fs/gfs2/super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 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 v.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 <asm/semaphore.h>
  16
  17 #include "gfs2.h"
  18 #include "bmap.h"
  19 #include "dir.h"
  20 #include "format.h"
  21 #include "glock.h"
  22 #include "glops.h"
  23 #include "inode.h"
  24 #include "log.h"
  25 #include "meta_io.h"
  26 #include "quota.h"
  27 #include "recovery.h"
  28 #include "rgrp.h"
  29 #include "super.h"
  30 #include "trans.h"
  31 #include "unlinked.h"
  32
  33 /**
  34  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  35  * @gt: tune
  36  *
  37  */
  38
  39 void gfs2_tune_init(struct gfs2_tune *gt)
  40 {
  41         spin_lock_init(&gt->gt_spin);
  42
  43         gt->gt_ilimit = 100;
  44         gt->gt_ilimit_tries = 3;
  45         gt->gt_ilimit_min = 1;
  46         gt->gt_demote_secs = 300;
  47         gt->gt_incore_log_blocks = 1024;
  48         gt->gt_log_flush_secs = 60;
  49         gt->gt_jindex_refresh_secs = 60;
  50         gt->gt_scand_secs = 15;
  51         gt->gt_recoverd_secs = 60;
  52         gt->gt_logd_secs = 1;
  53         gt->gt_quotad_secs = 5;
  54         gt->gt_inoded_secs = 15;
  55         gt->gt_quota_simul_sync = 64;
  56         gt->gt_quota_warn_period = 10;
  57         gt->gt_quota_scale_num = 1;
  58         gt->gt_quota_scale_den = 1;
  59         gt->gt_quota_cache_secs = 300;
  60         gt->gt_quota_quantum = 60;
  61         gt->gt_atime_quantum = 3600;
  62         gt->gt_new_files_jdata = 0;
  63         gt->gt_new_files_directio = 0;
  64         gt->gt_max_atomic_write = 4 << 20;
  65         gt->gt_max_readahead = 1 << 18;
  66         gt->gt_lockdump_size = 131072;
  67         gt->gt_stall_secs = 600;
  68         gt->gt_complain_secs = 10;
  69         gt->gt_reclaim_limit = 5000;
  70         gt->gt_entries_per_readdir = 32;
  71         gt->gt_prefetch_secs = 10;
  72         gt->gt_greedy_default = HZ / 10;
  73         gt->gt_greedy_quantum = HZ / 40;
  74         gt->gt_greedy_max = HZ / 4;
  75         gt->gt_statfs_quantum = 30;
  76         gt->gt_statfs_slow = 0;
  77 }
  78
  79 /**
  80  * gfs2_check_sb - Check superblock
  81  * @sdp: the filesystem
  82  * @sb: The superblock
  83  * @silent: Don't print a message if the check fails
  84  *
  85  * Checks the version code of the FS is one that we understand how to
  86  * read and that the sizes of the various on-disk structures have not
  87  * changed.
