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 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/kthread.h>
  17 #include <linux/namei.h>
  18 #include <linux/mount.h>
  19 #include <linux/gfs2_ondisk.h>
  20 #include <linux/quotaops.h>
  21
  22 #include "gfs2.h"
  23 #include "incore.h"
  24 #include "bmap.h"
  25 #include "glock.h"
  26 #include "glops.h"
  27 #include "inode.h"
  28 #include "recovery.h"
  29 #include "rgrp.h"
  30 #include "super.h"
  31 #include "sys.h"
  32 #include "util.h"
  33 #include "log.h"
  34 #include "quota.h"
  35 #include "dir.h"
  36 #include "trace_gfs2.h"
  37
  38 #define DO 0
  39 #define UNDO 1
  40
  41 /**
  42  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  43  * @gt: tune
  44  *
  45  */
  46
  47 static void gfs2_tune_init(struct gfs2_tune *gt)
  48 {
  49         spin_lock_init(&gt->gt_spin);
  50
  51         gt->gt_quota_simul_sync = 64;
  52         gt->gt_quota_warn_period = 10;
  53         gt->gt_quota_scale_num = 1;
  54         gt->gt_quota_scale_den = 1;
  55         gt->gt_new_files_jdata = 0;
  56         gt->gt_max_readahead = 1 << 18;
  57         gt->gt_complain_secs = 10;
  58 }
  59
  60 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  61 {
  62         struct gfs2_sbd *sdp;
  63
  64         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  65         if (!sdp)
  66                 return NULL;
  67
  68         sb->s_fs_info = sdp;
  69         sdp->sd_vfs = sb;
  70         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  71         gfs2_tune_init(&sdp->sd_tune);
  72
  73         init_waitqueue_head(&sdp->sd_glock_wait);
  74         atomic_set(&sdp->sd_glock_disposal, 0);
  75         spin_lock_init(&sdp->sd_statfs_spin);
  76
  77         spin_lock_init(&sdp->sd_rindex_spin);
  78         mutex_init(&sdp->sd_rindex_mutex);
  79         INIT_LIST_HEAD(&sdp->sd_rindex_list);
  80         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
  81
  82         INIT_LIST_HEAD(&sdp->sd_jindex_list);
  83         spin_lock_init(&sdp->sd_jindex_spin);
  84         mutex_init(&sdp->sd_jindex_mutex);
  85
  86         INIT_LIST_HEAD(&sdp->sd_quota_list);
  87         mutex_init(&sdp->sd_quota_mutex);
  88         init_waitqueue_head(&sdp->sd_quota_wait);
  89         INIT_LIST_HEAD(&sdp->sd_trunc_list);
  90         spin_lock_init(&sdp->sd_trunc_lock);
  91
  92         spin_lock_init(&sdp->sd_log_lock);
  93         atomic_set(&sdp->sd_log_pinned, 0);
  94         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
  95         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
  96         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
  97         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
  98         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
  99
 100         init_waitqueue_head(&sdp->sd_log_waitq);
 101         init_waitqueue_head(&sdp->sd_logd_waitq);
 102         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 103         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 104
 105         init_rwsem(&sdp->sd_log_flush_lock);
 106         atomic_set(&sdp->sd_log_in_flight, 0);
 107         init_waitqueue_head(&sdp->sd_log_flush_wait);
 108
 109         INIT_LIST_HEAD(&sdp->sd_revoke_list);
 110
 111         mutex_init(&sdp->sd_freeze_lock);
 112
 113         return sdp;
 114 }
 115
 116
 117 /**
 118  * gfs2_check_sb - Check superblock
 119  * @sdp: the filesystem
 120  * @sb: The superblock
 121  * @silent: Don't print a message if the check fails
 122  *
 123  * Checks the version code of the FS is one that we understand how to
 124  * read and that the sizes of the various on-disk structures have not
 125  * changed.
 126  */
 127
 128 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
 129 {
 130         if (sb->sb_magic != GFS2_MAGIC ||
 131             sb->sb_type != GFS2_METATYPE_SB) {
 132                 if (!silent)
 133                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 134                 return -EINVAL;
 135         }
 136
 137         /*  If format numbers match exactly, we're done.  */
 138
 139         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 140             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 141                 return 0;
 142
 143         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 144
 145         return -EINVAL;
 146 }
 147
 148 static void end_bio_io_page(struct bio *bio, int error)
 149 {
 150         struct page *page = bio->bi_private;
 151
 152         if (!error)
 153                 SetPageUptodate(page);
 154         else
 155                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 156         unlock_page(page);
 157 }
 158
 159 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 160 {
 161         const struct gfs2_sb *str = buf;
 162
 163         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 164         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 165         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 166         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 167         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 168         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 169         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 170         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 171         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 172         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 173         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 174
 175         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 176         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 177         memcpy(sb->sb_uuid, str->sb_uuid, 16);
 178 }
 179
 180 /**
 181  * gfs2_read_super - Read the gfs2 super block from disk
 182  * @sdp: The GFS2 super block
 183  * @sector: The location of the super block
 184  * @error: The error code to return
 185  *
 186  * This uses the bio functions to read the super block from disk
 187  * because we want to be 100% sure that we never read cached data.
