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_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 302         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 303         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 304         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 305         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 306                                 sizeof(struct gfs2_meta_header)) /
 307                                 sizeof(struct gfs2_quota_change);
 308         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 309                                      sizeof(struct gfs2_meta_header))
 310                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 311
 312         /* Compute maximum reservation required to add a entry to a directory */
 313
 314         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 315                              sdp->sd_jbsize);
 316
 317         ind_blocks = 0;
 318         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 319                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 320                 ind_blocks += tmp_blocks;
 321         }
 322
 323         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 324
 325         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 326
 327         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 328                                 sizeof(struct gfs2_dinode);
 329         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 330         for (x = 2;; x++) {
 331                 u64 space, d;
 332                 u32 m;
 333
 334                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 335                 d = space;
 336                 m = do_div(d, sdp->sd_inptrs);
 337
 338                 if (d != sdp->sd_heightsize[x - 1] || m)
 339                         break;
 340                 sdp->sd_heightsize[x] = space;
 341         }
 342         sdp->sd_max_height = x;
 343         sdp->sd_heightsize[x] = ~0;
 344         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 345
 346         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 347                                       sizeof(struct gfs2_leaf)) /
 348                                      GFS2_MIN_DIRENT_SIZE;
 349         return 0;
 350 }
 351
 352 static int init_names(struct gfs2_sbd *sdp, int silent)
 353 {
 354         char *proto, *table;
 355         int error = 0;
 356
 357         proto = sdp->sd_args.ar_lockproto;
 358         table = sdp->sd_args.ar_locktable;
 359
 360         /*  Try to autodetect  */
 361
 362         if (!proto[0] || !table[0]) {
 363                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 364                 if (error)
 365                         return error;
 366
 367                 if (!proto[0])
 368                         proto = sdp->sd_sb.sb_lockproto;
 369                 if (!table[0])
 370                         table = sdp->sd_sb.sb_locktable;
 371         }
 372
 373         if (!table[0])
 374                 table = sdp->sd_vfs->s_id;
 375
 376         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 377         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 378
 379         table = sdp->sd_table_name;
 380         while ((table = strchr(table, '/')))
 381                 *table = '_';
 382
 383         return error;
 384 }
 385
 386 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 387                         int undo)
 388 {
 389         int error = 0;
 390
 391         if (undo)
 392                 goto fail_trans;
 393
 394         error = gfs2_glock_nq_num(sdp,
 395                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 396                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 397                                   mount_gh);
 398         if (error) {
 399                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 400                 goto fail;
 401         }
 402
 403         error = gfs2_glock_nq_num(sdp,
 404                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 405                                   LM_ST_SHARED,
 406                                   LM_FLAG_NOEXP | GL_EXACT,
 407                                   &sdp->sd_live_gh);
 408         if (error) {
 409                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 410                 goto fail_mount;
 411         }
 412
 413         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 414                                CREATE, &sdp->sd_rename_gl);
 415         if (error) {
 416                 fs_err(sdp, "can't create rename glock: %d\n", error);
 417                 goto fail_live;
 418         }
 419
 420         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 421                                CREATE, &sdp->sd_freeze_gl);
 422         if (error) {
 423                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 424                 goto fail_rename;
 425         }
 426
 427         return 0;
 428
 429 fail_trans:
 430         gfs2_glock_put(sdp->sd_freeze_gl);
 431 fail_rename:
 432         gfs2_glock_put(sdp->sd_rename_gl);
 433 fail_live:
 434         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 435 fail_mount:
 436         gfs2_glock_dq_uninit(mount_gh);
 437 fail:
 438         return error;
 439 }
 440
 441 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 442                             u64 no_addr, const char *name)
 443 {
 444         struct gfs2_sbd *sdp = sb->s_fs_info;
 445         struct dentry *dentry;
 446         struct inode *inode;
 447
 448         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 449                                   GFS2_BLKST_FREE /* ignore */);
 450         if (IS_ERR(inode)) {
 451                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 452                 return PTR_ERR(inode);
 453         }
 454         dentry = d_make_root(inode);
 455         if (!dentry) {
 456                 fs_err(sdp, "can't alloc %s dentry\n", name);
 457                 return -ENOMEM;
 458         }
 459         *dptr = dentry;
 460         return 0;
 461 }
 462
 463 static int init_sb(struct gfs2_sbd *sdp, int silent)
 464 {
 465         struct super_block *sb = sdp->sd_vfs;
 466         struct gfs2_holder sb_gh;
 467         u64 no_addr;
 468         int ret;
 469
 470         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 471                                 LM_ST_SHARED, 0, &sb_gh);
