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)
 175 {
 176         struct page *page = bio->bi_private;
 177
 178         if (!bio->bi_error)
 179                 SetPageUptodate(page);
 180         else
 181                 pr_warn("error %d reading superblock\n", bio->bi_error);
 182         unlock_page(page);
 183 }
 184
 185 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 186 {
 187         struct gfs2_sb_host *sb = &sdp->sd_sb;
 188         struct super_block *s = sdp->sd_vfs;
 189         const struct gfs2_sb *str = buf;
 190
 191         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 192         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 193         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 194         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 195         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 196         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 197         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 198         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 199         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 200         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 201         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 202
 203         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 204         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 205         memcpy(s->s_uuid, str->sb_uuid, 16);
 206 }
 207
 208 /**
 209  * gfs2_read_super - Read the gfs2 super block from disk
 210  * @sdp: The GFS2 super block
 211  * @sector: The location of the super block
 212  * @error: The error code to return
 213  *
 214  * This uses the bio functions to read the super block from disk
 215  * because we want to be 100% sure that we never read cached data.
 216  * A super block is read twice only during each GFS2 mount and is
 217  * never written to by the filesystem. The first time its read no
 218  * locks are held, and the only details which are looked at are those
 219  * relating to the locking protocol. Once locking is up and working,
 220  * the sb is read again under the lock to establish the location of
 221  * the master directory (contains pointers to journals etc) and the
 222  * root directory.
 223  *
 224  * Returns: 0 on success or error
 225  */
 226
 227 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 228 {
 229         struct super_block *sb = sdp->sd_vfs;
 230         struct gfs2_sb *p;
 231         struct page *page;
 232         struct bio *bio;
 233
 234         page = alloc_page(GFP_NOFS);
 235         if (unlikely(!page))
 236                 return -ENOMEM;
 237
 238         ClearPageUptodate(page);
 239         ClearPageDirty(page);
 240         lock_page(page);
 241
 242         bio = bio_alloc(GFP_NOFS, 1);
 243         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 244         bio->bi_bdev = sb->s_bdev;
 245         bio_add_page(bio, page, PAGE_SIZE, 0);
 246
 247         bio->bi_end_io = end_bio_io_page;
 248         bio->bi_private = page;
 249         submit_bio(READ_SYNC | REQ_META, bio);
 250         wait_on_page_locked(page);
 251         bio_put(bio);
 252         if (!PageUptodate(page)) {
 253                 __free_page(page);
 254                 return -EIO;
 255         }
 256         p = kmap(page);
 257         gfs2_sb_in(sdp, p);
 258         kunmap(page);
 259         __free_page(page);
 260         return gfs2_check_sb(sdp, silent);
 261 }
 262
 263 /**
 264  * gfs2_read_sb - Read super block
 265  * @sdp: The GFS2 superblock
 266  * @silent: Don't print message if mount fails
 267  *
 268  */
 269
 270 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 271 {
 272         u32 hash_blocks, ind_blocks, leaf_blocks;
 273         u32 tmp_blocks;
 274         unsigned int x;
 275         int error;
 276
 277         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 278         if (error) {
 279                 if (!silent)
 280                         fs_err(sdp, "can't read superblock\n");
 281                 return error;
 282         }
 283
 284         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 285                                GFS2_BASIC_BLOCK_SHIFT;
 286         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 287         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 288                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 289         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 290                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 291         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 292         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 293         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 294         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 295         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 296                                 sizeof(struct gfs2_meta_header)) /
 297                                 sizeof(struct gfs2_quota_change);
 298         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 299                                      sizeof(struct gfs2_meta_header))
 300                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 301
 302         /* Compute maximum reservation required to add a entry to a directory */
 303
