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_entries_per_readdir = 32;
  80         gt->gt_statfs_quantum = 30;
  81         gt->gt_statfs_slow = 0;
  82 }
  83
  84 /**
  85  * gfs2_check_sb - Check superblock
  86  * @sdp: the filesystem
  87  * @sb: The superblock
  88  * @silent: Don't print a message if the check fails
  89  *
  90  * Checks the version code of the FS is one that we understand how to
  91  * read and that the sizes of the various on-disk structures have not
  92  * changed.
  93  */
  94
  95 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  96 {
  97         unsigned int x;
  98
  99         if (sb->sb_header.mh_magic != GFS2_MAGIC ||
 100             sb->sb_header.mh_type != GFS2_METATYPE_SB) {
 101                 if (!silent)
 102                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 103                 return -EINVAL;
 104         }
 105
 106         /*  If format numbers match exactly, we're done.  */
 107
 108         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 109             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 110                 return 0;
 111
 112         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 113                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 114                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 115                                 break;
 116
 117                 if (!gfs2_old_fs_formats[x]) {
 118                         printk(KERN_WARNING
 119                                "GFS2: code version (%u, %u) is incompatible "
 120                                "with ondisk format (%u, %u)\n",
 121                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 122                                sb->sb_fs_format, sb->sb_multihost_format);
 123                         printk(KERN_WARNING
 124                                "GFS2: I don't know how to upgrade this FS\n");
 125                         return -EINVAL;
 126                 }
 127         }
 128
 129         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 130                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 131                         if (gfs2_old_multihost_formats[x] ==
 132                             sb->sb_multihost_format)
 133                                 break;
 134
 135                 if (!gfs2_old_multihost_formats[x]) {
 136                         printk(KERN_WARNING
 137                                "GFS2: code version (%u, %u) is incompatible "
 138                                "with ondisk format (%u, %u)\n",
 139                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 140                                sb->sb_fs_format, sb->sb_multihost_format);
 141                         printk(KERN_WARNING
 142                                "GFS2: I don't know how to upgrade this FS\n");
 143                         return -EINVAL;
 144                 }
 145         }
 146
 147         if (!sdp->sd_args.ar_upgrade) {
 148                 printk(KERN_WARNING
 149                        "GFS2: code version (%u, %u) is incompatible "
 150                        "with ondisk format (%u, %u)\n",
 151                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 152                        sb->sb_fs_format, sb->sb_multihost_format);
 153                 printk(KERN_INFO
 154                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 155                        "the FS\n");
 156                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 157                 return -EINVAL;
 158         }
 159
 160         return 0;
 161 }
 162
 163
 164 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
 165 {
 166         struct page *page = bio->bi_private;
 167         if (bio->bi_size)
 168                 return 1;
 169
 170         if (!error)
 171                 SetPageUptodate(page);
 172         else
 173                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 174         unlock_page(page);
 175         return 0;
 176 }
 177
 178 /**
 179  * gfs2_read_super - Read the gfs2 super block from disk
 180  * @sb: The VFS super block
 181  * @sector: The location of the super block
 182  *
 183  * This uses the bio functions to read the super block from disk
 184  * because we want to be 100% sure that we never read cached data.
 185  * A super block is read twice only during each GFS2 mount and is
 186  * never written to by the filesystem. The first time its read no
 187  * locks are held, and the only details which are looked at are those
 188  * relating to the locking protocol. Once locking is up and working,
 189  * the sb is read again under the lock to establish the location of
 190  * the master directory (contains pointers to journals etc) and the
 191  * root directory.
