kernel/fs/gfs2/ops_fstype.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/completion.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/blkdev.h>
  18 #include <linux/kthread.h>
  19 #include <linux/export.h>
  20 #include <linux/namei.h>
  21 #include <linux/mount.h>
  22 #include <linux/gfs2_ondisk.h>
  23 #include <linux/quotaops.h>
  24 #include <linux/lockdep.h>
  25 #include <linux/module.h>
  26
  27 #include "gfs2.h"
  28 #include "incore.h"
  29 #include "bmap.h"
  30 #include "glock.h"
  31 #include "glops.h"
  32 #include "inode.h"
  33 #include "recovery.h"
  34 #include "rgrp.h"
  35 #include "super.h"
  36 #include "sys.h"
  37 #include "util.h"
  38 #include "log.h"
  39 #include "quota.h"
  40 #include "dir.h"
  41 #include "meta_io.h"
  42 #include "trace_gfs2.h"
  43
  44 #define DO 0
  45 #define UNDO 1
  46
  47 /**
  48  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  49  * @gt: tune
  50  *
  51  */
  52
  53 static void gfs2_tune_init(struct gfs2_tune *gt)
  54 {
  55         spin_lock_init(&gt->gt_spin);
  56
  57         gt->gt_quota_warn_period = 10;
  58         gt->gt_quota_scale_num = 1;
  59         gt->gt_quota_scale_den = 1;
  60         gt->gt_new_files_jdata = 0;
  61         gt->gt_max_readahead = 1 << 18;
  62         gt->gt_complain_secs = 10;
  63 }
  64
  65 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  66 {
  67         struct gfs2_sbd *sdp;
  68         struct address_space *mapping;
  69
  70         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  71         if (!sdp)
  72                 return NULL;
  73
  74         sb->s_fs_info = sdp;
  75         sdp->sd_vfs = sb;
  76         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  77         if (!sdp->sd_lkstats) {
  78                 kfree(sdp);
  79                 return NULL;
  80         }
  81
  82         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  83         gfs2_tune_init(&sdp->sd_tune);
  84
  85         init_waitqueue_head(&sdp->sd_glock_wait);
  86         atomic_set(&sdp->sd_glock_disposal, 0);
  87         init_completion(&sdp->sd_locking_init);
  88         init_completion(&sdp->sd_wdack);
  89         spin_lock_init(&sdp->sd_statfs_spin);
  90
  91         spin_lock_init(&sdp->sd_rindex_spin);
  92         sdp->sd_rindex_tree.rb_node = NULL;
  93
  94         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  95         spin_lock_init(&sdp->sd_jindex_spin);
  96         mutex_init(&sdp->sd_jindex_mutex);
  97         init_completion(&sdp->sd_journal_ready);
  98
  99         INIT_LIST_HEAD(&sdp->sd_quota_list);
 100         mutex_init(&sdp->sd_quota_mutex);
 101         mutex_init(&sdp->sd_quota_sync_mutex);
 102         init_waitqueue_head(&sdp->sd_quota_wait);
 103         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 104         spin_lock_init(&sdp->sd_trunc_lock);
 105         spin_lock_init(&sdp->sd_bitmap_lock);
 106
 107         mapping = &sdp->sd_aspace;
 108
 109         address_space_init_once(mapping);
 110         mapping->a_ops = &gfs2_rgrp_aops;
 111         mapping->host = sb->s_bdev->bd_inode;
 112         mapping->flags = 0;
 113         mapping_set_gfp_mask(mapping, GFP_NOFS);
 114         mapping->private_data = NULL;
 115         mapping->writeback_index = 0;
 116
 117         spin_lock_init(&sdp->sd_log_lock);
 118         atomic_set(&sdp->sd_log_pinned, 0);
 119         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
 120         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
 121         spin_lock_init(&sdp->sd_ordered_lock);
 122
 123         init_waitqueue_head(&sdp->sd_log_waitq);
 124         init_waitqueue_head(&sdp->sd_logd_waitq);
 125         spin_lock_init(&sdp->sd_ail_lock);
 126         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 127         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 128
 129         init_rwsem(&sdp->sd_log_flush_lock);
 130         atomic_set(&sdp->sd_log_in_flight, 0);
 131         atomic_set(&sdp->sd_reserving_log, 0);
 132         init_waitqueue_head(&sdp->sd_reserving_log_wait);
 133         init_waitqueue_head(&sdp->sd_log_flush_wait);
 134         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 135         mutex_init(&sdp->sd_freeze_mutex);
 136
 137         return sdp;
 138 }
 139
 140
 141 /**
 142  * gfs2_check_sb - Check superblock
 143  * @sdp: the filesystem
 144  * @sb: The superblock
 145  * @silent: Don't print a message if the check fails
 146  *
 147  * Checks the version code of the FS is one that we understand how to
 148  * read and that the sizes of the various on-disk structures have not
 149  * changed.
