fs/gfs2/super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/crc32.h>
  16 #include <linux/gfs2_ondisk.h>
  17 #include <linux/bio.h>
  18 #include <linux/lm_interface.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "dir.h"
  24 #include "glock.h"
  25 #include "glops.h"
  26 #include "inode.h"
  27 #include "log.h"
  28 #include "meta_io.h"
  29 #include "quota.h"
  30 #include "recovery.h"
  31 #include "rgrp.h"
  32 #include "super.h"
  33 #include "trans.h"
  34 #include "util.h"
  35
  36 static const u32 gfs2_old_fs_formats[] = {
  37         0
  38 };
  39
  40 static const u32 gfs2_old_multihost_formats[] = {
  41         0
  42 };
  43
  44 /**
  45  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  46  * @gt: tune
  47  *
  48  */
  49
  50 void gfs2_tune_init(struct gfs2_tune *gt)
  51 {
  52         spin_lock_init(&gt->gt_spin);
  53
  54         gt->gt_ilimit = 100;
  55         gt->gt_ilimit_tries = 3;
  56         gt->gt_ilimit_min = 1;
  57         gt->gt_demote_secs = 300;
  58         gt->gt_incore_log_blocks = 1024;
  59         gt->gt_log_flush_secs = 60;
  60         gt->gt_jindex_refresh_secs = 60;
  61         gt->gt_scand_secs = 15;
  62         gt->gt_recoverd_secs = 60;
  63         gt->gt_logd_secs = 1;
  64         gt->gt_quotad_secs = 5;
  65         gt->gt_quota_simul_sync = 64;
  66         gt->gt_quota_warn_period = 10;
  67         gt->gt_quota_scale_num = 1;
  68         gt->gt_quota_scale_den = 1;
  69         gt->gt_quota_cache_secs = 300;
  70         gt->gt_quota_quantum = 60;
  71         gt->gt_atime_quantum = 3600;
  72         gt->gt_new_files_jdata = 0;
  73         gt->gt_new_files_directio = 0;
  74         gt->gt_max_readahead = 1 << 18;
  75         gt->gt_lockdump_size = 131072;
  76         gt->gt_stall_secs = 600;
  77         gt->gt_complain_secs = 10;
  78         gt->gt_reclaim_limit = 5000;
  79         gt->gt_statfs_quantum = 30;
  80         gt->gt_statfs_slow = 0;
  81 }
  82
  83 /**
  84  * gfs2_check_sb - Check superblock
  85  * @sdp: the filesystem
  86  * @sb: The superblock
  87  * @silent: Don't print a message if the check fails
  88  *
  89  * Checks the version code of the FS is one that we understand how to
  90  * read and that the sizes of the various on-disk structures have not
  91  * changed.
  92  */
  93
  94 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  95 {
  96         unsigned int x;
  97
  98         if (sb->sb_magic != GFS2_MAGIC ||
  99             sb->sb_type != GFS2_METATYPE_SB) {
 100                 if (!silent)
 101                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 102                 return -EINVAL;
 103         }
 104
 105         /*  If format numbers match exactly, we're done.  */
 106
 107         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 108             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 109                 return 0;
 110
 111         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 112                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 113                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 114                                 break;
 115
 116                 if (!gfs2_old_fs_formats[x]) {
 117                         printk(KERN_WARNING
 118                                "GFS2: code version (%u, %u) is incompatible "
 119                                "with ondisk format (%u, %u)\n",
 120                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 121                                sb->sb_fs_format, sb->sb_multihost_format);
 122                         printk(KERN_WARNING
 123                                "GFS2: I don't know how to upgrade this FS\n");
 124                         return -EINVAL;
 125                 }
 126         }
 127
 128         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 129                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 130                         if (gfs2_old_multihost_formats[x] ==
 131                             sb->sb_multihost_format)
 132                                 break;
 133
 134                 if (!gfs2_old_multihost_formats[x]) {
 135                         printk(KERN_WARNING
 136                                "GFS2: code version (%u, %u) is incompatible "
 137                                "with ondisk format (%u, %u)\n",
 138                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 139                                sb->sb_fs_format, sb->sb_multihost_format);
 140                         printk(KERN_WARNING
 141                                "GFS2: I don't know how to upgrade this FS\n");
 142                         return -EINVAL;
 143                 }
 144         }
 145
 146         if (!sdp->sd_args.ar_upgrade) {
 147                 printk(KERN_WARNING
 148                        "GFS2: code version (%u, %u) is incompatible "
 149                        "with ondisk format (%u, %u)\n",
