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(GFP_NOFS, 1);
 255         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
 256         bio_set_dev(bio, sb->s_bdev);
 257         bio_add_page(bio, page, PAGE_SIZE, 0);
 258
 259         bio->bi_end_io = end_bio_io_page;
 260         bio->bi_private = page;
 261         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
 262         submit_bio(bio);
 263         wait_on_page_locked(page);
 264         bio_put(bio);
 265         if (!PageUptodate(page)) {
 266                 __free_page(page);
 267                 return -EIO;
 268         }
 269         p = kmap(page);
 270         gfs2_sb_in(sdp, p);
 271         kunmap(page);
 272         __free_page(page);
 273         return gfs2_check_sb(sdp, silent);
 274 }
 275
 276 /**
 277  * gfs2_read_sb - Read super block
 278  * @sdp: The GFS2 superblock
 279  * @silent: Don't print message if mount fails
 280  *
 281  */
 282
 283 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 284 {
 285         u32 hash_blocks, ind_blocks, leaf_blocks;
 286         u32 tmp_blocks;
 287         unsigned int x;
 288         int error;
 289
 290         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 291         if (error) {
 292                 if (!silent)
 293                         fs_err(sdp, "can't read superblock\n");
 294                 return error;
 295         }
 296
 297         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 298                                GFS2_BASIC_BLOCK_SHIFT;
 299         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
 300         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 301                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 302         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 303                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 304         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
 305                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
 306         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 307         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 308         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 309         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 310         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 311                                 sizeof(struct gfs2_meta_header)) /
 312                                 sizeof(struct gfs2_quota_change);
 313         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
 314                                      sizeof(struct gfs2_meta_header))
 315                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
 316
 317         /*
 318          * We always keep at least one block reserved for revokes in
 319          * transactions.  This greatly simplifies allocating additional
 320          * revoke blocks.
 321          */
 322         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
 323
 324         /* Compute maximum reservation required to add a entry to a directory */
 325
 326         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
 327                              sdp->sd_jbsize);
 328
 329         ind_blocks = 0;
 330         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 331                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 332                 ind_blocks += tmp_blocks;
 333         }
 334
 335         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 336
 337         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 338
 339         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 340                                 sizeof(struct gfs2_dinode);
 341         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 342         for (x = 2;; x++) {
 343                 u64 space, d;
 344                 u32 m;
 345
 346                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 347                 d = space;
 348                 m = do_div(d, sdp->sd_inptrs);
 349
 350                 if (d != sdp->sd_heightsize[x - 1] || m)
 351                         break;
 352                 sdp->sd_heightsize[x] = space;
 353         }
 354         sdp->sd_max_height = x;
 355         sdp->sd_heightsize[x] = ~0;
 356         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 357
 358         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
 359                                       sizeof(struct gfs2_leaf)) /
 360                                      GFS2_MIN_DIRENT_SIZE;
 361         return 0;
 362 }
 363
 364 static int init_names(struct gfs2_sbd *sdp, int silent)
 365 {
 366         char *proto, *table;
 367         int error = 0;
 368
 369         proto = sdp->sd_args.ar_lockproto;
 370         table = sdp->sd_args.ar_locktable;
 371
 372         /*  Try to autodetect  */
 373
 374         if (!proto[0] || !table[0]) {
 375                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
 376                 if (error)
 377                         return error;
 378
 379                 if (!proto[0])
 380                         proto = sdp->sd_sb.sb_lockproto;
 381                 if (!table[0])
 382                         table = sdp->sd_sb.sb_locktable;
 383         }
 384
 385         if (!table[0])
 386                 table = sdp->sd_vfs->s_id;
 387
 388         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
 389         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
 390
 391         table = sdp->sd_table_name;
 392         while ((table = strchr(table, '/')))
 393                 *table = '_';
 394
 395         return error;
 396 }
 397
 398 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
 399                         int undo)
 400 {
 401         int error = 0;
 402
 403         if (undo)
 404                 goto fail_trans;
 405
 406         error = gfs2_glock_nq_num(sdp,
 407                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
 408                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
 409                                   mount_gh);
 410         if (error) {
 411                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
 412                 goto fail;
 413         }
 414
 415         error = gfs2_glock_nq_num(sdp,
 416                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
 417                                   LM_ST_SHARED,
 418                                   LM_FLAG_NOEXP | GL_EXACT,
 419                                   &sdp->sd_live_gh);
 420         if (error) {
 421                 fs_err(sdp, "can't acquire live glock: %d\n", error);
 422                 goto fail_mount;
 423         }
 424
 425         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
 426                                CREATE, &sdp->sd_rename_gl);
 427         if (error) {
 428                 fs_err(sdp, "can't create rename glock: %d\n", error);
 429                 goto fail_live;
 430         }
 431
 432         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
 433                                CREATE, &sdp->sd_freeze_gl);
 434         if (error) {
 435                 fs_err(sdp, "can't create transaction glock: %d\n", error);
