fs/gfs2/ops_fstype.c

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