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         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  * @silent: Don't print a message if the check fails
 154  *
 155  * Checks the version code of the FS is one that we understand how to
 156  * read and that the sizes of the various on-disk structures have not
 157  * changed.
 158  */
 159
 160 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
 161 {
 162         struct gfs2_sb_host *sb = &sdp->sd_sb;
 163
 164         if (sb->sb_magic != GFS2_MAGIC ||
 165             sb->sb_type != GFS2_METATYPE_SB) {
 166                 if (!silent)
 167                         pr_warn("not a GFS2 filesystem\n");
 168                 return -EINVAL;
 169         }
 170
 171         if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
 172             sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
 173             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 174                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
 175                 return -EINVAL;
 176         }
 177
 178         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
 179             (sb->sb_bsize & (sb->sb_bsize - 1))) {
 180                 pr_warn("Invalid block size\n");
 181                 return -EINVAL;
 182         }
 183
 184         return 0;
 185 }
 186
 187 static void end_bio_io_page(struct bio *bio)
 188 {
 189         struct page *page = bio->bi_private;
 190
 191         if (!bio->bi_status)
 192                 SetPageUptodate(page);
 193         else
 194                 pr_warn("error %d reading superblock\n", bio->bi_status);
 195         unlock_page(page);
 196 }
 197
 198 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
 199 {
 200         struct gfs2_sb_host *sb = &sdp->sd_sb;
 201         struct super_block *s = sdp->sd_vfs;
 202         const struct gfs2_sb *str = buf;
 203
 204         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 205         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 206         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 207         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 208         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 209         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 210         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 211         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 212         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 213         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 214
 215         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 216         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 217         memcpy(&s->s_uuid, str->sb_uuid, 16);
 218 }
 219
 220 /**
 221  * gfs2_read_super - Read the gfs2 super block from disk
 222  * @sdp: The GFS2 super block
 223  * @sector: The location of the super block
 224  * @silent: Don't print a message if the check fails
 225  *
 226  * This uses the bio functions to read the super block from disk
 227  * because we want to be 100% sure that we never read cached data.
 228  * A super block is read twice only during each GFS2 mount and is
 229  * never written to by the filesystem. The first time its read no
 230  * locks are held, and the only details which are looked at are those
 231  * relating to the locking protocol. Once locking is up and working,
 232  * the sb is read again under the lock to establish the location of
 233  * the master directory (contains pointers to journals etc) and the
 234  * root directory.
 235  *
 236  * Returns: 0 on success or error
 237  */
 238
 239 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
 240 {
 241         struct super_block *sb = sdp->sd_vfs;
 242         struct gfs2_sb *p;
 243         struct page *page;
 244         struct bio *bio;
 245
 246         page = alloc_page(GFP_NOFS);
 247         if (unlikely(!page))
 248                 return -ENOMEM;
 249
 250         ClearPageUptodate(page);
 251         ClearPageDirty(page);
 252         lock_page(page);
 253
 254         bio = bio_alloc(sb->s_bdev, 1, REQ_OP_READ | REQ_META, GFP_NOFS);
 255         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 256         bio_add_page(bio, page, PAGE_SIZE, 0);
 257
 258         bio->bi_end_io = end_bio_io_page;
 259         bio->bi_private = page;
 260         submit_bio(bio);
 261         wait_on_page_locked(page);
 262         bio_put(bio);
 263         if (!PageUptodate(page)) {
 264                 __free_page(page);
 265                 return -EIO;
 266         }
 267         p = kmap(page);
 268         gfs2_sb_in(sdp, p);
 269         kunmap(page);
 270         __free_page(page);
 271         return gfs2_check_sb(sdp, silent);
 272 }
 273
 274 /**
 275  * gfs2_read_sb - Read super block
 276  * @sdp: The GFS2 superblock
 277  * @silent: Don't print message if mount fails
 278  *
 279  */
 280
 281 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 282 {
 283         u32 hash_blocks, ind_blocks, leaf_blocks;
 284         u32 tmp_blocks;
 285         unsigned int x;
 286         int error;
 287
 288         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 289         if (error) {
 290                 if (!silent)
 291                         fs_err(sdp, "can't read superblock\n");
 292                 return error;
 293         }
 294
 295         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 296                                GFS2_BASIC_BLOCK_SHIFT;
 297         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 298         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 299                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 300         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 301                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 302         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
 303                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 304         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 305         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 306         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 307         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 308         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 309                                 sizeof(struct gfs2_meta_header)) /
 310                                 sizeof(struct gfs2_quota_change);
 311         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 312                                      sizeof(struct gfs2_meta_header))
 313                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 314
 315         /*
 316          * We always keep at least one block reserved for revokes in
 317          * transactions.  This greatly simplifies allocating additional
 318          * revoke blocks.