 436                 goto fail_rename;
 437         }
 438
 439         return 0;
 440
 441 fail_trans:
 442         gfs2_glock_put(sdp->sd_freeze_gl);
 443 fail_rename:
 444         gfs2_glock_put(sdp->sd_rename_gl);
 445 fail_live:
 446         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
 447 fail_mount:
 448         gfs2_glock_dq_uninit(mount_gh);
 449 fail:
 450         return error;
 451 }
 452
 453 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
 454                             u64 no_addr, const char *name)
 455 {
 456         struct gfs2_sbd *sdp = sb->s_fs_info;
 457         struct dentry *dentry;
 458         struct inode *inode;
 459
 460         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
 461                                   GFS2_BLKST_FREE /* ignore */);
 462         if (IS_ERR(inode)) {
 463                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
 464                 return PTR_ERR(inode);
 465         }
 466         dentry = d_make_root(inode);
 467         if (!dentry) {
 468                 fs_err(sdp, "can't alloc %s dentry\n", name);
 469                 return -ENOMEM;
 470         }
 471         *dptr = dentry;
 472         return 0;
 473 }
 474
 475 static int init_sb(struct gfs2_sbd *sdp, int silent)
 476 {
 477         struct super_block *sb = sdp->sd_vfs;
 478         struct gfs2_holder sb_gh;
 479         u64 no_addr;
 480         int ret;
 481
 482         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
 483                                 LM_ST_SHARED, 0, &sb_gh);
 484         if (ret) {
 485                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
 486                 return ret;
 487         }
 488
 489         ret = gfs2_read_sb(sdp, silent);
 490         if (ret) {
 491                 fs_err(sdp, "can't read superblock: %d\n", ret);
 492                 goto out;
 493         }
 494
 495         switch(sdp->sd_sb.sb_fs_format) {
 496         case GFS2_FS_FORMAT_MAX:
 497                 sb->s_xattr = gfs2_xattr_handlers_max;
 498                 break;
 499
 500         case GFS2_FS_FORMAT_MIN:
 501                 sb->s_xattr = gfs2_xattr_handlers_min;
 502                 break;
 503
 504         default:
 505                 BUG();
 506         }
 507
 508         /* Set up the buffer cache and SB for real */
 509         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
 510                 ret = -EINVAL;
 511                 fs_err(sdp, "FS block size (%u) is too small for device "
 512                        "block size (%u)\n",
 513                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
 514                 goto out;
 515         }
 516         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
 517                 ret = -EINVAL;
 518                 fs_err(sdp, "FS block size (%u) is too big for machine "
 519                        "page size (%u)\n",
 520                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
 521                 goto out;
 522         }
 523         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
 524
 525         /* Get the root inode */
 526         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
 527         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
 528         if (ret)
 529                 goto out;
 530
 531         /* Get the master inode */
 532         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
 533         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
 534         if (ret) {
 535                 dput(sdp->sd_root_dir);
 536                 goto out;
 537         }
 538         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
 539 out:
 540         gfs2_glock_dq_uninit(&sb_gh);
 541         return ret;
 542 }
 543
 544 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 545 {
 546         char *message = "FIRSTMOUNT=Done";
 547         char *envp[] = { message, NULL };
 548
 549         fs_info(sdp, "first mount done, others may mount\n");
 550
 551         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
 552                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
 553
 554         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 555 }
 556
 557 /**
 558  * gfs2_jindex_hold - Grab a lock on the jindex
 559  * @sdp: The GFS2 superblock
 560  * @ji_gh: the holder for the jindex glock
 561  *
 562  * Returns: errno
 563  */
 564
 565 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 566 {
 567         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 568         struct qstr name;
 569         char buf[20];
 570         struct gfs2_jdesc *jd;
 571         int error;
 572
 573         name.name = buf;
 574
 575         mutex_lock(&sdp->sd_jindex_mutex);
 576
 577         for (;;) {
 578                 struct gfs2_inode *jip;
 579
 580                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 581                 if (error)
 582                         break;
 583
 584                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 585                 name.hash = gfs2_disk_hash(name.name, name.len);
 586
 587                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 588                 if (error == -ENOENT) {
 589                         error = 0;
 590                         break;
 591                 }
 592
 593                 gfs2_glock_dq_uninit(ji_gh);
 594
 595                 if (error)
 596                         break;
 597
 598                 error = -ENOMEM;
 599                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 600                 if (!jd)
 601                         break;
 602
 603                 INIT_LIST_HEAD(&jd->extent_list);
 604                 INIT_LIST_HEAD(&jd->jd_revoke_list);
 605
 606                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
 607                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
 608                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
 609                         if (!jd->jd_inode)
 610                                 error = -ENOENT;
 611                         else
 612                                 error = PTR_ERR(jd->jd_inode);
 613                         kfree(jd);
 614                         break;
 615                 }
 616
 617                 d_mark_dontcache(jd->jd_inode);
 618                 spin_lock(&sdp->sd_jindex_spin);
 619                 jd->jd_jid = sdp->sd_journals++;
 620                 jip = GFS2_I(jd->jd_inode);
 621                 jd->jd_no_addr = jip->i_no_addr;
 622                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 623                 spin_unlock(&sdp->sd_jindex_spin);
 624         }
 625
 626         mutex_unlock(&sdp->sd_jindex_mutex);
 627
 628         return error;
 629 }
 630
 631 /**
 632  * init_statfs - look up and initialize master and local (per node) statfs inodes
 633  * @sdp: The GFS2 superblock
 634  *
 635  * This should be called after the jindex is initialized in init_journal() and
 636  * before gfs2_journal_recovery() is called because we need to be able to write
 637  * to these inodes during recovery.