 188  * A super block is read twice only during each GFS2 mount and is
 189  * never written to by the filesystem. The first time its read no
 190  * locks are held, and the only details which are looked at are those
 191  * relating to the locking protocol. Once locking is up and working,
 192  * the sb is read again under the lock to establish the location of
 193  * the master directory (contains pointers to journals etc) and the
 194  * root directory.
 195  *
 196  * Returns: 0 on success or error
 197  */
 198
 199 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 200 {
 201         struct super_block *sb = sdp->sd_vfs;
 202         struct gfs2_sb *p;
 203         struct page *page;
 204         struct bio *bio;
 205
 206         page = alloc_page(GFP_NOFS);
 207         if (unlikely(!page))
 208                 return -ENOBUFS;
 209
 210         ClearPageUptodate(page);
 211         ClearPageDirty(page);
 212         lock_page(page);
 213
 214         bio = bio_alloc(GFP_NOFS, 1);
 215         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 216         bio->bi_bdev = sb->s_bdev;
 217         bio_add_page(bio, page, PAGE_SIZE, 0);
 218
 219         bio->bi_end_io = end_bio_io_page;
 220         bio->bi_private = page;
 221         submit_bio(READ_SYNC | REQ_META, bio);
 222         wait_on_page_locked(page);
 223         bio_put(bio);
 224         if (!PageUptodate(page)) {
 225                 __free_page(page);
 226                 return -EIO;
 227         }
 228         p = kmap(page);
 229         gfs2_sb_in(&sdp->sd_sb, p);
 230         kunmap(page);
 231         __free_page(page);
 232         return 0;
 233 }
 234
 235 /**
 236  * gfs2_read_sb - Read super block
 237  * @sdp: The GFS2 superblock
 238  * @silent: Don't print message if mount fails
 239  *
 240  */
 241
 242 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 243 {
 244         u32 hash_blocks, ind_blocks, leaf_blocks;
 245         u32 tmp_blocks;
 246         unsigned int x;
 247         int error;
 248
 249         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 250         if (error) {
 251                 if (!silent)
 252                         fs_err(sdp, "can't read superblock\n");
 253                 return error;
 254         }
 255
 256         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 257         if (error)
 258                 return error;
 259
 260         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 261                                GFS2_BASIC_BLOCK_SHIFT;
 262         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 263         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 264                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 265         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 266                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 267         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 268         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 269         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 270         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 271         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 272                                 sizeof(struct gfs2_meta_header)) /
 273                                 sizeof(struct gfs2_quota_change);
 274
 275         /* Compute maximum reservation required to add a entry to a directory */
 276
 277         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 278                              sdp->sd_jbsize);
 279
 280         ind_blocks = 0;
 281         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 282                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 283                 ind_blocks += tmp_blocks;
 284         }
 285
 286         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 287
 288         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 289
 290         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 291                                 sizeof(struct gfs2_dinode);
 292         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 293         for (x = 2;; x++) {
 294                 u64 space, d;
 295                 u32 m;
 296
 297                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 298                 d = space;
 299                 m = do_div(d, sdp->sd_inptrs);
 300
 301                 if (d != sdp->sd_heightsize[x - 1] || m)
 302                         break;
 303                 sdp->sd_heightsize[x] = space;
 304         }
 305         sdp->sd_max_height = x;
 306         sdp->sd_heightsize[x] = ~0;
 307         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 308
 309         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 310                                  sizeof(struct gfs2_dinode);
 311         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 312         for (x = 2;; x++) {
 313                 u64 space, d;
 314                 u32 m;
 315
 316                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 317                 d = space;
 318                 m = do_div(d, sdp->sd_inptrs);
 319
 320                 if (d != sdp->sd_jheightsize[x - 1] || m)
 321                         break;
 322                 sdp->sd_jheightsize[x] = space;
 323         }
 324         sdp->sd_max_jheight = x;
 325         sdp->sd_jheightsize[x] = ~0;
 326         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 327
 328         return 0;
 329 }
 330
 331 static int init_names(struct gfs2_sbd *sdp, int silent)
 332 {
 333         char *proto, *table;
 334         int error = 0;
 335
 336         proto = sdp->sd_args.ar_lockproto;
 337         table = sdp->sd_args.ar_locktable;
 338
 339         /*  Try to autodetect  */
 340
 341         if (!proto[0] || !table[0]) {
 342                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 343                 if (error)
 344                         return error;
 345
 346                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 347                 if (error)
 348                         goto out;
 349
 350                 if (!proto[0])
 351                         proto = sdp->sd_sb.sb_lockproto;
 352                 if (!table[0])
 353                         table = sdp->sd_sb.sb_locktable;
 354         }
