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         INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
 114
 115         mapping = &sdp->sd_aspace;
 116
 117         address_space_init_once(mapping);
 118         mapping->a_ops = &gfs2_rgrp_aops;
 119         mapping->host = sb->s_bdev->bd_inode;
 120         mapping->flags = 0;
 121         mapping_set_gfp_mask(mapping, GFP_NOFS);
 122         mapping->private_data = NULL;
 123         mapping->writeback_index = 0;
 124
 125         spin_lock_init(&sdp->sd_log_lock);
 126         atomic_set(&sdp->sd_log_pinned, 0);
 127         INIT_LIST_HEAD(&sdp->sd_log_revokes);
 128         INIT_LIST_HEAD(&sdp->sd_log_ordered);
 129         spin_lock_init(&sdp->sd_ordered_lock);
 130
 131         init_waitqueue_head(&sdp->sd_log_waitq);
 132         init_waitqueue_head(&sdp->sd_logd_waitq);
 133         spin_lock_init(&sdp->sd_ail_lock);
 134         INIT_LIST_HEAD(&sdp->sd_ail1_list);
 135         INIT_LIST_HEAD(&sdp->sd_ail2_list);
 136
 137         init_rwsem(&sdp->sd_log_flush_lock);
 138         atomic_set(&sdp->sd_log_in_flight, 0);
 139         atomic_set(&sdp->sd_reserving_log, 0);
 140         init_waitqueue_head(&sdp->sd_reserving_log_wait);
 141         init_waitqueue_head(&sdp->sd_log_flush_wait);
 142         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
 143         mutex_init(&sdp->sd_freeze_mutex);
 144
 145         return sdp;
 146
 147 fail:
 148         free_sbd(sdp);
 149         return NULL;
 150 }
 151
 152 /**
 153  * gfs2_check_sb - Check superblock
 154  * @sdp: the filesystem
 155  * @sb: The superblock
 156  * @silent: Don't print a message if the check fails
 157  *
 158  * Checks the version code of the FS is one that we understand how to
 159  * read and that the sizes of the various on-disk structures have not
 160  * changed.
 161  */
 162
 163 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 164 {
 165         struct gfs2_sb_host *sb = &sdp->sd_sb;
 166
 167         if (sb->sb_magic != GFS2_MAGIC ||
 168             sb->sb_type != GFS2_METATYPE_SB) {
 169                 if (!silent)
 170                         pr_warn("not a GFS2 filesystem\n");
 171                 return -EINVAL;
 172         }
 173
 174         if (sb->sb_fs_format != GFS2_FORMAT_FS ||
 175             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 176                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 177                 return -EINVAL;
 178         }
 179
 180         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
 181             (sb->sb_bsize & (sb->sb_bsize - 1))) {
 182                 pr_warn("Invalid superblock size\n");
 183                 return -EINVAL;
 184         }
 185
 186         return 0;
 187 }
 188
 189 static void end_bio_io_page(struct bio *bio)
 190 {
 191         struct page *page = bio->bi_private;
 192
 193         if (!bio->bi_status)
 194                 SetPageUptodate(page);
 195         else
 196                 pr_warn("error %d reading superblock\n", bio->bi_status);
 197         unlock_page(page);
 198 }
 199
 200 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 201 {
 202         struct gfs2_sb_host *sb = &sdp->sd_sb;
 203         struct super_block *s = sdp->sd_vfs;
 204         const struct gfs2_sb *str = buf;
 205
 206         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 207         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 208         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 209         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 210         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 211         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 212         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 213         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 214         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 215         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 216         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 217
 218         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 219         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 220         memcpy(&s->s_uuid, str->sb_uuid, 16);
 221 }
 222
 223 /**
 224  * gfs2_read_super - Read the gfs2 super block from disk
 225  * @sdp: The GFS2 super block
 226  * @sector: The location of the super block
 227  * @error: The error code to return
 228  *
 229  * This uses the bio functions to read the super block from disk
 230  * because we want to be 100% sure that we never read cached data.
 231  * A super block is read twice only during each GFS2 mount and is
 232  * never written to by the filesystem. The first time its read no
 233  * locks are held, and the only details which are looked at are those
 234  * relating to the locking protocol. Once locking is up and working,
 235  * the sb is read again under the lock to establish the location of
 236  * the master directory (contains pointers to journals etc) and the
 237  * root directory.
