh;
 895         }
 896
 897         return 0;
 898
 899 fail_qc_gh:
 900         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 901 fail_ut_gh:
 902         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 903 fail_qc_i:
 904         iput(sdp->sd_qc_inode);
 905 fail_ut_i:
 906         iput(sdp->sd_sc_inode);
 907 fail:
 908         iput(pn);
 909         return error;
 910 }
 911
 912 static const match_table_t nolock_tokens = {
 913         { Opt_jid, "jid=%d\n", },
 914         { Opt_err, NULL },
 915 };
 916
 917 static const struct lm_lockops nolock_ops = {
 918         .lm_proto_name = "lock_nolock",
 919         .lm_put_lock = gfs2_glock_free,
 920         .lm_tokens = &nolock_tokens,
 921 };
 922
 923 /**
 924  * gfs2_lm_mount - mount a locking protocol
 925  * @sdp: the filesystem
 926  * @args: mount arguments
 927  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 928  *
 929  * Returns: errno
 930  */
 931
 932 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 933 {
 934         const struct lm_lockops *lm;
 935         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 936         struct gfs2_args *args = &sdp->sd_args;
 937         const char *proto = sdp->sd_proto_name;
 938         const char *table = sdp->sd_table_name;
 939         char *o, *options;
 940         int ret;
 941
 942         if (!strcmp("lock_nolock", proto)) {
 943                 lm = &nolock_ops;
 944                 sdp->sd_args.ar_localflocks = 1;
 945 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 946         } else if (!strcmp("lock_dlm", proto)) {
 947                 lm = &gfs2_dlm_ops;
 948 #endif
 949         } else {
 950                 pr_info("can't find protocol %s\n", proto);
 951                 return -ENOENT;
 952         }
 953
 954         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 955
 956         ls->ls_ops = lm;
 957         ls->ls_first = 1;
 958
 959         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 960                 substring_t tmp[MAX_OPT_ARGS];
 961                 int token, option;
 962
 963                 if (!o || !*o)
 964                         continue;
 965
 966                 token = match_token(o, *lm->lm_tokens, tmp);
 967                 switch (token) {
 968                 case Opt_jid:
 969                         ret = match_int(&tmp[0], &option);
 970                         if (ret || option < 0)
 971                                 goto hostdata_error;
 972                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 973                                 ls->ls_jid = option;
 974                         break;
 975                 case Opt_id:
 976                 case Opt_nodir:
 977                         /* Obsolete, but left for backward compat purposes */
 978                         break;
 979                 case Opt_first:
 980                         ret = match_int(&tmp[0], &option);
 981                         if (ret || (option != 0 && option != 1))
 982                                 goto hostdata_error;
 983                         ls->ls_first = option;
 984                         break;
 985                 case Opt_err:
 986                 default:
 987 hostdata_error:
 988                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 989                         return -EINVAL;
 990                 }
 991         }
 992
 993         if (lm->lm_mount == NULL) {
 994                 fs_info(sdp, "Now mounting FS...\n");
 995                 complete_all(&sdp->sd_locking_init);
 996                 return 0;
 997         }
 998         ret = lm->lm_mount(sdp, table);
 999         if (ret == 0)
1000                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1001         complete_all(&sdp->sd_locking_init);
1002         return ret;
1003 }
1004
1005 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1006 {
1007         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1008         if (likely(!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) &&
1009             lm->lm_unmount)
1010                 lm->lm_unmount(sdp);
1011 }
1012
1013 static int wait_on_journal(struct gfs2_sbd *sdp)
1014 {
1015         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1016                 return 0;
1017
1018         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1019                 ? -EINTR : 0;
1020 }
1021
1022 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1023 {
1024         struct super_block *sb = sdp->sd_vfs;
1025         char ro[20];
1026         char spectator[20];
1027         char *envp[] = { ro, spectator, NULL };