 355
 356         if (!table[0])
 357                 table = sdp->sd_vfs->s_id;
 358
 359         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 360         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 361
 362         table = sdp->sd_table_name;
 363         while ((table = strchr(table, '/')))
 364                 *table = '_';
 365
 366 out:
 367         return error;
 368 }
 369
 370 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 371                         int undo)
 372 {
 373         int error = 0;
 374
 375         if (undo)
 376                 goto fail_trans;
 377
 378         error = gfs2_glock_nq_num(sdp,
 379                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 380                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 381                                   mount_gh);
 382         if (error) {
 383                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 384                 goto fail;
 385         }
 386
 387         error = gfs2_glock_nq_num(sdp,
 388                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 389                                   LM_ST_SHARED,
 390                                   LM_FLAG_NOEXP | GL_EXACT,
 391                                   &sdp->sd_live_gh);
 392         if (error) {
 393                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 394                 goto fail_mount;
 395         }
 396
 397         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 398                                CREATE, &sdp->sd_rename_gl);
 399         if (error) {
 400                 fs_err(sdp, "can't create rename glock: %d\n", error);
 401                 goto fail_live;
 402         }
 403
 404         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 405                                CREATE, &sdp->sd_trans_gl);
 406         if (error) {
 407                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 408                 goto fail_rename;
 409         }
 410
 411         return 0;
 412
 413 fail_trans:
 414         gfs2_glock_put(sdp->sd_trans_gl);
 415 fail_rename:
 416         gfs2_glock_put(sdp->sd_rename_gl);
 417 fail_live:
 418         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 419 fail_mount:
 420         gfs2_glock_dq_uninit(mount_gh);
 421 fail:
 422         return error;
 423 }
 424
 425 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 426                             u64 no_addr, const char *name)
 427 {
 428         struct gfs2_sbd *sdp = sb->s_fs_info;
 429         struct dentry *dentry;
 430         struct inode *inode;
 431
 432         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0);
 433         if (IS_ERR(inode)) {
 434                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 435                 return PTR_ERR(inode);
 436         }
 437         dentry = d_alloc_root(inode);
 438         if (!dentry) {
 439                 fs_err(sdp, "can't alloc %s dentry\n", name);
 440                 iput(inode);
 441                 return -ENOMEM;
 442         }
 443         dentry->d_op = &gfs2_dops;
 444         *dptr = dentry;
 445         return 0;
 446 }
 447
 448 static int init_sb(struct gfs2_sbd *sdp, int silent)
 449 {
 450         struct super_block *sb = sdp->sd_vfs;
 451         struct gfs2_holder sb_gh;
 452         u64 no_addr;
 453         int ret;
 454
 455         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 456                                 LM_ST_SHARED, 0, &sb_gh);
 457         if (ret) {
 458                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 459                 return ret;
 460         }
 461
 462         ret = gfs2_read_sb(sdp, silent);
 463         if (ret) {
 464                 fs_err(sdp, "can't read superblock: %d\n", ret);
 465                 goto out;
 466         }
 467
 468         /* Set up the buffer cache and SB for real */
 469         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 470                 ret = -EINVAL;
 471                 fs_err(sdp, "FS block size (%u) is too small for device "
 472                        "block size (%u)\n",
 473                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 474                 goto out;
 475         }
 476         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 477                 ret = -EINVAL;
 478                 fs_err(sdp, "FS block size (%u) is too big for machine "
 479                        "page size (%u)\n",
 480                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 481                 goto out;
 482         }
 483         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 484
 485         /* Get the root inode */
 486         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 487         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 488         if (ret)
 489                 goto out;
 490
 491         /* Get the master inode */
 492         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 493         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 494         if (ret) {
 495                 dput(sdp->sd_root_dir);
 496                 goto out;
 497         }
 498         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 499 out:
 500         gfs2_glock_dq_uninit(&sb_gh);
 501         return ret;
 502 }
 503
 504 /**
 505  * map_journal_extents - create a reusable "extent" mapping from all logical
 506  * blocks to all physical blocks for the given journal.  This will save
 507  * us time when writing journal blocks.  Most journals will have only one
 508  * extent that maps all their logical blocks.  That's because gfs2.mkfs
 509  * arranges the journal blocks sequentially to maximize performance.
 510  * So the extent would map the first block for the entire file length.
 511  * However, gfs2_jadd can happen while file activity is happening, so
 512  * those journals may not be sequential.  Less likely is the case where
 513  * the users created their own journals by mounting the metafs and
 514  * laying it out.  But it's still possible.  These journals might have
 515  * several extents.
 516  *
 517  * TODO: This should be done in bigger chunks rather than one block at a time,
 518  *       but since it's only done at mount time, I'm not worried about the
 519  *       time it takes.