 150                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 151                        sb->sb_fs_format, sb->sb_multihost_format);
 152                 printk(KERN_INFO
 153                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 154                        "the FS\n");
 155                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 156                 return -EINVAL;
 157         }
 158
 159         return 0;
 160 }
 161
 162
 163 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
 164 {
 165         struct page *page = bio->bi_private;
 166         if (bio->bi_size)
 167                 return 1;
 168
 169         if (!error)
 170                 SetPageUptodate(page);
 171         else
 172                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 173         unlock_page(page);
 174         return 0;
 175 }
 176
 177 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 178 {
 179         const struct gfs2_sb *str = buf;
 180
 181         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 182         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 183         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 184         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 185         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 186         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 187         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 188         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 189         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 190         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 191         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 192
 193         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 194         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 195 }
 196
 197 /**
 198  * gfs2_read_super - Read the gfs2 super block from disk
 199  * @sdp: The GFS2 super block
 200  * @sector: The location of the super block
 201  * @error: The error code to return
 202  *
 203  * This uses the bio functions to read the super block from disk
 204  * because we want to be 100% sure that we never read cached data.
 205  * A super block is read twice only during each GFS2 mount and is
 206  * never written to by the filesystem. The first time its read no
 207  * locks are held, and the only details which are looked at are those
 208  * relating to the locking protocol. Once locking is up and working,
 209  * the sb is read again under the lock to establish the location of
 210  * the master directory (contains pointers to journals etc) and the
 211  * root directory.
 212  *
 213  * Returns: 0 on success or error
 214  */
 215
 216 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 217 {
 218         struct super_block *sb = sdp->sd_vfs;
 219         struct gfs2_sb *p;
 220         struct page *page;
 221         struct bio *bio;
 222
 223         page = alloc_page(GFP_KERNEL);
 224         if (unlikely(!page))
 225                 return -ENOBUFS;
 226
 227         ClearPageUptodate(page);
 228         ClearPageDirty(page);
 229         lock_page(page);
 230
 231         bio = bio_alloc(GFP_KERNEL, 1);
 232         if (unlikely(!bio)) {
 233                 __free_page(page);
 234                 return -ENOBUFS;
 235         }
 236
 237         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 238         bio->bi_bdev = sb->s_bdev;
 239         bio_add_page(bio, page, PAGE_SIZE, 0);
 240
 241         bio->bi_end_io = end_bio_io_page;
 242         bio->bi_private = page;
 243         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 244         wait_on_page_locked(page);
 245         bio_put(bio);
 246         if (!PageUptodate(page)) {
 247                 __free_page(page);
 248                 return -EIO;
 249         }
 250         p = kmap(page);
 251         gfs2_sb_in(&sdp->sd_sb, p);
 252         kunmap(page);
 253         __free_page(page);
 254         return 0;
 255 }
 256
 257 /**
 258  * gfs2_read_sb - Read super block
 259  * @sdp: The GFS2 superblock
 260  * @gl: the glock for the superblock (assumed to be held)
 261  * @silent: Don't print message if mount fails
 262  *
 263  */
 264
 265 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 266 {
 267         u32 hash_blocks, ind_blocks, leaf_blocks;
 268         u32 tmp_blocks;
 269         unsigned int x;
 270         int error;
 271
 272         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 273         if (error) {
 274                 if (!silent)
 275                         fs_err(sdp, "can't read superblock\n");
 276                 return error;
 277         }
 278
 279         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 280         if (error)
 281                 return error;
 282
 283         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 284                                GFS2_BASIC_BLOCK_SHIFT;