 150  */
 151
 152 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 153 {
 154         struct gfs2_sb_host *sb = &sdp->sd_sb;
 155
 156         if (sb->sb_magic != GFS2_MAGIC ||
 157             sb->sb_type != GFS2_METATYPE_SB) {
 158                 if (!silent)
 159                         pr_warn("not a GFS2 filesystem\n");
 160                 return -EINVAL;
 161         }
 162
 163         /*  If format numbers match exactly, we're done.  */
 164
 165         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 166             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 167                 return 0;
 168
 169         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 170
 171         return -EINVAL;
 172 }
 173
 174 static void end_bio_io_page(struct bio *bio, int error)
 175 {
 176         struct page *page = bio->bi_private;
 177
 178         if (!error)
 179                 SetPageUptodate(page);
 180         else
 181                 pr_warn("error %d reading superblock\n", error);
 182         unlock_page(page);
 183 }
 184
 185 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 186 {
 187         struct gfs2_sb_host *sb = &sdp->sd_sb;
 188         struct super_block *s = sdp->sd_vfs;
 189         const struct gfs2_sb *str = buf;
 190
 191         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 192         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 193         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 194         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 195         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 196         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 197         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 198         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 199         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 200         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 201         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 202
 203         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 204         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 205         memcpy(s->s_uuid, str->sb_uuid, 16);
 206 }
 207
 208 /**
 209  * gfs2_read_super - Read the gfs2 super block from disk
 210  * @sdp: The GFS2 super block
 211  * @sector: The location of the super block
 212  * @error: The error code to return
 213  *
 214  * This uses the bio functions to read the super block from disk
 215  * because we want to be 100% sure that we never read cached data.
 216  * A super block is read twice only during each GFS2 mount and is
 217  * never written to by the filesystem. The first time its read no
 218  * locks are held, and the only details which are looked at are those
 219  * relating to the locking protocol. Once locking is up and working,
 220  * the sb is read again under the lock to establish the location of
 221  * the master directory (contains pointers to journals etc) and the
 222  * root directory.
 223  *
 224  * Returns: 0 on success or error
 225  */
 226
 227 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 228 {
 229         struct super_block *sb = sdp->sd_vfs;
 230         struct gfs2_sb *p;
 231         struct page *page;
 232         struct bio *bio;
 233
 234         page = alloc_page(GFP_NOFS);
 235         if (unlikely(!page))
 236                 return -ENOMEM;
 237
 238         ClearPageUptodate(page);
 239         ClearPageDirty(page);
 240         lock_page(page);
 241
 242         bio = bio_alloc(GFP_NOFS, 1);
 243         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 244         bio->bi_bdev = sb->s_bdev;
 245         bio_add_page(bio, page, PAGE_SIZE, 0);
 246
 247         bio->bi_end_io = end_bio_io_page;
 248         bio->bi_private = page;
 249         submit_bio(READ_SYNC | REQ_META, bio);
 250         wait_on_page_locked(page);
 251         bio_put(bio);
 252         if (!PageUptodate(page)) {
 253                 __free_page(page);
 254                 return -EIO;
 255         }
 256         p = kmap(page);
 257         gfs2_sb_in(sdp, p);
 258         kunmap(page);
 259         __free_page(page);
 260         return gfs2_check_sb(sdp, silent);
 261 }
 262
 263 /**
 264  * gfs2_read_sb - Read super block
 265  * @sdp: The GFS2 superblock
 266  * @silent: Don't print message if mount fails
 267  *
 268  */
 269
 270 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 271 {
 272         u32 hash_blocks, ind_blocks, leaf_blocks;
 273         u32 tmp_blocks;
 274         unsigned int x;
 275         int error;
 276
 277         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 278         if (error) {
 279                 if (!silent)
 280                         fs_err(sdp, "can't read superblock\n");
 281                 return error;
 282         }
 283
 284         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 285                                GFS2_BASIC_BLOCK_SHIFT;
 286         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 287         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 288                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 289         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 290                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 291         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 292         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 293         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 294         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 295         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 296                                 sizeof(struct gfs2_meta_header)) /
 297                                 sizeof(struct gfs2_quota_change);
 298         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 299                                      sizeof(struct gfs2_meta_header))
 300                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 301
 302         /* Compute maximum reservation required to add a entry to a directory */
 303
 304         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 305                              sdp->sd_jbsize);
 306
 307         ind_blocks = 0;
 308         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 309                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 310                 ind_blocks += tmp_blocks;
 311         }
 312
 313         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 314
 315         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 316
 317         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 318                                 sizeof(struct gfs2_dinode);
 319         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 320         for (x = 2;; x++) {
 321                 u64 space, d;
 322                 u32 m;
 323
 324                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 325                 d = space;
 326                 m = do_div(d, sdp->sd_inptrs);
 327
 328                 if (d != sdp->sd_heightsize[x - 1] || m)
 329                         break;
 330                 sdp->sd_heightsize[x] = space;
 331         }
 332         sdp->sd_max_height = x;
 333         sdp->sd_heightsize[x] = ~0;
 334         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 335
