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
  98         INIT_LIST_HEAD(&sdp->sd_quota_list);
  99         mutex_init(&sdp->sd_quota_mutex);
 100         mutex_init(&sdp->sd_quota_sync_mutex);
 101         init_waitqueue_head(&sdp->sd_quota_wait);
 102         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 103         spin_lock_init(&sdp->sd_trunc_lock);
 104         spin_lock_init(&sdp->sd_bitmap_lock);
 105
 106         mapping = &sdp->sd_aspace;
 107
 108         address_space_init_once(mapping);
 109         mapping->a_ops = &gfs2_rgrp_aops;
 110         mapping->host = sb->s_bdev->bd_inode;
 111         mapping->flags = 0;
 112         mapping_set_gfp_mask(mapping, GFP_NOFS);
 113         mapping->private_data = NULL;
 114         mapping->backing_dev_info = sb->s_bdi;
 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         init_waitqueue_head(&sdp->sd_log_flush_wait);
 132
 133         return sdp;
 134 }
 135
 136
 137 /**
 138  * gfs2_check_sb - Check superblock
 139  * @sdp: the filesystem
 140  * @sb: The superblock
 141  * @silent: Don't print a message if the check fails
 142  *
 143  * Checks the version code of the FS is one that we understand how to
 144  * read and that the sizes of the various on-disk structures have not
 145  * changed.
 146  */
 147
 148 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 149 {
 150         struct gfs2_sb_host *sb = &sdp->sd_sb;
 151
 152         if (sb->sb_magic != GFS2_MAGIC ||
 153             sb->sb_type != GFS2_METATYPE_SB) {
 154                 if (!silent)
 155                         pr_warn("not a GFS2 filesystem\n");
 156                 return -EINVAL;
 157         }
 158
 159         /*  If format numbers match exactly, we're done.  */
 160
 161         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 162             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 163                 return 0;
 164
 165         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 166
 167         return -EINVAL;
 168 }
 169
 170 static void end_bio_io_page(struct bio *bio, int error)
 171 {
 172         struct page *page = bio->bi_private;
 173
 174         if (!error)
 175                 SetPageUptodate(page);
 176         else
 177                 pr_warn("error %d reading superblock\n", error);
 178         unlock_page(page);
 179 }
 180
 181 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 182 {
 183         struct gfs2_sb_host *sb = &sdp->sd_sb;
 184         struct super_block *s = sdp->sd_vfs;
 185         const struct gfs2_sb *str = buf;
 186
 187         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 188         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 189         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 190         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 191         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 192         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 193         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 194         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 195         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 196         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 197         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 198
 199         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 200         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 201         memcpy(s->s_uuid, str->sb_uuid, 16);
 202 }
 203
 204 /**
 205  * gfs2_read_super - Read the gfs2 super block from disk
 206  * @sdp: The GFS2 super block
 207  * @sector: The location of the super block
 208  * @error: The error code to return
 209  *
 210  * This uses the bio functions to read the super block from disk
 211  * because we want to be 100% sure that we never read cached data.
 212  * A super block is read twice only during each GFS2 mount and is
 213  * never written to by the filesystem. The first time its read no
 214  * locks are held, and the only details which are looked at are those
 215  * relating to the locking protocol. Once locking is up and working,
 216  * the sb is read again under the lock to establish the location of
 217  * the master directory (contains pointers to journals etc) and the
 218  * root directory.