 520  */
 521 static int map_journal_extents(struct gfs2_sbd *sdp)
 522 {
 523         struct gfs2_jdesc *jd = sdp->sd_jdesc;
 524         unsigned int lb;
 525         u64 db, prev_db; /* logical block, disk block, prev disk block */
 526         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 527         struct gfs2_journal_extent *jext = NULL;
 528         struct buffer_head bh;
 529         int rc = 0;
 530
 531         prev_db = 0;
 532
 533         for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
 534                 bh.b_state = 0;
 535                 bh.b_blocknr = 0;
 536                 bh.b_size = 1 << ip->i_inode.i_blkbits;
 537                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
 538                 db = bh.b_blocknr;
 539                 if (rc || !db) {
 540                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
 541                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
 542                         break;
 543                 }
 544                 if (!prev_db || db != prev_db + 1) {
 545                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
 546                                        GFP_KERNEL);
 547                         if (!jext) {
 548                                 printk(KERN_INFO "GFS2 error: out of memory "
 549                                        "mapping journal extents.\n");
 550                                 rc = -ENOMEM;
 551                                 break;
 552                         }
 553                         jext->dblock = db;
 554                         jext->lblock = lb;
 555                         jext->blocks = 1;
 556                         list_add_tail(&jext->extent_list, &jd->extent_list);
 557                 } else {
 558                         jext->blocks++;
 559                 }
 560                 prev_db = db;
 561         }
 562         return rc;
 563 }
 564
 565 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 566 {
 567         char *message = "FIRSTMOUNT=Done";
 568         char *envp[] = { message, NULL };
 569         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 570         ls->ls_first_done = 1;
 571         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 572 }
 573
 574 /**
 575  * gfs2_jindex_hold - Grab a lock on the jindex
 576  * @sdp: The GFS2 superblock
 577  * @ji_gh: the holder for the jindex glock
 578  *
 579  * Returns: errno
 580  */
 581
 582 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 583 {
 584         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 585         struct qstr name;
 586         char buf[20];
 587         struct gfs2_jdesc *jd;
 588         int error;
 589
 590         name.name = buf;
 591
 592         mutex_lock(&sdp->sd_jindex_mutex);
 593
 594         for (;;) {
 595                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 596                 if (error)
 597                         break;
 598
 599                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 600                 name.hash = gfs2_disk_hash(name.name, name.len);
 601
 602                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 603                 if (error == -ENOENT) {
 604                         error = 0;
 605                         break;
 606                 }
 607
 608                 gfs2_glock_dq_uninit(ji_gh);
 609
 610                 if (error)
 611                         break;
 612
 613                 error = -ENOMEM;
 614                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 615                 if (!jd)
 616                         break;
 617
 618                 INIT_LIST_HEAD(&jd->extent_list);
 619                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 620                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 621                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 622                         if (!jd->jd_inode)
 623                                 error = -ENOENT;
 624                         else
 625                                 error = PTR_ERR(jd->jd_inode);
 626                         kfree(jd);
 627                         break;
 628                 }
 629
 630                 spin_lock(&sdp->sd_jindex_spin);
 631                 jd->jd_jid = sdp->sd_journals++;
 632                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 633                 spin_unlock(&sdp->sd_jindex_spin);
 634         }
 635
 636         mutex_unlock(&sdp->sd_jindex_mutex);
 637
 638         return error;
 639 }
 640
 641 static int init_journal(struct gfs2_sbd *sdp, int undo)
 642 {
 643         struct inode *master = sdp->sd_master_dir->d_inode;
 644         struct gfs2_holder ji_gh;
 645         struct gfs2_inode *ip;
 646         int jindex = 1;
 647         int error = 0;
 648
 649         if (undo) {
 650                 jindex = 0;
 651                 goto fail_jinode_gh;
 652         }
 653
 654         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 655         if (IS_ERR(sdp->sd_jindex)) {
 656                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 657                 return PTR_ERR(sdp->sd_jindex);
 658         }
 659         ip = GFS2_I(sdp->sd_jindex);
 660
 661         /* Load in the journal index special file */
 662
 663         error = gfs2_jindex_hold(sdp, &ji_gh);
 664         if (error) {
 665                 fs_err(sdp, "can't read journal index: %d\n", error);
 666                 goto fail;
 667         }
 668
 669         error = -EUSERS;
 670         if (!gfs2_jindex_size(sdp)) {
 671                 fs_err(sdp, "no journals!\n");
 672                 goto fail_jindex;
 673         }
 674
 675         if (sdp->sd_args.ar_spectator) {
 676                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 677                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 678                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 679                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 680         } else {
 681                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 682                         fs_err(sdp, "can't mount journal #%u\n",
 683                                sdp->sd_lockstruct.ls_jid);
 684                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 685                                gfs2_jindex_size(sdp),
 686                                gfs2_jindex_size(sdp) - 1);
 687                         goto fail_jindex;
 688                 }
 689                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 690
 691                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 692                                           &gfs2_journal_glops,