 472         if (ret) {
 473                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 474                 return ret;
 475         }
 476
 477         ret = gfs2_read_sb(sdp, silent);
 478         if (ret) {
 479                 fs_err(sdp, "can't read superblock: %d\n", ret);
 480                 goto out;
 481         }
 482
 483         /* Set up the buffer cache and SB for real */
 484         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 485                 ret = -EINVAL;
 486                 fs_err(sdp, "FS block size (%u) is too small for device "
 487                        "block size (%u)\n",
 488                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 489                 goto out;
 490         }
 491         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 492                 ret = -EINVAL;
 493                 fs_err(sdp, "FS block size (%u) is too big for machine "
 494                        "page size (%u)\n",
 495                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 496                 goto out;
 497         }
 498         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 499
 500         /* Get the root inode */
 501         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 502         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 503         if (ret)
 504                 goto out;
 505
 506         /* Get the master inode */
 507         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 508         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 509         if (ret) {
 510                 dput(sdp->sd_root_dir);
 511                 goto out;
 512         }
 513         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 514 out:
 515         gfs2_glock_dq_uninit(&sb_gh);
 516         return ret;
 517 }
 518
 519 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 520 {
 521         char *message = "FIRSTMOUNT=Done";
 522         char *envp[] = { message, NULL };
 523
 524         fs_info(sdp, "first mount done, others may mount\n");
 525
 526         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 527                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 528
 529         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 530 }
 531
 532 /**
 533  * gfs2_jindex_hold - Grab a lock on the jindex
 534  * @sdp: The GFS2 superblock
 535  * @ji_gh: the holder for the jindex glock
 536  *
 537  * Returns: errno
 538  */
 539
 540 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 541 {
 542         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 543         struct qstr name;
 544         char buf[20];
 545         struct gfs2_jdesc *jd;
 546         int error;
 547
 548         name.name = buf;
 549
 550         mutex_lock(&sdp->sd_jindex_mutex);
 551
 552         for (;;) {
 553                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 554                 if (error)
 555                         break;
 556
 557                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 558                 name.hash = gfs2_disk_hash(name.name, name.len);
 559
 560                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 561                 if (error == -ENOENT) {
 562                         error = 0;
 563                         break;
 564                 }
 565
 566                 gfs2_glock_dq_uninit(ji_gh);
 567
 568                 if (error)
 569                         break;
 570
 571                 error = -ENOMEM;
 572                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 573                 if (!jd)
 574                         break;
 575
 576                 INIT_LIST_HEAD(&jd->extent_list);
 577                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 578
 579                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 580                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 581                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 582                         if (!jd->jd_inode)
 583                                 error = -ENOENT;
 584                         else
 585                                 error = PTR_ERR(jd->jd_inode);
 586                         kfree(jd);
 587                         break;
 588                 }
 589
 590                 spin_lock(&sdp->sd_jindex_spin);
 591                 jd->jd_jid = sdp->sd_journals++;
 592                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 593                 spin_unlock(&sdp->sd_jindex_spin);
 594         }
 595
 596         mutex_unlock(&sdp->sd_jindex_mutex);
 597
 598         return error;
 599 }
 600
 601 /**
 602  * check_journal_clean - Make sure a journal is clean for a spectator mount
 603  * @sdp: The GFS2 superblock
 604  * @jd: The journal descriptor
 605  *
 606  * Returns: 0 if the journal is clean or locked, else an error
 607  */
 608 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
 609 {
 610         int error;
 611         struct gfs2_holder j_gh;
 612         struct gfs2_log_header_host head;
 613         struct gfs2_inode *ip;
 614
 615         ip = GFS2_I(jd->jd_inode);
 616         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
 617                                    GL_EXACT | GL_NOCACHE, &j_gh);
 618         if (error) {
 619                 fs_err(sdp, "Error locking journal for spectator mount.\n");
 620                 return -EPERM;
 621         }
 622         error = gfs2_jdesc_check(jd);
 623         if (error) {
 624                 fs_err(sdp, "Error checking journal for spectator mount.\n");
 625                 goto out_unlock;
 626         }
 627         error = gfs2_find_jhead(jd, &head, false);
 628         if (error) {
 629                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
 630                 goto out_unlock;
 631         }
 632         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 633                 error = -EPERM;
 634                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
 635                        "must not be a spectator.\n", jd->jd_jid);
 636         }
 637
 638 out_unlock:
 639         gfs2_glock_dq_uninit(&j_gh);
 640         return error;
 641 }
 642
 643 static int init_journal(struct gfs2_sbd *sdp, int undo)
 644 {
 645         struct inode *master = d_inode(sdp->sd_master_dir);
 646         struct gfs2_holder ji_gh;
 647         struct gfs2_inode *ip;
 648         int jindex = 1;