 192  *
 193  * Returns: A page containing the sb or NULL
 194  */
 195
 196 struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
 197 {
 198         struct page *page;
 199         struct bio *bio;
 200
 201         page = alloc_page(GFP_KERNEL);
 202         if (unlikely(!page))
 203                 return NULL;
 204
 205         ClearPageUptodate(page);
 206         ClearPageDirty(page);
 207         lock_page(page);
 208
 209         bio = bio_alloc(GFP_KERNEL, 1);
 210         if (unlikely(!bio)) {
 211                 __free_page(page);
 212                 return NULL;
 213         }
 214
 215         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 216         bio->bi_bdev = sb->s_bdev;
 217         bio_add_page(bio, page, PAGE_SIZE, 0);
 218
 219         bio->bi_end_io = end_bio_io_page;
 220         bio->bi_private = page;
 221         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 222         wait_on_page_locked(page);
 223         bio_put(bio);
 224         if (!PageUptodate(page)) {
 225                 __free_page(page);
 226                 return NULL;
 227         }
 228         return page;
 229 }
 230
 231 /**
 232  * gfs2_read_sb - Read super block
 233  * @sdp: The GFS2 superblock
 234  * @gl: the glock for the superblock (assumed to be held)
 235  * @silent: Don't print message if mount fails
 236  *
 237  */
 238
 239 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 240 {
 241         u32 hash_blocks, ind_blocks, leaf_blocks;
 242         u32 tmp_blocks;
 243         unsigned int x;
 244         int error;
 245         struct page *page;
 246         char *sb;
 247
 248         page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 249         if (!page) {
 250                 if (!silent)
 251                         fs_err(sdp, "can't read superblock\n");
 252                 return -EIO;
 253         }
 254         sb = kmap(page);
 255         gfs2_sb_in(&sdp->sd_sb, sb);
 256         kunmap(page);
 257         __free_page(page);
 258
 259         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 260         if (error)
 261                 return error;
 262
 263         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 264                                GFS2_BASIC_BLOCK_SHIFT;
 265         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 266         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 267                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 268         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 269                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 270         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 271         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 272         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 273         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 274         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 275                                 sizeof(struct gfs2_meta_header)) /
 276                                 sizeof(struct gfs2_quota_change);
 277
 278         /* Compute maximum reservation required to add a entry to a directory */
 279
 280         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 281                              sdp->sd_jbsize);
 282
 283         ind_blocks = 0;
 284         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 285                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 286                 ind_blocks += tmp_blocks;
 287         }
 288
 289         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 290
 291         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 292
 293         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 294                                 sizeof(struct gfs2_dinode);
 295         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 296         for (x = 2;; x++) {
 297                 u64 space, d;
 298                 u32 m;
 299
 300                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 301                 d = space;
 302                 m = do_div(d, sdp->sd_inptrs);
 303
 304                 if (d != sdp->sd_heightsize[x - 1] || m)
 305                         break;
 306                 sdp->sd_heightsize[x] = space;
 307         }
 308         sdp->sd_max_height = x;
 309         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 310
 311         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 312                                  sizeof(struct gfs2_dinode);
 313         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 314         for (x = 2;; x++) {
 315                 u64 space, d;
 316                 u32 m;
 317
 318                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 319                 d = space;
 320                 m = do_div(d, sdp->sd_inptrs);
 321
 322                 if (d != sdp->sd_jheightsize[x - 1] || m)
 323                         break;
 324                 sdp->sd_jheightsize[x] = space;
 325         }
 326         sdp->sd_max_jheight = x;
 327         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 328
 329         return 0;
 330 }
 331
 332 /**
 333  * gfs2_jindex_hold - Grab a lock on the jindex
 334  * @sdp: The GFS2 superblock
 335  * @ji_gh: the holder for the jindex glock
 336  *
 337  * This is very similar to the gfs2_rindex_hold() function, except that
 338  * in general we hold the jindex lock for longer periods of time and
 339  * we grab it far less frequently (in general) then the rgrp lock.