 319          */
 320         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
 321
 322         /* Compute maximum reservation required to add a entry to a directory */
 323
 324         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 325                              sdp->sd_jbsize);
 326
 327         ind_blocks = 0;
 328         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 329                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 330                 ind_blocks += tmp_blocks;
 331         }
 332
 333         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 334
 335         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 336
 337         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 338                                 sizeof(struct gfs2_dinode);
 339         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 340         for (x = 2;; x++) {
 341                 u64 space, d;
 342                 u32 m;
 343
 344                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 345                 d = space;
 346                 m = do_div(d, sdp->sd_inptrs);
 347
 348                 if (d != sdp->sd_heightsize[x - 1] || m)
 349                         break;
 350                 sdp->sd_heightsize[x] = space;
 351         }
 352         sdp->sd_max_height = x;
 353         sdp->sd_heightsize[x] = ~0;
 354         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 355
 356         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 357                                       sizeof(struct gfs2_leaf)) /
 358                                      GFS2_MIN_DIRENT_SIZE;
 359         return 0;
 360 }
 361
 362 static int init_names(struct gfs2_sbd *sdp, int silent)
 363 {
 364         char *proto, *table;
 365         int error = 0;
 366
 367         proto = sdp->sd_args.ar_lockproto;
 368         table = sdp->sd_args.ar_locktable;
 369
 370         /*  Try to autodetect  */
 371
 372         if (!proto[0] || !table[0]) {
 373                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 374                 if (error)
 375                         return error;
 376
 377                 if (!proto[0])
 378                         proto = sdp->sd_sb.sb_lockproto;
 379                 if (!table[0])
 380                         table = sdp->sd_sb.sb_locktable;
 381         }
 382
 383         if (!table[0])
 384                 table = sdp->sd_vfs->s_id;
 385
 386         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 387         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 388
 389         table = sdp->sd_table_name;
 390         while ((table = strchr(table, '/')))
 391                 *table = '_';
 392
 393         return error;
 394 }
 395
 396 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 397                         int undo)
 398 {
 399         int error = 0;
 400
 401         if (undo)
 402                 goto fail_trans;
 403
 404         error = gfs2_glock_nq_num(sdp,
 405                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 406                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 407                                   mount_gh);
 408         if (error) {
 409                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 410                 goto fail;
 411         }
 412
 413         error = gfs2_glock_nq_num(sdp,
 414                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 415                                   LM_ST_SHARED,
 416                                   LM_FLAG_NOEXP | GL_EXACT,
 417                                   &sdp->sd_live_gh);
 418         if (error) {
 419                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 420                 goto fail_mount;
 421         }
 422
 423         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 424                                CREATE, &sdp->sd_rename_gl);
 425         if (error) {
 426                 fs_err(sdp, "can't create rename glock: %d\n", error);
 427                 goto fail_live;
 428         }
 429
 430         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 431                                CREATE, &sdp->sd_freeze_gl);
 432         if (error) {
 433                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 434                 goto fail_rename;
 435         }
 436
 437         return 0;
 438
 439 fail_trans:
 440         gfs2_glock_put(sdp->sd_freeze_gl);
 441 fail_rename:
 442         gfs2_glock_put(sdp->sd_rename_gl);
 443 fail_live:
 444         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 445 fail_mount:
 446         gfs2_glock_dq_uninit(mount_gh);
 447 fail:
 448         return error;
 449 }
 450
 451 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 452                             u64 no_addr, const char *name)
 453 {
 454         struct gfs2_sbd *sdp = sb->s_fs_info;
 455         struct dentry *dentry;
 456         struct inode *inode;
 457
 458         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 459                                   GFS2_BLKST_FREE /* ignore */);
 460         if (IS_ERR(inode)) {
 461                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 462                 return PTR_ERR(inode);
 463         }
 464         dentry = d_make_root(inode);
 465         if (!dentry) {
 466                 fs_err(sdp, "can't alloc %s dentry\n", name);
 467                 return -ENOMEM;
 468         }
 469         *dptr = dentry;
 470         return 0;
 471 }
 472
 473 static int init_sb(struct gfs2_sbd *sdp, int silent)
 474 {
 475         struct super_block *sb = sdp->sd_vfs;
 476         struct gfs2_holder sb_gh;
 477         u64 no_addr;
 478         int ret;
 479
 480         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 481                                 LM_ST_SHARED, 0, &sb_gh);
 482         if (ret) {
 483                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 484                 return ret;
 485         }
 486
 487         ret = gfs2_read_sb(sdp, silent);
 488         if (ret) {
 489                 fs_err(sdp, "can't read superblock: %d\n", ret);
 490                 goto out;
 491         }
 492
 493         switch(sdp->sd_sb.sb_fs_format) {
 494         case GFS2_FS_FORMAT_MAX:
 495                 sb->s_xattr = gfs2_xattr_handlers_max;
 496                 break;
 497
 498         case GFS2_FS_FORMAT_MIN:
 499                 sb->s_xattr = gfs2_xattr_handlers_min;
 500                 break;
 501
 502         default:
 503                 BUG();
 504         }
 505
 506         /* Set up the buffer cache and SB for real */
 507         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 508                 ret = -EINVAL;
 509                 fs_err(sdp, "FS block size (%u) is too small for device "
 510                        "block size (%u)\n",
 511                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 512                 goto out;
 513         }
 514         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 515                 ret = -EINVAL;
 516                 fs_err(sdp, "FS block size (%u) is too big for machine "
 517                        "page size (%u)\n",
 518                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 519                 goto out;
 520         }
 521         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 522
 523         /* Get the root inode */
 524         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 525         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 526         if (ret)
 527                 goto out;
 528
 529         /* Get the master inode */
 530         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 531         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 532         if (ret) {
 533                 dput(sdp->sd_root_dir);
 534                 goto out;
 535         }
 536         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 537 out:
 538         gfs2_glock_dq_uninit(&sb_gh);
 539         return ret;
 540 }
 541
 542 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 543 {
 544         char *message = "FIRSTMOUNT=Done";
 545         char *envp[] = { message, NULL };
 546
 547         fs_info(sdp, "first mount done, others may mount\n");
 548
 549         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 550                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 551
 552         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 553 }
 554
 555 /**
 556  * gfs2_jindex_hold - Grab a lock on the jindex
 557  * @sdp: The GFS2 superblock
 558  * @ji_gh: the holder for the jindex glock
 559  *
 560  * Returns: errno
 561  */
 562
 563 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 564 {
 565         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 566         struct qstr name;
 567         char buf[20];
 568         struct gfs2_jdesc *jd;
 569         int error;
 570
 571         name.name = buf;
 572
 573         mutex_lock(&sdp->sd_jindex_mutex);
 574
 575         for (;;) {
 576                 struct gfs2_inode *jip;
 577
 578                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 579                 if (error)
 580                         break;
 581
 582                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 583                 name.hash = gfs2_disk_hash(name.name, name.len);
 584
 585                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 586                 if (error == -ENOENT) {
 587                         error = 0;
 588                         break;
 589                 }
 590
 591                 gfs2_glock_dq_uninit(ji_gh);
 592
 593                 if (error)
 594                         break;
 595
 596                 error = -ENOMEM;
 597                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 598                 if (!jd)
 599                         break;
 600
 601                 INIT_LIST_HEAD(&jd->extent_list);
 602                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 603
 604                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 605                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 606                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 607                         if (!jd->jd_inode)
 608                                 error = -ENOENT;
 609                         else
 610                                 error = PTR_ERR(jd->jd_inode);
 611                         kfree(jd);
 612                         break;
 613                 }
 614
 615                 d_mark_dontcache(jd->jd_inode);
 616                 spin_lock(&sdp->sd_jindex_spin);
 617                 jd->jd_jid = sdp->sd_journals++;
 618                 jip = GFS2_I(jd->jd_inode);
 619                 jd->jd_no_addr = jip->i_no_addr;
 620                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 621                 spin_unlock(&sdp->sd_jindex_spin);
 622         }
 623
 624         mutex_unlock(&sdp->sd_jindex_mutex);
 625
 626         return error;
 627 }
 628
 629 /**
 630  * init_statfs - look up and initialize master and local (per node) statfs inodes
 631  * @sdp: The GFS2 superblock
 632  *
 633  * This should be called after the jindex is initialized in init_journal() and
 634  * before gfs2_journal_recovery() is called because we need to be able to write
 635  * to these inodes during recovery.
