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 #include <linux/fs_parser.h>
  25
  26 #include "gfs2.h"
  27 #include "incore.h"
  28 #include "bmap.h"
  29 #include "glock.h"
  30 #include "glops.h"
  31 #include "inode.h"
  32 #include "recovery.h"
  33 #include "rgrp.h"
  34 #include "super.h"
  35 #include "sys.h"
  36 #include "util.h"
  37 #include "log.h"
  38 #include "quota.h"
  39 #include "dir.h"
  40 #include "meta_io.h"
  41 #include "trace_gfs2.h"
  42 #include "lops.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 = BIT(18);
  62         gt->gt_complain_secs = 10;
  63 }
  64
  65 void free_sbd(struct gfs2_sbd *sdp)
  66 {
  67         if (sdp->sd_lkstats)
  68                 free_percpu(sdp->sd_lkstats);
  69         kfree(sdp);
  70 }
  71
  72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
  73 {
  74         struct gfs2_sbd *sdp;
  75         struct address_space *mapping;
  76
  77         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
  78         if (!sdp)
  79                 return NULL;
  80
  81         sdp->sd_vfs = sb;
  82         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
  83         if (!sdp->sd_lkstats)
  84                 goto fail;
  85         sb->s_fs_info = sdp;
  86
  87         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
  88         gfs2_tune_init(&sdp->sd_tune);
  89
  90         init_waitqueue_head(&sdp->sd_glock_wait);
  91         init_waitqueue_head(&sdp->sd_async_glock_wait);
  92         atomic_set(&sdp->sd_glock_disposal, 0);
  93         init_completion(&sdp->sd_locking_init);
  94         init_completion(&sdp->sd_wdack);
  95         spin_lock_init(&sdp->sd_statfs_spin);
  96
  97         spin_lock_init(&sdp->sd_rindex_spin);
  98         sdp->sd_rindex_tree.rb_node = NULL;
  99
 100         INIT_LIST_HEAD(&sdp->sd_jindex_list);
 101         spin_lock_init(&sdp->sd_jindex_spin);
 102         mutex_init(&sdp->sd_jindex_mutex);
 103         init_completion(&sdp->sd_journal_ready);
 104
 105         INIT_LIST_HEAD(&sdp->sd_quota_list);
 106         mutex_init(&sdp->sd_quota_mutex);
 107         mutex_init(&sdp->sd_quota_sync_mutex);
 108         init_waitqueue_head(&sdp->sd_quota_wait);
 109         INIT_LIST_HEAD(&sdp->sd_trunc_list);
 110         spin_lock_init(&sdp->sd_trunc_lock);
 111         spin_lock_init(&sdp->sd_bitmap_lock);
 112
 113         mapping = &sdp->sd_aspace;
 114
 115         address_space_init_once(mapping);
 116         mapping->a_ops = &gfs2_rgrp_aops;
 117         mapping->host = sb->s_bdev->bd_inode;
 118         mapping->flags = 0;
 119         mapping_set_gfp_mask(mapping, GFP_NOFS);
 120         mapping->private_data = NULL;
 121         mapping->writeback_index = 0;
 122
 123         spin_lock_init(&sdp->sd_log_lock);
 124         atomic_set(&sdp->sd_log_pinned, 0);
 125         INIT_LIST_HEAD(&sdp->sd_log_revokes);
 126         INIT_LIST_HEAD(&sdp->sd_log_ordered);
 127         spin_lock_init(&sdp->sd_ordered_lock);
 128
 129         init_waitqueue_head(&sdp->sd_log_waitq);
 130         init_waitqueue_head(&sdp->sd_logd_waitq);
 131         spin_lock_init(&sdp->sd_ail_lock);
 132         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 133         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 134
 135         init_rwsem(&sdp->sd_log_flush_lock);
 136         atomic_set(&sdp->sd_log_in_flight, 0);
 137         atomic_set(&sdp->sd_reserving_log, 0);
 138         init_waitqueue_head(&sdp->sd_reserving_log_wait);
 139         init_waitqueue_head(&sdp->sd_log_flush_wait);
 140         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 141         mutex_init(&sdp->sd_freeze_mutex);
 142
 143         return sdp;
 144
 145 fail:
 146         free_sbd(sdp);
 147         return NULL;
 148 }
 149
 150 /**
 151  * gfs2_check_sb - Check superblock
 152  * @sdp: the filesystem
 153  * @sb: The superblock
 154  * @silent: Don't print a message if the check fails
 155  *
 156  * Checks the version code of the FS is one that we understand how to
 157  * read and that the sizes of the various on-disk structures have not
 158  * changed.
 159  */
 160
 161 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 162 {
 163         struct gfs2_sb_host *sb = &sdp->sd_sb;
 164
 165         if (sb->sb_magic != GFS2_MAGIC ||
 166             sb->sb_type != GFS2_METATYPE_SB) {
 167                 if (!silent)
 168                         pr_warn("not a GFS2 filesystem\n");
 169                 return -EINVAL;
 170         }
 171
 172         /*  If format numbers match exactly, we're done.  */
 173
 174         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 175             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 176                 return 0;
 177
 178         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 179
 180         return -EINVAL;
 181 }
 182
 183 static void end_bio_io_page(struct bio *bio)
 184 {
 185         struct page *page = bio->bi_private;
 186
 187         if (!bio->bi_status)
 188                 SetPageUptodate(page);
 189         else
 190                 pr_warn("error %d reading superblock\n", bio->bi_status);
 191         unlock_page(page);
 192 }
 193
 194 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 195 {
 196         struct gfs2_sb_host *sb = &sdp->sd_sb;
 197         struct super_block *s = sdp->sd_vfs;
 198         const struct gfs2_sb *str = buf;
 199
 200         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 201         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 202         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 203         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 204         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 205         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 206         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 207         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 208         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 209         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 210         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 211
 212         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 213         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 214         memcpy(&s->s_uuid, str->sb_uuid, 16);
 215 }
 216
 217 /**
 218  * gfs2_read_super - Read the gfs2 super block from disk
 219  * @sdp: The GFS2 super block
 220  * @sector: The location of the super block
 221  * @error: The error code to return
 222  *
 223  * This uses the bio functions to read the super block from disk
 224  * because we want to be 100% sure that we never read cached data.