 219  *
 220  * Returns: 0 on success or error
 221  */
 222
 223 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 224 {
 225         struct super_block *sb = sdp->sd_vfs;
 226         struct gfs2_sb *p;
 227         struct page *page;
 228         struct bio *bio;
 229
 230         page = alloc_page(GFP_NOFS);
 231         if (unlikely(!page))
 232                 return -ENOMEM;
 233
 234         ClearPageUptodate(page);
 235         ClearPageDirty(page);
 236         lock_page(page);
 237
 238         bio = bio_alloc(GFP_NOFS, 1);
 239         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 240         bio->bi_bdev = sb->s_bdev;
 241         bio_add_page(bio, page, PAGE_SIZE, 0);
 242
 243         bio->bi_end_io = end_bio_io_page;
 244         bio->bi_private = page;
 245         submit_bio(READ_SYNC | REQ_META, bio);
 246         wait_on_page_locked(page);
 247         bio_put(bio);
 248         if (!PageUptodate(page)) {
 249                 __free_page(page);
 250                 return -EIO;
 251         }
 252         p = kmap(page);
 253         gfs2_sb_in(sdp, p);
 254         kunmap(page);
 255         __free_page(page);
 256         return gfs2_check_sb(sdp, silent);
 257 }
 258
 259 /**
 260  * gfs2_read_sb - Read super block
 261  * @sdp: The GFS2 superblock
 262  * @silent: Don't print message if mount fails
 263  *
 264  */
 265
 266 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 267 {
 268         u32 hash_blocks, ind_blocks, leaf_blocks;
 269         u32 tmp_blocks;
 270         unsigned int x;
 271         int error;
 272
 273         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 274         if (error) {
 275                 if (!silent)
 276                         fs_err(sdp, "can't read superblock\n");
 277                 return error;
 278         }
 279
 280         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 281                                GFS2_BASIC_BLOCK_SHIFT;
 282         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 283         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 284                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 285         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 286                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 287         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 288         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 289         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 290         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 291         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 292                                 sizeof(struct gfs2_meta_header)) /
 293                                 sizeof(struct gfs2_quota_change);
 294         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 295                                      sizeof(struct gfs2_meta_header))
 296                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 297
 298         /* Compute maximum reservation required to add a entry to a directory */
 299
 300         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 301                              sdp->sd_jbsize);
 302
 303         ind_blocks = 0;
 304         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 305                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 306                 ind_blocks += tmp_blocks;
 307         }
 308
 309         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 310
 311         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 312
 313         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 314                                 sizeof(struct gfs2_dinode);
 315         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 316         for (x = 2;; x++) {
 317                 u64 space, d;
 318                 u32 m;
 319
 320                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 321                 d = space;
 322                 m = do_div(d, sdp->sd_inptrs);
 323
 324                 if (d != sdp->sd_heightsize[x - 1] || m)
 325                         break;
 326                 sdp->sd_heightsize[x] = space;
 327         }
 328         sdp->sd_max_height = x;
 329         sdp->sd_heightsize[x] = ~0;
 330         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 331
 332         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 333                                  sizeof(struct gfs2_dinode);
 334         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 335         for (x = 2;; x++) {
 336                 u64 space, d;
 337                 u32 m;
 338
 339                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 340                 d = space;
 341                 m = do_div(d, sdp->sd_inptrs);
 342
 343                 if (d != sdp->sd_jheightsize[x - 1] || m)
 344                         break;
 345                 sdp->sd_jheightsize[x] = space;
 346         }
 347         sdp->sd_max_jheight = x;
 348         sdp->sd_jheightsize[x] = ~0;
 349         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 350
 351         return 0;
 352 }
 353
 354 static int init_names(struct gfs2_sbd *sdp, int silent)
 355 {
 356         char *proto, *table;
 357         int error = 0;
 358
 359         proto = sdp->sd_args.ar_lockproto;
 360         table = sdp->sd_args.ar_locktable;
 361
 362         /*  Try to autodetect  */
 363
 364         if (!proto[0] || !table[0]) {
 365                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 366                 if (error)
 367                         return error;
 368
 369                 if (!proto[0])
 370                         proto = sdp->sd_sb.sb_lockproto;
 371                 if (!table[0])
 372                         table = sdp->sd_sb.sb_locktable;