 285         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 286         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 287                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 288         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 289                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 290         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 291         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 292         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 293         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 294         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 295                                 sizeof(struct gfs2_meta_header)) /
 296                                 sizeof(struct gfs2_quota_change);
 297
 298         /* Compute maximum reservation required to add a entry to a directory */
 299
 300         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 301                              sdp->sd_jbsize);
 302
 303         ind_blocks = 0;
 304         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 305                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 306                 ind_blocks += tmp_blocks;
 307         }
 308
 309         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 310
 311         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 312
 313         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 314                                 sizeof(struct gfs2_dinode);
 315         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 316         for (x = 2;; x++) {
 317                 u64 space, d;
 318                 u32 m;
 319
 320                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 321                 d = space;
 322                 m = do_div(d, sdp->sd_inptrs);
 323
 324                 if (d != sdp->sd_heightsize[x - 1] || m)
 325                         break;
 326                 sdp->sd_heightsize[x] = space;
 327         }
 328         sdp->sd_max_height = x;
 329         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 330
 331         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 332                                  sizeof(struct gfs2_dinode);
 333         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 334         for (x = 2;; x++) {
 335                 u64 space, d;
 336                 u32 m;
 337
 338                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 339                 d = space;
 340                 m = do_div(d, sdp->sd_inptrs);
 341
 342                 if (d != sdp->sd_jheightsize[x - 1] || m)
 343                         break;
 344                 sdp->sd_jheightsize[x] = space;
 345         }
 346         sdp->sd_max_jheight = x;
 347         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 348
 349         return 0;
 350 }
 351
 352 /**
 353  * gfs2_jindex_hold - Grab a lock on the jindex
 354  * @sdp: The GFS2 superblock
 355  * @ji_gh: the holder for the jindex glock
 356  *
 357  * This is very similar to the gfs2_rindex_hold() function, except that
 358  * in general we hold the jindex lock for longer periods of time and
 359  * we grab it far less frequently (in general) then the rgrp lock.
 360  *
 361  * Returns: errno
 362  */
 363
 364 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 365 {
 366         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 367         struct qstr name;
 368         char buf[20];
 369         struct gfs2_jdesc *jd;
 370         int error;
 371
 372         name.name = buf;
 373
 374         mutex_lock(&sdp->sd_jindex_mutex);
 375
 376         for (;;) {
 377                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 378                 if (error)
 379                         break;
 380
 381                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 382                 name.hash = gfs2_disk_hash(name.name, name.len);
 383
 384                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 385                 if (error == -ENOENT) {
 386                         error = 0;
 387                         break;
 388                 }
 389
 390                 gfs2_glock_dq_uninit(ji_gh);
 391
 392                 if (error)
 393                         break;
 394
 395                 error = -ENOMEM;
 396                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 397                 if (!jd)
 398                         break;
 399
 400                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 401                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 402                         if (!jd->jd_inode)
 403                                 error = -ENOENT;
 404                         else
 405                                 error = PTR_ERR(jd->jd_inode);
 406                         kfree(jd);
 407                         break;
 408                 }
 409
 410                 spin_lock(&sdp->sd_jindex_spin);
 411                 jd->jd_jid = sdp->sd_journals++;