 225  * A super block is read twice only during each GFS2 mount and is
 226  * never written to by the filesystem. The first time its read no
 227  * locks are held, and the only details which are looked at are those
 228  * relating to the locking protocol. Once locking is up and working,
 229  * the sb is read again under the lock to establish the location of
 230  * the master directory (contains pointers to journals etc) and the
 231  * root directory.
 232  *
 233  * Returns: 0 on success or error
 234  */
 235
 236 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 237 {
 238         struct super_block *sb = sdp->sd_vfs;
 239         struct gfs2_sb *p;
 240         struct page *page;
 241         struct bio *bio;
 242
 243         page = alloc_page(GFP_NOFS);
 244         if (unlikely(!page))
 245                 return -ENOMEM;
 246
 247         ClearPageUptodate(page);
 248         ClearPageDirty(page);
 249         lock_page(page);
 250
 251         bio = bio_alloc(GFP_NOFS, 1);
 252         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 253         bio_set_dev(bio, sb->s_bdev);
 254         bio_add_page(bio, page, PAGE_SIZE, 0);
 255
 256         bio->bi_end_io = end_bio_io_page;
 257         bio->bi_private = page;
 258         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
 259         submit_bio(bio);
 260         wait_on_page_locked(page);
 261         bio_put(bio);
 262         if (!PageUptodate(page)) {
 263                 __free_page(page);
 264                 return -EIO;
 265         }
 266         p = kmap(page);
 267         gfs2_sb_in(sdp, p);
 268         kunmap(page);
 269         __free_page(page);
 270         return gfs2_check_sb(sdp, silent);
 271 }
 272
 273 /**
 274  * gfs2_read_sb - Read super block
 275  * @sdp: The GFS2 superblock
 276  * @silent: Don't print message if mount fails
 277  *
 278  */
 279
 280 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 281 {
 282         u32 hash_blocks, ind_blocks, leaf_blocks;
 283         u32 tmp_blocks;
 284         unsigned int x;
 285         int error;
 286
 287         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 288         if (error) {
 289                 if (!silent)
 290                         fs_err(sdp, "can't read superblock\n");
 291                 return error;
 292         }
 293
 294         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 295                                GFS2_BASIC_BLOCK_SHIFT;
 296         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 297         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 298                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 299         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 300                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 301         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
 302                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 303         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 304         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 305         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 306         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 307         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 308                                 sizeof(struct gfs2_meta_header)) /
 309                                 sizeof(struct gfs2_quota_change);
 310         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 311                                      sizeof(struct gfs2_meta_header))
 312                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 313
 314         /* Compute maximum reservation required to add a entry to a directory */
 315
 316         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 317                              sdp->sd_jbsize);
 318
 319         ind_blocks = 0;
 320         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 321                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 322                 ind_blocks += tmp_blocks;
 323         }
 324
 325         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 326
 327         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 328
 329         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 330                                 sizeof(struct gfs2_dinode);
 331         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 332         for (x = 2;; x++) {
 333                 u64 space, d;
 334                 u32 m;
 335
 336                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 337                 d = space;
 338                 m = do_div(d, sdp->sd_inptrs);
 339
 340                 if (d != sdp->sd_heightsize[x - 1] || m)
 341                         break;
 342                 sdp->sd_heightsize[x] = space;
 343         }
 344         sdp->sd_max_height = x;
 345         sdp->sd_heightsize[x] = ~0;
 346         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 347
 348         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 349                                       sizeof(struct gfs2_leaf)) /
 350                                      GFS2_MIN_DIRENT_SIZE;
 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_FREEZE_LOCK, &gfs2_freeze_glops,
 423                                CREATE, &sdp->sd_freeze_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_freeze_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,
 451                                   GFS2_BLKST_FREE /* ignore */);
 452         if (IS_ERR(inode)) {
 453                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 454                 return PTR_ERR(inode);
 455         }
 456         dentry = d_make_root(inode);
 457         if (!dentry) {
 458                 fs_err(sdp, "can't alloc %s dentry\n", name);
 459                 return -ENOMEM;
 460         }
 461         *dptr = dentry;
 462         return 0;
 463 }
 464
 465 static int init_sb(struct gfs2_sbd *sdp, int silent)
 466 {
 467         struct super_block *sb = sdp->sd_vfs;
 468         struct gfs2_holder sb_gh;
 469         u64 no_addr;
 470         int ret;
 471
 472         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 473                                 LM_ST_SHARED, 0, &sb_gh);
 474         if (ret) {
 475                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 476                 return ret;
 477         }
 478
 479         ret = gfs2_read_sb(sdp, silent);
 480         if (ret) {
 481                 fs_err(sdp, "can't read superblock: %d\n", ret);
 482                 goto out;
 483         }
 484
 485         /* Set up the buffer cache and SB for real */
 486         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 487                 ret = -EINVAL;
 488                 fs_err(sdp, "FS block size (%u) is too small for device "
 489                        "block size (%u)\n",
 490                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 491                 goto out;
 492         }
 493         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 494                 ret = -EINVAL;
 495                 fs_err(sdp, "FS block size (%u) is too big for machine "
 496                        "page size (%u)\n",
 497                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 498                 goto out;
 499         }
 500         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 501