 238  *
 239  * Returns: 0 on success or error
 240  */
 241
 242 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 243 {
 244         struct super_block *sb = sdp->sd_vfs;
 245         struct gfs2_sb *p;
 246         struct page *page;
 247         struct bio *bio;
 248
 249         page = alloc_page(GFP_NOFS);
 250         if (unlikely(!page))
 251                 return -ENOMEM;
 252
 253         ClearPageUptodate(page);
 254         ClearPageDirty(page);
 255         lock_page(page);
 256
 257         bio = bio_alloc(GFP_NOFS, 1);
 258         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 259         bio_set_dev(bio, sb->s_bdev);
 260         bio_add_page(bio, page, PAGE_SIZE, 0);
 261
 262         bio->bi_end_io = end_bio_io_page;
 263         bio->bi_private = page;
 264         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
 265         submit_bio(bio);
 266         wait_on_page_locked(page);
 267         bio_put(bio);
 268         if (!PageUptodate(page)) {
 269                 __free_page(page);
 270                 return -EIO;
 271         }
 272         p = kmap(page);
 273         gfs2_sb_in(sdp, p);
 274         kunmap(page);
 275         __free_page(page);
 276         return gfs2_check_sb(sdp, silent);
 277 }
 278
 279 /**
 280  * gfs2_read_sb - Read super block
 281  * @sdp: The GFS2 superblock
 282  * @silent: Don't print message if mount fails
 283  *
 284  */
 285
 286 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 287 {
 288         u32 hash_blocks, ind_blocks, leaf_blocks;
 289         u32 tmp_blocks;
 290         unsigned int x;
 291         int error;
 292
 293         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 294         if (error) {
 295                 if (!silent)
 296                         fs_err(sdp, "can't read superblock\n");
 297                 return error;
 298         }
 299
 300         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 301                                GFS2_BASIC_BLOCK_SHIFT;
 302         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 303         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 304                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 305         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 306                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 307         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
 308                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 309         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 310         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 311         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 312         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 313         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 314                                 sizeof(struct gfs2_meta_header)) /
 315                                 sizeof(struct gfs2_quota_change);
 316         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 317                                      sizeof(struct gfs2_meta_header))
 318                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 319
 320         /* Compute maximum reservation required to add a entry to a directory */
 321
 322         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 323                              sdp->sd_jbsize);
 324
 325         ind_blocks = 0;
 326         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 327                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 328                 ind_blocks += tmp_blocks;
 329         }
 330
 331         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 332
 333         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 334
 335         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 336                                 sizeof(struct gfs2_dinode);
 337         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 338         for (x = 2;; x++) {
 339                 u64 space, d;
 340                 u32 m;
 341
 342                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 343                 d = space;
 344                 m = do_div(d, sdp->sd_inptrs);
 345
 346                 if (d != sdp->sd_heightsize[x - 1] || m)
 347                         break;
 348                 sdp->sd_heightsize[x] = space;
 349         }
 350         sdp->sd_max_height = x;
 351         sdp->sd_heightsize[x] = ~0;
 352         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 353
 354         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 355                                       sizeof(struct gfs2_leaf)) /
 356                                      GFS2_MIN_DIRENT_SIZE;
 357         return 0;
 358 }
 359
 360 static int init_names(struct gfs2_sbd *sdp, int silent)
 361 {
 362         char *proto, *table;
 363         int error = 0;
 364
 365         proto = sdp->sd_args.ar_lockproto;
 366         table = sdp->sd_args.ar_locktable;
 367
 368         /*  Try to autodetect  */
 369
 370         if (!proto[0] || !table[0]) {
 371                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 372                 if (error)
 373                         return error;
 374
 375                 if (!proto[0])
 376                         proto = sdp->sd_sb.sb_lockproto;
 377                 if (!table[0])
 378                         table = sdp->sd_sb.sb_locktable;
 379         }
 380
 381         if (!table[0])
 382                 table = sdp->sd_vfs->s_id;
 383
 384         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 385         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 386
 387         table = sdp->sd_table_name;
 388         while ((table = strchr(table, '/')))
 389                 *table = '_';
 390
 391         return error;
 392 }
 393
 394 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 395                         int undo)
 396 {
 397         int error = 0;
 398
 399         if (undo)
 400                 goto fail_trans;
 401
 402         error = gfs2_glock_nq_num(sdp,
 403                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 404                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 405                                   mount_gh);
 406         if (error) {
 407                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 408                 goto fail;
 409         }
 410
 411         error = gfs2_glock_nq_num(sdp,
 412                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 413                                   LM_ST_SHARED,
 414                                   LM_FLAG_NOEXP | GL_EXACT,
 415                                   &sdp->sd_live_gh);
 416         if (error) {
 417                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 418                 goto fail_mount;
 419         }
 420
 421         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 422                                CREATE, &sdp->sd_rename_gl);
 423         if (error) {
 424                 fs_err(sdp, "can't create rename glock: %d\n", error);
 425                 goto fail_live;
 426         }
 427
 428         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 429                                CREATE, &sdp->sd_freeze_gl);
 430         if (error) {
 431                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 432                 goto fail_rename;
 433         }
 434
 435         return 0;
 436
 437 fail_trans:
 438         gfs2_glock_put(sdp->sd_freeze_gl);
 439 fail_rename:
 440         gfs2_glock_put(sdp->sd_rename_gl);
 441 fail_live:
 442         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 443 fail_mount:
 444         gfs2_glock_dq_uninit(mount_gh);
 445 fail:
 446         return error;
 447 }
 448
 449 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 450                             u64 no_addr, const char *name)
 451 {
 452         struct gfs2_sbd *sdp = sb->s_fs_info;
 453         struct dentry *dentry;
 454         struct inode *inode;
 455
 456         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 457                                   GFS2_BLKST_FREE /* ignore */);
 458         if (IS_ERR(inode)) {