 340  *
 341  * Returns: errno
 342  */
 343
 344 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 345 {
 346         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 347         struct qstr name;
 348         char buf[20];
 349         struct gfs2_jdesc *jd;
 350         int error;
 351
 352         name.name = buf;
 353
 354         mutex_lock(&sdp->sd_jindex_mutex);
 355
 356         for (;;) {
 357                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED,
 358                                            GL_LOCAL_EXCL, ji_gh);
 359                 if (error)
 360                         break;
 361
 362                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 363                 name.hash = gfs2_disk_hash(name.name, name.len);
 364
 365                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 366                 if (error == -ENOENT) {
 367                         error = 0;
 368                         break;
 369                 }
 370
 371                 gfs2_glock_dq_uninit(ji_gh);
 372
 373                 if (error)
 374                         break;
 375
 376                 error = -ENOMEM;
 377                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 378                 if (!jd)
 379                         break;
 380
 381                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 382                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 383                         if (!jd->jd_inode)
 384                                 error = -ENOENT;
 385                         else
 386                                 error = PTR_ERR(jd->jd_inode);
 387                         kfree(jd);
 388                         break;
 389                 }
 390
 391                 spin_lock(&sdp->sd_jindex_spin);
 392                 jd->jd_jid = sdp->sd_journals++;
 393                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 394                 spin_unlock(&sdp->sd_jindex_spin);
 395         }
 396
 397         mutex_unlock(&sdp->sd_jindex_mutex);
 398
 399         return error;
 400 }
 401
 402 /**
 403  * gfs2_jindex_free - Clear all the journal index information
 404  * @sdp: The GFS2 superblock
 405  *
 406  */
 407
 408 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 409 {
 410         struct list_head list;
 411         struct gfs2_jdesc *jd;
 412
 413         spin_lock(&sdp->sd_jindex_spin);
 414         list_add(&list, &sdp->sd_jindex_list);
 415         list_del_init(&sdp->sd_jindex_list);
 416         sdp->sd_journals = 0;
 417         spin_unlock(&sdp->sd_jindex_spin);
 418
 419         while (!list_empty(&list)) {
 420                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 421                 list_del(&jd->jd_list);
 422                 iput(jd->jd_inode);
 423                 kfree(jd);
 424         }
 425 }
 426
 427 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 428 {
 429         struct gfs2_jdesc *jd;
 430         int found = 0;
 431
 432         list_for_each_entry(jd, head, jd_list) {
 433                 if (jd->jd_jid == jid) {
 434                         found = 1;
 435                         break;
 436                 }
 437         }
 438
 439         if (!found)
 440                 jd = NULL;
 441
 442         return jd;
 443 }
 444
 445 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 446 {
 447         struct gfs2_jdesc *jd;
 448
 449         spin_lock(&sdp->sd_jindex_spin);
 450         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 451         spin_unlock(&sdp->sd_jindex_spin);
 452
 453         return jd;
 454 }
 455
 456 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 457 {
 458         struct gfs2_jdesc *jd;
 459
 460         spin_lock(&sdp->sd_jindex_spin);
 461         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 462         if (jd)
 463                 jd->jd_dirty = 1;
 464         spin_unlock(&sdp->sd_jindex_spin);
 465 }
 466
 467 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 468 {
 469         struct gfs2_jdesc *jd;
 470         int found = 0;
 471
 472         spin_lock(&sdp->sd_jindex_spin);
 473
 474         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 475                 if (jd->jd_dirty) {
 476                         jd->jd_dirty = 0;
 477                         found = 1;
 478                         break;
 479                 }
 480         }
 481         spin_unlock(&sdp->sd_jindex_spin);
 482
 483         if (!found)
 484                 jd = NULL;
 485
 486         return jd;
 487 }
 488
 489 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 490 {
 491         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 492         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 493         int ar;
 494         int error;
 495
 496         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 497             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 498                 gfs2_consist_inode(ip);