 502         /* Get the root inode */
 503         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 504         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 505         if (ret)
 506                 goto out;
 507
 508         /* Get the master inode */
 509         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 510         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 511         if (ret) {
 512                 dput(sdp->sd_root_dir);
 513                 goto out;
 514         }
 515         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 516 out:
 517         gfs2_glock_dq_uninit(&sb_gh);
 518         return ret;
 519 }
 520
 521 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 522 {
 523         char *message = "FIRSTMOUNT=Done";
 524         char *envp[] = { message, NULL };
 525
 526         fs_info(sdp, "first mount done, others may mount\n");
 527
 528         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 529                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 530
 531         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 532 }
 533
 534 /**
 535  * gfs2_jindex_hold - Grab a lock on the jindex
 536  * @sdp: The GFS2 superblock
 537  * @ji_gh: the holder for the jindex glock
 538  *
 539  * Returns: errno
 540  */
 541
 542 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 543 {
 544         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 545         struct qstr name;
 546         char buf[20];
 547         struct gfs2_jdesc *jd;
 548         int error;
 549
 550         name.name = buf;
 551
 552         mutex_lock(&sdp->sd_jindex_mutex);
 553
 554         for (;;) {
 555                 struct gfs2_inode *jip;
 556
 557                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 558                 if (error)
 559                         break;
 560
 561                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 562                 name.hash = gfs2_disk_hash(name.name, name.len);
 563
 564                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 565                 if (error == -ENOENT) {
 566                         error = 0;
 567                         break;
 568                 }
 569
 570                 gfs2_glock_dq_uninit(ji_gh);
 571
 572                 if (error)
 573                         break;
 574
 575                 error = -ENOMEM;
 576                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 577                 if (!jd)
 578                         break;
 579
 580                 INIT_LIST_HEAD(&jd->extent_list);
 581                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 582
 583                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 584                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 585                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 586                         if (!jd->jd_inode)
 587                                 error = -ENOENT;
 588                         else
 589                                 error = PTR_ERR(jd->jd_inode);
 590                         kfree(jd);
 591                         break;
 592                 }
 593
 594                 spin_lock(&sdp->sd_jindex_spin);
 595                 jd->jd_jid = sdp->sd_journals++;
 596                 jip = GFS2_I(jd->jd_inode);
 597                 jd->jd_no_addr = jip->i_no_addr;
 598                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 599                 spin_unlock(&sdp->sd_jindex_spin);
 600         }
 601
 602         mutex_unlock(&sdp->sd_jindex_mutex);
 603
 604         return error;
 605 }
 606
 607 static int init_journal(struct gfs2_sbd *sdp, int undo)
 608 {
 609         struct inode *master = d_inode(sdp->sd_master_dir);
 610         struct gfs2_holder ji_gh;
 611         struct gfs2_inode *ip;
 612         int jindex = 1;
 613         int error = 0;
 614
 615         if (undo) {
 616                 jindex = 0;
 617                 goto fail_jinode_gh;
 618         }
 619
 620         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 621         if (IS_ERR(sdp->sd_jindex)) {
 622                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 623                 return PTR_ERR(sdp->sd_jindex);
 624         }
 625
 626         /* Load in the journal index special file */
 627
 628         error = gfs2_jindex_hold(sdp, &ji_gh);
 629         if (error) {
 630                 fs_err(sdp, "can't read journal index: %d\n", error);
 631                 goto fail;
 632         }
 633
 634         error = -EUSERS;
 635         if (!gfs2_jindex_size(sdp)) {
 636                 fs_err(sdp, "no journals!\n");
 637                 goto fail_jindex;
 638         }
 639
 640         atomic_set(&sdp->sd_log_blks_needed, 0);
 641         if (sdp->sd_args.ar_spectator) {
 642                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 643                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 644                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 645                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 646         } else {
 647                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 648                         fs_err(sdp, "can't mount journal #%u\n",
 649                                sdp->sd_lockstruct.ls_jid);
 650                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 651                                gfs2_jindex_size(sdp),
 652                                gfs2_jindex_size(sdp) - 1);
 653                         goto fail_jindex;
 654                 }
 655                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 656
 657                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 658                                           &gfs2_journal_glops,
 659                                           LM_ST_EXCLUSIVE,
 660                                           LM_FLAG_NOEXP | GL_NOCACHE,
 661                                           &sdp->sd_journal_gh);
 662                 if (error) {
 663                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 664                         goto fail_jindex;
 665                 }
 666
 667                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 668                 sdp->sd_jinode_gl = ip->i_gl;
 669                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 670                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 671                                            &sdp->sd_jinode_gh);
 672                 if (error) {
 673                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 674                                error);
 675                         goto fail_journal_gh;
 676                 }
 677
 678                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 679                 if (error) {
 680                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 681                                sdp->sd_jdesc->jd_jid, error);
 682                         goto fail_jinode_gh;
 683                 }
 684                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 685                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 686                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 687
 688                 /* Map the extents for this journal's blocks */