 373         }
 374
 375         if (!table[0])
 376                 table = sdp->sd_vfs->s_id;
 377
 378         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 379         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 380
 381         table = sdp->sd_table_name;
 382         while ((table = strchr(table, '/')))
 383                 *table = '_';
 384
 385         return error;
 386 }
 387
 388 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 389                         int undo)
 390 {
 391         int error = 0;
 392
 393         if (undo)
 394                 goto fail_trans;
 395
 396         error = gfs2_glock_nq_num(sdp,
 397                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 398                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 399                                   mount_gh);
 400         if (error) {
 401                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 402                 goto fail;
 403         }
 404
 405         error = gfs2_glock_nq_num(sdp,
 406                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 407                                   LM_ST_SHARED,
 408                                   LM_FLAG_NOEXP | GL_EXACT,
 409                                   &sdp->sd_live_gh);
 410         if (error) {
 411                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 412                 goto fail_mount;
 413         }
 414
 415         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 416                                CREATE, &sdp->sd_rename_gl);
 417         if (error) {
 418                 fs_err(sdp, "can't create rename glock: %d\n", error);
 419                 goto fail_live;
 420         }
 421
 422         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
 423                                CREATE, &sdp->sd_trans_gl);
 424         if (error) {
 425                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 426                 goto fail_rename;
 427         }
 428
 429         return 0;
 430
 431 fail_trans:
 432         gfs2_glock_put(sdp->sd_trans_gl);
 433 fail_rename:
 434         gfs2_glock_put(sdp->sd_rename_gl);
 435 fail_live:
 436         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 437 fail_mount:
 438         gfs2_glock_dq_uninit(mount_gh);
 439 fail:
 440         return error;
 441 }
 442
 443 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 444                             u64 no_addr, const char *name)
 445 {
 446         struct gfs2_sbd *sdp = sb->s_fs_info;
 447         struct dentry *dentry;
 448         struct inode *inode;
 449
 450         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
 451         if (IS_ERR(inode)) {
 452                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 453                 return PTR_ERR(inode);
 454         }
 455         dentry = d_make_root(inode);
 456         if (!dentry) {
 457                 fs_err(sdp, "can't alloc %s dentry\n", name);
 458                 return -ENOMEM;
 459         }
 460         *dptr = dentry;
 461         return 0;
 462 }
 463
 464 static int init_sb(struct gfs2_sbd *sdp, int silent)
 465 {
 466         struct super_block *sb = sdp->sd_vfs;
 467         struct gfs2_holder sb_gh;
 468         u64 no_addr;
 469         int ret;
 470
 471         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 472                                 LM_ST_SHARED, 0, &sb_gh);
 473         if (ret) {
 474                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 475                 return ret;
 476         }
 477
 478         ret = gfs2_read_sb(sdp, silent);
 479         if (ret) {
 480                 fs_err(sdp, "can't read superblock: %d\n", ret);
 481                 goto out;
 482         }
 483
 484         /* Set up the buffer cache and SB for real */
 485         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 486                 ret = -EINVAL;
 487                 fs_err(sdp, "FS block size (%u) is too small for device "
 488                        "block size (%u)\n",
 489                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 490                 goto out;
 491         }
 492         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 493                 ret = -EINVAL;
 494                 fs_err(sdp, "FS block size (%u) is too big for machine "
 495                        "page size (%u)\n",
 496                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 497                 goto out;
 498         }
 499         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 500
 501         /* Get the root inode */
 502         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 503         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 504         if (ret)
 505                 goto out;
 506
 507         /* Get the master inode */
 508         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 509         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 510         if (ret) {
 511                 dput(sdp->sd_root_dir);
 512                 goto out;
 513         }
 514         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 515 out:
 516         gfs2_glock_dq_uninit(&sb_gh);
 517         return ret;
 518 }
 519
 520 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 521 {
 522         char *message = "FIRSTMOUNT=Done";
 523         char *envp[] = { message, NULL };
 524
 525         fs_info(sdp, "first mount done, others may mount\n");
 526
 527         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 528                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 529
 530         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 531 }
 532
 533 /**
 534  * gfs2_jindex_hold - Grab a lock on the jindex
 535  * @sdp: The GFS2 superblock