 499                 return -EIO;
 500         }
 501         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 502
 503         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 504         if (!error && ar) {
 505                 gfs2_consist_inode(ip);
 506                 error = -EIO;
 507         }
 508
 509         return error;
 510 }
 511
 512 /**
 513  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 514  * @sdp: the filesystem
 515  *
 516  * Returns: errno
 517  */
 518
 519 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 520 {
 521         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 522         struct gfs2_glock *j_gl = ip->i_gl;
 523         struct gfs2_holder t_gh;
 524         struct gfs2_log_header_host head;
 525         int error;
 526
 527         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 528                                    GL_LOCAL_EXCL, &t_gh);
 529         if (error)
 530                 return error;
 531
 532         gfs2_meta_cache_flush(ip);
 533         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 534
 535         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 536         if (error)
 537                 goto fail;
 538
 539         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 540                 gfs2_consist(sdp);
 541                 error = -EIO;
 542                 goto fail;
 543         }
 544
 545         /*  Initialize some head of the log stuff  */
 546         sdp->sd_log_sequence = head.lh_sequence + 1;
 547         gfs2_log_pointers_init(sdp, head.lh_blkno);
 548
 549         error = gfs2_quota_init(sdp);
 550         if (error)
 551                 goto fail;
 552
 553         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 554
 555         gfs2_glock_dq_uninit(&t_gh);
 556
 557         return 0;
 558
 559 fail:
 560         t_gh.gh_flags |= GL_NOCACHE;
 561         gfs2_glock_dq_uninit(&t_gh);
 562
 563         return error;
 564 }
 565
 566 /**
 567  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 568  * @sdp: the filesystem
 569  *
 570  * Returns: errno
 571  */
 572
 573 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 574 {
 575         struct gfs2_holder t_gh;
 576         int error;
 577
 578         gfs2_quota_sync(sdp);
 579         gfs2_statfs_sync(sdp);
 580
 581         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 582                                 GL_LOCAL_EXCL | GL_NOCACHE,
 583                                 &t_gh);
 584         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 585                 return error;
 586
 587         gfs2_meta_syncfs(sdp);
 588         gfs2_log_shutdown(sdp);
 589
 590         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 591
 592         if (t_gh.gh_gl)
 593                 gfs2_glock_dq_uninit(&t_gh);
 594
 595         gfs2_quota_cleanup(sdp);
 596
 597         return error;
 598 }
 599
 600 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 601 {
 602         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 603         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 604         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 605         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 606         struct buffer_head *m_bh, *l_bh;
 607         struct gfs2_holder gh;
 608         int error;
 609
 610         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 611                                    &gh);
 612         if (error)
 613                 return error;
 614
 615         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 616         if (error)
 617                 goto out;
 618
 619         if (sdp->sd_args.ar_spectator) {
 620                 spin_lock(&sdp->sd_statfs_spin);
 621                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 622                                       sizeof(struct gfs2_dinode));
 623                 spin_unlock(&sdp->sd_statfs_spin);
 624         } else {
 625                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 626                 if (error)
 627                         goto out_m_bh;
 628
 629                 spin_lock(&sdp->sd_statfs_spin);
 630                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 631                                       sizeof(struct gfs2_dinode));
 632                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 633                                       sizeof(struct gfs2_dinode));
 634                 spin_unlock(&sdp->sd_statfs_spin);
 635
 636                 brelse(l_bh);
 637         }
 638
 639 out_m_bh:
 640         brelse(m_bh);
 641 out:
 642         gfs2_glock_dq_uninit(&gh);
 643         return 0;
 644 }
 645
 646 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 647                         s64 dinodes)
 648 {