 638  *
 639  * Returns: errno
 640  */
 641 static int init_statfs(struct gfs2_sbd *sdp)
 642 {
 643         int error = 0;
 644         struct inode *master = d_inode(sdp->sd_master_dir);
 645         struct inode *pn = NULL;
 646         char buf[30];
 647         struct gfs2_jdesc *jd;
 648         struct gfs2_inode *ip;
 649
 650         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
 651         if (IS_ERR(sdp->sd_statfs_inode)) {
 652                 error = PTR_ERR(sdp->sd_statfs_inode);
 653                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
 654                 goto out;
 655         }
 656         if (sdp->sd_args.ar_spectator)
 657                 goto out;
 658
 659         pn = gfs2_lookup_simple(master, "per_node");
 660         if (IS_ERR(pn)) {
 661                 error = PTR_ERR(pn);
 662                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 663                 goto put_statfs;
 664         }
 665
 666         /* For each jid, lookup the corresponding local statfs inode in the
 667          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
 668         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 669                 struct local_statfs_inode *lsi =
 670                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
 671                 if (!lsi) {
 672                         error = -ENOMEM;
 673                         goto free_local;
 674                 }
 675                 sprintf(buf, "statfs_change%u", jd->jd_jid);
 676                 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
 677                 if (IS_ERR(lsi->si_sc_inode)) {
 678                         error = PTR_ERR(lsi->si_sc_inode);
 679                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
 680                                jd->jd_jid, error);
 681                         kfree(lsi);
 682                         goto free_local;
 683                 }
 684                 lsi->si_jid = jd->jd_jid;
 685                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
 686                         sdp->sd_sc_inode = lsi->si_sc_inode;
 687
 688                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
 689         }
 690
 691         iput(pn);
 692         pn = NULL;
 693         ip = GFS2_I(sdp->sd_sc_inode);
 694         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 695                                    &sdp->sd_sc_gh);
 696         if (error) {
 697                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
 698                 goto free_local;
 699         }
 700         /* read in the local statfs buffer - other nodes don't change it. */
 701         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
 702         if (error) {
 703                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
 704                 goto unlock_sd_gh;
 705         }
 706         return 0;
 707
 708 unlock_sd_gh:
 709         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 710 free_local:
 711         free_local_statfs_inodes(sdp);
 712         iput(pn);
 713 put_statfs:
 714         iput(sdp->sd_statfs_inode);
 715 out:
 716         return error;
 717 }
 718
 719 /* Uninitialize and free up memory used by the list of statfs inodes */
 720 static void uninit_statfs(struct gfs2_sbd *sdp)
 721 {
 722         if (!sdp->sd_args.ar_spectator) {
 723                 brelse(sdp->sd_sc_bh);
 724                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
 725                 free_local_statfs_inodes(sdp);
 726         }
 727         iput(sdp->sd_statfs_inode);
 728 }
 729
 730 static int init_journal(struct gfs2_sbd *sdp, int undo)
 731 {
 732         struct inode *master = d_inode(sdp->sd_master_dir);
 733         struct gfs2_holder ji_gh;
 734         struct gfs2_inode *ip;
 735         int jindex = 1;
 736         int error = 0;
 737
 738         if (undo) {
 739                 jindex = 0;
 740                 goto fail_statfs;
 741         }
 742
 743         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
 744         if (IS_ERR(sdp->sd_jindex)) {
 745                 fs_err(sdp, "can't lookup journal index: %d\n", error);
 746                 return PTR_ERR(sdp->sd_jindex);
 747         }
 748
 749         /* Load in the journal index special file */
 750
 751         error = gfs2_jindex_hold(sdp, &ji_gh);
 752         if (error) {
 753                 fs_err(sdp, "can't read journal index: %d\n", error);
 754                 goto fail;
 755         }
 756
 757         error = -EUSERS;
 758         if (!gfs2_jindex_size(sdp)) {
 759                 fs_err(sdp, "no journals!\n");
 760                 goto fail_jindex;
 761         }
 762
 763         atomic_set(&sdp->sd_log_blks_needed, 0);
 764         if (sdp->sd_args.ar_spectator) {
 765                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
 766                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 767                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 768                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 769         } else {
 770                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
 771                         fs_err(sdp, "can't mount journal #%u\n",
 772                                sdp->sd_lockstruct.ls_jid);
 773                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
 774                                gfs2_jindex_size(sdp),
 775                                gfs2_jindex_size(sdp) - 1);
 776                         goto fail_jindex;
 777                 }
 778                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
 779
 780                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
 781                                           &gfs2_journal_glops,
 782                                           LM_ST_EXCLUSIVE,
 783                                           LM_FLAG_NOEXP | GL_NOCACHE,
 784                                           &sdp->sd_journal_gh);
 785                 if (error) {
 786                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
 787                         goto fail_jindex;
 788                 }
 789
 790                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 791                 sdp->sd_jinode_gl = ip->i_gl;
 792                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
 793                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
 794                                            &sdp->sd_jinode_gh);
 795                 if (error) {
 796                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
 797                                error);
 798                         goto fail_journal_gh;
 799                 }
 800
 801                 error = gfs2_jdesc_check(sdp->sd_jdesc);
 802                 if (error) {
 803                         fs_err(sdp, "my journal (%u) is bad: %d\n",
 804                                sdp->sd_jdesc->jd_jid, error);
 805                         goto fail_jinode_gh;