 693                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 694                                           &sdp->sd_journal_gh);
 695                 if (error) {
 696                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 697                         goto fail_jindex;
 698                 }
 699
 700                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 701                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 702                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 703                                            &sdp->sd_jinode_gh);
 704                 if (error) {
 705                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 706                                error);
 707                         goto fail_journal_gh;
 708                 }
 709
 710                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 711                 if (error) {
 712                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 713                                sdp->sd_jdesc->jd_jid, error);
 714                         goto fail_jinode_gh;
 715                 }
 716                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 717                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 718                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 719
 720                 /* Map the extents for this journal's blocks */
 721                 map_journal_extents(sdp);
 722         }
 723         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 724
 725         if (sdp->sd_lockstruct.ls_first) {
 726                 unsigned int x;
 727                 for (x = 0; x < sdp->sd_journals; x++) {
 728                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
 729                                                      true);
 730                         if (error) {
 731                                 fs_err(sdp, "error recovering journal %u: %d\n",
 732                                        x, error);
 733                                 goto fail_jinode_gh;
 734                         }
 735                 }
 736
 737                 gfs2_others_may_mount(sdp);
 738         } else if (!sdp->sd_args.ar_spectator) {
 739                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 740                 if (error) {
 741                         fs_err(sdp, "error recovering my journal: %d\n", error);
 742                         goto fail_jinode_gh;
 743                 }
 744         }
 745
 746         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 747         gfs2_glock_dq_uninit(&ji_gh);
 748         jindex = 0;
 749
 750         return 0;
 751
 752 fail_jinode_gh:
 753         if (!sdp->sd_args.ar_spectator)
 754                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 755 fail_journal_gh:
 756         if (!sdp->sd_args.ar_spectator)
 757                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 758 fail_jindex:
 759         gfs2_jindex_free(sdp);
 760         if (jindex)
 761                 gfs2_glock_dq_uninit(&ji_gh);
 762 fail:
 763         iput(sdp->sd_jindex);
 764         return error;
 765 }
 766
 767
 768 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 769 {
 770         int error = 0;
 771         struct gfs2_inode *ip;
 772         struct inode *master = sdp->sd_master_dir->d_inode;
 773
 774         if (undo)
 775                 goto fail_qinode;
 776
 777         error = init_journal(sdp, undo);
 778         if (error)
 779                 goto fail;
 780
 781         /* Read in the master statfs inode */
 782         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 783         if (IS_ERR(sdp->sd_statfs_inode)) {
 784                 error = PTR_ERR(sdp->sd_statfs_inode);
 785                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 786                 goto fail_journal;
 787         }
 788
 789         /* Read in the resource index inode */
 790         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 791         if (IS_ERR(sdp->sd_rindex)) {
 792                 error = PTR_ERR(sdp->sd_rindex);
 793                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 794                 goto fail_statfs;
 795         }
 796         ip = GFS2_I(sdp->sd_rindex);
 797         sdp->sd_rindex_uptodate = 0;
 798
 799         /* Read in the quota inode */
 800         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 801         if (IS_ERR(sdp->sd_quota_inode)) {
 802                 error = PTR_ERR(sdp->sd_quota_inode);
 803                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 804                 goto fail_rindex;
 805         }
 806         return 0;
 807
 808 fail_qinode:
 809         iput(sdp->sd_quota_inode);
 810 fail_rindex:
 811         gfs2_clear_rgrpd(sdp);
 812         iput(sdp->sd_rindex);
 813 fail_statfs:
 814         iput(sdp->sd_statfs_inode);
 815 fail_journal:
 816         init_journal(sdp, UNDO);
 817 fail:
 818         return error;
 819 }
 820
 821 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 822 {
 823         struct inode *pn = NULL;
 824         char buf[30];
 825         int error = 0;
 826         struct gfs2_inode *ip;
 827         struct inode *master = sdp->sd_master_dir->d_inode;
 828
 829         if (sdp->sd_args.ar_spectator)
 830                 return 0;
 831
 832         if (undo)
 833                 goto fail_qc_gh;
 834
 835         pn = gfs2_lookup_simple(master, "per_node");
 836         if (IS_ERR(pn)) {
 837                 error = PTR_ERR(pn);
 838                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 839                 return error;
 840         }
 841
 842         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 843         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 844         if (IS_ERR(sdp->sd_sc_inode)) {
 845                 error = PTR_ERR(sdp->sd_sc_inode);
 846                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 847                 goto fail;
 848         }
 849
 850         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 851         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 852         if (IS_ERR(sdp->sd_qc_inode)) {
 853                 error = PTR_ERR(sdp->sd_qc_inode);
 854                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 855                 goto fail_ut_i;
 856         }
 857
 858         iput(pn);
 859         pn = NULL;
 860
 861         ip = GFS2_I(sdp->sd_sc_inode);
 862         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 863                                    &sdp->sd_sc_gh);
 864         if (error) {
 865                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 866                 goto fail_qc_i;