 649         int error = 0;
 650
 651         if (undo) {
 652                 jindex = 0;
 653                 goto fail_jinode_gh;
 654         }
 655
 656         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 657         if (IS_ERR(sdp->sd_jindex)) {
 658                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 659                 return PTR_ERR(sdp->sd_jindex);
 660         }
 661
 662         /* Load in the journal index special file */
 663
 664         error = gfs2_jindex_hold(sdp, &ji_gh);
 665         if (error) {
 666                 fs_err(sdp, "can't read journal index: %d\n", error);
 667                 goto fail;
 668         }
 669
 670         error = -EUSERS;
 671         if (!gfs2_jindex_size(sdp)) {
 672                 fs_err(sdp, "no journals!\n");
 673                 goto fail_jindex;
 674         }
 675
 676         atomic_set(&sdp->sd_log_blks_needed, 0);
 677         if (sdp->sd_args.ar_spectator) {
 678                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 679                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 680                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 681                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 682         } else {
 683                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 684                         fs_err(sdp, "can't mount journal #%u\n",
 685                                sdp->sd_lockstruct.ls_jid);
 686                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 687                                gfs2_jindex_size(sdp),
 688                                gfs2_jindex_size(sdp) - 1);
 689                         goto fail_jindex;
 690                 }
 691                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 692
 693                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 694                                           &gfs2_journal_glops,
 695                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
 696                                           &sdp->sd_journal_gh);
 697                 if (error) {
 698                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 699                         goto fail_jindex;
 700                 }
 701
 702                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 703                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 704                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 705                                            &sdp->sd_jinode_gh);
 706                 if (error) {
 707                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 708                                error);
 709                         goto fail_journal_gh;
 710                 }
 711
 712                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 713                 if (error) {
 714                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 715                                sdp->sd_jdesc->jd_jid, error);
 716                         goto fail_jinode_gh;
 717                 }
 718                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 719                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 720                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 721
 722                 /* Map the extents for this journal's blocks */
 723                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 724         }
 725         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 726
 727         if (sdp->sd_lockstruct.ls_first) {
 728                 unsigned int x;
 729                 for (x = 0; x < sdp->sd_journals; x++) {
 730                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 731
 732                         if (sdp->sd_args.ar_spectator) {
 733                                 error = check_journal_clean(sdp, jd);
 734                                 if (error)
 735                                         goto fail_jinode_gh;
 736                                 continue;
 737                         }
 738                         error = gfs2_recover_journal(jd, true);
 739                         if (error) {
 740                                 fs_err(sdp, "error recovering journal %u: %d\n",
 741                                        x, error);
 742                                 goto fail_jinode_gh;
 743                         }
 744                 }
 745
 746                 gfs2_others_may_mount(sdp);
 747         } else if (!sdp->sd_args.ar_spectator) {
 748                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 749                 if (error) {
 750                         fs_err(sdp, "error recovering my journal: %d\n", error);
 751                         goto fail_jinode_gh;
 752                 }
 753         }
 754
 755         sdp->sd_log_idle = 1;
 756         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 757         gfs2_glock_dq_uninit(&ji_gh);
 758         jindex = 0;
 759         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 760         return 0;
 761
 762 fail_jinode_gh:
 763         if (!sdp->sd_args.ar_spectator)
 764                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 765 fail_journal_gh:
 766         if (!sdp->sd_args.ar_spectator)
 767                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 768 fail_jindex:
 769         gfs2_jindex_free(sdp);
 770         if (jindex)
 771                 gfs2_glock_dq_uninit(&ji_gh);
 772 fail:
 773         iput(sdp->sd_jindex);
 774         return error;
 775 }
 776
 777 static struct lock_class_key gfs2_quota_imutex_key;
 778
 779 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 780 {
 781         int error = 0;
 782         struct inode *master = d_inode(sdp->sd_master_dir);
 783
 784         if (undo)
 785                 goto fail_qinode;
 786
 787         error = init_journal(sdp, undo);
 788         complete_all(&sdp->sd_journal_ready);
 789         if (error)
 790                 goto fail;
 791
 792         /* Read in the master statfs inode */
 793         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 794         if (IS_ERR(sdp->sd_statfs_inode)) {
 795                 error = PTR_ERR(sdp->sd_statfs_inode);
 796                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 797                 goto fail_journal;
 798         }
 799
 800         /* Read in the resource index inode */