1028         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1029         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1030         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1031 }
1032
1033 /**
1034  * fill_super - Read in superblock
1035  * @sb: The VFS superblock
1036  * @data: Mount options
1037  * @silent: Don't complain if it's not a GFS2 filesystem
1038  *
1039  * Returns: errno
1040  */
1041
1042 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1043 {
1044         struct gfs2_sbd *sdp;
1045         struct gfs2_holder mount_gh;
1046         int error;
1047
1048         sdp = init_sbd(sb);
1049         if (!sdp) {
1050                 pr_warn("can't alloc struct gfs2_sbd\n");
1051                 return -ENOMEM;
1052         }
1053         sdp->sd_args = *args;
1054
1055         if (sdp->sd_args.ar_spectator) {
1056                 sb->s_flags |= SB_RDONLY;
1057                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1058         }
1059         if (sdp->sd_args.ar_posix_acl)
1060                 sb->s_flags |= SB_POSIXACL;
1061         if (sdp->sd_args.ar_nobarrier)
1062                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1063
1064         sb->s_flags |= SB_NOSEC;
1065         sb->s_magic = GFS2_MAGIC;
1066         sb->s_op = &gfs2_super_ops;
1067         sb->s_d_op = &gfs2_dops;
1068         sb->s_export_op = &gfs2_export_ops;
1069         sb->s_xattr = gfs2_xattr_handlers;
1070         sb->s_qcop = &gfs2_quotactl_ops;
1071         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1072         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1073         sb->s_time_gran = 1;
1074         sb->s_maxbytes = MAX_LFS_FILESIZE;
1075
1076         /* Set up the buffer cache and fill in some fake block size values
1077            to allow us to read-in the on-disk superblock. */
1078         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1079         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1080         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1081                                GFS2_BASIC_BLOCK_SHIFT;
1082         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1083
1084         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1085         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1086         if (sdp->sd_args.ar_statfs_quantum) {
1087                 sdp->sd_tune.gt_statfs_slow = 0;
1088                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1089         } else {
1090                 sdp->sd_tune.gt_statfs_slow = 1;
1091                 sdp->sd_tune.gt_statfs_quantum = 30;
1092         }
1093
1094         error = init_names(sdp, silent);
1095         if (error) {
1096                 /* In this case, we haven't initialized sysfs, so we have to
1097                    manually free the sdp. */
1098                 free_sbd(sdp);
1099                 sb->s_fs_info = NULL;
1100                 return error;
1101         }
1102
1103         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1104
1105         error = gfs2_sys_fs_add(sdp);
1106         /*
1107          * If we hit an error here, gfs2_sys_fs_add will have called function
1108          * kobject_put which causes the sysfs usage count to go to zero, which
1109          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1110          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1111          * kobject_put to free sdp.
1112          */
1113         if (error)
1114                 return error;
1115
1116         gfs2_create_debugfs_file(sdp);
1117
1118         error = gfs2_lm_mount(sdp, silent);
1119         if (error)
1120                 goto fail_debug;
1121
1122         error = init_locking(sdp, &mount_gh, DO);
1123         if (error)
1124                 goto fail_lm;
1125
1126         error = init_sb(sdp, silent);
1127         if (error)
1128                 goto fail_locking;
1129
1130         error = wait_on_journal(sdp);
1131         if (error)
1132                 goto fail_sb;
1133
1134         /*
1135          * If user space has failed to join the cluster or some similar
1136          * failure has occurred, then the journal id will contain a
1137          * negative (error) number. This will then be returned to the
1138          * caller (of the mount syscall). We do this even for spectator
1139          * mounts (which just write a jid of 0 to indicate "ok" even though
1140          * the jid is unused in the spectator case)
1141          */
1142         if (sdp->sd_lockstruct.ls_jid < 0) {
1143                 error = sdp->sd_lockstruct.ls_jid;
1144                 sdp->sd_lockstruct.ls_jid = 0;