 336         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 337                                  sizeof(struct gfs2_dinode);
 338         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 339         for (x = 2;; x++) {
 340                 u64 space, d;
 341                 u32 m;
 342
 343                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 344                 d = space;
 345                 m = do_div(d, sdp->sd_inptrs);
 346
 347                 if (d != sdp->sd_jheightsize[x - 1] || m)
 348                         break;
 349                 sdp->sd_jheightsize[x] = space;
 350         }
 351         sdp->sd_max_jheight = x;
 352         sdp->sd_jheightsize[x] = ~0;
 353         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 354
 355         return 0;
 356 }
 357
 358 static int init_names(struct gfs2_sbd *sdp, int silent)
 359 {
 360         char *proto, *table;
 361         int error = 0;
 362
 363         proto = sdp->sd_args.ar_lockproto;
 364         table = sdp->sd_args.ar_locktable;
 365
 366         /*  Try to autodetect  */
 367
 368         if (!proto[0] || !table[0]) {
 369                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 370                 if (error)
 371                         return error;
 372
 373                 if (!proto[0])
 374                         proto = sdp->sd_sb.sb_lockproto;
 375                 if (!table[0])
 376                         table = sdp->sd_sb.sb_locktable;
 377         }
 378
 379         if (!table[0])
 380                 table = sdp->sd_vfs->s_id;
 381
 382         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 383         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 384
 385         table = sdp->sd_table_name;
 386         while ((table = strchr(table, '/')))
 387                 *table = '_';
 388
 389         return error;
 390 }
 391
 392 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 393                         int undo)
 394 {
 395         int error = 0;
 396
 397         if (undo)
 398                 goto fail_trans;
 399
 400         error = gfs2_glock_nq_num(sdp,
 401                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 402                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 403                                   mount_gh);
 404         if (error) {
 405                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 406                 goto fail;
 407         }
 408
 409         error = gfs2_glock_nq_num(sdp,
 410                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 411                                   LM_ST_SHARED,
 412                                   LM_FLAG_NOEXP | GL_EXACT,
 413                                   &sdp->sd_live_gh);
 414         if (error) {
 415                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 416                 goto fail_mount;
 417         }
 418
 419         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 420                                CREATE, &sdp->sd_rename_gl);
 421         if (error) {
 422                 fs_err(sdp, "can't create rename glock: %d\n", error);
 423                 goto fail_live;
 424         }
 425
 426         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 427                                CREATE, &sdp->sd_freeze_gl);
 428         if (error) {
 429                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 430                 goto fail_rename;
 431         }
 432
 433         return 0;
 434
 435 fail_trans:
 436         gfs2_glock_put(sdp->sd_freeze_gl);
 437 fail_rename:
 438         gfs2_glock_put(sdp->sd_rename_gl);
 439 fail_live:
 440         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 441 fail_mount:
 442         gfs2_glock_dq_uninit(mount_gh);
 443 fail:
 444         return error;
 445 }
 446
 447 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 448                             u64 no_addr, const char *name)
 449 {
 450         struct gfs2_sbd *sdp = sb->s_fs_info;
 451         struct dentry *dentry;
 452         struct inode *inode;
 453
 454         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 455         if (IS_ERR(inode)) {
 456                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 457                 return PTR_ERR(inode);
 458         }
 459         dentry = d_make_root(inode);
 460         if (!dentry) {
 461                 fs_err(sdp, "can't alloc %s dentry\n", name);
 462                 return -ENOMEM;
 463         }
 464         *dptr = dentry;
 465         return 0;
 466 }
 467
 468 static int init_sb(struct gfs2_sbd *sdp, int silent)
 469 {
 470         struct super_block *sb = sdp->sd_vfs;
 471         struct gfs2_holder sb_gh;
 472         u64 no_addr;
 473         int ret;
 474
 475         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 476                                 LM_ST_SHARED, 0, &sb_gh);
 477         if (ret) {
 478                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 479                 return ret;
 480         }
 481
 482         ret = gfs2_read_sb(sdp, silent);
 483         if (ret) {
 484                 fs_err(sdp, "can't read superblock: %d\n", ret);
 485                 goto out;
 486         }
 487
 488         /* Set up the buffer cache and SB for real */
 489         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 490                 ret = -EINVAL;
 491                 fs_err(sdp, "FS block size (%u) is too small for device "
 492                        "block size (%u)\n",
 493                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 494                 goto out;
 495         }
 496         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 497                 ret = -EINVAL;
 498                 fs_err(sdp, "FS block size (%u) is too big for machine "
 499                        "page size (%u)\n",
 500                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 501                 goto out;
 502         }
 503         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 504
 505         /* Get the root inode */
 506         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 507         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 508         if (ret)
 509                 goto out;
 510
 511         /* Get the master inode */
 512         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 513         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 514         if (ret) {
 515                 dput(sdp->sd_root_dir);
 516                 goto out;
 517         }
 518         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 519 out:
 520         gfs2_glock_dq_uninit(&sb_gh);
 521         return ret;
 522 }
 523
 524 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 525 {
 526         char *message = "FIRSTMOUNT=Done";
 527         char *envp[] = { message, NULL };
 528
 529         fs_info(sdp, "first mount done, others may mount\n");
 530
 531         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 532                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 533
 534         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 535 }
 536
 537 /**
 538  * gfs2_jindex_hold - Grab a lock on the jindex
 539  * @sdp: The GFS2 superblock
 540  * @ji_gh: the holder for the jindex glock