 801         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 802         if (IS_ERR(sdp->sd_rindex)) {
 803                 error = PTR_ERR(sdp->sd_rindex);
 804                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 805                 goto fail_statfs;
 806         }
 807         sdp->sd_rindex_uptodate = 0;
 808
 809         /* Read in the quota inode */
 810         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 811         if (IS_ERR(sdp->sd_quota_inode)) {
 812                 error = PTR_ERR(sdp->sd_quota_inode);
 813                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 814                 goto fail_rindex;
 815         }
 816         /*
 817          * i_rwsem on quota files is special. Since this inode is hidden system
 818          * file, we are safe to define locking ourselves.
 819          */
 820         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 821                           &gfs2_quota_imutex_key);
 822
 823         error = gfs2_rindex_update(sdp);
 824         if (error)
 825                 goto fail_qinode;
 826
 827         return 0;
 828
 829 fail_qinode:
 830         iput(sdp->sd_quota_inode);
 831 fail_rindex:
 832         gfs2_clear_rgrpd(sdp);
 833         iput(sdp->sd_rindex);
 834 fail_statfs:
 835         iput(sdp->sd_statfs_inode);
 836 fail_journal:
 837         init_journal(sdp, UNDO);
 838 fail:
 839         return error;
 840 }
 841
 842 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 843 {
 844         struct inode *pn = NULL;
 845         char buf[30];
 846         int error = 0;
 847         struct gfs2_inode *ip;
 848         struct inode *master = d_inode(sdp->sd_master_dir);
 849
 850         if (sdp->sd_args.ar_spectator)
 851                 return 0;
 852
 853         if (undo)
 854                 goto fail_qc_gh;
 855
 856         pn = gfs2_lookup_simple(master, "per_node");
 857         if (IS_ERR(pn)) {
 858                 error = PTR_ERR(pn);
 859                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 860                 return error;
 861         }
 862
 863         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
 864         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
 865         if (IS_ERR(sdp->sd_sc_inode)) {
 866                 error = PTR_ERR(sdp->sd_sc_inode);
 867                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
 868                 goto fail;
 869         }
 870
 871         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 872         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 873         if (IS_ERR(sdp->sd_qc_inode)) {
 874                 error = PTR_ERR(sdp->sd_qc_inode);
 875                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 876                 goto fail_ut_i;
 877         }
 878
 879         iput(pn);
 880         pn = NULL;
 881
 882         ip = GFS2_I(sdp->sd_sc_inode);
 883         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 884                                    &sdp->sd_sc_gh);
 885         if (error) {
 886                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 887                 goto fail_qc_i;
 888         }
 889
 890         ip = GFS2_I(sdp->sd_qc_inode);
 891         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 892                                    &sdp->sd_qc_gh);
 893         if (error) {
 894                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 895                 goto fail_ut_gh;
 896         }
 897
 898         return 0;
 899
 900 fail_qc_gh:
 901         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 902 fail_ut_gh:
 903         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 904 fail_qc_i:
 905         iput(sdp->sd_qc_inode);
 906 fail_ut_i:
 907         iput(sdp->sd_sc_inode);
 908 fail:
 909         iput(pn);
 910         return error;
 911 }
 912
 913 static const match_table_t nolock_tokens = {
 914         { Opt_jid, "jid=%d\n", },
 915         { Opt_err, NULL },
 916 };
 917
 918 static const struct lm_lockops nolock_ops = {
 919         .lm_proto_name = "lock_nolock",
 920         .lm_put_lock = gfs2_glock_free,
 921         .lm_tokens = &nolock_tokens,
 922 };
 923
 924 /**
 925  * gfs2_lm_mount - mount a locking protocol
 926  * @sdp: the filesystem
 927  * @args: mount arguments
 928  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 929  *
 930  * Returns: errno
 931  */
 932
 933 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 934 {
 935         const struct lm_lockops *lm;
 936         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 937         struct gfs2_args *args = &sdp->sd_args;
 938         const char *proto = sdp->sd_proto_name;
 939         const char *table = sdp->sd_table_name;
 940         char *o, *options;
 941         int ret;
 942
 943         if (!strcmp("lock_nolock", proto)) {
 944                 lm = &nolock_ops;
 945                 sdp->sd_args.ar_localflocks = 1;
 946 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 947         } else if (!strcmp("lock_dlm", proto)) {
 948                 lm = &gfs2_dlm_ops;
 949 #endif
 950         } else {
 951                 pr_info("can't find protocol %s\n", proto);
 952                 return -ENOENT;
 953         }
 954
 955         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 956
 957         ls->ls_ops = lm;
 958         ls->ls_first = 1;
 959
 960         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 961                 substring_t tmp[MAX_OPT_ARGS];
 962                 int token, option;
 963
 964                 if (!o || !*o)
 965                         continue;
 966
 967                 token = match_token(o, *lm->lm_tokens, tmp);
 968                 switch (token) {
 969                 case Opt_jid:
 970                         ret = match_int(&tmp[0], &option);
 971                         if (ret || option < 0)
 972                                 goto hostdata_error;
 973                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
 974                                 ls->ls_jid = option;
 975                         break;
 976                 case Opt_id:
 977                 case Opt_nodir:
 978                         /* Obsolete, but left for backward compat purposes */
 979                         break;
 980                 case Opt_first:
 981                         ret = match_int(&tmp[0], &option);
 982                         if (ret || (option != 0 && option != 1))