 806                 }
 807                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
 808                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
 809                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
 810
 811                 /* Map the extents for this journal's blocks */
 812                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
 813         }
 814         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 815
 816         /* Lookup statfs inodes here so journal recovery can use them. */
 817         error = init_statfs(sdp);
 818         if (error)
 819                 goto fail_jinode_gh;
 820
 821         if (sdp->sd_lockstruct.ls_first) {
 822                 unsigned int x;
 823                 for (x = 0; x < sdp->sd_journals; x++) {
 824                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
 825
 826                         if (sdp->sd_args.ar_spectator) {
 827                                 error = check_journal_clean(sdp, jd, true);
 828                                 if (error)
 829                                         goto fail_statfs;
 830                                 continue;
 831                         }
 832                         error = gfs2_recover_journal(jd, true);
 833                         if (error) {
 834                                 fs_err(sdp, "error recovering journal %u: %d\n",
 835                                        x, error);
 836                                 goto fail_statfs;
 837                         }
 838                 }
 839
 840                 gfs2_others_may_mount(sdp);
 841         } else if (!sdp->sd_args.ar_spectator) {
 842                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
 843                 if (error) {
 844                         fs_err(sdp, "error recovering my journal: %d\n", error);
 845                         goto fail_statfs;
 846                 }
 847         }
 848
 849         sdp->sd_log_idle = 1;
 850         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
 851         gfs2_glock_dq_uninit(&ji_gh);
 852         jindex = 0;
 853         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
 854         return 0;
 855
 856 fail_statfs:
 857         uninit_statfs(sdp);
 858 fail_jinode_gh:
 859         /* A withdraw may have done dq/uninit so now we need to check it */
 860         if (!sdp->sd_args.ar_spectator &&
 861             gfs2_holder_initialized(&sdp->sd_jinode_gh))
 862                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
 863 fail_journal_gh:
 864         if (!sdp->sd_args.ar_spectator &&
 865             gfs2_holder_initialized(&sdp->sd_journal_gh))
 866                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
 867 fail_jindex:
 868         gfs2_jindex_free(sdp);
 869         if (jindex)
 870                 gfs2_glock_dq_uninit(&ji_gh);
 871 fail:
 872         iput(sdp->sd_jindex);
 873         return error;
 874 }
 875
 876 static struct lock_class_key gfs2_quota_imutex_key;
 877
 878 static int init_inodes(struct gfs2_sbd *sdp, int undo)
 879 {
 880         int error = 0;
 881         struct inode *master = d_inode(sdp->sd_master_dir);
 882
 883         if (undo)
 884                 goto fail_qinode;
 885
 886         error = init_journal(sdp, undo);
 887         complete_all(&sdp->sd_journal_ready);
 888         if (error)
 889                 goto fail;
 890
 891         /* Read in the resource index inode */
 892         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
 893         if (IS_ERR(sdp->sd_rindex)) {
 894                 error = PTR_ERR(sdp->sd_rindex);
 895                 fs_err(sdp, "can't get resource index inode: %d\n", error);
 896                 goto fail_journal;
 897         }
 898         sdp->sd_rindex_uptodate = 0;
 899
 900         /* Read in the quota inode */
 901         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
 902         if (IS_ERR(sdp->sd_quota_inode)) {
 903                 error = PTR_ERR(sdp->sd_quota_inode);
 904                 fs_err(sdp, "can't get quota file inode: %d\n", error);
 905                 goto fail_rindex;
 906         }
 907         /*
 908          * i_rwsem on quota files is special. Since this inode is hidden system
 909          * file, we are safe to define locking ourselves.
 910          */
 911         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
 912                           &gfs2_quota_imutex_key);
 913
 914         error = gfs2_rindex_update(sdp);
 915         if (error)
 916                 goto fail_qinode;
 917
 918         return 0;
 919
 920 fail_qinode:
 921         iput(sdp->sd_quota_inode);
 922 fail_rindex:
 923         gfs2_clear_rgrpd(sdp);
 924         iput(sdp->sd_rindex);
 925 fail_journal:
 926         init_journal(sdp, UNDO);
 927 fail:
 928         return error;
 929 }
 930
 931 static int init_per_node(struct gfs2_sbd *sdp, int undo)
 932 {
 933         struct inode *pn = NULL;
 934         char buf[30];
 935         int error = 0;
 936         struct gfs2_inode *ip;
 937         struct inode *master = d_inode(sdp->sd_master_dir);
 938
 939         if (sdp->sd_args.ar_spectator)
 940                 return 0;
 941
 942         if (undo)
 943                 goto fail_qc_gh;
 944
 945         pn = gfs2_lookup_simple(master, "per_node");
 946         if (IS_ERR(pn)) {
 947                 error = PTR_ERR(pn);
 948                 fs_err(sdp, "can't find per_node directory: %d\n", error);
 949                 return error;
 950         }
 951
 952         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
 953         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
 954         if (IS_ERR(sdp->sd_qc_inode)) {
 955                 error = PTR_ERR(sdp->sd_qc_inode);
 956                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
 957                 goto fail_ut_i;
 958         }
 959
 960         iput(pn);
 961         pn = NULL;
 962
 963         ip = GFS2_I(sdp->sd_qc_inode);
 964         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
 965                                    &sdp->sd_qc_gh);
 966         if (error) {
 967                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
 968                 goto fail_qc_i;
 969         }
 970
 971         return 0;
 972
 973 fail_qc_gh:
 974         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 975 fail_qc_i:
 976         iput(sdp->sd_qc_inode);
 977 fail_ut_i:
 978         iput(pn);
 979         return error;
 980 }
 981
 982 static const match_table_t nolock_tokens = {
 983         { Opt_jid, "jid=%d", },
 984         { Opt_err, NULL },
 985 };
 986
 987 static const struct lm_lockops nolock_ops = {
 988         .lm_proto_name = "lock_nolock",
 989         .lm_put_lock = gfs2_glock_free,
 990         .lm_tokens = &nolock_tokens,
 991 };
 992
 993 /**
 994  * gfs2_lm_mount - mount a locking protocol
 995  * @sdp: the filesystem
 996  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
 997  *
 998  * Returns: errno