 867         }
 868
 869         ip = GFS2_I(sdp->sd_qc_inode);
 870         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 871                                    &sdp->sd_qc_gh);
 872         if (error) {
 873                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 874                 goto fail_ut_gh;
 875         }
 876
 877         return 0;
 878
 879 fail_qc_gh:
 880         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 881 fail_ut_gh:
 882         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 883 fail_qc_i:
 884         iput(sdp->sd_qc_inode);
 885 fail_ut_i:
 886         iput(sdp->sd_sc_inode);
 887 fail:
 888         if (pn)
 889                 iput(pn);
 890         return error;
 891 }
 892
 893 static int init_threads(struct gfs2_sbd *sdp, int undo)
 894 {
 895         struct task_struct *p;
 896         int error = 0;
 897
 898         if (undo)
 899                 goto fail_quotad;
 900
 901         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
 902         error = IS_ERR(p);
 903         if (error) {
 904                 fs_err(sdp, "can't start logd thread: %d\n", error);
 905                 return error;
 906         }
 907         sdp->sd_logd_process = p;
 908
 909         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
 910         error = IS_ERR(p);
 911         if (error) {
 912                 fs_err(sdp, "can't start quotad thread: %d\n", error);
 913                 goto fail;
 914         }
 915         sdp->sd_quotad_process = p;
 916
 917         return 0;
 918
 919
 920 fail_quotad:
 921         kthread_stop(sdp->sd_quotad_process);
 922 fail:
 923         kthread_stop(sdp->sd_logd_process);
 924         return error;
 925 }
 926
 927 static const match_table_t nolock_tokens = {
 928         { Opt_jid, "jid=%d\n", },
 929         { Opt_err, NULL },
 930 };
 931
 932 static void nolock_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
 933 {
 934         struct gfs2_sbd *sdp = gl->gl_sbd;
 935         kmem_cache_free(cachep, gl);
 936         if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 937                 wake_up(&sdp->sd_glock_wait);
 938 }
 939
 940 static const struct lm_lockops nolock_ops = {
 941         .lm_proto_name = "lock_nolock",
 942         .lm_put_lock = nolock_put_lock,
 943         .lm_tokens = &nolock_tokens,
 944 };
 945
 946 /**
 947  * gfs2_lm_mount - mount a locking protocol
 948  * @sdp: the filesystem
 949  * @args: mount arguments
 950  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 951  *
 952  * Returns: errno
 953  */
 954
 955 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 956 {
 957         const struct lm_lockops *lm;
 958         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 959         struct gfs2_args *args = &sdp->sd_args;
 960         const char *proto = sdp->sd_proto_name;
 961         const char *table = sdp->sd_table_name;
 962         const char *fsname;
 963         char *o, *options;
 964         int ret;
 965
 966         if (!strcmp("lock_nolock", proto)) {
 967                 lm = &nolock_ops;
 968                 sdp->sd_args.ar_localflocks = 1;
 969 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 970         } else if (!strcmp("lock_dlm", proto)) {
 971                 lm = &gfs2_dlm_ops;
 972 #endif
 973         } else {
 974                 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
 975                 return -ENOENT;
 976         }
 977
 978         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 979
 980         ls->ls_ops = lm;
 981         ls->ls_first = 1;
 982
 983         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 984                 substring_t tmp[MAX_OPT_ARGS];
 985                 int token, option;
 986
 987                 if (!o || !*o)
 988                         continue;
 989
 990                 token = match_token(o, *lm->lm_tokens, tmp);
 991                 switch (token) {
 992                 case Opt_jid:
 993                         ret = match_int(&tmp[0], &option);
 994                         if (ret || option < 0)
 995                                 goto hostdata_error;
 996                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 997                                 ls->ls_jid = option;
 998                         break;
 999                 case Opt_id:
1000                         /* Obsolete, but left for backward compat purposes */
1001                         break;
1002                 case Opt_first:
1003                         ret = match_int(&tmp[0], &option);
1004                         if (ret || (option != 0 && option != 1))
1005                                 goto hostdata_error;
1006                         ls->ls_first = option;
1007                         break;
1008                 case Opt_nodir:
1009                         ret = match_int(&tmp[0], &option);
1010                         if (ret || (option != 0 && option != 1))
1011                                 goto hostdata_error;
1012                         ls->ls_nodir = option;
1013                         break;
1014                 case Opt_err:
1015                 default:
1016 hostdata_error:
1017                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1018                         return -EINVAL;
1019                 }
1020         }
1021
1022         if (sdp->sd_args.ar_spectator)
1023                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1024         else
1025                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1026                          sdp->sd_lockstruct.ls_jid);
1027
1028         fsname = strchr(table, ':');
1029         if (fsname)
1030                 fsname++;
1031         if (lm->lm_mount == NULL) {
1032                 fs_info(sdp, "Now mounting FS...\n");
1033                 return 0;
1034         }
1035         ret = lm->lm_mount(sdp, fsname);
1036         if (ret == 0)
1037                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1038         return ret;
1039 }
1040
1041 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1042 {
1043         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1044         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1045             lm->lm_unmount)
1046                 lm->lm_unmount(sdp);
1047 }
1048
1049 static int gfs2_journalid_wait(void *word)
1050 {
1051         if (signal_pending(current))
1052                 return -EINTR;
1053         schedule();
1054         return 0;
1055 }
1056
1057 static int wait_on_journal(struct gfs2_sbd *sdp)
1058 {
1059         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1060                 return 0;
1061
1062         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1063 }