 459                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 460                 return PTR_ERR(inode);
 461         }
 462         dentry = d_make_root(inode);
 463         if (!dentry) {
 464                 fs_err(sdp, "can't alloc %s dentry\n", name);
 465                 return -ENOMEM;
 466         }
 467         *dptr = dentry;
 468         return 0;
 469 }
 470
 471 static int init_sb(struct gfs2_sbd *sdp, int silent)
 472 {
 473         struct super_block *sb = sdp->sd_vfs;
 474         struct gfs2_holder sb_gh;
 475         u64 no_addr;
 476         int ret;
 477
 478         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 479                                 LM_ST_SHARED, 0, &sb_gh);
 480         if (ret) {
 481                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 482                 return ret;
 483         }
 484
 485         ret = gfs2_read_sb(sdp, silent);
 486         if (ret) {
 487                 fs_err(sdp, "can't read superblock: %d\n", ret);
 488                 goto out;
 489         }
 490
 491         /* Set up the buffer cache and SB for real */
 492         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 493                 ret = -EINVAL;
 494                 fs_err(sdp, "FS block size (%u) is too small for device "
 495                        "block size (%u)\n",
 496                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 497                 goto out;
 498         }
 499         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 500                 ret = -EINVAL;
 501                 fs_err(sdp, "FS block size (%u) is too big for machine "
 502                        "page size (%u)\n",
 503                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 504                 goto out;
 505         }
 506         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 507
 508         /* Get the root inode */
 509         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 510         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 511         if (ret)
 512                 goto out;
 513
 514         /* Get the master inode */
 515         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 516         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 517         if (ret) {
 518                 dput(sdp->sd_root_dir);
 519                 goto out;
 520         }
 521         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 522 out:
 523         gfs2_glock_dq_uninit(&sb_gh);
 524         return ret;
 525 }
 526
 527 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 528 {
 529         char *message = "FIRSTMOUNT=Done";
 530         char *envp[] = { message, NULL };
 531
 532         fs_info(sdp, "first mount done, others may mount\n");
 533
 534         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 535                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 536
 537         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 538 }
 539
 540 /**
 541  * gfs2_jindex_hold - Grab a lock on the jindex
 542  * @sdp: The GFS2 superblock
 543  * @ji_gh: the holder for the jindex glock
 544  *
 545  * Returns: errno
 546  */
 547
 548 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 549 {
 550         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 551         struct qstr name;
 552         char buf[20];
 553         struct gfs2_jdesc *jd;
 554         int error;
 555
 556         name.name = buf;
 557
 558         mutex_lock(&sdp->sd_jindex_mutex);
 559
 560         for (;;) {
 561                 struct gfs2_inode *jip;
 562
 563                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 564                 if (error)
 565                         break;
 566
 567                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 568                 name.hash = gfs2_disk_hash(name.name, name.len);
 569
 570                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 571                 if (error == -ENOENT) {
 572                         error = 0;
 573                         break;
 574                 }
 575
 576                 gfs2_glock_dq_uninit(ji_gh);
 577
 578                 if (error)
 579                         break;
 580
 581                 error = -ENOMEM;
 582                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 583                 if (!jd)
 584                         break;
 585
 586                 INIT_LIST_HEAD(&jd->extent_list);
 587                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 588
 589                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 590                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 591                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 592                         if (!jd->jd_inode)
 593                                 error = -ENOENT;
 594                         else
 595                                 error = PTR_ERR(jd->jd_inode);
 596                         kfree(jd);
 597                         break;
 598                 }
 599
 600                 spin_lock(&sdp->sd_jindex_spin);
 601                 jd->jd_jid = sdp->sd_journals++;
 602                 jip = GFS2_I(jd->jd_inode);
 603                 jd->jd_no_addr = jip->i_no_addr;
 604                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 605                 spin_unlock(&sdp->sd_jindex_spin);
 606         }
 607
 608         mutex_unlock(&sdp->sd_jindex_mutex);
 609
 610         return error;
 611 }
 612
 613 /**
 614  * init_statfs - look up and initialize master and local (per node) statfs inodes
 615  * @sdp: The GFS2 superblock
 616  *
 617  * This should be called after the jindex is initialized in init_journal() and
 618  * before gfs2_journal_recovery() is called because we need to be able to write
 619  * to these inodes during recovery.
 620  *
 621  * Returns: errno
 622  */
 623 static int init_statfs(struct gfs2_sbd *sdp)
 624 {
 625         int error = 0;
 626         struct inode *master = d_inode(sdp->sd_master_dir);
 627         struct inode *pn = NULL;
 628         char buf[30];
 629         struct gfs2_jdesc *jd;
 630         struct gfs2_inode *ip;
 631
 632         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 633         if (IS_ERR(sdp->sd_statfs_inode)) {
 634                 error = PTR_ERR(sdp->sd_statfs_inode);
 635                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 636                 goto fail;
 637         }
 638
 639         pn = gfs2_lookup_simple(master, "per_node");
 640         if (IS_ERR(pn)) {
 641                 error = PTR_ERR(pn);
 642                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 643                 goto put_statfs;
 644         }
 645
 646         /* For each jid, lookup the corresponding local statfs inode in the
 647          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
 648         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 649                 struct local_statfs_inode *lsi =
 650                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
 651                 if (!lsi) {
 652                         error = -ENOMEM;
 653                         goto free_local;
 654                 }
 655                 sprintf(buf, "statfs_change%u", jd->jd_jid);
 656                 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
 657                 if (IS_ERR(lsi->si_sc_inode)) {
 658                         error = PTR_ERR(lsi->si_sc_inode);
 659                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
 660                                jd->jd_jid, error);
 661                         goto free_local;
 662                 }
 663                 lsi->si_jid = jd->jd_jid;
 664                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
 665                         sdp->sd_sc_inode = lsi->si_sc_inode;
 666
 667                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
 668         }
 669
 670         iput(pn);
 671         ip = GFS2_I(sdp->sd_sc_inode);
 672         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 673                                    &sdp->sd_sc_gh);
 674         if (error) {
 675                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 676                 goto free_local;
 677         }