 689                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 690         }
 691         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 692
 693         if (sdp->sd_lockstruct.ls_first) {
 694                 unsigned int x;
 695                 for (x = 0; x < sdp->sd_journals; x++) {
 696                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 697
 698                         if (sdp->sd_args.ar_spectator) {
 699                                 error = check_journal_clean(sdp, jd, true);
 700                                 if (error)
 701                                         goto fail_jinode_gh;
 702                                 continue;
 703                         }
 704                         error = gfs2_recover_journal(jd, true);
 705                         if (error) {
 706                                 fs_err(sdp, "error recovering journal %u: %d\n",
 707                                        x, error);
 708                                 goto fail_jinode_gh;
 709                         }
 710                 }
 711
 712                 gfs2_others_may_mount(sdp);
 713         } else if (!sdp->sd_args.ar_spectator) {
 714                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 715                 if (error) {
 716                         fs_err(sdp, "error recovering my journal: %d\n", error);
 717                         goto fail_jinode_gh;
 718                 }
 719         }
 720
 721         sdp->sd_log_idle = 1;
 722         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 723         gfs2_glock_dq_uninit(&ji_gh);
 724         jindex = 0;
 725         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 726         return 0;
 727
 728 fail_jinode_gh:
 729         /* A withdraw may have done dq/uninit so now we need to check it */
 730         if (!sdp->sd_args.ar_spectator &&
 731             gfs2_holder_initialized(&sdp->sd_jinode_gh))
 732                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 733 fail_journal_gh:
 734         if (!sdp->sd_args.ar_spectator &&
 735             gfs2_holder_initialized(&sdp->sd_journal_gh))
 736                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 737 fail_jindex:
 738         gfs2_jindex_free(sdp);
 739         if (jindex)
 740                 gfs2_glock_dq_uninit(&ji_gh);
 741 fail:
 742         iput(sdp->sd_jindex);
 743         return error;
 744 }
 745
 746 static struct lock_class_key gfs2_quota_imutex_key;
 747
 748 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 749 {
 750         int error = 0;
 751         struct inode *master = d_inode(sdp->sd_master_dir);
 752
 753         if (undo)
 754                 goto fail_qinode;
 755
 756         error = init_journal(sdp, undo);
 757         complete_all(&sdp->sd_journal_ready);
 758         if (error)
 759                 goto fail;
 760
 761         /* Read in the master statfs inode */
 762         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 763         if (IS_ERR(sdp->sd_statfs_inode)) {
 764                 error = PTR_ERR(sdp->sd_statfs_inode);
 765                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 766                 goto fail_journal;
 767         }
 768
 769         /* Read in the resource index inode */
 770         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 771         if (IS_ERR(sdp->sd_rindex)) {
 772                 error = PTR_ERR(sdp->sd_rindex);
 773                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 774                 goto fail_statfs;
 775         }
 776         sdp->sd_rindex_uptodate = 0;
 777
 778         /* Read in the quota inode */
 779         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 780         if (IS_ERR(sdp->sd_quota_inode)) {
 781                 error = PTR_ERR(sdp->sd_quota_inode);
 782                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 783                 goto fail_rindex;
 784         }
 785         /*
 786          * i_rwsem on quota files is special. Since this inode is hidden system
 787          * file, we are safe to define locking ourselves.
 788          */
 789         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 790                           &gfs2_quota_imutex_key);
 791
 792         error = gfs2_rindex_update(sdp);
 793         if (error)
 794                 goto fail_qinode;
 795
 796         return 0;
 797
 798 fail_qinode:
 799         iput(sdp->sd_quota_inode);
 800 fail_rindex:
 801         gfs2_clear_rgrpd(sdp);
 802         iput(sdp->sd_rindex);
 803 fail_statfs:
 804         iput(sdp->sd_statfs_inode);
 805 fail_journal:
 806         init_journal(sdp, UNDO);
 807 fail:
 808         return error;
 809 }
 810
 811 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 812 {
 813         struct inode *pn = NULL;
 814         char buf[30];
 815         int error = 0;
 816         struct gfs2_inode *ip;
 817         struct inode *master = d_inode(sdp->sd_master_dir);
 818
 819         if (sdp->sd_args.ar_spectator)
 820                 return 0;
 821
 822         if (undo)
 823                 goto fail_qc_gh;
 824
 825         pn = gfs2_lookup_simple(master, "per_node");
 826         if (IS_ERR(pn)) {
 827                 error = PTR_ERR(pn);
 828                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 829                 return error;
 830         }
 831
 832         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 833         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 834         if (IS_ERR(sdp->sd_sc_inode)) {
 835                 error = PTR_ERR(sdp->sd_sc_inode);
 836                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 837                 goto fail;
 838         }
 839
 840         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 841         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 842         if (IS_ERR(sdp->sd_qc_inode)) {
 843                 error = PTR_ERR(sdp->sd_qc_inode);
 844                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 845                 goto fail_ut_i;
 846         }
 847
 848         iput(pn);
 849         pn = NULL;
 850
 851         ip = GFS2_I(sdp->sd_sc_inode);
 852         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 853                                    &sdp->sd_sc_gh);
 854         if (error) {
 855                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 856                 goto fail_qc_i;
 857         }
 858
 859         ip = GFS2_I(sdp->sd_qc_inode);
 860         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 861                                    &sdp->sd_qc_gh);
 862         if (error) {
 863                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 864                 goto fail_ut_gh;
 865         }
 866
 867         return 0;
 868
 869 fail_qc_gh:
 870         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 871 fail_ut_gh:
 872         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 873 fail_qc_i:
 874         iput(sdp->sd_qc_inode);
 875 fail_ut_i:
 876         iput(sdp->sd_sc_inode);
 877 fail:
 878         iput(pn);
 879         return error;
 880 }
 881
 882 static const match_table_t nolock_tokens = {
 883         { Opt_jid, "jid=%d", },
 884         { Opt_err, NULL },
 885 };
 886
 887 static const struct lm_lockops nolock_ops = {
 888         .lm_proto_name = "lock_nolock",
 889         .lm_put_lock = gfs2_glock_free,
 890         .lm_tokens = &nolock_tokens,