 304         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 305                              sdp->sd_jbsize);
 306
 307         ind_blocks = 0;
 308         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 309                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 310                 ind_blocks += tmp_blocks;
 311         }
 312
 313         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 314
 315         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 316
 317         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 318                                 sizeof(struct gfs2_dinode);
 319         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 320         for (x = 2;; x++) {
 321                 u64 space, d;
 322                 u32 m;
 323
 324                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 325                 d = space;
 326                 m = do_div(d, sdp->sd_inptrs);
 327
 328                 if (d != sdp->sd_heightsize[x - 1] || m)
 329                         break;
 330                 sdp->sd_heightsize[x] = space;
 331         }
 332         sdp->sd_max_height = x;
 333         sdp->sd_heightsize[x] = ~0;
 334         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 335
 336         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 337                                  sizeof(struct gfs2_dinode);
 338         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 339         for (x = 2;; x++) {
 340                 u64 space, d;
 341                 u32 m;
 342
 343                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 344                 d = space;
 345                 m = do_div(d, sdp->sd_inptrs);
 346
 347                 if (d != sdp->sd_jheightsize[x - 1] || m)
 348                         break;
 349                 sdp->sd_jheightsize[x] = space;
 350         }
 351         sdp->sd_max_jheight = x;
 352         sdp->sd_jheightsize[x] = ~0;
 353         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 354
 355         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         sdp->sd_log_idle = 1;
 760         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 761         gfs2_glock_dq_uninit(&ji_gh);
 762         jindex = 0;
 763         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 764         return 0;
 765
 766 fail_jinode_gh:
 767         if (!sdp->sd_args.ar_spectator)
 768                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 769 fail_journal_gh:
 770         if (!sdp->sd_args.ar_spectator)
 771                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 772 fail_jindex:
 773         gfs2_jindex_free(sdp);
 774         if (jindex)
 775                 gfs2_glock_dq_uninit(&ji_gh);
 776 fail:
 777         iput(sdp->sd_jindex);
 778         return error;
 779 }
 780
 781 static struct lock_class_key gfs2_quota_imutex_key;
 782
 783 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 784 {
 785         int error = 0;
 786         struct inode *master = d_inode(sdp->sd_master_dir);
 787
 788         if (undo)
 789                 goto fail_qinode;
 790
 791         error = init_journal(sdp, undo);
 792         complete_all(&sdp->sd_journal_ready);
 793         if (error)
 794                 goto fail;
 795
 796         /* Read in the master statfs inode */
 797         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 798         if (IS_ERR(sdp->sd_statfs_inode)) {
 799                 error = PTR_ERR(sdp->sd_statfs_inode);
 800                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 801                 goto fail_journal;
 802         }
 803
 804         /* Read in the resource index inode */
 805         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 806         if (IS_ERR(sdp->sd_rindex)) {
 807                 error = PTR_ERR(sdp->sd_rindex);
 808                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 809                 goto fail_statfs;
 810         }
 811         sdp->sd_rindex_uptodate = 0;
 812
 813         /* Read in the quota inode */
 814         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 815         if (IS_ERR(sdp->sd_quota_inode)) {
 816                 error = PTR_ERR(sdp->sd_quota_inode);
 817                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 818                 goto fail_rindex;
 819         }
 820         /*
 821          * i_mutex on quota files is special. Since this inode is hidden system
 822          * file, we are safe to define locking ourselves.
 823          */
 824         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
 825                           &gfs2_quota_imutex_key);
 826
 827         error = gfs2_rindex_update(sdp);
 828         if (error)
 829                 goto fail_qinode;
 830
 831         return 0;
 832
 833 fail_qinode:
 834         iput(sdp->sd_quota_inode);
 835 fail_rindex:
 836         gfs2_clear_rgrpd(sdp);
 837         iput(sdp->sd_rindex);
 838 fail_statfs:
 839         iput(sdp->sd_statfs_inode);
 840 fail_journal:
 841         init_journal(sdp, UNDO);
 842 fail:
 843         return error;
 844 }
 845
 846 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 847 {
 848         struct inode *pn = NULL;
 849         char buf[30];
 850         int error = 0;
 851         struct gfs2_inode *ip;