 983                                 goto hostdata_error;
 984                         ls->ls_first = option;
 985                         break;
 986                 case Opt_err:
 987                 default:
 988 hostdata_error:
 989                         fs_info(sdp, "unknown hostdata (%s)\n", o);
 990                         return -EINVAL;
 991                 }
 992         }
 993
 994         if (lm->lm_mount == NULL) {
 995                 fs_info(sdp, "Now mounting FS...\n");
 996                 complete_all(&sdp->sd_locking_init);
 997                 return 0;
 998         }
 999         ret = lm->lm_mount(sdp, table);
1000         if (ret == 0)
1001                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1002         complete_all(&sdp->sd_locking_init);
1003         return ret;
1004 }
1005
1006 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1007 {
1008         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1009         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1010                 lm->lm_unmount(sdp);
1011 }
1012
1013 static int wait_on_journal(struct gfs2_sbd *sdp)
1014 {
1015         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1016                 return 0;
1017
1018         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1019                 ? -EINTR : 0;
1020 }
1021
1022 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1023 {
1024         struct super_block *sb = sdp->sd_vfs;
1025         char ro[20];
1026         char spectator[20];
1027         char *envp[] = { ro, spectator, NULL };
1028         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1029         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1030         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1031 }
1032
1033 /**
1034  * gfs2_fill_super - Read in superblock
1035  * @sb: The VFS superblock
1036  * @args: Mount options
1037  * @silent: Don't complain if it's not a GFS2 filesystem
1038  *
1039  * Returns: -errno
1040  */
1041 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1042 {
1043         struct gfs2_args *args = fc->fs_private;
1044         int silent = fc->sb_flags & SB_SILENT;
1045         struct gfs2_sbd *sdp;
1046         struct gfs2_holder mount_gh;
1047         int error;
1048
1049         sdp = init_sbd(sb);
1050         if (!sdp) {
1051                 pr_warn("can't alloc struct gfs2_sbd\n");
1052                 return -ENOMEM;
1053         }
1054         sdp->sd_args = *args;
1055
1056         if (sdp->sd_args.ar_spectator) {
1057                 sb->s_flags |= SB_RDONLY;
1058                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1059         }
1060         if (sdp->sd_args.ar_posix_acl)
1061                 sb->s_flags |= SB_POSIXACL;
1062         if (sdp->sd_args.ar_nobarrier)
1063                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1064
1065         sb->s_flags |= SB_NOSEC;
1066         sb->s_magic = GFS2_MAGIC;
1067         sb->s_op = &gfs2_super_ops;
1068         sb->s_d_op = &gfs2_dops;
1069         sb->s_export_op = &gfs2_export_ops;
1070         sb->s_xattr = gfs2_xattr_handlers;
1071         sb->s_qcop = &gfs2_quotactl_ops;
1072         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1073         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1074         sb->s_time_gran = 1;
1075         sb->s_maxbytes = MAX_LFS_FILESIZE;
1076
1077         /* Set up the buffer cache and fill in some fake block size values
1078            to allow us to read-in the on-disk superblock. */
1079         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1080         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1081         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1082                                GFS2_BASIC_BLOCK_SHIFT;
1083         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1084
1085         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1086         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1087         if (sdp->sd_args.ar_statfs_quantum) {
1088                 sdp->sd_tune.gt_statfs_slow = 0;
1089                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1090         } else {
1091                 sdp->sd_tune.gt_statfs_slow = 1;
1092                 sdp->sd_tune.gt_statfs_quantum = 30;
1093         }
1094
1095         error = init_names(sdp, silent);
1096         if (error) {
1097                 /* In this case, we haven't initialized sysfs, so we have to
1098                    manually free the sdp. */
1099                 free_sbd(sdp);
1100                 sb->s_fs_info = NULL;
1101                 return error;
1102         }
1103
1104         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1105
1106         error = gfs2_sys_fs_add(sdp);
1107         /*
1108          * If we hit an error here, gfs2_sys_fs_add will have called function
1109          * kobject_put which causes the sysfs usage count to go to zero, which
1110          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1111          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1112          * kobject_put to free sdp.
1113          */
1114         if (error)
1115                 return error;
1116
1117         gfs2_create_debugfs_file(sdp);
1118
1119         error = gfs2_lm_mount(sdp, silent);
1120         if (error)
1121                 goto fail_debug;
1122
1123         error = init_locking(sdp, &mount_gh, DO);
1124         if (error)
1125                 goto fail_lm;
1126
1127         error = init_sb(sdp, silent);
1128         if (error)
1129                 goto fail_locking;
1130
1131         error = wait_on_journal(sdp);
1132         if (error)
1133                 goto fail_sb;
1134
1135         /*
1136          * If user space has failed to join the cluster or some similar
1137          * failure has occurred, then the journal id will contain a
1138          * negative (error) number. This will then be returned to the
1139          * caller (of the mount syscall). We do this even for spectator
1140          * mounts (which just write a jid of 0 to indicate "ok" even though
1141          * the jid is unused in the spectator case)
1142          */
1143         if (sdp->sd_lockstruct.ls_jid < 0) {
1144                 error = sdp->sd_lockstruct.ls_jid;
1145                 sdp->sd_lockstruct.ls_jid = 0;
1146                 goto fail_sb;
1147         }
1148
1149         if (sdp->sd_args.ar_spectator)