 678         return 0;
 679
 680 free_local:
 681         free_local_statfs_inodes(sdp);
 682         iput(pn);
 683 put_statfs:
 684         iput(sdp->sd_statfs_inode);
 685 fail:
 686         return error;
 687 }
 688
 689 /* Uninitialize and free up memory used by the list of statfs inodes */
 690 static void uninit_statfs(struct gfs2_sbd *sdp)
 691 {
 692         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 693         free_local_statfs_inodes(sdp);
 694         iput(sdp->sd_statfs_inode);
 695 }
 696
 697 static int init_journal(struct gfs2_sbd *sdp, int undo)
 698 {
 699         struct inode *master = d_inode(sdp->sd_master_dir);
 700         struct gfs2_holder ji_gh;
 701         struct gfs2_inode *ip;
 702         int jindex = 1;
 703         int error = 0;
 704
 705         if (undo) {
 706                 jindex = 0;
 707                 goto fail_jinode_gh;
 708         }
 709
 710         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 711         if (IS_ERR(sdp->sd_jindex)) {
 712                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 713                 return PTR_ERR(sdp->sd_jindex);
 714         }
 715
 716         /* Load in the journal index special file */
 717
 718         error = gfs2_jindex_hold(sdp, &ji_gh);
 719         if (error) {
 720                 fs_err(sdp, "can't read journal index: %d\n", error);
 721                 goto fail;
 722         }
 723
 724         error = -EUSERS;
 725         if (!gfs2_jindex_size(sdp)) {
 726                 fs_err(sdp, "no journals!\n");
 727                 goto fail_jindex;
 728         }
 729
 730         atomic_set(&sdp->sd_log_blks_needed, 0);
 731         if (sdp->sd_args.ar_spectator) {
 732                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 733                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 734                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 735                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 736         } else {
 737                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 738                         fs_err(sdp, "can't mount journal #%u\n",
 739                                sdp->sd_lockstruct.ls_jid);
 740                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 741                                gfs2_jindex_size(sdp),
 742                                gfs2_jindex_size(sdp) - 1);
 743                         goto fail_jindex;
 744                 }
 745                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 746
 747                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 748                                           &gfs2_journal_glops,
 749                                           LM_ST_EXCLUSIVE,
 750                                           LM_FLAG_NOEXP | GL_NOCACHE,
 751                                           &sdp->sd_journal_gh);
 752                 if (error) {
 753                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 754                         goto fail_jindex;
 755                 }
 756
 757                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 758                 sdp->sd_jinode_gl = ip->i_gl;
 759                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 760                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 761                                            &sdp->sd_jinode_gh);
 762                 if (error) {
 763                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 764                                error);
 765                         goto fail_journal_gh;
 766                 }
 767
 768                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 769                 if (error) {
 770                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 771                                sdp->sd_jdesc->jd_jid, error);
 772                         goto fail_jinode_gh;
 773                 }
 774                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 775                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 776                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 777
 778                 /* Map the extents for this journal's blocks */
 779                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 780         }
 781         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 782
 783         /* Lookup statfs inodes here so journal recovery can use them. */
 784         error = init_statfs(sdp);
 785         if (error)
 786                 goto fail_jinode_gh;
 787
 788         if (sdp->sd_lockstruct.ls_first) {
 789                 unsigned int x;
 790                 for (x = 0; x < sdp->sd_journals; x++) {
 791                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 792
 793                         if (sdp->sd_args.ar_spectator) {
 794                                 error = check_journal_clean(sdp, jd, true);
 795                                 if (error)
 796                                         goto fail_statfs;
 797                                 continue;
 798                         }
 799                         error = gfs2_recover_journal(jd, true);
 800                         if (error) {
 801                                 fs_err(sdp, "error recovering journal %u: %d\n",
 802                                        x, error);
 803                                 goto fail_statfs;
 804                         }
 805                 }
 806
 807                 gfs2_others_may_mount(sdp);
 808         } else if (!sdp->sd_args.ar_spectator) {
 809                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 810                 if (error) {
 811                         fs_err(sdp, "error recovering my journal: %d\n", error);
 812                         goto fail_statfs;
 813                 }
 814         }
 815
 816         sdp->sd_log_idle = 1;
 817         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 818         gfs2_glock_dq_uninit(&ji_gh);
 819         jindex = 0;
 820         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 821         return 0;
 822
 823 fail_statfs:
 824         uninit_statfs(sdp);
 825 fail_jinode_gh:
 826         /* A withdraw may have done dq/uninit so now we need to check it */
 827         if (!sdp->sd_args.ar_spectator &&
 828             gfs2_holder_initialized(&sdp->sd_jinode_gh))
 829                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 830 fail_journal_gh:
 831         if (!sdp->sd_args.ar_spectator &&
 832             gfs2_holder_initialized(&sdp->sd_journal_gh))
 833                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 834 fail_jindex:
 835         gfs2_jindex_free(sdp);
 836         if (jindex)
 837                 gfs2_glock_dq_uninit(&ji_gh);
 838 fail:
 839         iput(sdp->sd_jindex);
 840         return error;
 841 }
 842
 843 static struct lock_class_key gfs2_quota_imutex_key;
 844
 845 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 846 {
 847         int error = 0;
 848         struct inode *master = d_inode(sdp->sd_master_dir);
 849
 850         if (undo)
 851                 goto fail_qinode;
 852
 853         error = init_journal(sdp, undo);
 854         complete_all(&sdp->sd_journal_ready);
 855         if (error)
 856                 goto fail;
 857
 858         /* Read in the resource index inode */
 859         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 860         if (IS_ERR(sdp->sd_rindex)) {
 861                 error = PTR_ERR(sdp->sd_rindex);
 862                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 863                 goto fail_journal;
 864         }
 865         sdp->sd_rindex_uptodate = 0;
 866
 867         /* Read in the quota inode */
 868         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 869         if (IS_ERR(sdp->sd_quota_inode)) {
 870                 error = PTR_ERR(sdp->sd_quota_inode);
 871                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 872                 goto fail_rindex;
 873         }
 874         /*
 875          * i_rwsem on quota files is special. Since this inode is hidden system
 876          * file, we are safe to define locking ourselves.