 636  *
 637  * Returns: errno
 638  */
 639 static int init_statfs(struct gfs2_sbd *sdp)
 640 {
 641         int error = 0;
 642         struct inode *master = d_inode(sdp->sd_master_dir);
 643         struct inode *pn = NULL;
 644         char buf[30];
 645         struct gfs2_jdesc *jd;
 646         struct gfs2_inode *ip;
 647
 648         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 649         if (IS_ERR(sdp->sd_statfs_inode)) {
 650                 error = PTR_ERR(sdp->sd_statfs_inode);
 651                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 652                 goto out;
 653         }
 654         if (sdp->sd_args.ar_spectator)
 655                 goto out;
 656
 657         pn = gfs2_lookup_simple(master, "per_node");
 658         if (IS_ERR(pn)) {
 659                 error = PTR_ERR(pn);
 660                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 661                 goto put_statfs;
 662         }
 663
 664         /* For each jid, lookup the corresponding local statfs inode in the
 665          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
 666         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 667                 struct local_statfs_inode *lsi =
 668                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
 669                 if (!lsi) {
 670                         error = -ENOMEM;
 671                         goto free_local;
 672                 }
 673                 sprintf(buf, "statfs_change%u", jd->jd_jid);
 674                 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
 675                 if (IS_ERR(lsi->si_sc_inode)) {
 676                         error = PTR_ERR(lsi->si_sc_inode);
 677                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
 678                                jd->jd_jid, error);
 679                         kfree(lsi);
 680                         goto free_local;
 681                 }
 682                 lsi->si_jid = jd->jd_jid;
 683                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
 684                         sdp->sd_sc_inode = lsi->si_sc_inode;
 685
 686                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
 687         }
 688
 689         iput(pn);
 690         pn = NULL;
 691         ip = GFS2_I(sdp->sd_sc_inode);
 692         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 693                                    &sdp->sd_sc_gh);
 694         if (error) {
 695                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 696                 goto free_local;
 697         }
 698         /* read in the local statfs buffer - other nodes don't change it. */
 699         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
 700         if (error) {
 701                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
 702                 goto unlock_sd_gh;
 703         }
 704         return 0;
 705
 706 unlock_sd_gh:
 707         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 708 free_local:
 709         free_local_statfs_inodes(sdp);
 710         iput(pn);
 711 put_statfs:
 712         iput(sdp->sd_statfs_inode);
 713 out:
 714         return error;
 715 }
 716
 717 /* Uninitialize and free up memory used by the list of statfs inodes */
 718 static void uninit_statfs(struct gfs2_sbd *sdp)
 719 {
 720         if (!sdp->sd_args.ar_spectator) {
 721                 brelse(sdp->sd_sc_bh);
 722                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 723                 free_local_statfs_inodes(sdp);
 724         }
 725         iput(sdp->sd_statfs_inode);
 726 }
 727
 728 static int init_journal(struct gfs2_sbd *sdp, int undo)
 729 {
 730         struct inode *master = d_inode(sdp->sd_master_dir);
 731         struct gfs2_holder ji_gh;
 732         struct gfs2_inode *ip;
 733         int jindex = 1;
 734         int error = 0;
 735
 736         if (undo) {
 737                 jindex = 0;
 738                 goto fail_statfs;
 739         }
 740
 741         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 742         if (IS_ERR(sdp->sd_jindex)) {
 743                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 744                 return PTR_ERR(sdp->sd_jindex);
 745         }
 746
 747         /* Load in the journal index special file */
 748
 749         error = gfs2_jindex_hold(sdp, &ji_gh);
 750         if (error) {
 751                 fs_err(sdp, "can't read journal index: %d\n", error);
 752                 goto fail;
 753         }
 754
 755         error = -EUSERS;
 756         if (!gfs2_jindex_size(sdp)) {
 757                 fs_err(sdp, "no journals!\n");
 758                 goto fail_jindex;
 759         }
 760
 761         atomic_set(&sdp->sd_log_blks_needed, 0);
 762         if (sdp->sd_args.ar_spectator) {
 763                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 764                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 765                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 766                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 767         } else {
 768                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 769                         fs_err(sdp, "can't mount journal #%u\n",
 770                                sdp->sd_lockstruct.ls_jid);
 771                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 772                                gfs2_jindex_size(sdp),
 773                                gfs2_jindex_size(sdp) - 1);
 774                         goto fail_jindex;
 775                 }
 776                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 777
 778                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 779                                           &gfs2_journal_glops,
 780                                           LM_ST_EXCLUSIVE,
 781                                           LM_FLAG_NOEXP | GL_NOCACHE,
 782                                           &sdp->sd_journal_gh);
 783                 if (error) {
 784                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 785                         goto fail_jindex;
 786                 }
 787
 788                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 789                 sdp->sd_jinode_gl = ip->i_gl;
 790                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 791                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 792                                            &sdp->sd_jinode_gh);
 793                 if (error) {
 794                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 795                                error);
 796                         goto fail_journal_gh;
 797                 }
 798
 799                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 800                 if (error) {
 801                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 802                                sdp->sd_jdesc->jd_jid, error);
 803                         goto fail_jinode_gh;