1145                 goto fail_sb;
1146         }
1147
1148         if (sdp->sd_args.ar_spectator)
1149                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1150                          sdp->sd_table_name);
1151         else
1152                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1153                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1154
1155         error = init_inodes(sdp, DO);
1156         if (error)
1157                 goto fail_sb;
1158
1159         error = init_per_node(sdp, DO);
1160         if (error)
1161                 goto fail_inodes;
1162
1163         error = gfs2_statfs_init(sdp);
1164         if (error) {
1165                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1166                 goto fail_per_node;
1167         }
1168
1169         if (!sb_rdonly(sb)) {
1170                 error = gfs2_make_fs_rw(sdp);
1171                 if (error) {
1172                         fs_err(sdp, "can't make FS RW: %d\n", error);
1173                         goto fail_per_node;
1174                 }
1175         }
1176
1177         gfs2_glock_dq_uninit(&mount_gh);
1178         gfs2_online_uevent(sdp);
1179         return 0;
1180
1181 fail_per_node:
1182         init_per_node(sdp, UNDO);
1183 fail_inodes:
1184         init_inodes(sdp, UNDO);
1185 fail_sb:
1186         if (sdp->sd_root_dir)
1187                 dput(sdp->sd_root_dir);
1188         if (sdp->sd_master_dir)
1189                 dput(sdp->sd_master_dir);
1190         if (sb->s_root)
1191                 dput(sb->s_root);
1192         sb->s_root = NULL;
1193 fail_locking:
1194         init_locking(sdp, &mount_gh, UNDO);
1195 fail_lm:
1196         complete_all(&sdp->sd_journal_ready);
1197         gfs2_gl_hash_clear(sdp);
1198         gfs2_lm_unmount(sdp);
1199 fail_debug:
1200         gfs2_delete_debugfs_file(sdp);
1201         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1202          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1203         gfs2_sys_fs_del(sdp);
1204         sb->s_fs_info = NULL;
1205         return error;
1206 }
1207
1208 static int set_gfs2_super(struct super_block *s, void *data)
1209 {
1210         s->s_bdev = data;
1211         s->s_dev = s->s_bdev->bd_dev;
1212         s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1213         return 0;
1214 }
1215
1216 static int test_gfs2_super(struct super_block *s, void *ptr)
1217 {
1218         struct block_device *bdev = ptr;
1219         return (bdev == s->s_bdev);
1220 }
1221
1222 /**
1223  * gfs2_mount - Get the GFS2 superblock
1224  * @fs_type: The GFS2 filesystem type
1225  * @flags: Mount flags
1226  * @dev_name: The name of the device
1227  * @data: The mount arguments
1228  *
1229  * Q. Why not use get_sb_bdev() ?
1230  * A. We need to select one of two root directories to mount, independent
1231  *    of whether this is the initial, or subsequent, mount of this sb
1232  *
1233  * Returns: 0 or -ve on error
1234  */
1235
1236 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1237                        const char *dev_name, void *data)
1238 {
1239         struct block_device *bdev;
1240         struct super_block *s;
1241         fmode_t mode = FMODE_READ | FMODE_EXCL;
1242         int error;
1243         struct gfs2_args args;
1244         struct gfs2_sbd *sdp;
1245
1246         if (!(flags & SB_RDONLY))
1247                 mode |= FMODE_WRITE;
1248
1249         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1250         if (IS_ERR(bdev))
1251                 return ERR_CAST(bdev);
1252
1253         /*
1254          * once the super is inserted into the list by sget, s_umount
1255          * will protect the lockfs code from trying to start a snapshot
1256          * while we are mounting
1257          */
1258         mutex_lock(&bdev->bd_fsfreeze_mutex);
1259         if (bdev->bd_fsfreeze_count > 0) {
1260                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1261                 error = -EBUSY;
1262                 goto error_bdev;
1263         }
1264         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1265         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1266         error = PTR_ERR(s);
1267         if (IS_ERR(s))
1268                 goto error_bdev;
1269
1270         if (s->s_root) {
1271                 /*
1272                  * s_umount nests inside bd_mutex during
1273                  * __invalidate_device().  blkdev_put() acquires
1274                  * bd_mutex and can't be called under s_umount.  Drop
1275                  * s_umount temporarily.  This is safe as we're
1276                  * holding an active reference.