 877          */
 878         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 879                           &gfs2_quota_imutex_key);
 880
 881         error = gfs2_rindex_update(sdp);
 882         if (error)
 883                 goto fail_qinode;
 884
 885         return 0;
 886
 887 fail_qinode:
 888         iput(sdp->sd_quota_inode);
 889 fail_rindex:
 890         gfs2_clear_rgrpd(sdp);
 891         iput(sdp->sd_rindex);
 892 fail_journal:
 893         init_journal(sdp, UNDO);
 894 fail:
 895         return error;
 896 }
 897
 898 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 899 {
 900         struct inode *pn = NULL;
 901         char buf[30];
 902         int error = 0;
 903         struct gfs2_inode *ip;
 904         struct inode *master = d_inode(sdp->sd_master_dir);
 905
 906         if (sdp->sd_args.ar_spectator)
 907                 return 0;
 908
 909         if (undo)
 910                 goto fail_qc_gh;
 911
 912         pn = gfs2_lookup_simple(master, "per_node");
 913         if (IS_ERR(pn)) {
 914                 error = PTR_ERR(pn);
 915                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 916                 return error;
 917         }
 918
 919         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 920         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 921         if (IS_ERR(sdp->sd_qc_inode)) {
 922                 error = PTR_ERR(sdp->sd_qc_inode);
 923                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 924                 goto fail_ut_i;
 925         }
 926
 927         iput(pn);
 928         pn = NULL;
 929
 930         ip = GFS2_I(sdp->sd_qc_inode);
 931         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 932                                    &sdp->sd_qc_gh);
 933         if (error) {
 934                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 935                 goto fail_qc_i;
 936         }
 937
 938         return 0;
 939
 940 fail_qc_gh:
 941         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 942 fail_qc_i:
 943         iput(sdp->sd_qc_inode);
 944 fail_ut_i:
 945         iput(pn);
 946         return error;
 947 }
 948
 949 static const match_table_t nolock_tokens = {
 950         { Opt_jid, "jid=%d", },
 951         { Opt_err, NULL },
 952 };
 953
 954 static const struct lm_lockops nolock_ops = {
 955         .lm_proto_name = "lock_nolock",
 956         .lm_put_lock = gfs2_glock_free,
 957         .lm_tokens = &nolock_tokens,
 958 };
 959
 960 /**
 961  * gfs2_lm_mount - mount a locking protocol
 962  * @sdp: the filesystem
 963  * @args: mount arguments
 964  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 965  *
 966  * Returns: errno
 967  */
 968
 969 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 970 {
 971         const struct lm_lockops *lm;
 972         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 973         struct gfs2_args *args = &sdp->sd_args;
 974         const char *proto = sdp->sd_proto_name;
 975         const char *table = sdp->sd_table_name;
 976         char *o, *options;
 977         int ret;
 978
 979         if (!strcmp("lock_nolock", proto)) {
 980                 lm = &nolock_ops;
 981                 sdp->sd_args.ar_localflocks = 1;
 982 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
 983         } else if (!strcmp("lock_dlm", proto)) {
 984                 lm = &gfs2_dlm_ops;
 985 #endif
 986         } else {
 987                 pr_info("can't find protocol %s\n", proto);
 988                 return -ENOENT;
 989         }
 990
 991         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 992
 993         ls->ls_ops = lm;
 994         ls->ls_first = 1;
 995
 996         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
 997                 substring_t tmp[MAX_OPT_ARGS];
 998                 int token, option;
 999
1000                 if (!o || !*o)
1001                         continue;
1002
1003                 token = match_token(o, *lm->lm_tokens, tmp);
1004                 switch (token) {
1005                 case Opt_jid:
1006                         ret = match_int(&tmp[0], &option);
1007                         if (ret || option < 0)
1008                                 goto hostdata_error;
1009                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1010                                 ls->ls_jid = option;
1011                         break;
1012                 case Opt_id:
1013                 case Opt_nodir:
1014                         /* Obsolete, but left for backward compat purposes */
1015                         break;
1016                 case Opt_first:
1017                         ret = match_int(&tmp[0], &option);
1018                         if (ret || (option != 0 && option != 1))
1019                                 goto hostdata_error;
1020                         ls->ls_first = option;
1021                         break;
1022                 case Opt_err:
1023                 default:
1024 hostdata_error:
1025                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1026                         return -EINVAL;
1027                 }
1028         }
1029
1030         if (lm->lm_mount == NULL) {
1031                 fs_info(sdp, "Now mounting FS...\n");
1032                 complete_all(&sdp->sd_locking_init);
1033                 return 0;
1034         }
1035         ret = lm->lm_mount(sdp, table);
1036         if (ret == 0)
1037                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1038         complete_all(&sdp->sd_locking_init);
1039         return ret;
1040 }
1041
1042 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1043 {
1044         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1045         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1046                 lm->lm_unmount(sdp);
1047 }
1048
1049 static int wait_on_journal(struct gfs2_sbd *sdp)
1050 {
1051         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1052                 return 0;
1053
1054         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1055                 ? -EINTR : 0;
1056 }
1057
1058 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1059 {
1060         struct super_block *sb = sdp->sd_vfs;
1061         char ro[20];
1062         char spectator[20];
1063         char *envp[] = { ro, spectator, NULL };
1064         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1065         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1066         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1067 }
1068
1069 /**
1070  * gfs2_fill_super - Read in superblock
1071  * @sb: The VFS superblock
1072  * @args: Mount options
1073  * @silent: Don't complain if it's not a GFS2 filesystem
1074  *
1075  * Returns: -errno
1076  */
1077 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1078 {
1079         struct gfs2_args *args = fc->fs_private;
1080         int silent = fc->sb_flags & SB_SILENT;
1081         struct gfs2_sbd *sdp;
1082         struct gfs2_holder mount_gh;
1083         int error;
1084
1085         sdp = init_sbd(sb);
1086         if (!sdp) {
1087                 pr_warn("can't alloc struct gfs2_sbd\n");
1088                 return -ENOMEM;
1089         }
1090         sdp->sd_args = *args;
1091
1092         if (sdp->sd_args.ar_spectator) {
1093                 sb->s_flags |= SB_RDONLY;