 804                 }
 805                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 806                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 807                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 808
 809                 /* Map the extents for this journal's blocks */
 810                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 811         }
 812         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 813
 814         /* Lookup statfs inodes here so journal recovery can use them. */
 815         error = init_statfs(sdp);
 816         if (error)
 817                 goto fail_jinode_gh;
 818
 819         if (sdp->sd_lockstruct.ls_first) {
 820                 unsigned int x;
 821                 for (x = 0; x < sdp->sd_journals; x++) {
 822                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 823
 824                         if (sdp->sd_args.ar_spectator) {
 825                                 error = check_journal_clean(sdp, jd, true);
 826                                 if (error)
 827                                         goto fail_statfs;
 828                                 continue;
 829                         }
 830                         error = gfs2_recover_journal(jd, true);
 831                         if (error) {
 832                                 fs_err(sdp, "error recovering journal %u: %d\n",
 833                                        x, error);
 834                                 goto fail_statfs;
 835                         }
 836                 }
 837
 838                 gfs2_others_may_mount(sdp);
 839         } else if (!sdp->sd_args.ar_spectator) {
 840                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 841                 if (error) {
 842                         fs_err(sdp, "error recovering my journal: %d\n", error);
 843                         goto fail_statfs;
 844                 }
 845         }
 846
 847         sdp->sd_log_idle = 1;
 848         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 849         gfs2_glock_dq_uninit(&ji_gh);
 850         jindex = 0;
 851         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 852         return 0;
 853
 854 fail_statfs:
 855         uninit_statfs(sdp);
 856 fail_jinode_gh:
 857         /* A withdraw may have done dq/uninit so now we need to check it */
 858         if (!sdp->sd_args.ar_spectator &&
 859             gfs2_holder_initialized(&sdp->sd_jinode_gh))
 860                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 861 fail_journal_gh:
 862         if (!sdp->sd_args.ar_spectator &&
 863             gfs2_holder_initialized(&sdp->sd_journal_gh))
 864                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 865 fail_jindex:
 866         gfs2_jindex_free(sdp);
 867         if (jindex)
 868                 gfs2_glock_dq_uninit(&ji_gh);
 869 fail:
 870         iput(sdp->sd_jindex);
 871         return error;
 872 }
 873
 874 static struct lock_class_key gfs2_quota_imutex_key;
 875
 876 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 877 {
 878         int error = 0;
 879         struct inode *master = d_inode(sdp->sd_master_dir);
 880
 881         if (undo)
 882                 goto fail_qinode;
 883
 884         error = init_journal(sdp, undo);
 885         complete_all(&sdp->sd_journal_ready);
 886         if (error)
 887                 goto fail;
 888
 889         /* Read in the resource index inode */
 890         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 891         if (IS_ERR(sdp->sd_rindex)) {
 892                 error = PTR_ERR(sdp->sd_rindex);
 893                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 894                 goto fail_journal;
 895         }
 896         sdp->sd_rindex_uptodate = 0;
 897
 898         /* Read in the quota inode */
 899         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 900         if (IS_ERR(sdp->sd_quota_inode)) {
 901                 error = PTR_ERR(sdp->sd_quota_inode);
 902                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 903                 goto fail_rindex;
 904         }
 905         /*
 906          * i_rwsem on quota files is special. Since this inode is hidden system
 907          * file, we are safe to define locking ourselves.
 908          */
 909         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 910                           &gfs2_quota_imutex_key);
 911
 912         error = gfs2_rindex_update(sdp);
 913         if (error)
 914                 goto fail_qinode;
 915
 916         return 0;
 917
 918 fail_qinode:
 919         iput(sdp->sd_quota_inode);
 920 fail_rindex:
 921         gfs2_clear_rgrpd(sdp);
 922         iput(sdp->sd_rindex);
 923 fail_journal:
 924         init_journal(sdp, UNDO);
 925 fail:
 926         return error;
 927 }
 928
 929 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 930 {
 931         struct inode *pn = NULL;
 932         char buf[30];
 933         int error = 0;
 934         struct gfs2_inode *ip;
 935         struct inode *master = d_inode(sdp->sd_master_dir);
 936
 937         if (sdp->sd_args.ar_spectator)
 938                 return 0;
 939
 940         if (undo)
 941                 goto fail_qc_gh;
 942
 943         pn = gfs2_lookup_simple(master, "per_node");
 944         if (IS_ERR(pn)) {
 945                 error = PTR_ERR(pn);
 946                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 947                 return error;
 948         }
 949
 950         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 951         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 952         if (IS_ERR(sdp->sd_qc_inode)) {
 953                 error = PTR_ERR(sdp->sd_qc_inode);
 954                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 955                 goto fail_ut_i;
 956         }
 957
 958         iput(pn);
 959         pn = NULL;
 960
 961         ip = GFS2_I(sdp->sd_qc_inode);
 962         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 963                                    &sdp->sd_qc_gh);
 964         if (error) {
 965                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 966                 goto fail_qc_i;
 967         }
 968
 969         return 0;
 970
 971 fail_qc_gh:
 972         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 973 fail_qc_i:
 974         iput(sdp->sd_qc_inode);
 975 fail_ut_i:
 976         iput(pn);
 977         return error;
 978 }
 979
 980 static const match_table_t nolock_tokens = {
 981         { Opt_jid, "jid=%d", },
 982         { Opt_err, NULL },
 983 };
 984
 985 static const struct lm_lockops nolock_ops = {
 986         .lm_proto_name = "lock_nolock",
 987         .lm_put_lock = gfs2_glock_free,
 988         .lm_tokens = &nolock_tokens,
 989 };
 990
 991 /**
 992  * gfs2_lm_mount - mount a locking protocol
 993  * @sdp: the filesystem
 994  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 995  *
 996  * Returns: errno
 997  */
 998