 999  */
1000
1001 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1002 {
1003         const struct lm_lockops *lm;
1004         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1005         struct gfs2_args *args = &sdp->sd_args;
1006         const char *proto = sdp->sd_proto_name;
1007         const char *table = sdp->sd_table_name;
1008         char *o, *options;
1009         int ret;
1010
1011         if (!strcmp("lock_nolock", proto)) {
1012                 lm = &nolock_ops;
1013                 sdp->sd_args.ar_localflocks = 1;
1014 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1015         } else if (!strcmp("lock_dlm", proto)) {
1016                 lm = &gfs2_dlm_ops;
1017 #endif
1018         } else {
1019                 pr_info("can't find protocol %s\n", proto);
1020                 return -ENOENT;
1021         }
1022
1023         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1024
1025         ls->ls_ops = lm;
1026         ls->ls_first = 1;
1027
1028         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1029                 substring_t tmp[MAX_OPT_ARGS];
1030                 int token, option;
1031
1032                 if (!o || !*o)
1033                         continue;
1034
1035                 token = match_token(o, *lm->lm_tokens, tmp);
1036                 switch (token) {
1037                 case Opt_jid:
1038                         ret = match_int(&tmp[0], &option);
1039                         if (ret || option < 0)
1040                                 goto hostdata_error;
1041                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1042                                 ls->ls_jid = option;
1043                         break;
1044                 case Opt_id:
1045                 case Opt_nodir:
1046                         /* Obsolete, but left for backward compat purposes */
1047                         break;
1048                 case Opt_first:
1049                         ret = match_int(&tmp[0], &option);
1050                         if (ret || (option != 0 && option != 1))
1051                                 goto hostdata_error;
1052                         ls->ls_first = option;
1053                         break;
1054                 case Opt_err:
1055                 default:
1056 hostdata_error:
1057                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1058                         return -EINVAL;
1059                 }
1060         }
1061
1062         if (lm->lm_mount == NULL) {
1063                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1064                 complete_all(&sdp->sd_locking_init);
1065                 return 0;
1066         }
1067         ret = lm->lm_mount(sdp, table);
1068         if (ret == 0)
1069                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1070                         sdp->sd_sb.sb_fs_format);
1071         complete_all(&sdp->sd_locking_init);
1072         return ret;
1073 }
1074
1075 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1076 {
1077         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1078         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1079                 lm->lm_unmount(sdp);
1080 }
1081
1082 static int wait_on_journal(struct gfs2_sbd *sdp)
1083 {
1084         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1085                 return 0;
1086
1087         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1088                 ? -EINTR : 0;
1089 }
1090
1091 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1092 {
1093         struct super_block *sb = sdp->sd_vfs;
1094         char ro[20];
1095         char spectator[20];
1096         char *envp[] = { ro, spectator, NULL };
1097         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1098         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1099         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1100 }
1101
1102 static int init_threads(struct gfs2_sbd *sdp)
1103 {
1104         struct task_struct *p;
1105         int error = 0;
1106
1107         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1108         if (IS_ERR(p)) {
1109                 error = PTR_ERR(p);
1110                 fs_err(sdp, "can't start logd thread: %d\n", error);
1111                 return error;
1112         }
1113         sdp->sd_logd_process = p;
1114
1115         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1116         if (IS_ERR(p)) {
1117                 error = PTR_ERR(p);
1118                 fs_err(sdp, "can't start quotad thread: %d\n", error);
1119                 goto fail;
1120         }
1121         sdp->sd_quotad_process = p;
1122         return 0;
1123
1124 fail:
1125         kthread_stop(sdp->sd_logd_process);
1126         sdp->sd_logd_process = NULL;
1127         return error;
1128 }
1129
1130 /**
1131  * gfs2_fill_super - Read in superblock
1132  * @sb: The VFS superblock
1133  * @fc: Mount options and flags
1134  *
1135  * Returns: -errno
1136  */
1137 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1138 {
1139         struct gfs2_args *args = fc->fs_private;
1140         int silent = fc->sb_flags & SB_SILENT;
1141         struct gfs2_sbd *sdp;
1142         struct gfs2_holder mount_gh;
1143         struct gfs2_holder freeze_gh;
1144         int error;
1145
1146         sdp = init_sbd(sb);
1147         if (!sdp) {
1148                 pr_warn("can't alloc struct gfs2_sbd\n");
1149                 return -ENOMEM;
1150         }
1151         sdp->sd_args = *args;
1152
1153         if (sdp->sd_args.ar_spectator) {
1154                 sb->s_flags |= SB_RDONLY;
1155                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1156         }
1157         if (sdp->sd_args.ar_posix_acl)
1158                 sb->s_flags |= SB_POSIXACL;
1159         if (sdp->sd_args.ar_nobarrier)
1160                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1161
1162         sb->s_flags |= SB_NOSEC;
1163         sb->s_magic = GFS2_MAGIC;
1164         sb->s_op = &gfs2_super_ops;
1165         sb->s_d_op = &gfs2_dops;
1166         sb->s_export_op = &gfs2_export_ops;
1167         sb->s_qcop = &gfs2_quotactl_ops;
1168         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1169         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1170         sb->s_time_gran = 1;
1171         sb->s_maxbytes = MAX_LFS_FILESIZE;
1172
1173         /* Set up the buffer cache and fill in some fake block size values
1174            to allow us to read-in the on-disk superblock. */
1175         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1176         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1177         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1178                                GFS2_BASIC_BLOCK_SHIFT;
1179         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1180
1181         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1182         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1183         if (sdp->sd_args.ar_statfs_quantum) {