 649         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 650         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 651         struct buffer_head *l_bh;
 652         int error;
 653
 654         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 655         if (error)
 656                 return;
 657
 658         mutex_lock(&sdp->sd_statfs_mutex);
 659         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 660         mutex_unlock(&sdp->sd_statfs_mutex);
 661
 662         spin_lock(&sdp->sd_statfs_spin);
 663         l_sc->sc_total += total;
 664         l_sc->sc_free += free;
 665         l_sc->sc_dinodes += dinodes;
 666         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 667         spin_unlock(&sdp->sd_statfs_spin);
 668
 669         brelse(l_bh);
 670 }
 671
 672 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 673 {
 674         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 675         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 676         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 677         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 678         struct gfs2_holder gh;
 679         struct buffer_head *m_bh, *l_bh;
 680         int error;
 681
 682         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 683                                    &gh);
 684         if (error)
 685                 return error;
 686
 687         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 688         if (error)
 689                 goto out;
 690
 691         spin_lock(&sdp->sd_statfs_spin);
 692         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 693                               sizeof(struct gfs2_dinode));
 694         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 695                 spin_unlock(&sdp->sd_statfs_spin);
 696                 goto out_bh;
 697         }
 698         spin_unlock(&sdp->sd_statfs_spin);
 699
 700         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 701         if (error)
 702                 goto out_bh;
 703
 704         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 705         if (error)
 706                 goto out_bh2;
 707
 708         mutex_lock(&sdp->sd_statfs_mutex);
 709         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 710         mutex_unlock(&sdp->sd_statfs_mutex);
 711
 712         spin_lock(&sdp->sd_statfs_spin);
 713         m_sc->sc_total += l_sc->sc_total;
 714         m_sc->sc_free += l_sc->sc_free;
 715         m_sc->sc_dinodes += l_sc->sc_dinodes;
 716         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 717         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 718                0, sizeof(struct gfs2_statfs_change));
 719         spin_unlock(&sdp->sd_statfs_spin);
 720
 721         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 722         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 723
 724         gfs2_trans_end(sdp);
 725
 726 out_bh2:
 727         brelse(l_bh);
 728 out_bh:
 729         brelse(m_bh);
 730 out:
 731         gfs2_glock_dq_uninit(&gh);
 732         return error;
 733 }
 734
 735 /**
 736  * gfs2_statfs_i - Do a statfs
 737  * @sdp: the filesystem
 738  * @sg: the sg structure
 739  *
 740  * Returns: errno
 741  */
 742
 743 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 744 {
 745         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 746         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 747
 748         spin_lock(&sdp->sd_statfs_spin);
 749
 750         *sc = *m_sc;
 751         sc->sc_total += l_sc->sc_total;
 752         sc->sc_free += l_sc->sc_free;
 753         sc->sc_dinodes += l_sc->sc_dinodes;
 754
 755         spin_unlock(&sdp->sd_statfs_spin);
 756
 757         if (sc->sc_free < 0)
 758                 sc->sc_free = 0;
 759         if (sc->sc_free > sc->sc_total)
 760                 sc->sc_free = sc->sc_total;
 761         if (sc->sc_dinodes < 0)
 762                 sc->sc_dinodes = 0;
 763
 764         return 0;
 765 }
 766
 767 /**
 768  * statfs_fill - fill in the sg for a given RG
 769  * @rgd: the RG
 770  * @sc: the sc structure
 771  *
 772  * Returns: 0 on success, -ESTALE if the LVB is invalid
 773  */
 774
 775 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 776                             struct gfs2_statfs_change_host *sc)
 777 {
 778         gfs2_rgrp_verify(rgd);
 779         sc->sc_total += rgd->rd_ri.ri_data;
 780         sc->sc_free += rgd->rd_rg.rg_free;
 781         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 782         return 0;
 783 }
 784
 785 /**
 786  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 787  * @sdp: the filesystem
 788  * @sc: the sc info that will be returned
 789  *
 790  * Any error (other than a signal) will cause this routine to fall back
 791  * to the synchronous version.
 792  *
 793  * FIXME: This really shouldn't busy wait like this.