 999 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1000 {
1001         const struct lm_lockops *lm;
1002         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1003         struct gfs2_args *args = &sdp->sd_args;
1004         const char *proto = sdp->sd_proto_name;
1005         const char *table = sdp->sd_table_name;
1006         char *o, *options;
1007         int ret;
1008
1009         if (!strcmp("lock_nolock", proto)) {
1010                 lm = &nolock_ops;
1011                 sdp->sd_args.ar_localflocks = 1;
1012 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1013         } else if (!strcmp("lock_dlm", proto)) {
1014                 lm = &gfs2_dlm_ops;
1015 #endif
1016         } else {
1017                 pr_info("can't find protocol %s\n", proto);
1018                 return -ENOENT;
1019         }
1020
1021         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1022
1023         ls->ls_ops = lm;
1024         ls->ls_first = 1;
1025
1026         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1027                 substring_t tmp[MAX_OPT_ARGS];
1028                 int token, option;
1029
1030                 if (!o || !*o)
1031                         continue;
1032
1033                 token = match_token(o, *lm->lm_tokens, tmp);
1034                 switch (token) {
1035                 case Opt_jid:
1036                         ret = match_int(&tmp[0], &option);
1037                         if (ret || option < 0)
1038                                 goto hostdata_error;
1039                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1040                                 ls->ls_jid = option;
1041                         break;
1042                 case Opt_id:
1043                 case Opt_nodir:
1044                         /* Obsolete, but left for backward compat purposes */
1045                         break;
1046                 case Opt_first:
1047                         ret = match_int(&tmp[0], &option);
1048                         if (ret || (option != 0 && option != 1))
1049                                 goto hostdata_error;
1050                         ls->ls_first = option;
1051                         break;
1052                 case Opt_err:
1053                 default:
1054 hostdata_error:
1055                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1056                         return -EINVAL;
1057                 }
1058         }
1059
1060         if (lm->lm_mount == NULL) {
1061                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1062                 complete_all(&sdp->sd_locking_init);
1063                 return 0;
1064         }
1065         ret = lm->lm_mount(sdp, table);
1066         if (ret == 0)
1067                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1068                         sdp->sd_sb.sb_fs_format);
1069         complete_all(&sdp->sd_locking_init);
1070         return ret;
1071 }
1072
1073 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1074 {
1075         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1076         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1077                 lm->lm_unmount(sdp);
1078 }
1079
1080 static int wait_on_journal(struct gfs2_sbd *sdp)
1081 {
1082         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1083                 return 0;
1084
1085         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1086                 ? -EINTR : 0;
1087 }
1088
1089 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1090 {
1091         struct super_block *sb = sdp->sd_vfs;
1092         char ro[20];
1093         char spectator[20];
1094         char *envp[] = { ro, spectator, NULL };
1095         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1096         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1097         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1098 }
1099
1100 static int init_threads(struct gfs2_sbd *sdp)
1101 {
1102         struct task_struct *p;
1103         int error = 0;
1104
1105         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1106         if (IS_ERR(p)) {
1107                 error = PTR_ERR(p);
1108                 fs_err(sdp, "can't start logd thread: %d\n", error);
1109                 return error;
1110         }
1111         sdp->sd_logd_process = p;
1112
1113         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1114         if (IS_ERR(p)) {
1115                 error = PTR_ERR(p);
1116                 fs_err(sdp, "can't start quotad thread: %d\n", error);
1117                 goto fail;
1118         }
1119         sdp->sd_quotad_process = p;
1120         return 0;
1121
1122 fail:
1123         kthread_stop(sdp->sd_logd_process);
1124         sdp->sd_logd_process = NULL;
1125         return error;
1126 }
1127
1128 /**
1129  * gfs2_fill_super - Read in superblock
1130  * @sb: The VFS superblock
1131  * @fc: Mount options and flags
1132  *
1133  * Returns: -errno
1134  */
1135 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1136 {
1137         struct gfs2_args *args = fc->fs_private;
1138         int silent = fc->sb_flags & SB_SILENT;
1139         struct gfs2_sbd *sdp;
1140         struct gfs2_holder mount_gh;
1141         struct gfs2_holder freeze_gh;
1142         int error;
1143
1144         sdp = init_sbd(sb);
1145         if (!sdp) {
1146                 pr_warn("can't alloc struct gfs2_sbd\n");
1147                 return -ENOMEM;
1148         }
1149         sdp->sd_args = *args;
1150
1151         if (sdp->sd_args.ar_spectator) {
1152                 sb->s_flags |= SB_RDONLY;
1153                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1154         }
1155         if (sdp->sd_args.ar_posix_acl)
1156                 sb->s_flags |= SB_POSIXACL;
1157         if (sdp->sd_args.ar_nobarrier)
1158                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1159
1160         sb->s_flags |= SB_NOSEC;
1161         sb->s_magic = GFS2_MAGIC;
1162         sb->s_op = &gfs2_super_ops;
1163         sb->s_d_op = &gfs2_dops;
1164         sb->s_export_op = &gfs2_export_ops;
1165         sb->s_qcop = &gfs2_quotactl_ops;
1166         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1167         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1168         sb->s_time_gran = 1;
1169         sb->s_maxbytes = MAX_LFS_FILESIZE;
1170
1171         /* Set up the buffer cache and fill in some fake block size values
1172            to allow us to read-in the on-disk superblock. */
1173         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1174         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1175         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1176                                GFS2_BASIC_BLOCK_SHIFT;
1177         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1178
1179         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1180         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1181         if (sdp->sd_args.ar_statfs_quantum) {
1182                 sdp->sd_tune.gt_statfs_slow = 0;