1064
1065 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1066 {
1067         struct super_block *sb = sdp->sd_vfs;
1068         char ro[20];
1069         char spectator[20];
1070         char *envp[] = { ro, spectator, NULL };
1071         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1072         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1073         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1074 }
1075
1076 /**
1077  * fill_super - Read in superblock
1078  * @sb: The VFS superblock
1079  * @data: Mount options
1080  * @silent: Don't complain if it's not a GFS2 filesystem
1081  *
1082  * Returns: errno
1083  */
1084
1085 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1086 {
1087         struct gfs2_sbd *sdp;
1088         struct gfs2_holder mount_gh;
1089         int error;
1090
1091         sdp = init_sbd(sb);
1092         if (!sdp) {
1093                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1094                 return -ENOMEM;
1095         }
1096         sdp->sd_args = *args;
1097
1098         if (sdp->sd_args.ar_spectator) {
1099                 sb->s_flags |= MS_RDONLY;
1100                 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1101         }
1102         if (sdp->sd_args.ar_posix_acl)
1103                 sb->s_flags |= MS_POSIXACL;
1104         if (sdp->sd_args.ar_nobarrier)
1105                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1106
1107         sb->s_magic = GFS2_MAGIC;
1108         sb->s_op = &gfs2_super_ops;
1109         sb->s_export_op = &gfs2_export_ops;
1110         sb->s_xattr = gfs2_xattr_handlers;
1111         sb->s_qcop = &gfs2_quotactl_ops;
1112         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1113         sb->s_time_gran = 1;
1114         sb->s_maxbytes = MAX_LFS_FILESIZE;
1115
1116         /* Set up the buffer cache and fill in some fake block size values
1117            to allow us to read-in the on-disk superblock. */
1118         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1119         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1120         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1121                                GFS2_BASIC_BLOCK_SHIFT;
1122         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1123
1124         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1125         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1126         if (sdp->sd_args.ar_statfs_quantum) {
1127                 sdp->sd_tune.gt_statfs_slow = 0;
1128                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1129         }
1130         else {
1131                 sdp->sd_tune.gt_statfs_slow = 1;
1132                 sdp->sd_tune.gt_statfs_quantum = 30;
1133         }
1134
1135         error = init_names(sdp, silent);
1136         if (error)
1137                 goto fail;
1138
1139         gfs2_create_debugfs_file(sdp);
1140
1141         error = gfs2_sys_fs_add(sdp);
1142         if (error)
1143                 goto fail;
1144
1145         error = gfs2_lm_mount(sdp, silent);
1146         if (error)
1147                 goto fail_sys;
1148
1149         error = init_locking(sdp, &mount_gh, DO);
1150         if (error)
1151                 goto fail_lm;
1152
1153         error = init_sb(sdp, silent);
1154         if (error)
1155                 goto fail_locking;
1156
1157         error = wait_on_journal(sdp);
1158         if (error)
1159                 goto fail_sb;
1160
1161         /*
1162          * If user space has failed to join the cluster or some similar
1163          * failure has occurred, then the journal id will contain a
1164          * negative (error) number. This will then be returned to the
1165          * caller (of the mount syscall). We do this even for spectator
1166          * mounts (which just write a jid of 0 to indicate "ok" even though
1167          * the jid is unused in the spectator case)
1168          */
1169         if (sdp->sd_lockstruct.ls_jid < 0) {
1170                 error = sdp->sd_lockstruct.ls_jid;
1171                 sdp->sd_lockstruct.ls_jid = 0;
1172                 goto fail_sb;
1173         }
1174
1175         error = init_inodes(sdp, DO);
1176         if (error)
1177                 goto fail_sb;
1178
1179         error = init_per_node(sdp, DO);
1180         if (error)
1181                 goto fail_inodes;
1182
1183         error = gfs2_statfs_init(sdp);
1184         if (error) {
1185                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1186                 goto fail_per_node;
1187         }
1188
1189         error = init_threads(sdp, DO);
1190         if (error)
1191                 goto fail_per_node;
1192
1193         if (!(sb->s_flags & MS_RDONLY)) {
1194                 error = gfs2_make_fs_rw(sdp);
1195                 if (error) {
1196                         fs_err(sdp, "can't make FS RW: %d\n", error);
1197                         goto fail_threads;
1198                 }
1199         }
1200
1201         gfs2_glock_dq_uninit(&mount_gh);
1202         gfs2_online_uevent(sdp);
1203         return 0;
1204
1205 fail_threads:
1206         init_threads(sdp, UNDO);
1207 fail_per_node:
1208         init_per_node(sdp, UNDO);
1209 fail_inodes:
1210         init_inodes(sdp, UNDO);
1211 fail_sb:
1212         if (sdp->sd_root_dir)
1213                 dput(sdp->sd_root_dir);
1214         if (sdp->sd_master_dir)
1215                 dput(sdp->sd_master_dir);
1216         if (sb->s_root)
1217                 dput(sb->s_root);
1218         sb->s_root = NULL;
1219 fail_locking:
1220         init_locking(sdp, &mount_gh, UNDO);
1221 fail_lm:
1222         gfs2_gl_hash_clear(sdp);
1223         gfs2_lm_unmount(sdp);
1224 fail_sys:
1225         gfs2_sys_fs_del(sdp);
1226 fail:
1227         gfs2_delete_debugfs_file(sdp);
1228         kfree(sdp);
1229         sb->s_fs_info = NULL;
1230         return error;
1231 }
1232
1233 static int set_gfs2_super(struct super_block *s, void *data)
1234 {
1235         s->s_bdev = data;
1236         s->s_dev = s->s_bdev->bd_dev;
1237
1238         /*
1239          * We set the bdi here to the queue backing, file systems can
1240          * overwrite this in ->fill_super()
1241          */
1242         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1243         return 0;
1244 }
1245
1246 static int test_gfs2_super(struct super_block *s, void *ptr)
1247 {
1248         struct block_device *bdev = ptr;
1249         return (bdev == s->s_bdev);
1250 }
1251
1252 /**
1253  * gfs2_mount - Get the GFS2 superblock
1254  * @fs_type: The GFS2 filesystem type
1255  * @flags: Mount flags
1256  * @dev_name: The name of the device
1257  * @data: The mount arguments
1258  *
1259  * Q. Why not use get_sb_bdev() ?