1150                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1151                          sdp->sd_table_name);
1152         else
1153                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1154                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1155
1156         error = init_inodes(sdp, DO);
1157         if (error)
1158                 goto fail_sb;
1159
1160         error = init_per_node(sdp, DO);
1161         if (error)
1162                 goto fail_inodes;
1163
1164         error = gfs2_statfs_init(sdp);
1165         if (error) {
1166                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1167                 goto fail_per_node;
1168         }
1169
1170         if (!sb_rdonly(sb)) {
1171                 error = gfs2_make_fs_rw(sdp);
1172                 if (error) {
1173                         fs_err(sdp, "can't make FS RW: %d\n", error);
1174                         goto fail_per_node;
1175                 }
1176         }
1177
1178         gfs2_glock_dq_uninit(&mount_gh);
1179         gfs2_online_uevent(sdp);
1180         return 0;
1181
1182 fail_per_node:
1183         init_per_node(sdp, UNDO);
1184 fail_inodes:
1185         init_inodes(sdp, UNDO);
1186 fail_sb:
1187         if (sdp->sd_root_dir)
1188                 dput(sdp->sd_root_dir);
1189         if (sdp->sd_master_dir)
1190                 dput(sdp->sd_master_dir);
1191         if (sb->s_root)
1192                 dput(sb->s_root);
1193         sb->s_root = NULL;
1194 fail_locking:
1195         init_locking(sdp, &mount_gh, UNDO);
1196 fail_lm:
1197         complete_all(&sdp->sd_journal_ready);
1198         gfs2_gl_hash_clear(sdp);
1199         gfs2_lm_unmount(sdp);
1200 fail_debug:
1201         gfs2_delete_debugfs_file(sdp);
1202         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1203          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1204         gfs2_sys_fs_del(sdp);
1205         sb->s_fs_info = NULL;
1206         return error;
1207 }
1208
1209 /**
1210  * gfs2_get_tree - Get the GFS2 superblock and root directory
1211  * @fc: The filesystem context
1212  *
1213  * Returns: 0 or -errno on error
1214  */
1215 static int gfs2_get_tree(struct fs_context *fc)
1216 {
1217         struct gfs2_args *args = fc->fs_private;
1218         struct gfs2_sbd *sdp;
1219         int error;
1220
1221         error = get_tree_bdev(fc, gfs2_fill_super);
1222         if (error)
1223                 return error;
1224
1225         sdp = fc->root->d_sb->s_fs_info;
1226         dput(fc->root);
1227         if (args->ar_meta)
1228                 fc->root = dget(sdp->sd_master_dir);
1229         else
1230                 fc->root = dget(sdp->sd_root_dir);
1231         return 0;
1232 }
1233
1234 static void gfs2_fc_free(struct fs_context *fc)
1235 {
1236         struct gfs2_args *args = fc->fs_private;
1237
1238         kfree(args);
1239 }
1240
1241 enum gfs2_param {
1242         Opt_lockproto,
1243         Opt_locktable,
1244         Opt_hostdata,
1245         Opt_spectator,
1246         Opt_ignore_local_fs,
1247         Opt_localflocks,
1248         Opt_localcaching,
1249         Opt_debug,
1250         Opt_upgrade,
1251         Opt_acl,
1252         Opt_quota,
1253         Opt_suiddir,
1254         Opt_data,
1255         Opt_meta,
1256         Opt_discard,
1257         Opt_commit,
1258         Opt_errors,
1259         Opt_statfs_quantum,
1260         Opt_statfs_percent,
1261         Opt_quota_quantum,
1262         Opt_barrier,
1263         Opt_rgrplvb,
1264         Opt_loccookie,
1265 };
1266
1267 enum opt_quota {
1268         Opt_quota_unset = 0,
1269         Opt_quota_off,
1270         Opt_quota_account,
1271         Opt_quota_on,
1272 };
1273
1274 static const unsigned int opt_quota_values[] = {
1275         [Opt_quota_off]     = GFS2_QUOTA_OFF,
1276         [Opt_quota_account] = GFS2_QUOTA_ACCOUNT,
1277         [Opt_quota_on]      = GFS2_QUOTA_ON,
1278 };
1279
1280 enum opt_data {
1281         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1282         Opt_data_ordered   = GFS2_DATA_ORDERED,
1283 };
1284
1285 enum opt_errors {
1286         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1287         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1288 };
1289
1290 static const struct fs_parameter_spec gfs2_param_specs[] = {
1291         fsparam_string ("lockproto",          Opt_lockproto),
1292         fsparam_string ("locktable",          Opt_locktable),
1293         fsparam_string ("hostdata",           Opt_hostdata),
1294         fsparam_flag   ("spectator",          Opt_spectator),
1295         fsparam_flag   ("norecovery",         Opt_spectator),
1296         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1297         fsparam_flag   ("localflocks",        Opt_localflocks),
1298         fsparam_flag   ("localcaching",       Opt_localcaching),
1299         fsparam_flag_no("debug",              Opt_debug),
1300         fsparam_flag   ("upgrade",            Opt_upgrade),
1301         fsparam_flag_no("acl",                Opt_acl),
1302         fsparam_flag_no("suiddir",            Opt_suiddir),
1303         fsparam_enum   ("data",               Opt_data),
1304         fsparam_flag   ("meta",               Opt_meta),
1305         fsparam_flag_no("discard",            Opt_discard),
1306         fsparam_s32    ("commit",             Opt_commit),
1307         fsparam_enum   ("errors",             Opt_errors),
1308         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1309         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1310         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1311         fsparam_flag_no("barrier",            Opt_barrier),
1312         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1313         fsparam_flag_no("loccookie",          Opt_loccookie),
1314         /* quota can be a flag or an enum so it gets special treatment */
1315         __fsparam(fs_param_is_enum, "quota", Opt_quota, fs_param_neg_with_no|fs_param_v_optional),
1316         {}
1317 };
1318
1319 static const struct fs_parameter_enum gfs2_param_enums[] = {
1320         { Opt_quota,    "off",        Opt_quota_off },
1321         { Opt_quota,    "account",    Opt_quota_account },
1322         { Opt_quota,    "on",         Opt_quota_on },
1323         { Opt_data,     "writeback",  Opt_data_writeback },
1324         { Opt_data,     "ordered",    Opt_data_ordered },
1325         { Opt_errors,   "withdraw",   Opt_errors_withdraw },
1326         { Opt_errors,   "panic",      Opt_errors_panic },
1327         {}