 536  * @ji_gh: the holder for the jindex glock
 537  *
 538  * Returns: errno
 539  */
 540
 541 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 542 {
 543         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 544         struct qstr name;
 545         char buf[20];
 546         struct gfs2_jdesc *jd;
 547         int error;
 548
 549         name.name = buf;
 550
 551         mutex_lock(&sdp->sd_jindex_mutex);
 552
 553         for (;;) {
 554                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 555                 if (error)
 556                         break;
 557
 558                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 559                 name.hash = gfs2_disk_hash(name.name, name.len);
 560
 561                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 562                 if (error == -ENOENT) {
 563                         error = 0;
 564                         break;
 565                 }
 566
 567                 gfs2_glock_dq_uninit(ji_gh);
 568
 569                 if (error)
 570                         break;
 571
 572                 error = -ENOMEM;
 573                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 574                 if (!jd)
 575                         break;
 576
 577                 INIT_LIST_HEAD(&jd->extent_list);
 578                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 579
 580                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 581                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 582                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 583                         if (!jd->jd_inode)
 584                                 error = -ENOENT;
 585                         else
 586                                 error = PTR_ERR(jd->jd_inode);
 587                         kfree(jd);
 588                         break;
 589                 }
 590
 591                 spin_lock(&sdp->sd_jindex_spin);
 592                 jd->jd_jid = sdp->sd_journals++;
 593                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 594                 spin_unlock(&sdp->sd_jindex_spin);
 595         }
 596
 597         mutex_unlock(&sdp->sd_jindex_mutex);
 598
 599         return error;
 600 }
 601
 602 /**
 603  * check_journal_clean - Make sure a journal is clean for a spectator mount
 604  * @sdp: The GFS2 superblock
 605  * @jd: The journal descriptor
 606  *
 607  * Returns: 0 if the journal is clean or locked, else an error
 608  */
 609 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 610 {
 611         int error;
 612         struct gfs2_holder j_gh;
 613         struct gfs2_log_header_host head;
 614         struct gfs2_inode *ip;
 615
 616         ip = GFS2_I(jd->jd_inode);
 617         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 618                                    GL_EXACT | GL_NOCACHE, &j_gh);
 619         if (error) {
 620                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 621                 return -EPERM;
 622         }
 623         error = gfs2_jdesc_check(jd);
 624         if (error) {
 625                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 626                 goto out_unlock;
 627         }
 628         error = gfs2_find_jhead(jd, &head);
 629         if (error) {
 630                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 631                 goto out_unlock;
 632         }
 633         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 634                 error = -EPERM;
 635                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 636                        "must not be a spectator.\n", jd->jd_jid);
 637         }
 638
 639 out_unlock:
 640         gfs2_glock_dq_uninit(&j_gh);
 641         return error;
 642 }
 643
 644 static int init_journal(struct gfs2_sbd *sdp, int undo)
 645 {
 646         struct inode *master = sdp->sd_master_dir->d_inode;
 647         struct gfs2_holder ji_gh;
 648         struct gfs2_inode *ip;
 649         int jindex = 1;
 650         int error = 0;
 651
 652         if (undo) {
 653                 jindex = 0;
 654                 goto fail_jinode_gh;
 655         }
 656
 657         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 658         if (IS_ERR(sdp->sd_jindex)) {
 659                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 660                 return PTR_ERR(sdp->sd_jindex);
 661         }
 662
 663         /* Load in the journal index special file */
 664
 665         error = gfs2_jindex_hold(sdp, &ji_gh);
 666         if (error) {
 667                 fs_err(sdp, "can't read journal index: %d\n", error);
 668                 goto fail;
 669         }
 670
 671         error = -EUSERS;
 672         if (!gfs2_jindex_size(sdp)) {
 673                 fs_err(sdp, "no journals!\n");
 674                 goto fail_jindex;
 675         }
 676
 677         if (sdp->sd_args.ar_spectator) {
 678                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 679                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 680                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 681                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 682         } else {
 683                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 684                         fs_err(sdp, "can't mount journal #%u\n",
 685                                sdp->sd_lockstruct.ls_jid);
 686                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 687                                gfs2_jindex_size(sdp),
 688                                gfs2_jindex_size(sdp) - 1);
 689                         goto fail_jindex;
 690                 }
 691                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 692