 541  *
 542  * Returns: errno
 543  */
 544
 545 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 546 {
 547         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 548         struct qstr name;
 549         char buf[20];
 550         struct gfs2_jdesc *jd;
 551         int error;
 552
 553         name.name = buf;
 554
 555         mutex_lock(&sdp->sd_jindex_mutex);
 556
 557         for (;;) {
 558                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 559                 if (error)
 560                         break;
 561
 562                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 563                 name.hash = gfs2_disk_hash(name.name, name.len);
 564
 565                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 566                 if (error == -ENOENT) {
 567                         error = 0;
 568                         break;
 569                 }
 570
 571                 gfs2_glock_dq_uninit(ji_gh);
 572
 573                 if (error)
 574                         break;
 575
 576                 error = -ENOMEM;
 577                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 578                 if (!jd)
 579                         break;
 580
 581                 INIT_LIST_HEAD(&jd->extent_list);
 582                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 583
 584                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 585                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 586                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 587                         if (!jd->jd_inode)
 588                                 error = -ENOENT;
 589                         else
 590                                 error = PTR_ERR(jd->jd_inode);
 591                         kfree(jd);
 592                         break;
 593                 }
 594
 595                 spin_lock(&sdp->sd_jindex_spin);
 596                 jd->jd_jid = sdp->sd_journals++;
 597                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 598                 spin_unlock(&sdp->sd_jindex_spin);
 599         }
 600
 601         mutex_unlock(&sdp->sd_jindex_mutex);
 602
 603         return error;
 604 }
 605
 606 /**
 607  * check_journal_clean - Make sure a journal is clean for a spectator mount
 608  * @sdp: The GFS2 superblock
 609  * @jd: The journal descriptor
 610  *
 611  * Returns: 0 if the journal is clean or locked, else an error
 612  */
 613 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 614 {
 615         int error;
 616         struct gfs2_holder j_gh;
 617         struct gfs2_log_header_host head;
 618         struct gfs2_inode *ip;
 619
 620         ip = GFS2_I(jd->jd_inode);
 621         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 622                                    GL_EXACT | GL_NOCACHE, &j_gh);
 623         if (error) {
 624                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 625                 return -EPERM;
 626         }
 627         error = gfs2_jdesc_check(jd);
 628         if (error) {
 629                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 630                 goto out_unlock;
 631         }
 632         error = gfs2_find_jhead(jd, &head);
 633         if (error) {
 634                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 635                 goto out_unlock;
 636         }
 637         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 638                 error = -EPERM;
 639                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 640                        "must not be a spectator.\n", jd->jd_jid);
 641         }
 642
 643 out_unlock:
 644         gfs2_glock_dq_uninit(&j_gh);
 645         return error;
 646 }
 647
 648 static int init_journal(struct gfs2_sbd *sdp, int undo)
 649 {
 650         struct inode *master = d_inode(sdp->sd_master_dir);
 651         struct gfs2_holder ji_gh;
 652         struct gfs2_inode *ip;
 653         int jindex = 1;
 654         int error = 0;
 655
 656         if (undo) {
 657                 jindex = 0;
 658                 goto fail_jinode_gh;
 659         }
 660
 661         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 662         if (IS_ERR(sdp->sd_jindex)) {
 663                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 664                 return PTR_ERR(sdp->sd_jindex);
 665         }
 666
 667         /* Load in the journal index special file */
 668
 669         error = gfs2_jindex_hold(sdp, &ji_gh);
 670         if (error) {
 671                 fs_err(sdp, "can't read journal index: %d\n", error);
 672                 goto fail;
 673         }
 674
 675         error = -EUSERS;
 676         if (!gfs2_jindex_size(sdp)) {
 677                 fs_err(sdp, "no journals!\n");
 678                 goto fail_jindex;
 679         }
 680
 681         if (sdp->sd_args.ar_spectator) {
 682                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 683                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 684                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 685                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 686         } else {
 687                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 688                         fs_err(sdp, "can't mount journal #%u\n",
 689                                sdp->sd_lockstruct.ls_jid);
 690                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 691                                gfs2_jindex_size(sdp),
 692                                gfs2_jindex_size(sdp) - 1);
 693                         goto fail_jindex;
 694                 }
 695                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 696
 697                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 698                                           &gfs2_journal_glops,
 699                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 700                                           &sdp->sd_journal_gh);
 701                 if (error) {
 702                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 703                         goto fail_jindex;
 704                 }
 705
 706                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 707                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 708                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 709                                            &sdp->sd_jinode_gh);
 710                 if (error) {
 711                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 712                                error);
 713                         goto fail_journal_gh;
 714                 }
 715
 716                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 717                 if (error) {
 718                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 719                                sdp->sd_jdesc->jd_jid, error);
 720                         goto fail_jinode_gh;
 721                 }
 722                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 723                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 724                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 725
 726                 /* Map the extents for this journal's blocks */