 412                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 413                 spin_unlock(&sdp->sd_jindex_spin);
 414         }
 415
 416         mutex_unlock(&sdp->sd_jindex_mutex);
 417
 418         return error;
 419 }
 420
 421 /**
 422  * gfs2_jindex_free - Clear all the journal index information
 423  * @sdp: The GFS2 superblock
 424  *
 425  */
 426
 427 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 428 {
 429         struct list_head list;
 430         struct gfs2_jdesc *jd;
 431
 432         spin_lock(&sdp->sd_jindex_spin);
 433         list_add(&list, &sdp->sd_jindex_list);
 434         list_del_init(&sdp->sd_jindex_list);
 435         sdp->sd_journals = 0;
 436         spin_unlock(&sdp->sd_jindex_spin);
 437
 438         while (!list_empty(&list)) {
 439                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 440                 list_del(&jd->jd_list);
 441                 iput(jd->jd_inode);
 442                 kfree(jd);
 443         }
 444 }
 445
 446 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 447 {
 448         struct gfs2_jdesc *jd;
 449         int found = 0;
 450
 451         list_for_each_entry(jd, head, jd_list) {
 452                 if (jd->jd_jid == jid) {
 453                         found = 1;
 454                         break;
 455                 }
 456         }
 457
 458         if (!found)
 459                 jd = NULL;
 460
 461         return jd;
 462 }
 463
 464 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 465 {
 466         struct gfs2_jdesc *jd;
 467
 468         spin_lock(&sdp->sd_jindex_spin);
 469         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 470         spin_unlock(&sdp->sd_jindex_spin);
 471
 472         return jd;
 473 }
 474
 475 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 476 {
 477         struct gfs2_jdesc *jd;
 478
 479         spin_lock(&sdp->sd_jindex_spin);
 480         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 481         if (jd)
 482                 jd->jd_dirty = 1;
 483         spin_unlock(&sdp->sd_jindex_spin);
 484 }
 485
 486 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 487 {
 488         struct gfs2_jdesc *jd;
 489         int found = 0;
 490
 491         spin_lock(&sdp->sd_jindex_spin);
 492
 493         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 494                 if (jd->jd_dirty) {
 495                         jd->jd_dirty = 0;
 496                         found = 1;
 497                         break;
 498                 }
 499         }
 500         spin_unlock(&sdp->sd_jindex_spin);
 501
 502         if (!found)
 503                 jd = NULL;
 504
 505         return jd;
 506 }
 507
 508 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 509 {
 510         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 511         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 512         int ar;
 513         int error;
 514
 515         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 516             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 517                 gfs2_consist_inode(ip);
 518                 return -EIO;
 519         }
 520         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 521
 522         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 523         if (!error && ar) {
 524                 gfs2_consist_inode(ip);
 525                 error = -EIO;
 526         }
 527
 528         return error;
 529 }
 530
 531 /**
 532  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 533  * @sdp: the filesystem
 534  *
 535  * Returns: errno
 536  */
 537
 538 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 539 {
 540         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 541         struct gfs2_glock *j_gl = ip->i_gl;
 542         struct gfs2_holder t_gh;
 543         struct gfs2_log_header_host head;
 544         int error;
 545
 546         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 547         if (error)
 548                 return error;
 549
 550         gfs2_meta_cache_flush(ip);
 551         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 552
 553         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 554         if (error)
 555                 goto fail;
 556
 557         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 558                 gfs2_consist(sdp);
 559                 error = -EIO;
 560                 goto fail;
 561         }
 562
 563         /*  Initialize some head of the log stuff  */
 564         sdp->sd_log_sequence = head.lh_sequence + 1;
 565         gfs2_log_pointers_init(sdp, head.lh_blkno);
 566
 567         error = gfs2_quota_init(sdp);
 568         if (error)
 569                 goto fail;
 570
 571         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 572
 573         gfs2_glock_dq_uninit(&t_gh);
 574
 575         return 0;
 576
 577 fail:
 578         t_gh.gh_flags |= GL_NOCACHE;
 579         gfs2_glock_dq_uninit(&t_gh);
 580
 581         return error;