 693                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 694                                           &gfs2_journal_glops,
 695                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 696                                           &sdp->sd_journal_gh);
 697                 if (error) {
 698                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 699                         goto fail_jindex;
 700                 }
 701
 702                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 703                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 704                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 705                                            &sdp->sd_jinode_gh);
 706                 if (error) {
 707                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 708                                error);
 709                         goto fail_journal_gh;
 710                 }
 711
 712                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 713                 if (error) {
 714                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 715                                sdp->sd_jdesc->jd_jid, error);
 716                         goto fail_jinode_gh;
 717                 }
 718                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 719                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 720                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 721
 722                 /* Map the extents for this journal's blocks */
 723                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 724         }
 725         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 726
 727         if (sdp->sd_lockstruct.ls_first) {
 728                 unsigned int x;
 729                 for (x = 0; x < sdp->sd_journals; x++) {
 730                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 731
 732                         if (sdp->sd_args.ar_spectator) {
 733                                 error = check_journal_clean(sdp, jd);
 734                                 if (error)
 735                                         goto fail_jinode_gh;
 736                                 continue;
 737                         }
 738                         error = gfs2_recover_journal(jd, true);
 739                         if (error) {
 740                                 fs_err(sdp, "error recovering journal %u: %d\n",
 741                                        x, error);
 742                                 goto fail_jinode_gh;
 743                         }
 744                 }
 745
 746                 gfs2_others_may_mount(sdp);
 747         } else if (!sdp->sd_args.ar_spectator) {
 748                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 749                 if (error) {
 750                         fs_err(sdp, "error recovering my journal: %d\n", error);
 751                         goto fail_jinode_gh;
 752                 }
 753         }
 754
 755         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 756         gfs2_glock_dq_uninit(&ji_gh);
 757         jindex = 0;
 758
 759         return 0;
 760
 761 fail_jinode_gh:
 762         if (!sdp->sd_args.ar_spectator)
 763                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 764 fail_journal_gh:
 765         if (!sdp->sd_args.ar_spectator)
 766                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 767 fail_jindex:
 768         gfs2_jindex_free(sdp);
 769         if (jindex)
 770                 gfs2_glock_dq_uninit(&ji_gh);
 771 fail:
 772         iput(sdp->sd_jindex);
 773         return error;
 774 }
 775
 776 static struct lock_class_key gfs2_quota_imutex_key;
 777
 778 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 779 {
 780         int error = 0;
 781         struct inode *master = sdp->sd_master_dir->d_inode;
 782
 783         if (undo)
 784                 goto fail_qinode;
 785
 786         error = init_journal(sdp, undo);
 787         if (error)
 788                 goto fail;
 789
 790         /* Read in the master statfs inode */
 791         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 792         if (IS_ERR(sdp->sd_statfs_inode)) {
 793                 error = PTR_ERR(sdp->sd_statfs_inode);
 794                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 795                 goto fail_journal;
 796         }
 797
 798         /* Read in the resource index inode */
 799         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 800         if (IS_ERR(sdp->sd_rindex)) {
 801                 error = PTR_ERR(sdp->sd_rindex);
 802                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 803                 goto fail_statfs;
 804         }
 805         sdp->sd_rindex_uptodate = 0;
 806
 807         /* Read in the quota inode */
 808         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 809         if (IS_ERR(sdp->sd_quota_inode)) {
 810                 error = PTR_ERR(sdp->sd_quota_inode);
 811                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 812                 goto fail_rindex;
 813         }
 814         /*
 815          * i_mutex on quota files is special. Since this inode is hidden system
 816          * file, we are safe to define locking ourselves.
 817          */
 818         lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
 819                           &gfs2_quota_imutex_key);
 820
 821         error = gfs2_rindex_update(sdp);
 822         if (error)
 823                 goto fail_qinode;
 824
 825         return 0;
 826
 827 fail_qinode:
 828         iput(sdp->sd_quota_inode);
 829 fail_rindex:
 830         gfs2_clear_rgrpd(sdp);
 831         iput(sdp->sd_rindex);
 832 fail_statfs:
 833         iput(sdp->sd_statfs_inode);