 891 };
 892
 893 /**
 894  * gfs2_lm_mount - mount a locking protocol
 895  * @sdp: the filesystem
 896  * @args: mount arguments
 897  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 898  *
 899  * Returns: errno
 900  */
 901
 902 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 903 {
 904         const struct lm_lockops *lm;
 905         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 906         struct gfs2_args *args = &sdp->sd_args;
 907         const char *proto = sdp->sd_proto_name;
 908         const char *table = sdp->sd_table_name;
 909         char *o, *options;
 910         int ret;
 911
 912         if (!strcmp("lock_nolock", proto)) {
 913                 lm = &nolock_ops;
 914                 sdp->sd_args.ar_localflocks = 1;
 915 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 916         } else if (!strcmp("lock_dlm", proto)) {
 917                 lm = &gfs2_dlm_ops;
 918 #endif
 919         } else {
 920                 pr_info("can't find protocol %s\n", proto);
 921                 return -ENOENT;
 922         }
 923
 924         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 925
 926         ls->ls_ops = lm;
 927         ls->ls_first = 1;
 928
 929         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 930                 substring_t tmp[MAX_OPT_ARGS];
 931                 int token, option;
 932
 933                 if (!o || !*o)
 934                         continue;
 935
 936                 token = match_token(o, *lm->lm_tokens, tmp);
 937                 switch (token) {
 938                 case Opt_jid:
 939                         ret = match_int(&tmp[0], &option);
 940                         if (ret || option < 0)
 941                                 goto hostdata_error;
 942                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 943                                 ls->ls_jid = option;
 944                         break;
 945                 case Opt_id:
 946                 case Opt_nodir:
 947                         /* Obsolete, but left for backward compat purposes */
 948                         break;
 949                 case Opt_first:
 950                         ret = match_int(&tmp[0], &option);
 951                         if (ret || (option != 0 && option != 1))
 952                                 goto hostdata_error;
 953                         ls->ls_first = option;
 954                         break;
 955                 case Opt_err:
 956                 default:
 957 hostdata_error:
 958                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 959                         return -EINVAL;
 960                 }
 961         }
 962
 963         if (lm->lm_mount == NULL) {
 964                 fs_info(sdp, "Now mounting FS...\n");
 965                 complete_all(&sdp->sd_locking_init);
 966                 return 0;
 967         }
 968         ret = lm->lm_mount(sdp, table);
 969         if (ret == 0)
 970                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
 971         complete_all(&sdp->sd_locking_init);
 972         return ret;
 973 }
 974
 975 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
 976 {
 977         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
 978         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
 979                 lm->lm_unmount(sdp);
 980 }
 981
 982 static int wait_on_journal(struct gfs2_sbd *sdp)
 983 {
 984         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
 985                 return 0;
 986
 987         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
 988                 ? -EINTR : 0;
 989 }
 990
 991 void gfs2_online_uevent(struct gfs2_sbd *sdp)
 992 {
 993         struct super_block *sb = sdp->sd_vfs;
 994         char ro[20];
 995         char spectator[20];
 996         char *envp[] = { ro, spectator, NULL };
 997         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
 998         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
 999         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1000 }
1001
1002 /**
1003  * gfs2_fill_super - Read in superblock
1004  * @sb: The VFS superblock
1005  * @args: Mount options
1006  * @silent: Don't complain if it's not a GFS2 filesystem
1007  *
1008  * Returns: -errno
1009  */
1010 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1011 {
1012         struct gfs2_args *args = fc->fs_private;
1013         int silent = fc->sb_flags & SB_SILENT;
1014         struct gfs2_sbd *sdp;
1015         struct gfs2_holder mount_gh;
1016         int error;
1017
1018         sdp = init_sbd(sb);
1019         if (!sdp) {
1020                 pr_warn("can't alloc struct gfs2_sbd\n");
1021                 return -ENOMEM;
1022         }
1023         sdp->sd_args = *args;
1024
1025         if (sdp->sd_args.ar_spectator) {
1026                 sb->s_flags |= SB_RDONLY;
1027                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1028         }
1029         if (sdp->sd_args.ar_posix_acl)
1030                 sb->s_flags |= SB_POSIXACL;
1031         if (sdp->sd_args.ar_nobarrier)
1032                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1033
1034         sb->s_flags |= SB_NOSEC;
1035         sb->s_magic = GFS2_MAGIC;
1036         sb->s_op = &gfs2_super_ops;
1037         sb->s_d_op = &gfs2_dops;
1038         sb->s_export_op = &gfs2_export_ops;
1039         sb->s_xattr = gfs2_xattr_handlers;
1040         sb->s_qcop = &gfs2_quotactl_ops;
1041         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1042         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1043         sb->s_time_gran = 1;
1044         sb->s_maxbytes = MAX_LFS_FILESIZE;
1045
1046         /* Set up the buffer cache and fill in some fake block size values
1047            to allow us to read-in the on-disk superblock. */
1048         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1049         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1050         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1051                                GFS2_BASIC_BLOCK_SHIFT;
1052         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1053
1054         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1055         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1056         if (sdp->sd_args.ar_statfs_quantum) {
1057                 sdp->sd_tune.gt_statfs_slow = 0;
1058                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1059         } else {
1060                 sdp->sd_tune.gt_statfs_slow = 1;
1061                 sdp->sd_tune.gt_statfs_quantum = 30;
1062         }
1063
1064         error = init_names(sdp, silent);
1065         if (error) {
1066                 /* In this case, we haven't initialized sysfs, so we have to
1067                    manually free the sdp. */
1068                 free_sbd(sdp);
1069                 sb->s_fs_info = NULL;
1070                 return error;
1071         }
1072
1073         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1074
1075         error = gfs2_sys_fs_add(sdp);
1076         /*
1077          * If we hit an error here, gfs2_sys_fs_add will have called function
1078          * kobject_put which causes the sysfs usage count to go to zero, which
1079          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1080          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1081          * kobject_put to free sdp.