 582 }
 583
 584 /**
 585  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 586  * @sdp: the filesystem
 587  *
 588  * Returns: errno
 589  */
 590
 591 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 592 {
 593         struct gfs2_holder t_gh;
 594         int error;
 595
 596         gfs2_quota_sync(sdp);
 597         gfs2_statfs_sync(sdp);
 598
 599         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 600                                    &t_gh);
 601         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 602                 return error;
 603
 604         gfs2_meta_syncfs(sdp);
 605         gfs2_log_shutdown(sdp);
 606
 607         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 608
 609         if (t_gh.gh_gl)
 610                 gfs2_glock_dq_uninit(&t_gh);
 611
 612         gfs2_quota_cleanup(sdp);
 613
 614         return error;
 615 }
 616
 617 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 618 {
 619         const struct gfs2_statfs_change *str = buf;
 620
 621         sc->sc_total = be64_to_cpu(str->sc_total);
 622         sc->sc_free = be64_to_cpu(str->sc_free);
 623         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 624 }
 625
 626 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 627 {
 628         struct gfs2_statfs_change *str = buf;
 629
 630         str->sc_total = cpu_to_be64(sc->sc_total);
 631         str->sc_free = cpu_to_be64(sc->sc_free);
 632         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 633 }
 634
 635 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 636 {
 637         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 638         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 639         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 640         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 641         struct buffer_head *m_bh, *l_bh;
 642         struct gfs2_holder gh;
 643         int error;
 644
 645         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 646                                    &gh);
 647         if (error)
 648                 return error;
 649
 650         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 651         if (error)
 652                 goto out;
 653
 654         if (sdp->sd_args.ar_spectator) {
 655                 spin_lock(&sdp->sd_statfs_spin);
 656                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 657                                       sizeof(struct gfs2_dinode));
 658                 spin_unlock(&sdp->sd_statfs_spin);
 659         } else {
 660                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 661                 if (error)
 662                         goto out_m_bh;
 663
 664                 spin_lock(&sdp->sd_statfs_spin);
 665                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 666                                       sizeof(struct gfs2_dinode));
 667                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 668                                       sizeof(struct gfs2_dinode));
 669                 spin_unlock(&sdp->sd_statfs_spin);
 670
 671                 brelse(l_bh);
 672         }
 673
 674 out_m_bh:
 675         brelse(m_bh);
 676 out:
 677         gfs2_glock_dq_uninit(&gh);
 678         return 0;
 679 }
 680
 681 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 682                         s64 dinodes)
 683 {
 684         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 685         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 686         struct buffer_head *l_bh;
 687         int error;
 688
 689         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 690         if (error)
 691                 return;
 692
 693         mutex_lock(&sdp->sd_statfs_mutex);
 694         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 695         mutex_unlock(&sdp->sd_statfs_mutex);
 696
 697         spin_lock(&sdp->sd_statfs_spin);
 698         l_sc->sc_total += total;
 699         l_sc->sc_free += free;
 700         l_sc->sc_dinodes += dinodes;
 701         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 702         spin_unlock(&sdp->sd_statfs_spin);
 703
 704         brelse(l_bh);
 705 }
 706
 707 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 708 {
 709         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 710         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 711         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 712         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 713         struct gfs2_holder gh;
 714         struct buffer_head *m_bh, *l_bh;
 715         int error;
 716
 717         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 718                                    &gh);
 719         if (error)
 720                 return error;
 721
 722         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 723         if (error)
 724                 goto out;
 725
 726         spin_lock(&sdp->sd_statfs_spin);