1328 };
1329
1330 static const struct fs_parameter_description gfs2_fs_parameters = {
1331         .name = "gfs2",
1332         .specs = gfs2_param_specs,
1333         .enums = gfs2_param_enums,
1334 };
1335
1336 /* Parse a single mount parameter */
1337 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1338 {
1339         struct gfs2_args *args = fc->fs_private;
1340         struct fs_parse_result result;
1341         int o;
1342
1343         o = fs_parse(fc, &gfs2_fs_parameters, param, &result);
1344         if (o < 0)
1345                 return o;
1346
1347         switch (o) {
1348         case Opt_lockproto:
1349                 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1350                 break;
1351         case Opt_locktable:
1352                 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1353                 break;
1354         case Opt_hostdata:
1355                 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1356                 break;
1357         case Opt_spectator:
1358                 args->ar_spectator = 1;
1359                 break;
1360         case Opt_ignore_local_fs:
1361                 /* Retained for backwards compat only */
1362                 break;
1363         case Opt_localflocks:
1364                 args->ar_localflocks = 1;
1365                 break;
1366         case Opt_localcaching:
1367                 /* Retained for backwards compat only */
1368                 break;
1369         case Opt_debug:
1370                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1371                         return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1372                 args->ar_debug = result.boolean;
1373                 break;
1374         case Opt_upgrade:
1375                 /* Retained for backwards compat only */
1376                 break;
1377         case Opt_acl:
1378                 args->ar_posix_acl = result.boolean;
1379                 break;
1380         case Opt_quota:
1381                 /* The quota option can be a flag or an enum. A non-zero int_32
1382                    result means that we have an enum index. Otherwise we have
1383                    to rely on the 'negated' flag to tell us whether 'quota' or
1384                    'noquota' was specified. */
1385                 if (result.negated)
1386                         args->ar_quota = GFS2_QUOTA_OFF;
1387                 else if (result.int_32 > 0)
1388                         args->ar_quota = opt_quota_values[result.int_32];
1389                 else
1390                         args->ar_quota = GFS2_QUOTA_ON;
1391                 break;
1392         case Opt_suiddir:
1393                 args->ar_suiddir = result.boolean;
1394                 break;
1395         case Opt_data:
1396                 /* The uint_32 result maps directly to GFS2_DATA_* */
1397                 args->ar_data = result.uint_32;
1398                 break;
1399         case Opt_meta:
1400                 args->ar_meta = 1;
1401                 break;
1402         case Opt_discard:
1403                 args->ar_discard = result.boolean;
1404                 break;
1405         case Opt_commit:
1406                 if (result.int_32 <= 0)
1407                         return invalf(fc, "gfs2: commit mount option requires a positive numeric argument");
1408                 args->ar_commit = result.int_32;
1409                 break;
1410         case Opt_statfs_quantum:
1411                 if (result.int_32 < 0)
1412                         return invalf(fc, "gfs2: statfs_quantum mount option requires a non-negative numeric argument");
1413                 args->ar_statfs_quantum = result.int_32;
1414                 break;
1415         case Opt_quota_quantum:
1416                 if (result.int_32 <= 0)
1417                         return invalf(fc, "gfs2: quota_quantum mount option requires a positive numeric argument");
1418                 args->ar_quota_quantum = result.int_32;
1419                 break;
1420         case Opt_statfs_percent:
1421                 if (result.int_32 < 0 || result.int_32 > 100)
1422                         return invalf(fc, "gfs2: statfs_percent mount option requires a numeric argument between 0 and 100");
1423                 args->ar_statfs_percent = result.int_32;
1424                 break;
1425         case Opt_errors:
1426                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1427                         return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1428                 args->ar_errors = result.uint_32;
1429                 break;
1430         case Opt_barrier:
1431                 args->ar_nobarrier = result.boolean;
1432                 break;
1433         case Opt_rgrplvb:
1434                 args->ar_rgrplvb = result.boolean;
1435                 break;
1436         case Opt_loccookie:
1437                 args->ar_loccookie = result.boolean;
1438                 break;
1439         default:
1440                 return invalf(fc, "gfs2: invalid mount option: %s", param->key);
1441         }
1442         return 0;
1443 }
1444
1445 static int gfs2_reconfigure(struct fs_context *fc)
1446 {
1447         struct super_block *sb = fc->root->d_sb;
1448         struct gfs2_sbd *sdp = sb->s_fs_info;
1449         struct gfs2_args *oldargs = &sdp->sd_args;
1450         struct gfs2_args *newargs = fc->fs_private;
1451         struct gfs2_tune *gt = &sdp->sd_tune;
1452         int error = 0;
1453
1454         sync_filesystem(sb);
1455
1456         spin_lock(&gt->gt_spin);
1457         oldargs->ar_commit = gt->gt_logd_secs;
1458         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1459         if (gt->gt_statfs_slow)
1460                 oldargs->ar_statfs_quantum = 0;
1461         else
1462                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1463         spin_unlock(&gt->gt_spin);
1464
1465         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1466                 errorf(fc, "gfs2: reconfiguration of locking protocol not allowed");
1467                 return -EINVAL;
1468         }
1469         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1470                 errorf(fc, "gfs2: reconfiguration of lock table not allowed");
1471                 return -EINVAL;
1472         }
1473         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1474                 errorf(fc, "gfs2: reconfiguration of host data not allowed");
1475                 return -EINVAL;