 852         struct inode *master = d_inode(sdp->sd_master_dir);
 853
 854         if (sdp->sd_args.ar_spectator)
 855                 return 0;
 856
 857         if (undo)
 858                 goto fail_qc_gh;
 859
 860         pn = gfs2_lookup_simple(master, "per_node");
 861         if (IS_ERR(pn)) {
 862                 error = PTR_ERR(pn);
 863                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 864                 return error;
 865         }
 866
 867         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 868         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 869         if (IS_ERR(sdp->sd_sc_inode)) {
 870                 error = PTR_ERR(sdp->sd_sc_inode);
 871                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 872                 goto fail;
 873         }
 874
 875         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 876         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 877         if (IS_ERR(sdp->sd_qc_inode)) {
 878                 error = PTR_ERR(sdp->sd_qc_inode);
 879                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 880                 goto fail_ut_i;
 881         }
 882
 883         iput(pn);
 884         pn = NULL;
 885
 886         ip = GFS2_I(sdp->sd_sc_inode);
 887         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 888                                    &sdp->sd_sc_gh);
 889         if (error) {
 890                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 891                 goto fail_qc_i;
 892         }
 893
 894         ip = GFS2_I(sdp->sd_qc_inode);
 895         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 896                                    &sdp->sd_qc_gh);
 897         if (error) {
 898                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 899                 goto fail_ut_gh;
 900         }
 901
 902         return 0;
 903
 904 fail_qc_gh:
 905         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 906 fail_ut_gh:
 907         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 908 fail_qc_i:
 909         iput(sdp->sd_qc_inode);
 910 fail_ut_i:
 911         iput(sdp->sd_sc_inode);
 912 fail:
 913         if (pn)
 914                 iput(pn);
 915         return error;
 916 }
 917
 918 static const match_table_t nolock_tokens = {
 919         { Opt_jid, "jid=%d\n", },
 920         { Opt_err, NULL },
 921 };
 922
 923 static const struct lm_lockops nolock_ops = {
 924         .lm_proto_name = "lock_nolock",
 925         .lm_put_lock = gfs2_glock_free,
 926         .lm_tokens = &nolock_tokens,
 927 };
 928
 929 /**
 930  * gfs2_lm_mount - mount a locking protocol
 931  * @sdp: the filesystem
 932  * @args: mount arguments
 933  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 934  *
 935  * Returns: errno
 936  */
 937
 938 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 939 {
 940         const struct lm_lockops *lm;
 941         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 942         struct gfs2_args *args = &sdp->sd_args;
 943         const char *proto = sdp->sd_proto_name;
 944         const char *table = sdp->sd_table_name;
 945         char *o, *options;
 946         int ret;
 947
 948         if (!strcmp("lock_nolock", proto)) {
 949                 lm = &nolock_ops;
 950                 sdp->sd_args.ar_localflocks = 1;
 951 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 952         } else if (!strcmp("lock_dlm", proto)) {
 953                 lm = &gfs2_dlm_ops;
 954 #endif
 955         } else {
 956                 pr_info("can't find protocol %s\n", proto);
 957                 return -ENOENT;
 958         }
 959
 960         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 961
 962         ls->ls_ops = lm;
 963         ls->ls_first = 1;
 964
 965         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 966                 substring_t tmp[MAX_OPT_ARGS];
 967                 int token, option;
 968
 969                 if (!o || !*o)
 970                         continue;
 971
 972                 token = match_token(o, *lm->lm_tokens, tmp);
 973                 switch (token) {
 974                 case Opt_jid:
 975                         ret = match_int(&tmp[0], &option);
 976                         if (ret || option < 0)
 977                                 goto hostdata_error;
 978                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 979                                 ls->ls_jid = option;
 980                         break;
 981                 case Opt_id:
 982                 case Opt_nodir:
 983                         /* Obsolete, but left for backward compat purposes */
 984                         break;
 985                 case Opt_first:
 986                         ret = match_int(&tmp[0], &option);
 987                         if (ret || (option != 0 && option != 1))
 988                                 goto hostdata_error;
 989                         ls->ls_first = option;
 990                         break;
 991                 case Opt_err:
 992                 default:
 993 hostdata_error:
 994                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 995                         return -EINVAL;
 996                 }
 997         }
 998
 999         if (lm->lm_mount == NULL) {
1000                 fs_info(sdp, "Now mounting FS...\n");