 727                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 728         }
 729         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 730
 731         if (sdp->sd_lockstruct.ls_first) {
 732                 unsigned int x;
 733                 for (x = 0; x < sdp->sd_journals; x++) {
 734                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 735
 736                         if (sdp->sd_args.ar_spectator) {
 737                                 error = check_journal_clean(sdp, jd);
 738                                 if (error)
 739                                         goto fail_jinode_gh;
 740                                 continue;
 741                         }
 742                         error = gfs2_recover_journal(jd, true);
 743                         if (error) {
 744                                 fs_err(sdp, "error recovering journal %u: %d\n",
 745                                        x, error);
 746                                 goto fail_jinode_gh;
 747                         }
 748                 }
 749
 750                 gfs2_others_may_mount(sdp);
 751         } else if (!sdp->sd_args.ar_spectator) {
 752                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 753                 if (error) {
 754                         fs_err(sdp, "error recovering my journal: %d\n", error);
 755                         goto fail_jinode_gh;
 756                 }
 757         }
 758
 759         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 760         gfs2_glock_dq_uninit(&ji_gh);
 761         jindex = 0;
 762         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 763         return 0;
 764
 765 fail_jinode_gh:
 766         if (!sdp->sd_args.ar_spectator)
 767                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 768 fail_journal_gh:
 769         if (!sdp->sd_args.ar_spectator)
 770                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 771 fail_jindex:
 772         gfs2_jindex_free(sdp);
 773         if (jindex)
 774                 gfs2_glock_dq_uninit(&ji_gh);
 775 fail:
 776         iput(sdp->sd_jindex);
 777         return error;
 778 }
 779
 780 static struct lock_class_key gfs2_quota_imutex_key;
 781
 782 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 783 {
 784         int error = 0;
 785         struct inode *master = d_inode(sdp->sd_master_dir);
 786
 787         if (undo)
 788                 goto fail_qinode;
 789
 790         error = init_journal(sdp, undo);
 791         complete_all(&sdp->sd_journal_ready);
 792         if (error)
 793                 goto fail;
 794
 795         /* Read in the master statfs inode */
 796         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 797         if (IS_ERR(sdp->sd_statfs_inode)) {
 798                 error = PTR_ERR(sdp->sd_statfs_inode);
 799                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 800                 goto fail_journal;
 801         }
 802
 803         /* Read in the resource index inode */
 804         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 805         if (IS_ERR(sdp->sd_rindex)) {
 806                 error = PTR_ERR(sdp->sd_rindex);
 807                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 808                 goto fail_statfs;
 809         }
 810         sdp->sd_rindex_uptodate = 0;
 811
 812         /* Read in the quota inode */
 813         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 814         if (IS_ERR(sdp->sd_quota_inode)) {
 815                 error = PTR_ERR(sdp->sd_quota_inode);
 816                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 817                 goto fail_rindex;
 818         }
 819         /*
 820          * i_mutex on quota files is special. Since this inode is hidden system
 821          * file, we are safe to define locking ourselves.
 822          */
 823         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
 824                           &gfs2_quota_imutex_key);
 825
 826         error = gfs2_rindex_update(sdp);
 827         if (error)
 828                 goto fail_qinode;
 829
 830         return 0;
 831
 832 fail_qinode:
 833         iput(sdp->sd_quota_inode);
 834 fail_rindex:
 835         gfs2_clear_rgrpd(sdp);
 836         iput(sdp->sd_rindex);
 837 fail_statfs:
 838         iput(sdp->sd_statfs_inode);
 839 fail_journal:
 840         init_journal(sdp, UNDO);
 841 fail:
 842         return error;
 843 }
 844
 845 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 846 {
 847         struct inode *pn = NULL;
 848         char buf[30];
 849         int error = 0;
 850         struct gfs2_inode *ip;
 851         struct inode *master = d_inode(sdp->sd_master_dir);
 852
 853         if (sdp->sd_args.ar_spectator)
 854                 return 0;
 855
 856         if (undo)
 857                 goto fail_qc_gh;
 858
 859         pn = gfs2_lookup_simple(master, "per_node");
 860         if (IS_ERR(pn)) {
 861                 error = PTR_ERR(pn);
 862                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 863                 return error;
 864         }
 865
 866         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 867         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 868         if (IS_ERR(sdp->sd_sc_inode)) {
 869                 error = PTR_ERR(sdp->sd_sc_inode);
 870                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 871                 goto fail;
 872         }
 873
 874         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 875         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 876         if (IS_ERR(sdp->sd_qc_inode)) {
 877                 error = PTR_ERR(sdp->sd_qc_inode);
 878                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 879                 goto fail_ut_i;
 880         }
 881
 882         iput(pn);
 883         pn = NULL;
 884
 885         ip = GFS2_I(sdp->sd_sc_inode);
 886         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 887                                    &sdp->sd_sc_gh);
 888         if (error) {
 889                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 890                 goto fail_qc_i;
 891         }
 892
 893         ip = GFS2_I(sdp->sd_qc_inode);
 894         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 895                                    &sdp->sd_qc_gh);
 896         if (error) {
 897                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 898                 goto fail_ut_gh;
 899         }
 900
 901         return 0;
 902
 903 fail_qc_gh:
 904         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 905 fail_ut_gh:
 906         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 907 fail_qc_i:
 908         iput(sdp->sd_qc_inode);
 909 fail_ut_i:
 910         iput(sdp->sd_sc_inode);
 911 fail:
 912         if (pn)
 913                 iput(pn);
 914         return error;
 915 }
 916
 917 static const match_table_t nolock_tokens = {
 918         { Opt_jid, "jid=%d\n", },
 919         { Opt_err, NULL },
 920 };
 921
 922 static const struct lm_lockops nolock_ops = {
 923         .lm_proto_name = "lock_nolock",
 924         .lm_put_lock = gfs2_glock_free,
 925         .lm_tokens = &nolock_tokens,
 926 };
 927
 928 /**
 929  * gfs2_lm_mount - mount a locking protocol