 727         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 728                               sizeof(struct gfs2_dinode));
 729         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 730                 spin_unlock(&sdp->sd_statfs_spin);
 731                 goto out_bh;
 732         }
 733         spin_unlock(&sdp->sd_statfs_spin);
 734
 735         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 736         if (error)
 737                 goto out_bh;
 738
 739         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 740         if (error)
 741                 goto out_bh2;
 742
 743         mutex_lock(&sdp->sd_statfs_mutex);
 744         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 745         mutex_unlock(&sdp->sd_statfs_mutex);
 746
 747         spin_lock(&sdp->sd_statfs_spin);
 748         m_sc->sc_total += l_sc->sc_total;
 749         m_sc->sc_free += l_sc->sc_free;
 750         m_sc->sc_dinodes += l_sc->sc_dinodes;
 751         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 752         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 753                0, sizeof(struct gfs2_statfs_change));
 754         spin_unlock(&sdp->sd_statfs_spin);
 755
 756         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 757         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 758
 759         gfs2_trans_end(sdp);
 760
 761 out_bh2:
 762         brelse(l_bh);
 763 out_bh:
 764         brelse(m_bh);
 765 out:
 766         gfs2_glock_dq_uninit(&gh);
 767         return error;
 768 }
 769
 770 /**
 771  * gfs2_statfs_i - Do a statfs
 772  * @sdp: the filesystem
 773  * @sg: the sg structure
 774  *
 775  * Returns: errno
 776  */
 777
 778 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 779 {
 780         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 781         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 782
 783         spin_lock(&sdp->sd_statfs_spin);
 784
 785         *sc = *m_sc;
 786         sc->sc_total += l_sc->sc_total;
 787         sc->sc_free += l_sc->sc_free;
 788         sc->sc_dinodes += l_sc->sc_dinodes;
 789
 790         spin_unlock(&sdp->sd_statfs_spin);
 791
 792         if (sc->sc_free < 0)
 793                 sc->sc_free = 0;
 794         if (sc->sc_free > sc->sc_total)
 795                 sc->sc_free = sc->sc_total;
 796         if (sc->sc_dinodes < 0)
 797                 sc->sc_dinodes = 0;
 798
 799         return 0;
 800 }
 801
 802 /**
 803  * statfs_fill - fill in the sg for a given RG
 804  * @rgd: the RG
 805  * @sc: the sc structure
 806  *
 807  * Returns: 0 on success, -ESTALE if the LVB is invalid
 808  */
 809
 810 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 811                             struct gfs2_statfs_change_host *sc)
 812 {
 813         gfs2_rgrp_verify(rgd);
 814         sc->sc_total += rgd->rd_data;
 815         sc->sc_free += rgd->rd_rg.rg_free;
 816         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 817         return 0;
 818 }
 819
 820 /**
 821  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 822  * @sdp: the filesystem
 823  * @sc: the sc info that will be returned
 824  *
 825  * Any error (other than a signal) will cause this routine to fall back
 826  * to the synchronous version.
 827  *
 828  * FIXME: This really shouldn't busy wait like this.
 829  *
 830  * Returns: errno
 831  */
 832
 833 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 834 {
 835         struct gfs2_holder ri_gh;
 836         struct gfs2_rgrpd *rgd_next;
 837         struct gfs2_holder *gha, *gh;
 838         unsigned int slots = 64;
 839         unsigned int x;
 840         int done;
 841         int error = 0, err;
 842
 843         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 844         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 845         if (!gha)
 846                 return -ENOMEM;
 847
 848         error = gfs2_rindex_hold(sdp, &ri_gh);
 849         if (error)
 850                 goto out;
 851
 852         rgd_next = gfs2_rgrpd_get_first(sdp);
 853
 854         for (;;) {
 855                 done = 1;
 856
 857                 for (x = 0; x < slots; x++) {
 858                         gh = gha + x;
 859
 860                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 861                                 err = gfs2_glock_wait(gh);
 862                                 if (err) {
 863                                         gfs2_holder_uninit(gh);
 864                                         error = err;
 865                                 } else {
 866                                         if (!error)
 867                                                 error = statfs_slow_fill(
 868                                                         gh->gh_gl->gl_object, sc);
 869                                         gfs2_glock_dq_uninit(gh);
 870                                 }
 871                         }
 872
 873                         if (gh->gh_gl)