 834 fail_journal:
 835         init_journal(sdp, UNDO);
 836 fail:
 837         return error;
 838 }
 839
 840 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 841 {
 842         struct inode *pn = NULL;
 843         char buf[30];
 844         int error = 0;
 845         struct gfs2_inode *ip;
 846         struct inode *master = sdp->sd_master_dir->d_inode;
 847
 848         if (sdp->sd_args.ar_spectator)
 849                 return 0;
 850
 851         if (undo)
 852                 goto fail_qc_gh;
 853
 854         pn = gfs2_lookup_simple(master, "per_node");
 855         if (IS_ERR(pn)) {
 856                 error = PTR_ERR(pn);
 857                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 858                 return error;
 859         }
 860
 861         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 862         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 863         if (IS_ERR(sdp->sd_sc_inode)) {
 864                 error = PTR_ERR(sdp->sd_sc_inode);
 865                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 866                 goto fail;
 867         }
 868
 869         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 870         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 871         if (IS_ERR(sdp->sd_qc_inode)) {
 872                 error = PTR_ERR(sdp->sd_qc_inode);
 873                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 874                 goto fail_ut_i;
 875         }
 876
 877         iput(pn);
 878         pn = NULL;
 879
 880         ip = GFS2_I(sdp->sd_sc_inode);
 881         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 882                                    &sdp->sd_sc_gh);
 883         if (error) {
 884                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 885                 goto fail_qc_i;
 886         }
 887
 888         ip = GFS2_I(sdp->sd_qc_inode);
 889         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 890                                    &sdp->sd_qc_gh);
 891         if (error) {
 892                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 893                 goto fail_ut_gh;
 894         }
 895
 896         return 0;
 897
 898 fail_qc_gh:
 899         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 900 fail_ut_gh:
 901         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 902 fail_qc_i:
 903         iput(sdp->sd_qc_inode);
 904 fail_ut_i:
 905         iput(sdp->sd_sc_inode);
 906 fail:
 907         if (pn)
 908                 iput(pn);
 909         return error;
 910 }
 911
 912 static const match_table_t nolock_tokens = {
 913         { Opt_jid, "jid=%d\n", },
 914         { Opt_err, NULL },
 915 };
 916
 917 static const struct lm_lockops nolock_ops = {
 918         .lm_proto_name = "lock_nolock",
 919         .lm_put_lock = gfs2_glock_free,
 920         .lm_tokens = &nolock_tokens,
 921 };
 922
 923 /**
 924  * gfs2_lm_mount - mount a locking protocol
 925  * @sdp: the filesystem
 926  * @args: mount arguments
 927  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 928  *
 929  * Returns: errno
 930  */
 931
 932 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 933 {
 934         const struct lm_lockops *lm;
 935         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 936         struct gfs2_args *args = &sdp->sd_args;
 937         const char *proto = sdp->sd_proto_name;
 938         const char *table = sdp->sd_table_name;
 939         char *o, *options;
 940         int ret;
 941
 942         if (!strcmp("lock_nolock", proto)) {
 943                 lm = &nolock_ops;
 944                 sdp->sd_args.ar_localflocks = 1;
 945 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 946         } else if (!strcmp("lock_dlm", proto)) {
 947                 lm = &gfs2_dlm_ops;
 948 #endif
 949         } else {
 950                 pr_info("can't find protocol %s\n", proto);
 951                 return -ENOENT;
 952         }
 953
 954         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 955
 956         ls->ls_ops = lm;
 957         ls->ls_first = 1;
 958
 959         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 960                 substring_t tmp[MAX_OPT_ARGS];
 961                 int token, option;
 962
 963                 if (!o || !*o)
 964                         continue;
 965
 966                 token = match_token(o, *lm->lm_tokens, tmp);
 967                 switch (token) {
 968                 case Opt_jid:
 969                         ret = match_int(&tmp[0], &option);
 970                         if (ret || option < 0)
 971                                 goto hostdata_error;
 972                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 973                                 ls->ls_jid = option;
 974                         break;
 975                 case Opt_id:
 976                 case Opt_nodir:
 977                         /* Obsolete, but left for backward compat purposes */
 978                         break;
 979                 case Opt_first:
 980                         ret = match_int(&tmp[0], &option);
 981                         if (ret || (option != 0 && option != 1))
 982                                 goto hostdata_error;
 983                         ls->ls_first = option;
 984                         break;
 985                 case Opt_err:
 986                 default:
 987 hostdata_error:
 988                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 989                         return -EINVAL;
 990                 }
 991         }
 992
 993         if (lm->lm_mount == NULL) {
 994                 fs_info(sdp, "Now mounting FS...\n");
 995                 complete_all(&sdp->sd_locking_init);
 996                 return 0;
 997         }
 998         ret = lm->lm_mount(sdp, table);