1001                 complete_all(&sdp->sd_locking_init);
1002                 return 0;
1003         }
1004         ret = lm->lm_mount(sdp, table);
1005         if (ret == 0)
1006                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1007         complete_all(&sdp->sd_locking_init);
1008         return ret;
1009 }
1010
1011 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1012 {
1013         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1014         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1015             lm->lm_unmount)
1016                 lm->lm_unmount(sdp);
1017 }
1018
1019 static int wait_on_journal(struct gfs2_sbd *sdp)
1020 {
1021         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1022                 return 0;
1023
1024         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1025                 ? -EINTR : 0;
1026 }
1027
1028 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1029 {
1030         struct super_block *sb = sdp->sd_vfs;
1031         char ro[20];
1032         char spectator[20];
1033         char *envp[] = { ro, spectator, NULL };
1034         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1035         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1036         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1037 }
1038
1039 /**
1040  * fill_super - Read in superblock
1041  * @sb: The VFS superblock
1042  * @data: Mount options
1043  * @silent: Don't complain if it's not a GFS2 filesystem
1044  *
1045  * Returns: errno
1046  */
1047
1048 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1049 {
1050         struct gfs2_sbd *sdp;
1051         struct gfs2_holder mount_gh;
1052         int error;
1053
1054         sdp = init_sbd(sb);
1055         if (!sdp) {
1056                 pr_warn("can't alloc struct gfs2_sbd\n");
1057                 return -ENOMEM;
1058         }
1059         sdp->sd_args = *args;
1060
1061         if (sdp->sd_args.ar_spectator) {
1062                 sb->s_flags |= MS_RDONLY;
1063                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1064         }
1065         if (sdp->sd_args.ar_posix_acl)
1066                 sb->s_flags |= MS_POSIXACL;
1067         if (sdp->sd_args.ar_nobarrier)
1068                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1069
1070         sb->s_flags |= MS_NOSEC;
1071         sb->s_magic = GFS2_MAGIC;
1072         sb->s_op = &gfs2_super_ops;
1073         sb->s_d_op = &gfs2_dops;
1074         sb->s_export_op = &gfs2_export_ops;
1075         sb->s_xattr = gfs2_xattr_handlers;
1076         sb->s_qcop = &gfs2_quotactl_ops;
1077         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1078         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1079         sb->s_time_gran = 1;
1080         sb->s_maxbytes = MAX_LFS_FILESIZE;
1081
1082         /* Set up the buffer cache and fill in some fake block size values
1083            to allow us to read-in the on-disk superblock. */
1084         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1085         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1086         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1087                                GFS2_BASIC_BLOCK_SHIFT;
1088         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1089
1090         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1091         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1092         if (sdp->sd_args.ar_statfs_quantum) {
1093                 sdp->sd_tune.gt_statfs_slow = 0;
1094                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1095         } else {
1096                 sdp->sd_tune.gt_statfs_slow = 1;
1097                 sdp->sd_tune.gt_statfs_quantum = 30;
1098         }
1099
1100         error = init_names(sdp, silent);
1101         if (error) {
1102                 /* In this case, we haven't initialized sysfs, so we have to
1103                    manually free the sdp. */
1104                 free_percpu(sdp->sd_lkstats);
1105                 kfree(sdp);
1106                 sb->s_fs_info = NULL;
1107                 return error;
1108         }
1109
1110         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1111
1112         error = gfs2_sys_fs_add(sdp);
1113         /*
1114          * If we hit an error here, gfs2_sys_fs_add will have called function
1115          * kobject_put which causes the sysfs usage count to go to zero, which
1116          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1117          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1118          * kobject_put to free sdp.
1119          */
1120         if (error)
1121                 return error;
1122
1123         gfs2_create_debugfs_file(sdp);
1124
1125         error = gfs2_lm_mount(sdp, silent);
1126         if (error)
1127                 goto fail_debug;
1128
1129         error = init_locking(sdp, &mount_gh, DO);
1130         if (error)
1131                 goto fail_lm;
1132
1133         error = init_sb(sdp, silent);
1134         if (error)
1135                 goto fail_locking;
1136
1137         error = wait_on_journal(sdp);
1138         if (error)
1139                 goto fail_sb;
1140
1141         /*
1142          * If user space has failed to join the cluster or some similar
1143          * failure has occurred, then the journal id will contain a