1184                 sdp->sd_tune.gt_statfs_slow = 0;
1185                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1186         } else {
1187                 sdp->sd_tune.gt_statfs_slow = 1;
1188                 sdp->sd_tune.gt_statfs_quantum = 30;
1189         }
1190
1191         error = init_names(sdp, silent);
1192         if (error)
1193                 goto fail_free;
1194
1195         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1196
1197         error = gfs2_sys_fs_add(sdp);
1198         if (error)
1199                 goto fail_free;
1200
1201         gfs2_create_debugfs_file(sdp);
1202
1203         error = gfs2_lm_mount(sdp, silent);
1204         if (error)
1205                 goto fail_debug;
1206
1207         error = init_locking(sdp, &mount_gh, DO);
1208         if (error)
1209                 goto fail_lm;
1210
1211         error = init_sb(sdp, silent);
1212         if (error)
1213                 goto fail_locking;
1214
1215         /* Turn rgrplvb on by default if fs format is recent enough */
1216         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1217                 sdp->sd_args.ar_rgrplvb = 1;
1218
1219         error = wait_on_journal(sdp);
1220         if (error)
1221                 goto fail_sb;
1222
1223         /*
1224          * If user space has failed to join the cluster or some similar
1225          * failure has occurred, then the journal id will contain a
1226          * negative (error) number. This will then be returned to the
1227          * caller (of the mount syscall). We do this even for spectator
1228          * mounts (which just write a jid of 0 to indicate "ok" even though
1229          * the jid is unused in the spectator case)
1230          */
1231         if (sdp->sd_lockstruct.ls_jid < 0) {
1232                 error = sdp->sd_lockstruct.ls_jid;
1233                 sdp->sd_lockstruct.ls_jid = 0;
1234                 goto fail_sb;
1235         }
1236
1237         if (sdp->sd_args.ar_spectator)
1238                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1239                          sdp->sd_table_name);
1240         else
1241                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1242                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1243
1244         error = init_inodes(sdp, DO);
1245         if (error)
1246                 goto fail_sb;
1247
1248         error = init_per_node(sdp, DO);
1249         if (error)
1250                 goto fail_inodes;
1251
1252         error = gfs2_statfs_init(sdp);
1253         if (error) {
1254                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1255                 goto fail_per_node;
1256         }
1257
1258         if (!sb_rdonly(sb)) {
1259                 error = init_threads(sdp);
1260                 if (error) {
1261                         gfs2_withdraw_delayed(sdp);
1262                         goto fail_per_node;
1263                 }
1264         }
1265
1266         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1267         if (error)
1268                 goto fail_per_node;
1269
1270         if (!sb_rdonly(sb))
1271                 error = gfs2_make_fs_rw(sdp);
1272
1273         gfs2_freeze_unlock(&freeze_gh);
1274         if (error) {
1275                 if (sdp->sd_quotad_process)
1276                         kthread_stop(sdp->sd_quotad_process);
1277                 sdp->sd_quotad_process = NULL;
1278                 if (sdp->sd_logd_process)
1279                         kthread_stop(sdp->sd_logd_process);
1280                 sdp->sd_logd_process = NULL;
1281                 fs_err(sdp, "can't make FS RW: %d\n", error);
1282                 goto fail_per_node;
1283         }
1284         gfs2_glock_dq_uninit(&mount_gh);
1285         gfs2_online_uevent(sdp);
1286         return 0;
1287
1288 fail_per_node:
1289         init_per_node(sdp, UNDO);
1290 fail_inodes:
1291         init_inodes(sdp, UNDO);
1292 fail_sb:
1293         if (sdp->sd_root_dir)
1294                 dput(sdp->sd_root_dir);
1295         if (sdp->sd_master_dir)
1296                 dput(sdp->sd_master_dir);
1297         if (sb->s_root)
1298                 dput(sb->s_root);
1299         sb->s_root = NULL;
1300 fail_locking:
1301         init_locking(sdp, &mount_gh, UNDO);
1302 fail_lm:
1303         complete_all(&sdp->sd_journal_ready);
1304         gfs2_gl_hash_clear(sdp);
1305         gfs2_lm_unmount(sdp);
1306 fail_debug:
1307         gfs2_delete_debugfs_file(sdp);
1308         gfs2_sys_fs_del(sdp);
1309 fail_free:
1310         free_sbd(sdp);
1311         sb->s_fs_info = NULL;
1312         return error;
1313 }
1314
1315 /**
1316  * gfs2_get_tree - Get the GFS2 superblock and root directory
1317  * @fc: The filesystem context
1318  *
1319  * Returns: 0 or -errno on error
1320  */
1321 static int gfs2_get_tree(struct fs_context *fc)
1322 {
1323         struct gfs2_args *args = fc->fs_private;
1324         struct gfs2_sbd *sdp;
1325         int error;
1326
1327         error = get_tree_bdev(fc, gfs2_fill_super);
1328         if (error)
1329                 return error;
1330
1331         sdp = fc->root->d_sb->s_fs_info;
1332         dput(fc->root);
1333         if (args->ar_meta)
1334                 fc->root = dget(sdp->sd_master_dir);
1335         else
1336                 fc->root = dget(sdp->sd_root_dir);
1337         return 0;
1338 }
1339
1340 static void gfs2_fc_free(struct fs_context *fc)
1341 {
1342         struct gfs2_args *args = fc->fs_private;
1343
1344         kfree(args);
1345 }
1346
1347 enum gfs2_param {
1348         Opt_lockproto,
1349         Opt_locktable,
1350         Opt_hostdata,
1351         Opt_spectator,
1352         Opt_ignore_local_fs,
1353         Opt_localflocks,
1354         Opt_localcaching,
1355         Opt_debug,
1356         Opt_upgrade,
1357         Opt_acl,
1358         Opt_quota,
1359         Opt_quota_flag,
1360         Opt_suiddir,
1361         Opt_data,
1362         Opt_meta,
1363         Opt_discard,
1364         Opt_commit,
1365         Opt_errors,
1366         Opt_statfs_quantum,
1367         Opt_statfs_percent,
1368         Opt_quota_quantum,
1369         Opt_barrier,
1370         Opt_rgrplvb,
1371         Opt_loccookie,
1372 };
1373
1374 static const struct constant_table gfs2_param_quota[] = {
1375         {"off",        GFS2_QUOTA_OFF},
1376         {"account",    GFS2_QUOTA_ACCOUNT},
1377         {"on",         GFS2_QUOTA_ON},
1378         {}
1379 };
1380
1381 enum opt_data {
1382         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1383         Opt_data_ordered   = GFS2_DATA_ORDERED,
1384 };
1385
1386 static const struct constant_table gfs2_param_data[] = {
1387         {"writeback",  Opt_data_writeback },
1388         {"ordered",    Opt_data_ordered },
1389         {}
1390 };
1391
1392 enum opt_errors {
1393         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1394         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1395 };
1396
1397 static const struct constant_table gfs2_param_errors[] = {
1398         {"withdraw",   Opt_errors_withdraw },