 999         if (ret == 0)
1000                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1001         complete_all(&sdp->sd_locking_init);
1002         return ret;
1003 }
1004
1005 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1006 {
1007         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1008         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1009             lm->lm_unmount)
1010                 lm->lm_unmount(sdp);
1011 }
1012
1013 static int gfs2_journalid_wait(void *word)
1014 {
1015         if (signal_pending(current))
1016                 return -EINTR;
1017         schedule();
1018         return 0;
1019 }
1020
1021 static int wait_on_journal(struct gfs2_sbd *sdp)
1022 {
1023         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1024                 return 0;
1025
1026         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1027 }
1028
1029 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1030 {
1031         struct super_block *sb = sdp->sd_vfs;
1032         char ro[20];
1033         char spectator[20];
1034         char *envp[] = { ro, spectator, NULL };
1035         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1036         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1037         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1038 }
1039
1040 /**
1041  * fill_super - Read in superblock
1042  * @sb: The VFS superblock
1043  * @data: Mount options
1044  * @silent: Don't complain if it's not a GFS2 filesystem
1045  *
1046  * Returns: errno
1047  */
1048
1049 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1050 {
1051         struct gfs2_sbd *sdp;
1052         struct gfs2_holder mount_gh;
1053         int error;
1054
1055         sdp = init_sbd(sb);
1056         if (!sdp) {
1057                 pr_warn("can't alloc struct gfs2_sbd\n");
1058                 return -ENOMEM;
1059         }
1060         sdp->sd_args = *args;
1061
1062         if (sdp->sd_args.ar_spectator) {
1063                 sb->s_flags |= MS_RDONLY;
1064                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1065         }
1066         if (sdp->sd_args.ar_posix_acl)
1067                 sb->s_flags |= MS_POSIXACL;
1068         if (sdp->sd_args.ar_nobarrier)
1069                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1070
1071         sb->s_flags |= MS_NOSEC;
1072         sb->s_magic = GFS2_MAGIC;
1073         sb->s_op = &gfs2_super_ops;
1074         sb->s_d_op = &gfs2_dops;
1075         sb->s_export_op = &gfs2_export_ops;
1076         sb->s_xattr = gfs2_xattr_handlers;
1077         sb->s_qcop = &gfs2_quotactl_ops;
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         gfs2_gl_hash_clear(sdp);
1204         gfs2_lm_unmount(sdp);
1205 fail_debug:
1206         gfs2_delete_debugfs_file(sdp);
1207         free_percpu(sdp->sd_lkstats);
1208         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1209          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1210         gfs2_sys_fs_del(sdp);
1211         sb->s_fs_info = NULL;
1212         return error;
1213 }
1214
1215 static int set_gfs2_super(struct super_block *s, void *data)
1216 {
1217         s->s_bdev = data;
1218         s->s_dev = s->s_bdev->bd_dev;
1219
1220         /*
1221          * We set the bdi here to the queue backing, file systems can
1222          * overwrite this in ->fill_super()
1223          */
1224         s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1225         return 0;
1226 }
1227
1228 static int test_gfs2_super(struct super_block *s, void *ptr)
1229 {
1230         struct block_device *bdev = ptr;
1231         return (bdev == s->s_bdev);
1232 }
1233
1234 /**
1235  * gfs2_mount - Get the GFS2 superblock
1236  * @fs_type: The GFS2 filesystem type
1237  * @flags: Mount flags
1238  * @dev_name: The name of the device
1239  * @data: The mount arguments
1240  *
1241  * Q. Why not use get_sb_bdev() ?
1242  * A. We need to select one of two root directories to mount, independent
1243  *    of whether this is the initial, or subsequent, mount of this sb
1244  *
1245  * Returns: 0 or -ve on error
1246  */
1247
1248 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1249                        const char *dev_name, void *data)
1250 {
1251         struct block_device *bdev;
1252         struct super_block *s;
1253         fmode_t mode = FMODE_READ | FMODE_EXCL;
1254         int error;
1255         struct gfs2_args args;
1256         struct gfs2_sbd *sdp;
1257
1258         if (!(flags & MS_RDONLY))
1259                 mode |= FMODE_WRITE;
1260
1261         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1262         if (IS_ERR(bdev))
1263                 return ERR_CAST(bdev);
1264
1265         /*
1266          * once the super is inserted into the list by sget, s_umount
1267          * will protect the lockfs code from trying to start a snapshot
1268          * while we are mounting
1269          */
1270         mutex_lock(&bdev->bd_fsfreeze_mutex);
1271         if (bdev->bd_fsfreeze_count > 0) {
1272                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1273                 error = -EBUSY;
1274                 goto error_bdev;
1275         }
1276         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1277         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1278         error = PTR_ERR(s);
1279         if (IS_ERR(s))
1280                 goto error_bdev;
1281
1282         if (s->s_root) {
1283                 /*
1284                  * s_umount nests inside bd_mutex during
1285                  * __invalidate_device().  blkdev_put() acquires
1286                  * bd_mutex and can't be called under s_umount.  Drop
1287                  * s_umount temporarily.  This is safe as we're
1288                  * holding an active reference.