1144          * negative (error) number. This will then be returned to the
1145          * caller (of the mount syscall). We do this even for spectator
1146          * mounts (which just write a jid of 0 to indicate "ok" even though
1147          * the jid is unused in the spectator case)
1148          */
1149         if (sdp->sd_lockstruct.ls_jid < 0) {
1150                 error = sdp->sd_lockstruct.ls_jid;
1151                 sdp->sd_lockstruct.ls_jid = 0;
1152                 goto fail_sb;
1153         }
1154
1155         if (sdp->sd_args.ar_spectator)
1156                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1157                          sdp->sd_table_name);
1158         else
1159                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1160                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1161
1162         error = init_inodes(sdp, DO);
1163         if (error)
1164                 goto fail_sb;
1165
1166         error = init_per_node(sdp, DO);
1167         if (error)
1168                 goto fail_inodes;
1169
1170         error = gfs2_statfs_init(sdp);
1171         if (error) {
1172                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1173                 goto fail_per_node;
1174         }
1175
1176         if (!(sb->s_flags & MS_RDONLY)) {
1177                 error = gfs2_make_fs_rw(sdp);
1178                 if (error) {
1179                         fs_err(sdp, "can't make FS RW: %d\n", error);
1180                         goto fail_per_node;
1181                 }
1182         }
1183
1184         gfs2_glock_dq_uninit(&mount_gh);
1185         gfs2_online_uevent(sdp);
1186         return 0;
1187
1188 fail_per_node:
1189         init_per_node(sdp, UNDO);
1190 fail_inodes:
1191         init_inodes(sdp, UNDO);
1192 fail_sb:
1193         if (sdp->sd_root_dir)
1194                 dput(sdp->sd_root_dir);
1195         if (sdp->sd_master_dir)
1196                 dput(sdp->sd_master_dir);
1197         if (sb->s_root)
1198                 dput(sb->s_root);
1199         sb->s_root = NULL;
1200 fail_locking:
1201         init_locking(sdp, &mount_gh, UNDO);
1202 fail_lm:
1203         complete_all(&sdp->sd_journal_ready);
1204         gfs2_gl_hash_clear(sdp);
1205         gfs2_lm_unmount(sdp);
1206 fail_debug:
1207         gfs2_delete_debugfs_file(sdp);
1208         free_percpu(sdp->sd_lkstats);
1209         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1210          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1211         gfs2_sys_fs_del(sdp);
1212         sb->s_fs_info = NULL;
1213         return error;
1214 }
1215
1216 static int set_gfs2_super(struct super_block *s, void *data)
1217 {
1218         s->s_bdev = data;
1219         s->s_dev = s->s_bdev->bd_dev;
1220
1221         /*
1222          * We set the bdi here to the queue backing, file systems can
1223          * overwrite this in ->fill_super()
1224          */
1225         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1226         return 0;
1227 }
1228
1229 static int test_gfs2_super(struct super_block *s, void *ptr)
1230 {
1231         struct block_device *bdev = ptr;
1232         return (bdev == s->s_bdev);
1233 }
1234
1235 /**
1236  * gfs2_mount - Get the GFS2 superblock
1237  * @fs_type: The GFS2 filesystem type
1238  * @flags: Mount flags
1239  * @dev_name: The name of the device
1240  * @data: The mount arguments
1241  *
1242  * Q. Why not use get_sb_bdev() ?
1243  * A. We need to select one of two root directories to mount, independent
1244  *    of whether this is the initial, or subsequent, mount of this sb
1245  *
1246  * Returns: 0 or -ve on error
1247  */
1248
1249 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1250                        const char *dev_name, void *data)
1251 {
1252         struct block_device *bdev;
1253         struct super_block *s;
1254         fmode_t mode = FMODE_READ | FMODE_EXCL;
1255         int error;
1256         struct gfs2_args args;
1257         struct gfs2_sbd *sdp;
1258
1259         if (!(flags & MS_RDONLY))
1260                 mode |= FMODE_WRITE;
1261
1262         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1263         if (IS_ERR(bdev))
1264                 return ERR_CAST(bdev);
1265
1266         /*
1267          * once the super is inserted into the list by sget, s_umount
1268          * will protect the lockfs code from trying to start a snapshot
1269          * while we are mounting
1270          */
1271         mutex_lock(&bdev->bd_fsfreeze_mutex);
1272         if (bdev->bd_fsfreeze_count > 0) {
1273                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1274                 error = -EBUSY;
1275                 goto error_bdev;
1276         }
1277         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1278         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1279         error = PTR_ERR(s);
1280         if (IS_ERR(s))
1281                 goto error_bdev;
1282
1283         if (s->s_root) {
1284                 /*
1285                  * s_umount nests inside bd_mutex during
1286                  * __invalidate_device().  blkdev_put() acquires
1287                  * bd_mutex and can't be called under s_umount.  Drop
1288                  * s_umount temporarily.  This is safe as we're
1289                  * holding an active reference.