 930  * @sdp: the filesystem
 931  * @args: mount arguments
 932  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 933  *
 934  * Returns: errno
 935  */
 936
 937 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 938 {
 939         const struct lm_lockops *lm;
 940         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 941         struct gfs2_args *args = &sdp->sd_args;
 942         const char *proto = sdp->sd_proto_name;
 943         const char *table = sdp->sd_table_name;
 944         char *o, *options;
 945         int ret;
 946
 947         if (!strcmp("lock_nolock", proto)) {
 948                 lm = &nolock_ops;
 949                 sdp->sd_args.ar_localflocks = 1;
 950 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 951         } else if (!strcmp("lock_dlm", proto)) {
 952                 lm = &gfs2_dlm_ops;
 953 #endif
 954         } else {
 955                 pr_info("can't find protocol %s\n", proto);
 956                 return -ENOENT;
 957         }
 958
 959         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 960
 961         ls->ls_ops = lm;
 962         ls->ls_first = 1;
 963
 964         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 965                 substring_t tmp[MAX_OPT_ARGS];
 966                 int token, option;
 967
 968                 if (!o || !*o)
 969                         continue;
 970
 971                 token = match_token(o, *lm->lm_tokens, tmp);
 972                 switch (token) {
 973                 case Opt_jid:
 974                         ret = match_int(&tmp[0], &option);
 975                         if (ret || option < 0)
 976                                 goto hostdata_error;
 977                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 978                                 ls->ls_jid = option;
 979                         break;
 980                 case Opt_id:
 981                 case Opt_nodir:
 982                         /* Obsolete, but left for backward compat purposes */
 983                         break;
 984                 case Opt_first:
 985                         ret = match_int(&tmp[0], &option);
 986                         if (ret || (option != 0 && option != 1))
 987                                 goto hostdata_error;
 988                         ls->ls_first = option;
 989                         break;
 990                 case Opt_err:
 991                 default:
 992 hostdata_error:
 993                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 994                         return -EINVAL;
 995                 }
 996         }
 997
 998         if (lm->lm_mount == NULL) {
 999                 fs_info(sdp, "Now mounting FS...\n");
1000                 complete_all(&sdp->sd_locking_init);
1001                 return 0;
1002         }
1003         ret = lm->lm_mount(sdp, table);
1004         if (ret == 0)
1005                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1006         complete_all(&sdp->sd_locking_init);
1007         return ret;
1008 }
1009
1010 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1011 {
1012         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1013         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1014             lm->lm_unmount)
1015                 lm->lm_unmount(sdp);
1016 }
1017
1018 static int wait_on_journal(struct gfs2_sbd *sdp)
1019 {
1020         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1021                 return 0;
1022
1023         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1024                 ? -EINTR : 0;
1025 }
1026
1027 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1028 {
1029         struct super_block *sb = sdp->sd_vfs;
1030         char ro[20];
1031         char spectator[20];
1032         char *envp[] = { ro, spectator, NULL };
1033         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1034         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1035         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1036 }
1037
1038 /**
1039  * fill_super - Read in superblock
1040  * @sb: The VFS superblock
1041  * @data: Mount options
1042  * @silent: Don't complain if it's not a GFS2 filesystem
1043  *
1044  * Returns: errno
1045  */
1046
1047 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1048 {
1049         struct gfs2_sbd *sdp;
1050         struct gfs2_holder mount_gh;
1051         int error;
1052
1053         sdp = init_sbd(sb);
1054         if (!sdp) {
1055                 pr_warn("can't alloc struct gfs2_sbd\n");
1056                 return -ENOMEM;
1057         }
1058         sdp->sd_args = *args;
1059
1060         if (sdp->sd_args.ar_spectator) {
1061                 sb->s_flags |= MS_RDONLY;
1062                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1063         }
1064         if (sdp->sd_args.ar_posix_acl)
1065                 sb->s_flags |= MS_POSIXACL;
1066         if (sdp->sd_args.ar_nobarrier)
1067                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1068
1069         sb->s_flags |= MS_NOSEC;
1070         sb->s_magic = GFS2_MAGIC;
1071         sb->s_op = &gfs2_super_ops;
1072         sb->s_d_op = &gfs2_dops;
1073         sb->s_export_op = &gfs2_export_ops;
1074         sb->s_xattr = gfs2_xattr_handlers;
1075         sb->s_qcop = &gfs2_quotactl_ops;
1076         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1077         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1078         sb->s_time_gran = 1;
1079         sb->s_maxbytes = MAX_LFS_FILESIZE;
1080
1081         /* Set up the buffer cache and fill in some fake block size values
1082            to allow us to read-in the on-disk superblock. */
1083         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1084         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1085         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1086                                GFS2_BASIC_BLOCK_SHIFT;
1087         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1088
1089         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1090         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1091         if (sdp->sd_args.ar_statfs_quantum) {
1092                 sdp->sd_tune.gt_statfs_slow = 0;
1093                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1094         } else {
1095                 sdp->sd_tune.gt_statfs_slow = 1;
1096                 sdp->sd_tune.gt_statfs_quantum = 30;
1097         }
1098
1099         error = init_names(sdp, silent);
1100         if (error) {
1101                 /* In this case, we haven't initialized sysfs, so we have to
1102                    manually free the sdp. */
1103                 free_percpu(sdp->sd_lkstats);
1104                 kfree(sdp);
1105                 sb->s_fs_info = NULL;
1106                 return error;
1107         }
1108
1109         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1110
1111         error = gfs2_sys_fs_add(sdp);
1112         /*
1113          * If we hit an error here, gfs2_sys_fs_add will have called function
1114          * kobject_put which causes the sysfs usage count to go to zero, which
1115          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1116          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1117          * kobject_put to free sdp.