1082          */
1083         if (error)
1084                 return error;
1085
1086         gfs2_create_debugfs_file(sdp);
1087
1088         error = gfs2_lm_mount(sdp, silent);
1089         if (error)
1090                 goto fail_debug;
1091
1092         error = init_locking(sdp, &mount_gh, DO);
1093         if (error)
1094                 goto fail_lm;
1095
1096         error = init_sb(sdp, silent);
1097         if (error)
1098                 goto fail_locking;
1099
1100         error = wait_on_journal(sdp);
1101         if (error)
1102                 goto fail_sb;
1103
1104         /*
1105          * If user space has failed to join the cluster or some similar
1106          * failure has occurred, then the journal id will contain a
1107          * negative (error) number. This will then be returned to the
1108          * caller (of the mount syscall). We do this even for spectator
1109          * mounts (which just write a jid of 0 to indicate "ok" even though
1110          * the jid is unused in the spectator case)
1111          */
1112         if (sdp->sd_lockstruct.ls_jid < 0) {
1113                 error = sdp->sd_lockstruct.ls_jid;
1114                 sdp->sd_lockstruct.ls_jid = 0;
1115                 goto fail_sb;
1116         }
1117
1118         if (sdp->sd_args.ar_spectator)
1119                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1120                          sdp->sd_table_name);
1121         else
1122                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1123                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1124
1125         error = init_inodes(sdp, DO);
1126         if (error)
1127                 goto fail_sb;
1128
1129         error = init_per_node(sdp, DO);
1130         if (error)
1131                 goto fail_inodes;
1132
1133         error = gfs2_statfs_init(sdp);
1134         if (error) {
1135                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1136                 goto fail_per_node;
1137         }
1138
1139         if (sb_rdonly(sb)) {
1140                 struct gfs2_holder freeze_gh;
1141
1142                 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1143                                            LM_FLAG_NOEXP | GL_EXACT,
1144                                            &freeze_gh);
1145                 if (error) {
1146                         fs_err(sdp, "can't make FS RO: %d\n", error);
1147                         goto fail_per_node;
1148                 }
1149                 gfs2_glock_dq_uninit(&freeze_gh);
1150         } else {
1151                 error = gfs2_make_fs_rw(sdp);
1152                 if (error) {
1153                         fs_err(sdp, "can't make FS RW: %d\n", error);
1154                         goto fail_per_node;
1155                 }
1156         }
1157
1158         gfs2_glock_dq_uninit(&mount_gh);
1159         gfs2_online_uevent(sdp);
1160         return 0;
1161
1162 fail_per_node:
1163         init_per_node(sdp, UNDO);
1164 fail_inodes:
1165         init_inodes(sdp, UNDO);
1166 fail_sb:
1167         if (sdp->sd_root_dir)
1168                 dput(sdp->sd_root_dir);
1169         if (sdp->sd_master_dir)
1170                 dput(sdp->sd_master_dir);
1171         if (sb->s_root)
1172                 dput(sb->s_root);
1173         sb->s_root = NULL;
1174 fail_locking:
1175         init_locking(sdp, &mount_gh, UNDO);
1176 fail_lm:
1177         complete_all(&sdp->sd_journal_ready);
1178         gfs2_gl_hash_clear(sdp);
1179         gfs2_lm_unmount(sdp);
1180 fail_debug:
1181         gfs2_delete_debugfs_file(sdp);
1182         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1183          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1184         gfs2_sys_fs_del(sdp);
1185         sb->s_fs_info = NULL;
1186         return error;
1187 }
1188
1189 /**
1190  * gfs2_get_tree - Get the GFS2 superblock and root directory
1191  * @fc: The filesystem context
1192  *
1193  * Returns: 0 or -errno on error
1194  */
1195 static int gfs2_get_tree(struct fs_context *fc)
1196 {
1197         struct gfs2_args *args = fc->fs_private;
1198         struct gfs2_sbd *sdp;
1199         int error;
1200
1201         error = get_tree_bdev(fc, gfs2_fill_super);
1202         if (error)
1203                 return error;
1204
1205         sdp = fc->root->d_sb->s_fs_info;
1206         dput(fc->root);
1207         if (args->ar_meta)
1208                 fc->root = dget(sdp->sd_master_dir);
1209         else
1210                 fc->root = dget(sdp->sd_root_dir);
1211         return 0;
1212 }
1213
1214 static void gfs2_fc_free(struct fs_context *fc)
1215 {
1216         struct gfs2_args *args = fc->fs_private;
1217
1218         kfree(args);
1219 }
1220
1221 enum gfs2_param {
1222         Opt_lockproto,
1223         Opt_locktable,
1224         Opt_hostdata,
1225         Opt_spectator,
1226         Opt_ignore_local_fs,
1227         Opt_localflocks,
1228         Opt_localcaching,
1229         Opt_debug,
1230         Opt_upgrade,
1231         Opt_acl,
1232         Opt_quota,
1233         Opt_quota_flag,
1234         Opt_suiddir,
1235         Opt_data,
1236         Opt_meta,
1237         Opt_discard,
1238         Opt_commit,
1239         Opt_errors,
1240         Opt_statfs_quantum,
1241         Opt_statfs_percent,
1242         Opt_quota_quantum,
1243         Opt_barrier,
1244         Opt_rgrplvb,
1245         Opt_loccookie,
1246 };
1247
1248 static const struct constant_table gfs2_param_quota[] = {
1249         {"off",        GFS2_QUOTA_OFF},
1250         {"account",    GFS2_QUOTA_ACCOUNT},
1251         {"on",         GFS2_QUOTA_ON},
1252         {}
1253 };
1254
1255 enum opt_data {
1256         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1257         Opt_data_ordered   = GFS2_DATA_ORDERED,
1258 };
1259
1260 static const struct constant_table gfs2_param_data[] = {
1261         {"writeback",  Opt_data_writeback },
1262         {"ordered",    Opt_data_ordered },
1263         {}
1264 };
1265
1266 enum opt_errors {
1267         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1268         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1269 };
1270
1271 static const struct constant_table gfs2_param_errors[] = {
1272         {"withdraw",   Opt_errors_withdraw },
1273         {"panic",      Opt_errors_panic },
1274         {}
1275 };
1276
1277 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1278         fsparam_string ("lockproto",          Opt_lockproto),
1279         fsparam_string ("locktable",          Opt_locktable),
1280         fsparam_string ("hostdata",           Opt_hostdata),
1281         fsparam_flag   ("spectator",          Opt_spectator),
1282         fsparam_flag   ("norecovery",         Opt_spectator),