1277                  */
1278                 up_write(&s->s_umount);
1279                 blkdev_put(bdev, mode);
1280                 down_write(&s->s_umount);
1281         } else {
1282                 /* s_mode must be set before deactivate_locked_super calls */
1283                 s->s_mode = mode;
1284         }
1285
1286         memset(&args, 0, sizeof(args));
1287         args.ar_quota = GFS2_QUOTA_DEFAULT;
1288         args.ar_data = GFS2_DATA_DEFAULT;
1289         args.ar_commit = 30;
1290         args.ar_statfs_quantum = 30;
1291         args.ar_quota_quantum = 60;
1292         args.ar_errors = GFS2_ERRORS_DEFAULT;
1293
1294         error = gfs2_mount_args(&args, data);
1295         if (error) {
1296                 pr_warn("can't parse mount arguments\n");
1297                 goto error_super;
1298         }
1299
1300         if (s->s_root) {
1301                 error = -EBUSY;
1302                 if ((flags ^ s->s_flags) & SB_RDONLY)
1303                         goto error_super;
1304         } else {
1305                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1306                 sb_set_blocksize(s, block_size(bdev));
1307                 error = fill_super(s, &args, flags & SB_SILENT ? 1 : 0);
1308                 if (error)
1309                         goto error_super;
1310                 s->s_flags |= SB_ACTIVE;
1311                 bdev->bd_super = s;
1312         }
1313
1314         sdp = s->s_fs_info;
1315         if (args.ar_meta)
1316                 return dget(sdp->sd_master_dir);
1317         else
1318                 return dget(sdp->sd_root_dir);
1319
1320 error_super:
1321         deactivate_locked_super(s);
1322         return ERR_PTR(error);
1323 error_bdev:
1324         blkdev_put(bdev, mode);
1325         return ERR_PTR(error);
1326 }
1327
1328 static int set_meta_super(struct super_block *s, void *ptr)
1329 {
1330         return -EINVAL;
1331 }
1332
1333 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1334                         int flags, const char *dev_name, void *data)
1335 {
1336         struct super_block *s;
1337         struct gfs2_sbd *sdp;
1338         struct path path;
1339         int error;
1340
1341         if (!dev_name || !*dev_name)
1342                 return ERR_PTR(-EINVAL);
1343
1344         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1345         if (error) {
1346                 pr_warn("path_lookup on %s returned error %d\n",
1347                         dev_name, error);
1348                 return ERR_PTR(error);
1349         }
1350         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1351                  path.dentry->d_sb->s_bdev);
1352         path_put(&path);
1353         if (IS_ERR(s)) {
1354                 pr_warn("gfs2 mount does not exist\n");
1355                 return ERR_CAST(s);
1356         }
1357         if ((flags ^ s->s_flags) & SB_RDONLY) {
1358                 deactivate_locked_super(s);
1359                 return ERR_PTR(-EBUSY);
1360         }
1361         sdp = s->s_fs_info;
1362         return dget(sdp->sd_master_dir);
1363 }
1364
1365 static void gfs2_kill_sb(struct super_block *sb)
1366 {
1367         struct gfs2_sbd *sdp = sb->s_fs_info;
1368
1369         if (sdp == NULL) {
1370                 kill_block_super(sb);
1371                 return;
1372         }
1373
1374         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1375         dput(sdp->sd_root_dir);
1376         dput(sdp->sd_master_dir);
1377         sdp->sd_root_dir = NULL;
1378         sdp->sd_master_dir = NULL;
1379         shrink_dcache_sb(sb);
1380         kill_block_super(sb);
1381 }
1382
1383 struct file_system_type gfs2_fs_type = {
1384         .name = "gfs2",
1385         .fs_flags = FS_REQUIRES_DEV,
1386         .mount = gfs2_mount,
1387         .kill_sb = gfs2_kill_sb,
1388         .owner = THIS_MODULE,
1389 };
1390 MODULE_ALIAS_FS("gfs2");
1391
1392 struct file_system_type gfs2meta_fs_type = {
1393         .name = "gfs2meta",
1394         .fs_flags = FS_REQUIRES_DEV,
1395         .mount = gfs2_mount_meta,
1396         .owner = THIS_MODULE,
1397 };
1398 MODULE_ALIAS_FS("gfs2meta");