1183                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1184         } else {
1185                 sdp->sd_tune.gt_statfs_slow = 1;
1186                 sdp->sd_tune.gt_statfs_quantum = 30;
1187         }
1188
1189         error = init_names(sdp, silent);
1190         if (error)
1191                 goto fail_free;
1192
1193         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1194
1195         error = gfs2_sys_fs_add(sdp);
1196         if (error)
1197                 goto fail_free;
1198
1199         gfs2_create_debugfs_file(sdp);
1200
1201         error = gfs2_lm_mount(sdp, silent);
1202         if (error)
1203                 goto fail_debug;
1204
1205         error = init_locking(sdp, &mount_gh, DO);
1206         if (error)
1207                 goto fail_lm;
1208
1209         error = init_sb(sdp, silent);
1210         if (error)
1211                 goto fail_locking;
1212
1213         /* Turn rgrplvb on by default if fs format is recent enough */
1214         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1215                 sdp->sd_args.ar_rgrplvb = 1;
1216
1217         error = wait_on_journal(sdp);
1218         if (error)
1219                 goto fail_sb;
1220
1221         /*
1222          * If user space has failed to join the cluster or some similar
1223          * failure has occurred, then the journal id will contain a
1224          * negative (error) number. This will then be returned to the
1225          * caller (of the mount syscall). We do this even for spectator
1226          * mounts (which just write a jid of 0 to indicate "ok" even though
1227          * the jid is unused in the spectator case)
1228          */
1229         if (sdp->sd_lockstruct.ls_jid < 0) {
1230                 error = sdp->sd_lockstruct.ls_jid;
1231                 sdp->sd_lockstruct.ls_jid = 0;
1232                 goto fail_sb;
1233         }
1234
1235         if (sdp->sd_args.ar_spectator)
1236                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1237                          sdp->sd_table_name);
1238         else
1239                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1240                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1241
1242         error = init_inodes(sdp, DO);
1243         if (error)
1244                 goto fail_sb;
1245
1246         error = init_per_node(sdp, DO);
1247         if (error)
1248                 goto fail_inodes;
1249
1250         error = gfs2_statfs_init(sdp);
1251         if (error) {
1252                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1253                 goto fail_per_node;
1254         }
1255
1256         if (!sb_rdonly(sb)) {
1257                 error = init_threads(sdp);
1258                 if (error) {
1259                         gfs2_withdraw_delayed(sdp);
1260                         goto fail_per_node;
1261                 }
1262         }
1263
1264         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1265         if (error)
1266                 goto fail_per_node;
1267
1268         if (!sb_rdonly(sb))
1269                 error = gfs2_make_fs_rw(sdp);
1270
1271         gfs2_freeze_unlock(&freeze_gh);
1272         if (error) {
1273                 if (sdp->sd_quotad_process)
1274                         kthread_stop(sdp->sd_quotad_process);
1275                 sdp->sd_quotad_process = NULL;
1276                 if (sdp->sd_logd_process)
1277                         kthread_stop(sdp->sd_logd_process);
1278                 sdp->sd_logd_process = NULL;
1279                 fs_err(sdp, "can't make FS RW: %d\n", error);
1280                 goto fail_per_node;
1281         }
1282         gfs2_glock_dq_uninit(&mount_gh);
1283         gfs2_online_uevent(sdp);
1284         return 0;
1285
1286 fail_per_node:
1287         init_per_node(sdp, UNDO);
1288 fail_inodes:
1289         init_inodes(sdp, UNDO);
1290 fail_sb:
1291         if (sdp->sd_root_dir)
1292                 dput(sdp->sd_root_dir);
1293         if (sdp->sd_master_dir)
1294                 dput(sdp->sd_master_dir);
1295         if (sb->s_root)
1296                 dput(sb->s_root);
1297         sb->s_root = NULL;
1298 fail_locking:
1299         init_locking(sdp, &mount_gh, UNDO);
1300 fail_lm:
1301         complete_all(&sdp->sd_journal_ready);
1302         gfs2_gl_hash_clear(sdp);
1303         gfs2_lm_unmount(sdp);
1304 fail_debug:
1305         gfs2_delete_debugfs_file(sdp);
1306         gfs2_sys_fs_del(sdp);
1307 fail_free:
1308         free_sbd(sdp);
1309         sb->s_fs_info = NULL;
1310         return error;
1311 }
1312
1313 /**
1314  * gfs2_get_tree - Get the GFS2 superblock and root directory
1315  * @fc: The filesystem context
1316  *
1317  * Returns: 0 or -errno on error
1318  */
1319 static int gfs2_get_tree(struct fs_context *fc)
1320 {
1321         struct gfs2_args *args = fc->fs_private;
1322         struct gfs2_sbd *sdp;
1323         int error;
1324
1325         error = get_tree_bdev(fc, gfs2_fill_super);
1326         if (error)
1327                 return error;
1328
1329         sdp = fc->root->d_sb->s_fs_info;
1330         dput(fc->root);
1331         if (args->ar_meta)
1332                 fc->root = dget(sdp->sd_master_dir);
1333         else
1334                 fc->root = dget(sdp->sd_root_dir);
1335         return 0;
1336 }
1337
1338 static void gfs2_fc_free(struct fs_context *fc)
1339 {
1340         struct gfs2_args *args = fc->fs_private;
1341
1342         kfree(args);
1343 }
1344
1345 enum gfs2_param {
1346         Opt_lockproto,
1347         Opt_locktable,
1348         Opt_hostdata,
1349         Opt_spectator,
1350         Opt_ignore_local_fs,
1351         Opt_localflocks,
1352         Opt_localcaching,
1353         Opt_debug,
1354         Opt_upgrade,
1355         Opt_acl,
1356         Opt_quota,
1357         Opt_quota_flag,
1358         Opt_suiddir,
1359         Opt_data,
1360         Opt_meta,
1361         Opt_discard,
1362         Opt_commit,
1363         Opt_errors,
1364         Opt_statfs_quantum,
1365         Opt_statfs_percent,
1366         Opt_quota_quantum,
1367         Opt_barrier,
1368         Opt_rgrplvb,
1369         Opt_loccookie,
1370 };
1371
1372 static const struct constant_table gfs2_param_quota[] = {
1373         {"off",        GFS2_QUOTA_OFF},
1374         {"account",    GFS2_QUOTA_ACCOUNT},
1375         {"on",         GFS2_QUOTA_ON},
1376         {}
1377 };
1378
1379 enum opt_data {
1380         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1381         Opt_data_ordered   = GFS2_DATA_ORDERED,
1382 };
1383
1384 static const struct constant_table gfs2_param_data[] = {
1385         {"writeback",  Opt_data_writeback },
1386         {"ordered",    Opt_data_ordered },
1387         {}
1388 };
1389
1390 enum opt_errors {
1391         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1392         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1393 };
1394
1395 static const struct constant_table gfs2_param_errors[] = {
1396         {"withdraw",   Opt_errors_withdraw },