1289                  */
1290                 up_write(&s->s_umount);
1291                 blkdev_put(bdev, mode);
1292                 down_write(&s->s_umount);
1293         }
1294
1295         memset(&args, 0, sizeof(args));
1296         args.ar_quota = GFS2_QUOTA_DEFAULT;
1297         args.ar_data = GFS2_DATA_DEFAULT;
1298         args.ar_commit = 30;
1299         args.ar_statfs_quantum = 30;
1300         args.ar_quota_quantum = 60;
1301         args.ar_errors = GFS2_ERRORS_DEFAULT;
1302
1303         error = gfs2_mount_args(&args, data);
1304         if (error) {
1305                 pr_warn("can't parse mount arguments\n");
1306                 goto error_super;
1307         }
1308
1309         if (s->s_root) {
1310                 error = -EBUSY;
1311                 if ((flags ^ s->s_flags) & MS_RDONLY)
1312                         goto error_super;
1313         } else {
1314                 char b[BDEVNAME_SIZE];
1315
1316                 s->s_mode = mode;
1317                 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1318                 sb_set_blocksize(s, block_size(bdev));
1319                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1320                 if (error)
1321                         goto error_super;
1322                 s->s_flags |= MS_ACTIVE;
1323                 bdev->bd_super = s;
1324         }
1325
1326         sdp = s->s_fs_info;
1327         if (args.ar_meta)
1328                 return dget(sdp->sd_master_dir);
1329         else
1330                 return dget(sdp->sd_root_dir);
1331
1332 error_super:
1333         deactivate_locked_super(s);
1334         return ERR_PTR(error);
1335 error_bdev:
1336         blkdev_put(bdev, mode);
1337         return ERR_PTR(error);
1338 }
1339
1340 static int set_meta_super(struct super_block *s, void *ptr)
1341 {
1342         return -EINVAL;
1343 }
1344
1345 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1346                         int flags, const char *dev_name, void *data)
1347 {
1348         struct super_block *s;
1349         struct gfs2_sbd *sdp;
1350         struct path path;
1351         int error;
1352
1353         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1354         if (error) {
1355                 pr_warn("path_lookup on %s returned error %d\n",
1356                         dev_name, error);
1357                 return ERR_PTR(error);
1358         }
1359         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1360                  path.dentry->d_inode->i_sb->s_bdev);
1361         path_put(&path);
1362         if (IS_ERR(s)) {
1363                 pr_warn("gfs2 mount does not exist\n");
1364                 return ERR_CAST(s);
1365         }
1366         if ((flags ^ s->s_flags) & MS_RDONLY) {
1367                 deactivate_locked_super(s);
1368                 return ERR_PTR(-EBUSY);
1369         }
1370         sdp = s->s_fs_info;
1371         return dget(sdp->sd_master_dir);
1372 }
1373
1374 static void gfs2_kill_sb(struct super_block *sb)
1375 {
1376         struct gfs2_sbd *sdp = sb->s_fs_info;
1377
1378         if (sdp == NULL) {
1379                 kill_block_super(sb);
1380                 return;
1381         }
1382
1383         gfs2_meta_syncfs(sdp);
1384         dput(sdp->sd_root_dir);
1385         dput(sdp->sd_master_dir);
1386         sdp->sd_root_dir = NULL;
1387         sdp->sd_master_dir = NULL;
1388         shrink_dcache_sb(sb);
1389         gfs2_delete_debugfs_file(sdp);
1390         free_percpu(sdp->sd_lkstats);
1391         kill_block_super(sb);
1392 }
1393
1394 struct file_system_type gfs2_fs_type = {
1395         .name = "gfs2",
1396         .fs_flags = FS_REQUIRES_DEV,
1397         .mount = gfs2_mount,
1398         .kill_sb = gfs2_kill_sb,
1399         .owner = THIS_MODULE,
1400 };
1401 MODULE_ALIAS_FS("gfs2");
1402
1403 struct file_system_type gfs2meta_fs_type = {
1404         .name = "gfs2meta",
1405         .fs_flags = FS_REQUIRES_DEV,
1406         .mount = gfs2_mount_meta,
1407         .owner = THIS_MODULE,
1408 };
1409 MODULE_ALIAS_FS("gfs2meta");