1118          */
1119         if (error)
1120                 return error;
1121
1122         gfs2_create_debugfs_file(sdp);
1123
1124         error = gfs2_lm_mount(sdp, silent);
1125         if (error)
1126                 goto fail_debug;
1127
1128         error = init_locking(sdp, &mount_gh, DO);
1129         if (error)
1130                 goto fail_lm;
1131
1132         error = init_sb(sdp, silent);
1133         if (error)
1134                 goto fail_locking;
1135
1136         error = wait_on_journal(sdp);
1137         if (error)
1138                 goto fail_sb;
1139
1140         /*
1141          * If user space has failed to join the cluster or some similar
1142          * failure has occurred, then the journal id will contain a
1143          * negative (error) number. This will then be returned to the
1144          * caller (of the mount syscall). We do this even for spectator
1145          * mounts (which just write a jid of 0 to indicate "ok" even though
1146          * the jid is unused in the spectator case)
1147          */
1148         if (sdp->sd_lockstruct.ls_jid < 0) {
1149                 error = sdp->sd_lockstruct.ls_jid;
1150                 sdp->sd_lockstruct.ls_jid = 0;
1151                 goto fail_sb;
1152         }
1153
1154         if (sdp->sd_args.ar_spectator)
1155                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1156                          sdp->sd_table_name);
1157         else
1158                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1159                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1160
1161         error = init_inodes(sdp, DO);
1162         if (error)
1163                 goto fail_sb;
1164
1165         error = init_per_node(sdp, DO);
1166         if (error)
1167                 goto fail_inodes;
1168
1169         error = gfs2_statfs_init(sdp);
1170         if (error) {
1171                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1172                 goto fail_per_node;
1173         }
1174
1175         if (!(sb->s_flags & MS_RDONLY)) {
1176                 error = gfs2_make_fs_rw(sdp);
1177                 if (error) {
1178                         fs_err(sdp, "can't make FS RW: %d\n", error);
1179                         goto fail_per_node;
1180                 }
1181         }
1182
1183         gfs2_glock_dq_uninit(&mount_gh);
1184         gfs2_online_uevent(sdp);
1185         return 0;
1186
1187 fail_per_node:
1188         init_per_node(sdp, UNDO);
1189 fail_inodes:
1190         init_inodes(sdp, UNDO);
1191 fail_sb:
1192         if (sdp->sd_root_dir)
1193                 dput(sdp->sd_root_dir);
1194         if (sdp->sd_master_dir)
1195                 dput(sdp->sd_master_dir);
1196         if (sb->s_root)
1197                 dput(sb->s_root);
1198         sb->s_root = NULL;
1199 fail_locking:
1200         init_locking(sdp, &mount_gh, UNDO);
1201 fail_lm:
1202         complete_all(&sdp->sd_journal_ready);
1203         gfs2_gl_hash_clear(sdp);
1204         gfs2_lm_unmount(sdp);
1205 fail_debug:
1206         gfs2_delete_debugfs_file(sdp);
1207         free_percpu(sdp->sd_lkstats);
1208         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1209          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1210         gfs2_sys_fs_del(sdp);
1211         sb->s_fs_info = NULL;
1212         return error;
1213 }
1214
1215 static int set_gfs2_super(struct super_block *s, void *data)
1216 {
1217         s->s_bdev = data;
1218         s->s_dev = s->s_bdev->bd_dev;
1219
1220         /*
1221          * We set the bdi here to the queue backing, file systems can
1222          * overwrite this in ->fill_super()
1223          */
1224         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1225         return 0;
1226 }
1227
1228 static int test_gfs2_super(struct super_block *s, void *ptr)
1229 {
1230         struct block_device *bdev = ptr;
1231         return (bdev == s->s_bdev);
1232 }
1233
1234 /**
1235  * gfs2_mount - Get the GFS2 superblock
1236  * @fs_type: The GFS2 filesystem type
1237  * @flags: Mount flags
1238  * @dev_name: The name of the device
1239  * @data: The mount arguments
1240  *
1241  * Q. Why not use get_sb_bdev() ?