1476         }
1477         if (newargs->ar_spectator != oldargs->ar_spectator) {
1478                 errorf(fc, "gfs2: reconfiguration of spectator mode not allowed");
1479                 return -EINVAL;
1480         }
1481         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1482                 errorf(fc, "gfs2: reconfiguration of localflocks not allowed");
1483                 return -EINVAL;
1484         }
1485         if (newargs->ar_meta != oldargs->ar_meta) {
1486                 errorf(fc, "gfs2: switching between gfs2 and gfs2meta not allowed");
1487                 return -EINVAL;
1488         }
1489         if (oldargs->ar_spectator)
1490                 fc->sb_flags |= SB_RDONLY;
1491
1492         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1493                 if (fc->sb_flags & SB_RDONLY) {
1494                         error = gfs2_make_fs_ro(sdp);
1495                         if (error)
1496                                 errorf(fc, "gfs2: unable to remount read-only");
1497                 } else {
1498                         error = gfs2_make_fs_rw(sdp);
1499                         if (error)
1500                                 errorf(fc, "gfs2: unable to remount read-write");
1501                 }
1502         }
1503         sdp->sd_args = *newargs;
1504
1505         if (sdp->sd_args.ar_posix_acl)
1506                 sb->s_flags |= SB_POSIXACL;
1507         else
1508                 sb->s_flags &= ~SB_POSIXACL;
1509         if (sdp->sd_args.ar_nobarrier)
1510                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1511         else
1512                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1513         spin_lock(&gt->gt_spin);
1514         gt->gt_logd_secs = newargs->ar_commit;
1515         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1516         if (newargs->ar_statfs_quantum) {
1517                 gt->gt_statfs_slow = 0;
1518                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1519         }
1520         else {
1521                 gt->gt_statfs_slow = 1;
1522                 gt->gt_statfs_quantum = 30;
1523         }
1524         spin_unlock(&gt->gt_spin);
1525
1526         gfs2_online_uevent(sdp);
1527         return error;
1528 }
1529
1530 static const struct fs_context_operations gfs2_context_ops = {
1531         .free        = gfs2_fc_free,
1532         .parse_param = gfs2_parse_param,
1533         .get_tree    = gfs2_get_tree,
1534         .reconfigure = gfs2_reconfigure,
1535 };
1536
1537 /* Set up the filesystem mount context */
1538 static int gfs2_init_fs_context(struct fs_context *fc)
1539 {
1540         struct gfs2_args *args;
1541
1542         args = kmalloc(sizeof(*args), GFP_KERNEL);
1543         if (args == NULL)
1544                 return -ENOMEM;
1545
1546         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1547                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1548
1549                 *args = sdp->sd_args;
1550         } else {
1551                 memset(args, 0, sizeof(*args));
1552                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1553                 args->ar_data = GFS2_DATA_DEFAULT;
1554                 args->ar_commit = 30;
1555                 args->ar_statfs_quantum = 30;
1556                 args->ar_quota_quantum = 60;
1557                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1558         }
1559         fc->fs_private = args;
1560         fc->ops = &gfs2_context_ops;
1561         return 0;
1562 }
1563
1564 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1565 {
1566         return -EINVAL;
1567 }
1568
1569 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1570 {
1571         return (fc->sget_key == s->s_bdev);
1572 }
1573
1574 static int gfs2_meta_get_tree(struct fs_context *fc)
1575 {
1576         struct super_block *s;
1577         struct gfs2_sbd *sdp;
1578         struct path path;
1579         int error;
1580
1581         if (!fc->source || !*fc->source)
1582                 return -EINVAL;
1583
1584         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1585         if (error) {
1586                 pr_warn("path_lookup on %s returned error %d\n",
1587                         fc->source, error);
1588                 return error;
1589         }
1590         fc->fs_type = &gfs2_fs_type;
1591         fc->sget_key = path.dentry->d_sb->s_bdev;
1592         s = sget_fc(fc, test_meta_super, set_meta_super);
1593         path_put(&path);
1594         if (IS_ERR(s)) {
1595                 pr_warn("gfs2 mount does not exist\n");
1596                 return PTR_ERR(s);
1597         }
1598         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1599                 deactivate_locked_super(s);
1600                 return -EBUSY;
1601         }
1602         sdp = s->s_fs_info;
1603         fc->root = dget(sdp->sd_master_dir);
1604         return 0;
1605 }
1606
1607 static const struct fs_context_operations gfs2_meta_context_ops = {
1608         .free        = gfs2_fc_free,
1609         .get_tree    = gfs2_meta_get_tree,
1610 };
1611
1612 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1613 {
1614         int ret = gfs2_init_fs_context(fc);
1615
1616         if (ret)
1617                 return ret;
1618
1619         fc->ops = &gfs2_meta_context_ops;
1620         return 0;
1621 }
1622
1623 static void gfs2_kill_sb(struct super_block *sb)
1624 {
1625         struct gfs2_sbd *sdp = sb->s_fs_info;
1626
1627         if (sdp == NULL) {
1628                 kill_block_super(sb);
1629                 return;
1630         }
1631
1632         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1633         dput(sdp->sd_root_dir);
1634         dput(sdp->sd_master_dir);
1635         sdp->sd_root_dir = NULL;
1636         sdp->sd_master_dir = NULL;
1637         shrink_dcache_sb(sb);
1638         kill_block_super(sb);
1639 }
1640
1641 struct file_system_type gfs2_fs_type = {
1642         .name = "gfs2",
1643         .fs_flags = FS_REQUIRES_DEV,
1644         .init_fs_context = gfs2_init_fs_context,
1645         .parameters = &gfs2_fs_parameters,
1646         .kill_sb = gfs2_kill_sb,
1647         .owner = THIS_MODULE,
1648 };
1649 MODULE_ALIAS_FS("gfs2");
1650
1651 struct file_system_type gfs2meta_fs_type = {
1652         .name = "gfs2meta",
1653         .fs_flags = FS_REQUIRES_DEV,
1654         .init_fs_context = gfs2_meta_init_fs_context,
1655         .owner = THIS_MODULE,
1656 };
1657 MODULE_ALIAS_FS("gfs2meta");