1283         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1284         fsparam_flag   ("localflocks",        Opt_localflocks),
1285         fsparam_flag   ("localcaching",       Opt_localcaching),
1286         fsparam_flag_no("debug",              Opt_debug),
1287         fsparam_flag   ("upgrade",            Opt_upgrade),
1288         fsparam_flag_no("acl",                Opt_acl),
1289         fsparam_flag_no("suiddir",            Opt_suiddir),
1290         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1291         fsparam_flag   ("meta",               Opt_meta),
1292         fsparam_flag_no("discard",            Opt_discard),
1293         fsparam_s32    ("commit",             Opt_commit),
1294         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1295         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1296         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1297         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1298         fsparam_flag_no("barrier",            Opt_barrier),
1299         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1300         fsparam_flag_no("loccookie",          Opt_loccookie),
1301         /* quota can be a flag or an enum so it gets special treatment */
1302         fsparam_flag_no("quota",              Opt_quota_flag),
1303         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1304         {}
1305 };
1306
1307 /* Parse a single mount parameter */
1308 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1309 {
1310         struct gfs2_args *args = fc->fs_private;
1311         struct fs_parse_result result;
1312         int o;
1313
1314         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1315         if (o < 0)
1316                 return o;
1317
1318         switch (o) {
1319         case Opt_lockproto:
1320                 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1321                 break;
1322         case Opt_locktable:
1323                 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1324                 break;
1325         case Opt_hostdata:
1326                 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1327                 break;
1328         case Opt_spectator:
1329                 args->ar_spectator = 1;
1330                 break;
1331         case Opt_ignore_local_fs:
1332                 /* Retained for backwards compat only */
1333                 break;
1334         case Opt_localflocks:
1335                 args->ar_localflocks = 1;
1336                 break;
1337         case Opt_localcaching:
1338                 /* Retained for backwards compat only */
1339                 break;
1340         case Opt_debug:
1341                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1342                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1343                 args->ar_debug = result.boolean;
1344                 break;
1345         case Opt_upgrade:
1346                 /* Retained for backwards compat only */
1347                 break;
1348         case Opt_acl:
1349                 args->ar_posix_acl = result.boolean;
1350                 break;
1351         case Opt_quota_flag:
1352                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1353                 break;
1354         case Opt_quota:
1355                 args->ar_quota = result.int_32;
1356                 break;
1357         case Opt_suiddir:
1358                 args->ar_suiddir = result.boolean;
1359                 break;
1360         case Opt_data:
1361                 /* The uint_32 result maps directly to GFS2_DATA_* */
1362                 args->ar_data = result.uint_32;
1363                 break;
1364         case Opt_meta:
1365                 args->ar_meta = 1;
1366                 break;
1367         case Opt_discard:
1368                 args->ar_discard = result.boolean;
1369                 break;
1370         case Opt_commit:
1371                 if (result.int_32 <= 0)
1372                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1373                 args->ar_commit = result.int_32;
1374                 break;
1375         case Opt_statfs_quantum:
1376                 if (result.int_32 < 0)
1377                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1378                 args->ar_statfs_quantum = result.int_32;
1379                 break;
1380         case Opt_quota_quantum:
1381                 if (result.int_32 <= 0)
1382                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1383                 args->ar_quota_quantum = result.int_32;
1384                 break;
1385         case Opt_statfs_percent:
1386                 if (result.int_32 < 0 || result.int_32 > 100)
1387                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1388                 args->ar_statfs_percent = result.int_32;
1389                 break;
1390         case Opt_errors:
1391                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1392                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1393                 args->ar_errors = result.uint_32;
1394                 break;
1395         case Opt_barrier:
1396                 args->ar_nobarrier = result.boolean;
1397                 break;
1398         case Opt_rgrplvb:
1399                 args->ar_rgrplvb = result.boolean;
1400                 break;
1401         case Opt_loccookie:
1402                 args->ar_loccookie = result.boolean;
1403                 break;
1404         default:
1405                 return invalfc(fc, "invalid mount option: %s", param->key);
1406         }
1407         return 0;
1408 }
1409
1410 static int gfs2_reconfigure(struct fs_context *fc)
1411 {
1412         struct super_block *sb = fc->root->d_sb;
1413         struct gfs2_sbd *sdp = sb->s_fs_info;
1414         struct gfs2_args *oldargs = &sdp->sd_args;
1415         struct gfs2_args *newargs = fc->fs_private;
1416         struct gfs2_tune *gt = &sdp->sd_tune;
1417         int error = 0;
1418
1419         sync_filesystem(sb);
1420
1421         spin_lock(&gt->gt_spin);
1422         oldargs->ar_commit = gt->gt_logd_secs;
1423         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1424         if (gt->gt_statfs_slow)
1425                 oldargs->ar_statfs_quantum = 0;
1426         else
1427                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1428         spin_unlock(&gt->gt_spin);
1429
1430         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1431                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1432                 return -EINVAL;
1433         }
1434         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1435                 errorfc(fc, "reconfiguration of lock table not allowed");