1397         {"panic",      Opt_errors_panic },
1398         {}
1399 };
1400
1401 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1402         fsparam_string ("lockproto",          Opt_lockproto),
1403         fsparam_string ("locktable",          Opt_locktable),
1404         fsparam_string ("hostdata",           Opt_hostdata),
1405         fsparam_flag   ("spectator",          Opt_spectator),
1406         fsparam_flag   ("norecovery",         Opt_spectator),
1407         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1408         fsparam_flag   ("localflocks",        Opt_localflocks),
1409         fsparam_flag   ("localcaching",       Opt_localcaching),
1410         fsparam_flag_no("debug",              Opt_debug),
1411         fsparam_flag   ("upgrade",            Opt_upgrade),
1412         fsparam_flag_no("acl",                Opt_acl),
1413         fsparam_flag_no("suiddir",            Opt_suiddir),
1414         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1415         fsparam_flag   ("meta",               Opt_meta),
1416         fsparam_flag_no("discard",            Opt_discard),
1417         fsparam_s32    ("commit",             Opt_commit),
1418         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1419         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1420         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1421         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1422         fsparam_flag_no("barrier",            Opt_barrier),
1423         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1424         fsparam_flag_no("loccookie",          Opt_loccookie),
1425         /* quota can be a flag or an enum so it gets special treatment */
1426         fsparam_flag_no("quota",              Opt_quota_flag),
1427         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1428         {}
1429 };
1430
1431 /* Parse a single mount parameter */
1432 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1433 {
1434         struct gfs2_args *args = fc->fs_private;
1435         struct fs_parse_result result;
1436         int o;
1437
1438         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1439         if (o < 0)
1440                 return o;
1441
1442         switch (o) {
1443         case Opt_lockproto:
1444                 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1445                 break;
1446         case Opt_locktable:
1447                 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1448                 break;
1449         case Opt_hostdata:
1450                 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1451                 break;
1452         case Opt_spectator:
1453                 args->ar_spectator = 1;
1454                 break;
1455         case Opt_ignore_local_fs:
1456                 /* Retained for backwards compat only */
1457                 break;
1458         case Opt_localflocks:
1459                 args->ar_localflocks = 1;
1460                 break;
1461         case Opt_localcaching:
1462                 /* Retained for backwards compat only */
1463                 break;
1464         case Opt_debug:
1465                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1466                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1467                 args->ar_debug = result.boolean;
1468                 break;
1469         case Opt_upgrade:
1470                 /* Retained for backwards compat only */
1471                 break;
1472         case Opt_acl:
1473                 args->ar_posix_acl = result.boolean;
1474                 break;
1475         case Opt_quota_flag:
1476                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1477                 break;
1478         case Opt_quota:
1479                 args->ar_quota = result.int_32;
1480                 break;
1481         case Opt_suiddir:
1482                 args->ar_suiddir = result.boolean;
1483                 break;
1484         case Opt_data:
1485                 /* The uint_32 result maps directly to GFS2_DATA_* */
1486                 args->ar_data = result.uint_32;
1487                 break;
1488         case Opt_meta:
1489                 args->ar_meta = 1;
1490                 break;
1491         case Opt_discard:
1492                 args->ar_discard = result.boolean;
1493                 break;
1494         case Opt_commit:
1495                 if (result.int_32 <= 0)
1496                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1497                 args->ar_commit = result.int_32;
1498                 break;
1499         case Opt_statfs_quantum:
1500                 if (result.int_32 < 0)
1501                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1502                 args->ar_statfs_quantum = result.int_32;
1503                 break;
1504         case Opt_quota_quantum:
1505                 if (result.int_32 <= 0)
1506                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1507                 args->ar_quota_quantum = result.int_32;
1508                 break;
1509         case Opt_statfs_percent:
1510                 if (result.int_32 < 0 || result.int_32 > 100)
1511                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1512                 args->ar_statfs_percent = result.int_32;
1513                 break;
1514         case Opt_errors:
1515                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1516                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1517                 args->ar_errors = result.uint_32;
1518                 break;
1519         case Opt_barrier:
1520                 args->ar_nobarrier = result.boolean;
1521                 break;
1522         case Opt_rgrplvb:
1523                 args->ar_rgrplvb = result.boolean;
1524                 args->ar_got_rgrplvb = 1;
1525                 break;
1526         case Opt_loccookie:
1527                 args->ar_loccookie = result.boolean;
1528                 break;
1529         default:
1530                 return invalfc(fc, "invalid mount option: %s", param->key);
1531         }
1532         return 0;
1533 }
1534
1535 static int gfs2_reconfigure(struct fs_context *fc)
1536 {
1537         struct super_block *sb = fc->root->d_sb;
1538         struct gfs2_sbd *sdp = sb->s_fs_info;
1539         struct gfs2_args *oldargs = &sdp->sd_args;
1540         struct gfs2_args *newargs = fc->fs_private;
1541         struct gfs2_tune *gt = &sdp->sd_tune;
1542         int error = 0;
1543
1544         sync_filesystem(sb);
1545
1546         spin_lock(&gt->gt_spin);
1547         oldargs->ar_commit = gt->gt_logd_secs;
1548         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1549         if (gt->gt_statfs_slow)
1550                 oldargs->ar_statfs_quantum = 0;
1551         else
1552                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1553         spin_unlock(&gt->gt_spin);
1554
1555         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1556                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1557                 return -EINVAL;