 794  *
 795  * Returns: errno
 796  */
 797
 798 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 799 {
 800         struct gfs2_holder ri_gh;
 801         struct gfs2_rgrpd *rgd_next;
 802         struct gfs2_holder *gha, *gh;
 803         unsigned int slots = 64;
 804         unsigned int x;
 805         int done;
 806         int error = 0, err;
 807
 808         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 809         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 810         if (!gha)
 811                 return -ENOMEM;
 812
 813         error = gfs2_rindex_hold(sdp, &ri_gh);
 814         if (error)
 815                 goto out;
 816
 817         rgd_next = gfs2_rgrpd_get_first(sdp);
 818
 819         for (;;) {
 820                 done = 1;
 821
 822                 for (x = 0; x < slots; x++) {
 823                         gh = gha + x;
 824
 825                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 826                                 err = gfs2_glock_wait(gh);
 827                                 if (err) {
 828                                         gfs2_holder_uninit(gh);
 829                                         error = err;
 830                                 } else {
 831                                         if (!error)
 832                                                 error = statfs_slow_fill(
 833                                                         gh->gh_gl->gl_object, sc);
 834                                         gfs2_glock_dq_uninit(gh);
 835                                 }
 836                         }
 837
 838                         if (gh->gh_gl)
 839                                 done = 0;
 840                         else if (rgd_next && !error) {
 841                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 842                                                            LM_ST_SHARED,
 843                                                            GL_ASYNC,
 844                                                            gh);
 845                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 846                                 done = 0;
 847                         }
 848
 849                         if (signal_pending(current))
 850                                 error = -ERESTARTSYS;
 851                 }
 852
 853                 if (done)
 854                         break;
 855
 856                 yield();
 857         }
 858
 859         gfs2_glock_dq_uninit(&ri_gh);
 860
 861 out:
 862         kfree(gha);
 863         return error;
 864 }
 865
 866 struct lfcc {
 867         struct list_head list;
 868         struct gfs2_holder gh;
 869 };
 870
 871 /**
 872  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 873  *                            journals are clean
 874  * @sdp: the file system
 875  * @state: the state to put the transaction lock into
 876  * @t_gh: the hold on the transaction lock
 877  *
 878  * Returns: errno
 879  */
 880
 881 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 882                                     struct gfs2_holder *t_gh)
 883 {
 884         struct gfs2_inode *ip;
 885         struct gfs2_holder ji_gh;
 886         struct gfs2_jdesc *jd;
 887         struct lfcc *lfcc;
 888         LIST_HEAD(list);
 889         struct gfs2_log_header_host lh;
 890         int error;
 891
 892         error = gfs2_jindex_hold(sdp, &ji_gh);
 893         if (error)
 894                 return error;
 895
 896         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 897                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 898                 if (!lfcc) {
 899                         error = -ENOMEM;
 900                         goto out;
 901                 }
 902                 ip = GFS2_I(jd->jd_inode);
 903                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 904                 if (error) {
 905                         kfree(lfcc);
 906                         goto out;
 907                 }
 908                 list_add(&lfcc->list, &list);
 909         }
 910
 911         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 912                                LM_FLAG_PRIORITY | GL_NOCACHE,
 913                                t_gh);
 914
 915         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 916                 error = gfs2_jdesc_check(jd);
 917                 if (error)
 918                         break;
 919                 error = gfs2_find_jhead(jd, &lh);
 920                 if (error)
 921                         break;
 922                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 923                         error = -EBUSY;
 924                         break;
 925                 }
 926         }
 927
 928         if (error)
 929                 gfs2_glock_dq_uninit(t_gh);
 930
 931 out:
 932         while (!list_empty(&list)) {
 933                 lfcc = list_entry(list.next, struct lfcc, list);
 934                 list_del(&lfcc->list);
 935                 gfs2_glock_dq_uninit(&lfcc->gh);
 936                 kfree(lfcc);
 937         }
 938         gfs2_glock_dq_uninit(&ji_gh);
 939         return error;
 940 }
 941
 942 /**
 943  * gfs2_freeze_fs - freezes the file system
 944  * @sdp: the file system
 945  *
 946  * This function flushes data and meta data for all machines by
 947  * aquiring the transaction log exclusively.  All journals are
 948  * ensured to be in a clean state as well.
 949  *
 950  * Returns: errno
 951  */
 952
 953 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 954 {
 955         int error = 0;
 956
 957         mutex_lock(&sdp->sd_freeze_lock);
 958
 959         if (!sdp->sd_freeze_count++) {
 960                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 961                 if (error)
 962                         sdp->sd_freeze_count--;
 963         }
 964
 965         mutex_unlock(&sdp->sd_freeze_lock);
 966
 967         return error;
 968 }
 969
 970 /**
 971  * gfs2_unfreeze_fs - unfreezes the file system
 972  * @sdp: the file system
 973  *
 974  * This function allows the file system to proceed by unlocking
 975  * the exclusively held transaction lock.  Other GFS2 nodes are
 976  * now free to acquire the lock shared and go on with their lives.
 977  *
 978  */
 979
 980 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 981 {
 982         mutex_lock(&sdp->sd_freeze_lock);
 983
 984         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 985                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 986
 987         mutex_unlock(&sdp->sd_freeze_lock);
 988 }
 989