1290                  */
1291                 up_write(&s->s_umount);
1292                 blkdev_put(bdev, mode);
1293                 down_write(&s->s_umount);
1294         } else {
1295                 /* s_mode must be set before deactivate_locked_super calls */
1296                 s->s_mode = mode;
1297         }
1298
1299         memset(&args, 0, sizeof(args));
1300         args.ar_quota = GFS2_QUOTA_DEFAULT;
1301         args.ar_data = GFS2_DATA_DEFAULT;
1302         args.ar_commit = 30;
1303         args.ar_statfs_quantum = 30;
1304         args.ar_quota_quantum = 60;
1305         args.ar_errors = GFS2_ERRORS_DEFAULT;
1306
1307         error = gfs2_mount_args(&args, data);
1308         if (error) {
1309                 pr_warn("can't parse mount arguments\n");
1310                 goto error_super;
1311         }
1312
1313         if (s->s_root) {
1314                 error = -EBUSY;
1315                 if ((flags ^ s->s_flags) & MS_RDONLY)
1316                         goto error_super;
1317         } else {
1318                 char b[BDEVNAME_SIZE];
1319
1320                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1321                 sb_set_blocksize(s, block_size(bdev));
1322                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1323                 if (error)
1324                         goto error_super;
1325                 s->s_flags |= MS_ACTIVE;
1326                 bdev->bd_super = s;
1327         }
1328
1329         sdp = s->s_fs_info;
1330         if (args.ar_meta)
1331                 return dget(sdp->sd_master_dir);
1332         else
1333                 return dget(sdp->sd_root_dir);
1334
1335 error_super:
1336         deactivate_locked_super(s);
1337         return ERR_PTR(error);
1338 error_bdev:
1339         blkdev_put(bdev, mode);
1340         return ERR_PTR(error);
1341 }
1342
1343 static int set_meta_super(struct super_block *s, void *ptr)
1344 {
1345         return -EINVAL;
1346 }
1347
1348 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1349                         int flags, const char *dev_name, void *data)
1350 {
1351         struct super_block *s;
1352         struct gfs2_sbd *sdp;
1353         struct path path;
1354         int error;
1355
1356         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1357         if (error) {
1358                 pr_warn("path_lookup on %s returned error %d\n",
1359                         dev_name, error);
1360                 return ERR_PTR(error);
1361         }
1362         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1363                  d_inode(path.dentry)->i_sb->s_bdev);
1364         path_put(&path);
1365         if (IS_ERR(s)) {
1366                 pr_warn("gfs2 mount does not exist\n");
1367                 return ERR_CAST(s);
1368         }
1369         if ((flags ^ s->s_flags) & MS_RDONLY) {
1370                 deactivate_locked_super(s);
1371                 return ERR_PTR(-EBUSY);
1372         }
1373         sdp = s->s_fs_info;
1374         return dget(sdp->sd_master_dir);
1375 }
1376
1377 static void gfs2_kill_sb(struct super_block *sb)
1378 {
1379         struct gfs2_sbd *sdp = sb->s_fs_info;
1380
1381         if (sdp == NULL) {
1382                 kill_block_super(sb);
1383                 return;
1384         }
1385
1386         gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1387         dput(sdp->sd_root_dir);
1388         dput(sdp->sd_master_dir);
1389         sdp->sd_root_dir = NULL;
1390         sdp->sd_master_dir = NULL;
1391         shrink_dcache_sb(sb);
1392         gfs2_delete_debugfs_file(sdp);
1393         free_percpu(sdp->sd_lkstats);
1394         kill_block_super(sb);
1395 }
1396
1397 struct file_system_type gfs2_fs_type = {
1398         .name = "gfs2",
1399         .fs_flags = FS_REQUIRES_DEV,
1400         .mount = gfs2_mount,
1401         .kill_sb = gfs2_kill_sb,
1402         .owner = THIS_MODULE,
1403 };
1404 MODULE_ALIAS_FS("gfs2");
1405
1406 struct file_system_type gfs2meta_fs_type = {
1407         .name = "gfs2meta",
1408         .fs_flags = FS_REQUIRES_DEV,
1409         .mount = gfs2_mount_meta,
1410         .owner = THIS_MODULE,
1411 };
1412 MODULE_ALIAS_FS("gfs2meta");