1260  * A. We need to select one of two root directories to mount, independent
1261  *    of whether this is the initial, or subsequent, mount of this sb
1262  *
1263  * Returns: 0 or -ve on error
1264  */
1265
1266 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1267                        const char *dev_name, void *data)
1268 {
1269         struct block_device *bdev;
1270         struct super_block *s;
1271         fmode_t mode = FMODE_READ;
1272         int error;
1273         struct gfs2_args args;
1274         struct gfs2_sbd *sdp;
1275
1276         if (!(flags & MS_RDONLY))
1277                 mode |= FMODE_WRITE;
1278
1279         bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1280         if (IS_ERR(bdev))
1281                 return ERR_CAST(bdev);
1282
1283         /*
1284          * once the super is inserted into the list by sget, s_umount
1285          * will protect the lockfs code from trying to start a snapshot
1286          * while we are mounting
1287          */
1288         mutex_lock(&bdev->bd_fsfreeze_mutex);
1289         if (bdev->bd_fsfreeze_count > 0) {
1290                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1291                 error = -EBUSY;
1292                 goto error_bdev;
1293         }
1294         s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1295         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1296         error = PTR_ERR(s);
1297         if (IS_ERR(s))
1298                 goto error_bdev;
1299
1300         if (s->s_root)
1301                 close_bdev_exclusive(bdev, mode);
1302
1303         memset(&args, 0, sizeof(args));
1304         args.ar_quota = GFS2_QUOTA_DEFAULT;
1305         args.ar_data = GFS2_DATA_DEFAULT;
1306         args.ar_commit = 30;
1307         args.ar_statfs_quantum = 30;
1308         args.ar_quota_quantum = 60;
1309         args.ar_errors = GFS2_ERRORS_DEFAULT;
1310
1311         error = gfs2_mount_args(&args, data);
1312         if (error) {
1313                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1314                 goto error_super;
1315         }
1316
1317         if (s->s_root) {
1318                 error = -EBUSY;
1319                 if ((flags ^ s->s_flags) & MS_RDONLY)
1320                         goto error_super;
1321         } else {
1322                 char b[BDEVNAME_SIZE];
1323
1324                 s->s_flags = flags;
1325                 s->s_mode = mode;
1326                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1327                 sb_set_blocksize(s, block_size(bdev));
1328                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1329                 if (error)
1330                         goto error_super;
1331                 s->s_flags |= MS_ACTIVE;
1332                 bdev->bd_super = s;
1333         }
1334
1335         sdp = s->s_fs_info;
1336         if (args.ar_meta)
1337                 return dget(sdp->sd_master_dir);
1338         else
1339                 return dget(sdp->sd_root_dir);
1340
1341 error_super:
1342         deactivate_locked_super(s);
1343         return ERR_PTR(error);
1344 error_bdev:
1345         close_bdev_exclusive(bdev, mode);
1346         return ERR_PTR(error);
1347 }
1348
1349 static int set_meta_super(struct super_block *s, void *ptr)
1350 {
1351         return -EINVAL;
1352 }
1353
1354 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1355                         int flags, const char *dev_name, void *data)
1356 {
1357         struct super_block *s;
1358         struct gfs2_sbd *sdp;
1359         struct path path;
1360         int error;
1361
1362         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1363         if (error) {
1364                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1365                        dev_name, error);
1366                 return ERR_PTR(error);
1367         }
1368         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1369                  path.dentry->d_inode->i_sb->s_bdev);
1370         path_put(&path);
1371         if (IS_ERR(s)) {
1372                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1373                 return ERR_CAST(s);
1374         }
1375         if ((flags ^ s->s_flags) & MS_RDONLY) {
1376                 deactivate_locked_super(s);
1377                 return ERR_PTR(-EBUSY);
1378         }
1379         sdp = s->s_fs_info;
1380         return dget(sdp->sd_master_dir);
1381 }
1382
1383 static void gfs2_kill_sb(struct super_block *sb)
1384 {
1385         struct gfs2_sbd *sdp = sb->s_fs_info;
1386
1387         if (sdp == NULL) {
1388                 kill_block_super(sb);
1389                 return;
1390         }
1391
1392         gfs2_meta_syncfs(sdp);
1393         dput(sdp->sd_root_dir);
1394         dput(sdp->sd_master_dir);
1395         sdp->sd_root_dir = NULL;
1396         sdp->sd_master_dir = NULL;
1397         shrink_dcache_sb(sb);
1398         kill_block_super(sb);
1399         gfs2_delete_debugfs_file(sdp);
1400         kfree(sdp);
1401 }
1402
1403 struct file_system_type gfs2_fs_type = {
1404         .name = "gfs2",
1405         .fs_flags = FS_REQUIRES_DEV,
1406         .mount = gfs2_mount,
1407         .kill_sb = gfs2_kill_sb,
1408         .owner = THIS_MODULE,
1409 };
1410
1411 struct file_system_type gfs2meta_fs_type = {
1412         .name = "gfs2meta",
1413         .fs_flags = FS_REQUIRES_DEV,
1414         .mount = gfs2_mount_meta,
1415         .owner = THIS_MODULE,
1416 };
1417