1399         {"panic",      Opt_errors_panic },
1400         {}
1401 };
1402
1403 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1404         fsparam_string ("lockproto",          Opt_lockproto),
1405         fsparam_string ("locktable",          Opt_locktable),
1406         fsparam_string ("hostdata",           Opt_hostdata),
1407         fsparam_flag   ("spectator",          Opt_spectator),
1408         fsparam_flag   ("norecovery",         Opt_spectator),
1409         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1410         fsparam_flag   ("localflocks",        Opt_localflocks),
1411         fsparam_flag   ("localcaching",       Opt_localcaching),
1412         fsparam_flag_no("debug",              Opt_debug),
1413         fsparam_flag   ("upgrade",            Opt_upgrade),
1414         fsparam_flag_no("acl",                Opt_acl),
1415         fsparam_flag_no("suiddir",            Opt_suiddir),
1416         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1417         fsparam_flag   ("meta",               Opt_meta),
1418         fsparam_flag_no("discard",            Opt_discard),
1419         fsparam_s32    ("commit",             Opt_commit),
1420         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1421         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1422         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1423         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1424         fsparam_flag_no("barrier",            Opt_barrier),
1425         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1426         fsparam_flag_no("loccookie",          Opt_loccookie),
1427         /* quota can be a flag or an enum so it gets special treatment */
1428         fsparam_flag_no("quota",              Opt_quota_flag),
1429         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1430         {}
1431 };
1432
1433 /* Parse a single mount parameter */
1434 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1435 {
1436         struct gfs2_args *args = fc->fs_private;
1437         struct fs_parse_result result;
1438         int o;
1439
1440         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1441         if (o < 0)
1442                 return o;
1443
1444         switch (o) {
1445         case Opt_lockproto:
1446                 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1447                 break;
1448         case Opt_locktable:
1449                 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1450                 break;
1451         case Opt_hostdata:
1452                 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1453                 break;
1454         case Opt_spectator:
1455                 args->ar_spectator = 1;
1456                 break;
1457         case Opt_ignore_local_fs:
1458                 /* Retained for backwards compat only */
1459                 break;
1460         case Opt_localflocks:
1461                 args->ar_localflocks = 1;
1462                 break;
1463         case Opt_localcaching:
1464                 /* Retained for backwards compat only */
1465                 break;
1466         case Opt_debug:
1467                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1468                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1469                 args->ar_debug = result.boolean;
1470                 break;
1471         case Opt_upgrade:
1472                 /* Retained for backwards compat only */
1473                 break;
1474         case Opt_acl:
1475                 args->ar_posix_acl = result.boolean;
1476                 break;
1477         case Opt_quota_flag:
1478                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1479                 break;
1480         case Opt_quota:
1481                 args->ar_quota = result.int_32;
1482                 break;
1483         case Opt_suiddir:
1484                 args->ar_suiddir = result.boolean;
1485                 break;
1486         case Opt_data:
1487                 /* The uint_32 result maps directly to GFS2_DATA_* */
1488                 args->ar_data = result.uint_32;
1489                 break;
1490         case Opt_meta:
1491                 args->ar_meta = 1;
1492                 break;
1493         case Opt_discard:
1494                 args->ar_discard = result.boolean;
1495                 break;
1496         case Opt_commit:
1497                 if (result.int_32 <= 0)
1498                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1499                 args->ar_commit = result.int_32;
1500                 break;
1501         case Opt_statfs_quantum:
1502                 if (result.int_32 < 0)
1503                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1504                 args->ar_statfs_quantum = result.int_32;
1505                 break;
1506         case Opt_quota_quantum:
1507                 if (result.int_32 <= 0)
1508                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1509                 args->ar_quota_quantum = result.int_32;
1510                 break;
1511         case Opt_statfs_percent:
1512                 if (result.int_32 < 0 || result.int_32 > 100)
1513                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1514                 args->ar_statfs_percent = result.int_32;
1515                 break;
1516         case Opt_errors:
1517                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1518                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1519                 args->ar_errors = result.uint_32;
1520                 break;
1521         case Opt_barrier:
1522                 args->ar_nobarrier = result.boolean;
1523                 break;
1524         case Opt_rgrplvb:
1525                 args->ar_rgrplvb = result.boolean;
1526                 args->ar_got_rgrplvb = 1;
1527                 break;
1528         case Opt_loccookie:
1529                 args->ar_loccookie = result.boolean;
1530                 break;
1531         default:
1532                 return invalfc(fc, "invalid mount option: %s", param->key);
1533         }
1534         return 0;
1535 }
1536
1537 static int gfs2_reconfigure(struct fs_context *fc)
1538 {
1539         struct super_block *sb = fc->root->d_sb;
1540         struct gfs2_sbd *sdp = sb->s_fs_info;
1541         struct gfs2_args *oldargs = &sdp->sd_args;
1542         struct gfs2_args *newargs = fc->fs_private;
1543         struct gfs2_tune *gt = &sdp->sd_tune;
1544         int error = 0;
1545
1546         sync_filesystem(sb);
1547
1548         spin_lock(&gt->gt_spin);
1549         oldargs->ar_commit = gt->gt_logd_secs;
1550         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1551         if (gt->gt_statfs_slow)
1552                 oldargs->ar_statfs_quantum = 0;
1553         else
1554                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1555         spin_unlock(&gt->gt_spin);
1556
1557         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1558                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1559                 return -EINVAL;