1094                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1095         }
1096         if (sdp->sd_args.ar_posix_acl)
1097                 sb->s_flags |= SB_POSIXACL;
1098         if (sdp->sd_args.ar_nobarrier)
1099                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1100
1101         sb->s_flags |= SB_NOSEC;
1102         sb->s_magic = GFS2_MAGIC;
1103         sb->s_op = &gfs2_super_ops;
1104         sb->s_d_op = &gfs2_dops;
1105         sb->s_export_op = &gfs2_export_ops;
1106         sb->s_xattr = gfs2_xattr_handlers;
1107         sb->s_qcop = &gfs2_quotactl_ops;
1108         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1109         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1110         sb->s_time_gran = 1;
1111         sb->s_maxbytes = MAX_LFS_FILESIZE;
1112
1113         /* Set up the buffer cache and fill in some fake block size values
1114            to allow us to read-in the on-disk superblock. */
1115         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1116         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1117         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1118                                GFS2_BASIC_BLOCK_SHIFT;
1119         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1120
1121         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1122         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1123         if (sdp->sd_args.ar_statfs_quantum) {
1124                 sdp->sd_tune.gt_statfs_slow = 0;
1125                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1126         } else {
1127                 sdp->sd_tune.gt_statfs_slow = 1;
1128                 sdp->sd_tune.gt_statfs_quantum = 30;
1129         }
1130
1131         error = init_names(sdp, silent);
1132         if (error)
1133                 goto fail_free;
1134
1135         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1136
1137         error = gfs2_sys_fs_add(sdp);
1138         if (error)
1139                 goto fail_free;
1140
1141         gfs2_create_debugfs_file(sdp);
1142
1143         error = gfs2_lm_mount(sdp, silent);
1144         if (error)
1145                 goto fail_debug;
1146
1147         error = init_locking(sdp, &mount_gh, DO);
1148         if (error)
1149                 goto fail_lm;
1150
1151         error = init_sb(sdp, silent);
1152         if (error)
1153                 goto fail_locking;
1154
1155         error = wait_on_journal(sdp);
1156         if (error)
1157                 goto fail_sb;
1158
1159         /*
1160          * If user space has failed to join the cluster or some similar
1161          * failure has occurred, then the journal id will contain a
1162          * negative (error) number. This will then be returned to the
1163          * caller (of the mount syscall). We do this even for spectator
1164          * mounts (which just write a jid of 0 to indicate "ok" even though
1165          * the jid is unused in the spectator case)
1166          */
1167         if (sdp->sd_lockstruct.ls_jid < 0) {
1168                 error = sdp->sd_lockstruct.ls_jid;
1169                 sdp->sd_lockstruct.ls_jid = 0;
1170                 goto fail_sb;
1171         }
1172
1173         if (sdp->sd_args.ar_spectator)
1174                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1175                          sdp->sd_table_name);
1176         else
1177                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1178                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1179
1180         error = init_inodes(sdp, DO);
1181         if (error)
1182                 goto fail_sb;
1183
1184         error = init_per_node(sdp, DO);
1185         if (error)
1186                 goto fail_inodes;
1187
1188         error = gfs2_statfs_init(sdp);
1189         if (error) {
1190                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1191                 goto fail_per_node;
1192         }
1193
1194         if (sb_rdonly(sb)) {
1195                 struct gfs2_holder freeze_gh;
1196
1197                 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1198                                            LM_FLAG_NOEXP | GL_EXACT,
1199                                            &freeze_gh);
1200                 if (error) {
1201                         fs_err(sdp, "can't make FS RO: %d\n", error);
1202                         goto fail_per_node;
1203                 }
1204                 gfs2_glock_dq_uninit(&freeze_gh);
1205         } else {
1206                 error = gfs2_make_fs_rw(sdp);
1207                 if (error) {
1208                         fs_err(sdp, "can't make FS RW: %d\n", error);
1209                         goto fail_per_node;
1210                 }
1211         }
1212
1213         gfs2_glock_dq_uninit(&mount_gh);
1214         gfs2_online_uevent(sdp);
1215         return 0;
1216
1217 fail_per_node:
1218         init_per_node(sdp, UNDO);
1219 fail_inodes:
1220         init_inodes(sdp, UNDO);
1221 fail_sb:
1222         if (sdp->sd_root_dir)
1223                 dput(sdp->sd_root_dir);
1224         if (sdp->sd_master_dir)
1225                 dput(sdp->sd_master_dir);
1226         if (sb->s_root)
1227                 dput(sb->s_root);
1228         sb->s_root = NULL;
1229 fail_locking:
1230         init_locking(sdp, &mount_gh, UNDO);
1231 fail_lm:
1232         complete_all(&sdp->sd_journal_ready);
1233         gfs2_gl_hash_clear(sdp);
1234         gfs2_lm_unmount(sdp);
1235 fail_debug:
1236         gfs2_delete_debugfs_file(sdp);
1237         gfs2_sys_fs_del(sdp);
1238 fail_free:
1239         free_sbd(sdp);
1240         sb->s_fs_info = NULL;
1241         return error;
1242 }
1243
1244 /**
1245  * gfs2_get_tree - Get the GFS2 superblock and root directory
1246  * @fc: The filesystem context
1247  *
1248  * Returns: 0 or -errno on error
1249  */
1250 static int gfs2_get_tree(struct fs_context *fc)
1251 {
1252         struct gfs2_args *args = fc->fs_private;
1253         struct gfs2_sbd *sdp;
1254         int error;
1255
1256         error = get_tree_bdev(fc, gfs2_fill_super);
1257         if (error)
1258                 return error;
1259
1260         sdp = fc->root->d_sb->s_fs_info;
1261         dput(fc->root);
1262         if (args->ar_meta)
1263                 fc->root = dget(sdp->sd_master_dir);
1264         else
1265                 fc->root = dget(sdp->sd_root_dir);
1266         return 0;
1267 }
1268
1269 static void gfs2_fc_free(struct fs_context *fc)
1270 {
1271         struct gfs2_args *args = fc->fs_private;
1272
1273         kfree(args);
1274 }
1275
1276 enum gfs2_param {
1277         Opt_lockproto,
1278         Opt_locktable,
1279         Opt_hostdata,
1280         Opt_spectator,
1281         Opt_ignore_local_fs,
1282         Opt_localflocks,
1283         Opt_localcaching,
1284         Opt_debug,
1285         Opt_upgrade,
1286         Opt_acl,
1287         Opt_quota,
1288         Opt_quota_flag,
1289         Opt_suiddir,
1290         Opt_data,
1291         Opt_meta,
1292         Opt_discard,
1293         Opt_commit,
1294         Opt_errors,
1295         Opt_statfs_quantum,
1296         Opt_statfs_percent,
1297         Opt_quota_quantum,
1298         Opt_barrier,
1299         Opt_rgrplvb,
1300         Opt_loccookie,
1301 };
1302
1303 static const struct constant_table gfs2_param_quota[] = {