1558         }
1559         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1560                 errorfc(fc, "reconfiguration of lock table not allowed");
1561                 return -EINVAL;
1562         }
1563         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1564                 errorfc(fc, "reconfiguration of host data not allowed");
1565                 return -EINVAL;
1566         }
1567         if (newargs->ar_spectator != oldargs->ar_spectator) {
1568                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1569                 return -EINVAL;
1570         }
1571         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1572                 errorfc(fc, "reconfiguration of localflocks not allowed");
1573                 return -EINVAL;
1574         }
1575         if (newargs->ar_meta != oldargs->ar_meta) {
1576                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1577                 return -EINVAL;
1578         }
1579         if (oldargs->ar_spectator)
1580                 fc->sb_flags |= SB_RDONLY;
1581
1582         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1583                 struct gfs2_holder freeze_gh;
1584
1585                 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1586                 if (error)
1587                         return -EINVAL;
1588
1589                 if (fc->sb_flags & SB_RDONLY) {
1590                         gfs2_make_fs_ro(sdp);
1591                 } else {
1592                         error = gfs2_make_fs_rw(sdp);
1593                         if (error)
1594                                 errorfc(fc, "unable to remount read-write");
1595                 }
1596                 gfs2_freeze_unlock(&freeze_gh);
1597         }
1598         sdp->sd_args = *newargs;
1599
1600         if (sdp->sd_args.ar_posix_acl)
1601                 sb->s_flags |= SB_POSIXACL;
1602         else
1603                 sb->s_flags &= ~SB_POSIXACL;
1604         if (sdp->sd_args.ar_nobarrier)
1605                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1606         else
1607                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1608         spin_lock(&gt->gt_spin);
1609         gt->gt_logd_secs = newargs->ar_commit;
1610         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1611         if (newargs->ar_statfs_quantum) {
1612                 gt->gt_statfs_slow = 0;
1613                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1614         }
1615         else {
1616                 gt->gt_statfs_slow = 1;
1617                 gt->gt_statfs_quantum = 30;
1618         }
1619         spin_unlock(&gt->gt_spin);
1620
1621         gfs2_online_uevent(sdp);
1622         return error;
1623 }
1624
1625 static const struct fs_context_operations gfs2_context_ops = {
1626         .free        = gfs2_fc_free,
1627         .parse_param = gfs2_parse_param,
1628         .get_tree    = gfs2_get_tree,
1629         .reconfigure = gfs2_reconfigure,
1630 };
1631
1632 /* Set up the filesystem mount context */
1633 static int gfs2_init_fs_context(struct fs_context *fc)
1634 {
1635         struct gfs2_args *args;
1636
1637         args = kmalloc(sizeof(*args), GFP_KERNEL);
1638         if (args == NULL)
1639                 return -ENOMEM;
1640
1641         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1642                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1643
1644                 *args = sdp->sd_args;
1645         } else {
1646                 memset(args, 0, sizeof(*args));
1647                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1648                 args->ar_data = GFS2_DATA_DEFAULT;
1649                 args->ar_commit = 30;
1650                 args->ar_statfs_quantum = 30;
1651                 args->ar_quota_quantum = 60;
1652                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1653         }
1654         fc->fs_private = args;
1655         fc->ops = &gfs2_context_ops;
1656         return 0;
1657 }
1658
1659 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1660 {
1661         return -EINVAL;
1662 }
1663
1664 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1665 {
1666         return (fc->sget_key == s->s_bdev);
1667 }
1668
1669 static int gfs2_meta_get_tree(struct fs_context *fc)
1670 {
1671         struct super_block *s;
1672         struct gfs2_sbd *sdp;
1673         struct path path;
1674         int error;
1675
1676         if (!fc->source || !*fc->source)
1677                 return -EINVAL;
1678
1679         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1680         if (error) {
1681                 pr_warn("path_lookup on %s returned error %d\n",
1682                         fc->source, error);
1683                 return error;
1684         }
1685         fc->fs_type = &gfs2_fs_type;
1686         fc->sget_key = path.dentry->d_sb->s_bdev;
1687         s = sget_fc(fc, test_meta_super, set_meta_super);
1688         path_put(&path);
1689         if (IS_ERR(s)) {
1690                 pr_warn("gfs2 mount does not exist\n");
1691                 return PTR_ERR(s);
1692         }
1693         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1694                 deactivate_locked_super(s);
1695                 return -EBUSY;
1696         }
1697         sdp = s->s_fs_info;
1698         fc->root = dget(sdp->sd_master_dir);
1699         return 0;
1700 }
1701
1702 static const struct fs_context_operations gfs2_meta_context_ops = {
1703         .free        = gfs2_fc_free,
1704         .get_tree    = gfs2_meta_get_tree,
1705 };
1706
1707 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1708 {
1709         int ret = gfs2_init_fs_context(fc);
1710
1711         if (ret)
1712                 return ret;
1713
1714         fc->ops = &gfs2_meta_context_ops;
1715         return 0;
1716 }
1717
1718 static void gfs2_kill_sb(struct super_block *sb)
1719 {
1720         struct gfs2_sbd *sdp = sb->s_fs_info;
1721
1722         if (sdp == NULL) {
1723                 kill_block_super(sb);
1724                 return;
1725         }
1726
1727         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1728         dput(sdp->sd_root_dir);
1729         dput(sdp->sd_master_dir);
1730         sdp->sd_root_dir = NULL;
1731         sdp->sd_master_dir = NULL;
1732         shrink_dcache_sb(sb);
1733         kill_block_super(sb);
1734 }
1735
1736 struct file_system_type gfs2_fs_type = {
1737         .name = "gfs2",
1738         .fs_flags = FS_REQUIRES_DEV,
1739         .init_fs_context = gfs2_init_fs_context,
1740         .parameters = gfs2_fs_parameters,
1741         .kill_sb = gfs2_kill_sb,
1742         .owner = THIS_MODULE,
1743 };
1744 MODULE_ALIAS_FS("gfs2");
1745
1746 struct file_system_type gfs2meta_fs_type = {
1747         .name = "gfs2meta",
1748         .fs_flags = FS_REQUIRES_DEV,
1749         .init_fs_context = gfs2_meta_init_fs_context,
1750         .owner = THIS_MODULE,
1751 };
1752 MODULE_ALIAS_FS("gfs2meta");