 874                                 done = 0;
 875                         else if (rgd_next && !error) {
 876                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 877                                                            LM_ST_SHARED,
 878                                                            GL_ASYNC,
 879                                                            gh);
 880                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 881                                 done = 0;
 882                         }
 883
 884                         if (signal_pending(current))
 885                                 error = -ERESTARTSYS;
 886                 }
 887
 888                 if (done)
 889                         break;
 890
 891                 yield();
 892         }
 893
 894         gfs2_glock_dq_uninit(&ri_gh);
 895
 896 out:
 897         kfree(gha);
 898         return error;
 899 }
 900
 901 struct lfcc {
 902         struct list_head list;
 903         struct gfs2_holder gh;
 904 };
 905
 906 /**
 907  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 908  *                            journals are clean
 909  * @sdp: the file system
 910  * @state: the state to put the transaction lock into
 911  * @t_gh: the hold on the transaction lock
 912  *
 913  * Returns: errno
 914  */
 915
 916 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 917                                     struct gfs2_holder *t_gh)
 918 {
 919         struct gfs2_inode *ip;
 920         struct gfs2_holder ji_gh;
 921         struct gfs2_jdesc *jd;
 922         struct lfcc *lfcc;
 923         LIST_HEAD(list);
 924         struct gfs2_log_header_host lh;
 925         int error;
 926
 927         error = gfs2_jindex_hold(sdp, &ji_gh);
 928         if (error)
 929                 return error;
 930
 931         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 932                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 933                 if (!lfcc) {
 934                         error = -ENOMEM;
 935                         goto out;
 936                 }
 937                 ip = GFS2_I(jd->jd_inode);
 938                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 939                 if (error) {
 940                         kfree(lfcc);
 941                         goto out;
 942                 }
 943                 list_add(&lfcc->list, &list);
 944         }
 945
 946         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 947                                LM_FLAG_PRIORITY | GL_NOCACHE,
 948                                t_gh);
 949
 950         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 951                 error = gfs2_jdesc_check(jd);
 952                 if (error)
 953                         break;
 954                 error = gfs2_find_jhead(jd, &lh);
 955                 if (error)
 956                         break;
 957                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 958                         error = -EBUSY;
 959                         break;
 960                 }
 961         }
 962
 963         if (error)
 964                 gfs2_glock_dq_uninit(t_gh);
 965
 966 out:
 967         while (!list_empty(&list)) {
 968                 lfcc = list_entry(list.next, struct lfcc, list);
 969                 list_del(&lfcc->list);
 970                 gfs2_glock_dq_uninit(&lfcc->gh);
 971                 kfree(lfcc);
 972         }
 973         gfs2_glock_dq_uninit(&ji_gh);
 974         return error;
 975 }
 976
 977 /**
 978  * gfs2_freeze_fs - freezes the file system
 979  * @sdp: the file system
 980  *
 981  * This function flushes data and meta data for all machines by
 982  * aquiring the transaction log exclusively.  All journals are
 983  * ensured to be in a clean state as well.
 984  *
 985  * Returns: errno
 986  */
 987
 988 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 989 {
 990         int error = 0;
 991
 992         mutex_lock(&sdp->sd_freeze_lock);
 993
 994         if (!sdp->sd_freeze_count++) {
 995                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 996                 if (error)
 997                         sdp->sd_freeze_count--;
 998         }
 999
1000         mutex_unlock(&sdp->sd_freeze_lock);
1001
1002         return error;
1003 }
1004
1005 /**
1006  * gfs2_unfreeze_fs - unfreezes the file system
1007  * @sdp: the file system
1008  *
1009  * This function allows the file system to proceed by unlocking
1010  * the exclusively held transaction lock.  Other GFS2 nodes are
1011  * now free to acquire the lock shared and go on with their lives.
1012  *
1013  */
1014
1015 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1016 {
1017         mutex_lock(&sdp->sd_freeze_lock);
1018
1019         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1020                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1021
1022         mutex_unlock(&sdp->sd_freeze_lock);
1023 }
1024