1304         {"off",        GFS2_QUOTA_OFF},
1305         {"account",    GFS2_QUOTA_ACCOUNT},
1306         {"on",         GFS2_QUOTA_ON},
1307         {}
1308 };
1309
1310 enum opt_data {
1311         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1312         Opt_data_ordered   = GFS2_DATA_ORDERED,
1313 };
1314
1315 static const struct constant_table gfs2_param_data[] = {
1316         {"writeback",  Opt_data_writeback },
1317         {"ordered",    Opt_data_ordered },
1318         {}
1319 };
1320
1321 enum opt_errors {
1322         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1323         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1324 };
1325
1326 static const struct constant_table gfs2_param_errors[] = {
1327         {"withdraw",   Opt_errors_withdraw },
1328         {"panic",      Opt_errors_panic },
1329         {}
1330 };
1331
1332 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1333         fsparam_string ("lockproto",          Opt_lockproto),
1334         fsparam_string ("locktable",          Opt_locktable),
1335         fsparam_string ("hostdata",           Opt_hostdata),
1336         fsparam_flag   ("spectator",          Opt_spectator),
1337         fsparam_flag   ("norecovery",         Opt_spectator),
1338         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1339         fsparam_flag   ("localflocks",        Opt_localflocks),
1340         fsparam_flag   ("localcaching",       Opt_localcaching),
1341         fsparam_flag_no("debug",              Opt_debug),
1342         fsparam_flag   ("upgrade",            Opt_upgrade),
1343         fsparam_flag_no("acl",                Opt_acl),
1344         fsparam_flag_no("suiddir",            Opt_suiddir),
1345         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1346         fsparam_flag   ("meta",               Opt_meta),
1347         fsparam_flag_no("discard",            Opt_discard),
1348         fsparam_s32    ("commit",             Opt_commit),
1349         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1350         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1351         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1352         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1353         fsparam_flag_no("barrier",            Opt_barrier),
1354         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1355         fsparam_flag_no("loccookie",          Opt_loccookie),
1356         /* quota can be a flag or an enum so it gets special treatment */
1357         fsparam_flag_no("quota",              Opt_quota_flag),
1358         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1359         {}
1360 };
1361
1362 /* Parse a single mount parameter */
1363 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1364 {
1365         struct gfs2_args *args = fc->fs_private;
1366         struct fs_parse_result result;
1367         int o;
1368
1369         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1370         if (o < 0)
1371                 return o;
1372
1373         switch (o) {
1374         case Opt_lockproto:
1375                 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1376                 break;
1377         case Opt_locktable:
1378                 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1379                 break;
1380         case Opt_hostdata:
1381                 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1382                 break;
1383         case Opt_spectator:
1384                 args->ar_spectator = 1;
1385                 break;
1386         case Opt_ignore_local_fs:
1387                 /* Retained for backwards compat only */
1388                 break;
1389         case Opt_localflocks:
1390                 args->ar_localflocks = 1;
1391                 break;
1392         case Opt_localcaching:
1393                 /* Retained for backwards compat only */
1394                 break;
1395         case Opt_debug:
1396                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1397                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1398                 args->ar_debug = result.boolean;
1399                 break;
1400         case Opt_upgrade:
1401                 /* Retained for backwards compat only */
1402                 break;
1403         case Opt_acl:
1404                 args->ar_posix_acl = result.boolean;
1405                 break;
1406         case Opt_quota_flag:
1407                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1408                 break;
1409         case Opt_quota:
1410                 args->ar_quota = result.int_32;
1411                 break;
1412         case Opt_suiddir:
1413                 args->ar_suiddir = result.boolean;
1414                 break;
1415         case Opt_data:
1416                 /* The uint_32 result maps directly to GFS2_DATA_* */
1417                 args->ar_data = result.uint_32;
1418                 break;
1419         case Opt_meta:
1420                 args->ar_meta = 1;
1421                 break;
1422         case Opt_discard:
1423                 args->ar_discard = result.boolean;
1424                 break;
1425         case Opt_commit:
1426                 if (result.int_32 <= 0)
1427                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1428                 args->ar_commit = result.int_32;
1429                 break;
1430         case Opt_statfs_quantum:
1431                 if (result.int_32 < 0)
1432                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1433                 args->ar_statfs_quantum = result.int_32;
1434                 break;
1435         case Opt_quota_quantum:
1436                 if (result.int_32 <= 0)
1437                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1438                 args->ar_quota_quantum = result.int_32;
1439                 break;
1440         case Opt_statfs_percent:
1441                 if (result.int_32 < 0 || result.int_32 > 100)
1442                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1443                 args->ar_statfs_percent = result.int_32;
1444                 break;
1445         case Opt_errors:
1446                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1447                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1448                 args->ar_errors = result.uint_32;
1449                 break;
1450         case Opt_barrier:
1451                 args->ar_nobarrier = result.boolean;
1452                 break;
1453         case Opt_rgrplvb:
1454                 args->ar_rgrplvb = result.boolean;
1455                 break;
1456         case Opt_loccookie:
1457                 args->ar_loccookie = result.boolean;
1458                 break;
1459         default:
1460                 return invalfc(fc, "invalid mount option: %s", param->key);
1461         }
1462         return 0;
1463 }
1464
1465 static int gfs2_reconfigure(struct fs_context *fc)
1466 {
1467         struct super_block *sb = fc->root->d_sb;
1468         struct gfs2_sbd *sdp = sb->s_fs_info;
1469         struct gfs2_args *oldargs = &sdp->sd_args;
1470         struct gfs2_args *newargs = fc->fs_private;
1471         struct gfs2_tune *gt = &sdp->sd_tune;
1472         int error = 0;
1473
1474         sync_filesystem(sb);
1475
1476         spin_lock(&gt->gt_spin);