1436                 return -EINVAL;
1437         }
1438         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1439                 errorfc(fc, "reconfiguration of host data not allowed");
1440                 return -EINVAL;
1441         }
1442         if (newargs->ar_spectator != oldargs->ar_spectator) {
1443                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1444                 return -EINVAL;
1445         }
1446         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1447                 errorfc(fc, "reconfiguration of localflocks not allowed");
1448                 return -EINVAL;
1449         }
1450         if (newargs->ar_meta != oldargs->ar_meta) {
1451                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1452                 return -EINVAL;
1453         }
1454         if (oldargs->ar_spectator)
1455                 fc->sb_flags |= SB_RDONLY;
1456
1457         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1458                 if (fc->sb_flags & SB_RDONLY) {
1459                         error = gfs2_make_fs_ro(sdp);
1460                         if (error)
1461                                 errorfc(fc, "unable to remount read-only");
1462                 } else {
1463                         error = gfs2_make_fs_rw(sdp);
1464                         if (error)
1465                                 errorfc(fc, "unable to remount read-write");
1466                 }
1467         }
1468         sdp->sd_args = *newargs;
1469
1470         if (sdp->sd_args.ar_posix_acl)
1471                 sb->s_flags |= SB_POSIXACL;
1472         else
1473                 sb->s_flags &= ~SB_POSIXACL;
1474         if (sdp->sd_args.ar_nobarrier)
1475                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1476         else
1477                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1478         spin_lock(&gt->gt_spin);
1479         gt->gt_logd_secs = newargs->ar_commit;
1480         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1481         if (newargs->ar_statfs_quantum) {
1482                 gt->gt_statfs_slow = 0;
1483                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1484         }
1485         else {
1486                 gt->gt_statfs_slow = 1;
1487                 gt->gt_statfs_quantum = 30;
1488         }
1489         spin_unlock(&gt->gt_spin);
1490
1491         gfs2_online_uevent(sdp);
1492         return error;
1493 }
1494
1495 static const struct fs_context_operations gfs2_context_ops = {
1496         .free        = gfs2_fc_free,
1497         .parse_param = gfs2_parse_param,
1498         .get_tree    = gfs2_get_tree,
1499         .reconfigure = gfs2_reconfigure,
1500 };
1501
1502 /* Set up the filesystem mount context */
1503 static int gfs2_init_fs_context(struct fs_context *fc)
1504 {
1505         struct gfs2_args *args;
1506
1507         args = kmalloc(sizeof(*args), GFP_KERNEL);
1508         if (args == NULL)
1509                 return -ENOMEM;
1510
1511         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1512                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1513
1514                 *args = sdp->sd_args;
1515         } else {
1516                 memset(args, 0, sizeof(*args));
1517                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1518                 args->ar_data = GFS2_DATA_DEFAULT;
1519                 args->ar_commit = 30;
1520                 args->ar_statfs_quantum = 30;
1521                 args->ar_quota_quantum = 60;
1522                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1523         }
1524         fc->fs_private = args;
1525         fc->ops = &gfs2_context_ops;
1526         return 0;
1527 }
1528
1529 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1530 {
1531         return -EINVAL;
1532 }
1533
1534 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1535 {
1536         return (fc->sget_key == s->s_bdev);
1537 }
1538
1539 static int gfs2_meta_get_tree(struct fs_context *fc)
1540 {
1541         struct super_block *s;
1542         struct gfs2_sbd *sdp;
1543         struct path path;
1544         int error;
1545
1546         if (!fc->source || !*fc->source)
1547                 return -EINVAL;
1548
1549         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1550         if (error) {
1551                 pr_warn("path_lookup on %s returned error %d\n",
1552                         fc->source, error);
1553                 return error;
1554         }
1555         fc->fs_type = &gfs2_fs_type;
1556         fc->sget_key = path.dentry->d_sb->s_bdev;
1557         s = sget_fc(fc, test_meta_super, set_meta_super);
1558         path_put(&path);
1559         if (IS_ERR(s)) {
1560                 pr_warn("gfs2 mount does not exist\n");
1561                 return PTR_ERR(s);
1562         }
1563         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1564                 deactivate_locked_super(s);
1565                 return -EBUSY;
1566         }
1567         sdp = s->s_fs_info;
1568         fc->root = dget(sdp->sd_master_dir);
1569         return 0;
1570 }
1571
1572 static const struct fs_context_operations gfs2_meta_context_ops = {
1573         .free        = gfs2_fc_free,
1574         .get_tree    = gfs2_meta_get_tree,
1575 };
1576
1577 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1578 {
1579         int ret = gfs2_init_fs_context(fc);
1580
1581         if (ret)
1582                 return ret;
1583
1584         fc->ops = &gfs2_meta_context_ops;
1585         return 0;
1586 }
1587
1588 static void gfs2_kill_sb(struct super_block *sb)
1589 {
1590         struct gfs2_sbd *sdp = sb->s_fs_info;
1591
1592         if (sdp == NULL) {
1593                 kill_block_super(sb);
1594                 return;
1595         }
1596
1597         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1598         dput(sdp->sd_root_dir);
1599         dput(sdp->sd_master_dir);
1600         sdp->sd_root_dir = NULL;
1601         sdp->sd_master_dir = NULL;
1602         shrink_dcache_sb(sb);
1603         kill_block_super(sb);
1604 }
1605
1606 struct file_system_type gfs2_fs_type = {
1607         .name = "gfs2",
1608         .fs_flags = FS_REQUIRES_DEV,
1609         .init_fs_context = gfs2_init_fs_context,
1610         .parameters = gfs2_fs_parameters,
1611         .kill_sb = gfs2_kill_sb,
1612         .owner = THIS_MODULE,
1613 };
1614 MODULE_ALIAS_FS("gfs2");
1615
1616 struct file_system_type gfs2meta_fs_type = {
1617         .name = "gfs2meta",
1618         .fs_flags = FS_REQUIRES_DEV,
1619         .init_fs_context = gfs2_meta_init_fs_context,
1620         .owner = THIS_MODULE,
1621 };
1622 MODULE_ALIAS_FS("gfs2meta");