1560         }
1561         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1562                 errorfc(fc, "reconfiguration of lock table not allowed");
1563                 return -EINVAL;
1564         }
1565         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1566                 errorfc(fc, "reconfiguration of host data not allowed");
1567                 return -EINVAL;
1568         }
1569         if (newargs->ar_spectator != oldargs->ar_spectator) {
1570                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1571                 return -EINVAL;
1572         }
1573         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1574                 errorfc(fc, "reconfiguration of localflocks not allowed");
1575                 return -EINVAL;
1576         }
1577         if (newargs->ar_meta != oldargs->ar_meta) {
1578                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1579                 return -EINVAL;
1580         }
1581         if (oldargs->ar_spectator)
1582                 fc->sb_flags |= SB_RDONLY;
1583
1584         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1585                 struct gfs2_holder freeze_gh;
1586
1587                 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1588                 if (error)
1589                         return -EINVAL;
1590
1591                 if (fc->sb_flags & SB_RDONLY) {
1592                         gfs2_make_fs_ro(sdp);
1593                 } else {
1594                         error = gfs2_make_fs_rw(sdp);
1595                         if (error)
1596                                 errorfc(fc, "unable to remount read-write");
1597                 }
1598                 gfs2_freeze_unlock(&freeze_gh);
1599         }
1600         sdp->sd_args = *newargs;
1601
1602         if (sdp->sd_args.ar_posix_acl)
1603                 sb->s_flags |= SB_POSIXACL;
1604         else
1605                 sb->s_flags &= ~SB_POSIXACL;
1606         if (sdp->sd_args.ar_nobarrier)
1607                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1608         else
1609                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1610         spin_lock(&gt->gt_spin);
1611         gt->gt_logd_secs = newargs->ar_commit;
1612         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1613         if (newargs->ar_statfs_quantum) {
1614                 gt->gt_statfs_slow = 0;
1615                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1616         }
1617         else {
1618                 gt->gt_statfs_slow = 1;
1619                 gt->gt_statfs_quantum = 30;
1620         }
1621         spin_unlock(&gt->gt_spin);
1622
1623         gfs2_online_uevent(sdp);
1624         return error;
1625 }
1626
1627 static const struct fs_context_operations gfs2_context_ops = {
1628         .free        = gfs2_fc_free,
1629         .parse_param = gfs2_parse_param,
1630         .get_tree    = gfs2_get_tree,
1631         .reconfigure = gfs2_reconfigure,
1632 };
1633
1634 /* Set up the filesystem mount context */
1635 static int gfs2_init_fs_context(struct fs_context *fc)
1636 {
1637         struct gfs2_args *args;
1638
1639         args = kmalloc(sizeof(*args), GFP_KERNEL);
1640         if (args == NULL)
1641                 return -ENOMEM;
1642
1643         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1644                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1645
1646                 *args = sdp->sd_args;
1647         } else {
1648                 memset(args, 0, sizeof(*args));
1649                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1650                 args->ar_data = GFS2_DATA_DEFAULT;
1651                 args->ar_commit = 30;
1652                 args->ar_statfs_quantum = 30;
1653                 args->ar_quota_quantum = 60;
1654                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1655         }
1656         fc->fs_private = args;
1657         fc->ops = &gfs2_context_ops;
1658         return 0;
1659 }
1660
1661 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1662 {
1663         return -EINVAL;
1664 }
1665
1666 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1667 {
1668         return (fc->sget_key == s->s_bdev);
1669 }
1670
1671 static int gfs2_meta_get_tree(struct fs_context *fc)
1672 {
1673         struct super_block *s;
1674         struct gfs2_sbd *sdp;
1675         struct path path;
1676         int error;
1677
1678         if (!fc->source || !*fc->source)
1679                 return -EINVAL;
1680
1681         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1682         if (error) {
1683                 pr_warn("path_lookup on %s returned error %d\n",
1684                         fc->source, error);
1685                 return error;
1686         }
1687         fc->fs_type = &gfs2_fs_type;
1688         fc->sget_key = path.dentry->d_sb->s_bdev;
1689         s = sget_fc(fc, test_meta_super, set_meta_super);
1690         path_put(&path);
1691         if (IS_ERR(s)) {
1692                 pr_warn("gfs2 mount does not exist\n");
1693                 return PTR_ERR(s);
1694         }
1695         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1696                 deactivate_locked_super(s);
1697                 return -EBUSY;
1698         }
1699         sdp = s->s_fs_info;
1700         fc->root = dget(sdp->sd_master_dir);
1701         return 0;
1702 }
1703
1704 static const struct fs_context_operations gfs2_meta_context_ops = {
1705         .free        = gfs2_fc_free,
1706         .get_tree    = gfs2_meta_get_tree,
1707 };
1708
1709 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1710 {
1711         int ret = gfs2_init_fs_context(fc);
1712
1713         if (ret)
1714                 return ret;
1715
1716         fc->ops = &gfs2_meta_context_ops;
1717         return 0;
1718 }
1719
1720 static void gfs2_kill_sb(struct super_block *sb)
1721 {
1722         struct gfs2_sbd *sdp = sb->s_fs_info;
1723
1724         if (sdp == NULL) {
1725                 kill_block_super(sb);
1726                 return;
1727         }
1728
1729         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1730         dput(sdp->sd_root_dir);
1731         dput(sdp->sd_master_dir);
1732         sdp->sd_root_dir = NULL;
1733         sdp->sd_master_dir = NULL;
1734         shrink_dcache_sb(sb);
1735         kill_block_super(sb);
1736 }
1737
1738 struct file_system_type gfs2_fs_type = {
1739         .name = "gfs2",
1740         .fs_flags = FS_REQUIRES_DEV,
1741         .init_fs_context = gfs2_init_fs_context,
1742         .parameters = gfs2_fs_parameters,
1743         .kill_sb = gfs2_kill_sb,
1744         .owner = THIS_MODULE,
1745 };
1746 MODULE_ALIAS_FS("gfs2");
1747
1748 struct file_system_type gfs2meta_fs_type = {
1749         .name = "gfs2meta",
1750         .fs_flags = FS_REQUIRES_DEV,
1751         .init_fs_context = gfs2_meta_init_fs_context,
1752         .owner = THIS_MODULE,
1753 };
1754 MODULE_ALIAS_FS("gfs2meta");