1477         oldargs->ar_commit = gt->gt_logd_secs;
1478         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1479         if (gt->gt_statfs_slow)
1480                 oldargs->ar_statfs_quantum = 0;
1481         else
1482                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1483         spin_unlock(&gt->gt_spin);
1484
1485         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1486                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1487                 return -EINVAL;
1488         }
1489         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1490                 errorfc(fc, "reconfiguration of lock table not allowed");
1491                 return -EINVAL;
1492         }
1493         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1494                 errorfc(fc, "reconfiguration of host data not allowed");
1495                 return -EINVAL;
1496         }
1497         if (newargs->ar_spectator != oldargs->ar_spectator) {
1498                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1499                 return -EINVAL;
1500         }
1501         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1502                 errorfc(fc, "reconfiguration of localflocks not allowed");
1503                 return -EINVAL;
1504         }
1505         if (newargs->ar_meta != oldargs->ar_meta) {
1506                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1507                 return -EINVAL;
1508         }
1509         if (oldargs->ar_spectator)
1510                 fc->sb_flags |= SB_RDONLY;
1511
1512         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1513                 if (fc->sb_flags & SB_RDONLY) {
1514                         error = gfs2_make_fs_ro(sdp);
1515                         if (error)
1516                                 errorfc(fc, "unable to remount read-only");
1517                 } else {
1518                         error = gfs2_make_fs_rw(sdp);
1519                         if (error)
1520                                 errorfc(fc, "unable to remount read-write");
1521                 }
1522         }
1523         sdp->sd_args = *newargs;
1524
1525         if (sdp->sd_args.ar_posix_acl)
1526                 sb->s_flags |= SB_POSIXACL;
1527         else
1528                 sb->s_flags &= ~SB_POSIXACL;
1529         if (sdp->sd_args.ar_nobarrier)
1530                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1531         else
1532                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1533         spin_lock(&gt->gt_spin);
1534         gt->gt_logd_secs = newargs->ar_commit;
1535         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1536         if (newargs->ar_statfs_quantum) {
1537                 gt->gt_statfs_slow = 0;
1538                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1539         }
1540         else {
1541                 gt->gt_statfs_slow = 1;
1542                 gt->gt_statfs_quantum = 30;
1543         }
1544         spin_unlock(&gt->gt_spin);
1545
1546         gfs2_online_uevent(sdp);
1547         return error;
1548 }
1549
1550 static const struct fs_context_operations gfs2_context_ops = {
1551         .free        = gfs2_fc_free,
1552         .parse_param = gfs2_parse_param,
1553         .get_tree    = gfs2_get_tree,
1554         .reconfigure = gfs2_reconfigure,
1555 };
1556
1557 /* Set up the filesystem mount context */
1558 static int gfs2_init_fs_context(struct fs_context *fc)
1559 {
1560         struct gfs2_args *args;
1561
1562         args = kmalloc(sizeof(*args), GFP_KERNEL);
1563         if (args == NULL)
1564                 return -ENOMEM;
1565
1566         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1567                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1568
1569                 *args = sdp->sd_args;
1570         } else {
1571                 memset(args, 0, sizeof(*args));
1572                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1573                 args->ar_data = GFS2_DATA_DEFAULT;
1574                 args->ar_commit = 30;
1575                 args->ar_statfs_quantum = 30;
1576                 args->ar_quota_quantum = 60;
1577                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1578         }
1579         fc->fs_private = args;
1580         fc->ops = &gfs2_context_ops;
1581         return 0;
1582 }
1583
1584 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1585 {
1586         return -EINVAL;
1587 }
1588
1589 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1590 {
1591         return (fc->sget_key == s->s_bdev);
1592 }
1593
1594 static int gfs2_meta_get_tree(struct fs_context *fc)
1595 {
1596         struct super_block *s;
1597         struct gfs2_sbd *sdp;
1598         struct path path;
1599         int error;
1600
1601         if (!fc->source || !*fc->source)
1602                 return -EINVAL;
1603
1604         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1605         if (error) {
1606                 pr_warn("path_lookup on %s returned error %d\n",
1607                         fc->source, error);
1608                 return error;
1609         }
1610         fc->fs_type = &gfs2_fs_type;
1611         fc->sget_key = path.dentry->d_sb->s_bdev;
1612         s = sget_fc(fc, test_meta_super, set_meta_super);
1613         path_put(&path);
1614         if (IS_ERR(s)) {
1615                 pr_warn("gfs2 mount does not exist\n");
1616                 return PTR_ERR(s);
1617         }
1618         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1619                 deactivate_locked_super(s);
1620                 return -EBUSY;
1621         }
1622         sdp = s->s_fs_info;
1623         fc->root = dget(sdp->sd_master_dir);
1624         return 0;
1625 }
1626
1627 static const struct fs_context_operations gfs2_meta_context_ops = {
1628         .free        = gfs2_fc_free,
1629         .get_tree    = gfs2_meta_get_tree,
1630 };
1631
1632 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1633 {
1634         int ret = gfs2_init_fs_context(fc);
1635
1636         if (ret)
1637                 return ret;
1638
1639         fc->ops = &gfs2_meta_context_ops;
1640         return 0;
1641 }
1642
1643 static void gfs2_kill_sb(struct super_block *sb)
1644 {
1645         struct gfs2_sbd *sdp = sb->s_fs_info;
1646
1647         if (sdp == NULL) {
1648                 kill_block_super(sb);
1649                 return;
1650         }
1651
1652         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1653         dput(sdp->sd_root_dir);
1654         dput(sdp->sd_master_dir);
1655         sdp->sd_root_dir = NULL;
1656         sdp->sd_master_dir = NULL;
1657         shrink_dcache_sb(sb);
1658         kill_block_super(sb);
1659 }
1660
1661 struct file_system_type gfs2_fs_type = {
1662         .name = "gfs2",
1663         .fs_flags = FS_REQUIRES_DEV,
1664         .init_fs_context = gfs2_init_fs_context,
1665         .parameters = gfs2_fs_parameters,
1666         .kill_sb = gfs2_kill_sb,
1667         .owner = THIS_MODULE,
1668 };
1669 MODULE_ALIAS_FS("gfs2");
1670
1671 struct file_system_type gfs2meta_fs_type = {
1672         .name = "gfs2meta",
1673         .fs_flags = FS_REQUIRES_DEV,
1674         .init_fs_context = gfs2_meta_init_fs_context,
1675         .owner = THIS_MODULE,
1676 };
1677 MODULE_ALIAS_FS("gfs2meta");