1242  * A. We need to select one of two root directories to mount, independent
1243  *    of whether this is the initial, or subsequent, mount of this sb
1244  *
1245  * Returns: 0 or -ve on error
1246  */
1247
1248 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1249                        const char *dev_name, void *data)
1250 {
1251         struct block_device *bdev;
1252         struct super_block *s;
1253         fmode_t mode = FMODE_READ | FMODE_EXCL;
1254         int error;
1255         struct gfs2_args args;
1256         struct gfs2_sbd *sdp;
1257
1258         if (!(flags & MS_RDONLY))
1259                 mode |= FMODE_WRITE;
1260
1261         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1262         if (IS_ERR(bdev))
1263                 return ERR_CAST(bdev);
1264
1265         /*
1266          * once the super is inserted into the list by sget, s_umount
1267          * will protect the lockfs code from trying to start a snapshot
1268          * while we are mounting
1269          */
1270         mutex_lock(&bdev->bd_fsfreeze_mutex);
1271         if (bdev->bd_fsfreeze_count > 0) {
1272                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1273                 error = -EBUSY;
1274                 goto error_bdev;
1275         }
1276         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1277         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1278         error = PTR_ERR(s);
1279         if (IS_ERR(s))
1280                 goto error_bdev;
1281
1282         if (s->s_root) {
1283                 /*
1284                  * s_umount nests inside bd_mutex during
1285                  * __invalidate_device().  blkdev_put() acquires
1286                  * bd_mutex and can't be called under s_umount.  Drop
1287                  * s_umount temporarily.  This is safe as we're
1288                  * holding an active reference.
1289                  */
1290                 up_write(&s->s_umount);
1291                 blkdev_put(bdev, mode);
1292                 down_write(&s->s_umount);
1293         }
1294
1295         memset(&args, 0, sizeof(args));
1296         args.ar_quota = GFS2_QUOTA_DEFAULT;
1297         args.ar_data = GFS2_DATA_DEFAULT;
1298         args.ar_commit = 30;
1299         args.ar_statfs_quantum = 30;
1300         args.ar_quota_quantum = 60;
1301         args.ar_errors = GFS2_ERRORS_DEFAULT;
1302
1303         error = gfs2_mount_args(&args, data);
1304         if (error) {
1305                 pr_warn("can't parse mount arguments\n");
1306                 goto error_super;
1307         }
1308
1309         if (s->s_root) {
1310                 error = -EBUSY;
1311                 if ((flags ^ s->s_flags) & MS_RDONLY)
1312                         goto error_super;
1313         } else {
1314                 char b[BDEVNAME_SIZE];
1315
1316                 s->s_mode = mode;
1317                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1318                 sb_set_blocksize(s, block_size(bdev));
1319                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1320                 if (error)
1321                         goto error_super;
1322                 s->s_flags |= MS_ACTIVE;
1323                 bdev->bd_super = s;
1324         }
1325
1326         sdp = s->s_fs_info;
1327         if (args.ar_meta)
1328                 return dget(sdp->sd_master_dir);
1329         else
1330                 return dget(sdp->sd_root_dir);
1331
1332 error_super:
1333         deactivate_locked_super(s);
1334         return ERR_PTR(error);
1335 error_bdev:
1336         blkdev_put(bdev, mode);
1337         return ERR_PTR(error);
1338 }
1339
1340 static int set_meta_super(struct super_block *s, void *ptr)
1341 {
1342         return -EINVAL;
1343 }
1344
1345 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1346                         int flags, const char *dev_name, void *data)
1347 {
1348         struct super_block *s;
1349         struct gfs2_sbd *sdp;
1350         struct path path;
1351         int error;
1352
1353         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1354         if (error) {
1355                 pr_warn("path_lookup on %s returned error %d\n",
1356                         dev_name, error);
1357                 return ERR_PTR(error);
1358         }
1359         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1360                  d_inode(path.dentry)->i_sb->s_bdev);
1361         path_put(&path);
1362         if (IS_ERR(s)) {
1363                 pr_warn("gfs2 mount does not exist\n");
1364                 return ERR_CAST(s);
1365         }
1366         if ((flags ^ s->s_flags) & MS_RDONLY) {
1367                 deactivate_locked_super(s);
1368                 return ERR_PTR(-EBUSY);
1369         }
1370         sdp = s->s_fs_info;
1371         return dget(sdp->sd_master_dir);
1372 }
1373
1374 static void gfs2_kill_sb(struct super_block *sb)
1375 {
1376         struct gfs2_sbd *sdp = sb->s_fs_info;
1377
1378         if (sdp == NULL) {
1379                 kill_block_super(sb);
1380                 return;
1381         }
1382
1383         gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1384         dput(sdp->sd_root_dir);
1385         dput(sdp->sd_master_dir);
1386         sdp->sd_root_dir = NULL;
1387         sdp->sd_master_dir = NULL;
1388         shrink_dcache_sb(sb);
1389         gfs2_delete_debugfs_file(sdp);
1390         free_percpu(sdp->sd_lkstats);
1391         kill_block_super(sb);
1392 }
1393
1394 struct file_system_type gfs2_fs_type = {
1395         .name = "gfs2",
1396         .fs_flags = FS_REQUIRES_DEV,
1397         .mount = gfs2_mount,
1398         .kill_sb = gfs2_kill_sb,
1399         .owner = THIS_MODULE,
1400 };
1401 MODULE_ALIAS_FS("gfs2");
1402
1403 struct file_system_type gfs2meta_fs_type = {
1404         .name = "gfs2meta",
1405         .fs_flags = FS_REQUIRES_DEV,
1406         .mount = gfs2_mount_meta,
1407         .owner = THIS_MODULE,
1408 };
1409 MODULE_ALIAS_FS("gfs2meta");