  88  */
  89
  90 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb *sb, int silent)
  91 {
  92         unsigned int x;
  93
  94         if (sb->sb_header.mh_magic != GFS2_MAGIC ||
  95             sb->sb_header.mh_type != GFS2_METATYPE_SB) {
  96                 if (!silent)
  97                         printk("GFS2: not a GFS2 filesystem\n");
  98                 return -EINVAL;
  99         }
 100
 101         /*  If format numbers match exactly, we're done.  */
 102
 103         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 104             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 105                 return 0;
 106
 107         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 108                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 109                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 110                                 break;
 111
 112                 if (!gfs2_old_fs_formats[x]) {
 113                         printk("GFS2: code version (%u, %u) is incompatible "
 114                                "with ondisk format (%u, %u)\n",
 115                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 116                                sb->sb_fs_format, sb->sb_multihost_format);
 117                         printk("GFS2: I don't know how to upgrade this FS\n");
 118                         return -EINVAL;
 119                 }
 120         }
 121
 122         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 123                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 124                         if (gfs2_old_multihost_formats[x] == sb->sb_multihost_format)
 125                                 break;
 126
 127                 if (!gfs2_old_multihost_formats[x]) {
 128                         printk("GFS2: code version (%u, %u) is incompatible "
 129                                "with ondisk format (%u, %u)\n",
 130                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 131                                sb->sb_fs_format, sb->sb_multihost_format);
 132                         printk("GFS2: I don't know how to upgrade this FS\n");
 133                         return -EINVAL;
 134                 }
 135         }
 136
 137         if (!sdp->sd_args.ar_upgrade) {
 138                 printk("GFS2: code version (%u, %u) is incompatible "
 139                        "with ondisk format (%u, %u)\n",
 140                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 141                        sb->sb_fs_format, sb->sb_multihost_format);
 142                 printk("GFS2: Use the \"upgrade\" mount option to upgrade "
 143                        "the FS\n");
 144                 printk("GFS2: See the manual for more details\n");
 145                 return -EINVAL;
 146         }
 147
 148         return 0;
 149 }
 150
 151 /**
 152  * gfs2_read_sb - Read super block
 153  * @sdp: The GFS2 superblock
 154  * @gl: the glock for the superblock (assumed to be held)
 155  * @silent: Don't print message if mount fails
 156  *
 157  */
 158
 159 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 160 {
 161         struct buffer_head *bh;
 162         uint32_t hash_blocks, ind_blocks, leaf_blocks;
 163         uint32_t tmp_blocks;
 164         unsigned int x;
 165         int error;
 166
 167         error = gfs2_meta_read(gl, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift,
 168                                DIO_FORCE | DIO_START | DIO_WAIT, &bh);
 169         if (error) {
 170                 if (!silent)
 171                         fs_err(sdp, "can't read superblock\n");
 172                 return error;
 173         }
 174
 175         gfs2_assert(sdp, sizeof(struct gfs2_sb) <= bh->b_size);
 176         gfs2_sb_in(&sdp->sd_sb, bh->b_data);
 177         brelse(bh);
 178
 179         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 180         if (error)
 181                 return error;
 182
 183         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 184                                GFS2_BASIC_BLOCK_SHIFT;
 185         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 186         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 187                           sizeof(struct gfs2_dinode)) / sizeof(uint64_t);
 188         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 189                           sizeof(struct gfs2_meta_header)) / sizeof(uint64_t);
 190         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 191         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 192         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 193         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(uint64_t);
 194         sdp->sd_ut_per_block = (sdp->sd_sb.sb_bsize -
 195                                 sizeof(struct gfs2_meta_header)) /
 196                                sizeof(struct gfs2_unlinked_tag);
 197         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 198                                 sizeof(struct gfs2_meta_header)) /
 199                                sizeof(struct gfs2_quota_change);
 200
 201         /* Compute maximum reservation required to add a entry to a directory */
 202
 203         hash_blocks = DIV_RU(sizeof(uint64_t) * (1 << GFS2_DIR_MAX_DEPTH),
 204                              sdp->sd_jbsize);
 205
 206         ind_blocks = 0;
 207         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 208                 tmp_blocks = DIV_RU(tmp_blocks, sdp->sd_inptrs);
 209                 ind_blocks += tmp_blocks;
 210         }
 211
 212         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 213
 214         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 215
 216         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 217                                 sizeof(struct gfs2_dinode);
 218         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 219         for (x = 2;; x++) {
 220                 uint64_t space, d;
 221                 uint32_t m;
 222
 223                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 224                 d = space;
 225                 m = do_div(d, sdp->sd_inptrs);
 226
 227                 if (d != sdp->sd_heightsize[x - 1] || m)
 228                         break;
 229                 sdp->sd_heightsize[x] = space;
 230         }
 231         sdp->sd_max_height = x;
 232         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 233
 234         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 235                                  sizeof(struct gfs2_dinode);
 236         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 237         for (x = 2;; x++) {
 238                 uint64_t space, d;
 239                 uint32_t m;
 240
 241                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 242                 d = space;
 243                 m = do_div(d, sdp->sd_inptrs);
 244
 245                 if (d != sdp->sd_jheightsize[x - 1] || m)
 246                         break;
 247                 sdp->sd_jheightsize[x] = space;
 248         }
 249         sdp->sd_max_jheight = x;
 250         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 251
 252         return 0;
 253 }
 254
 255 int gfs2_do_upgrade(struct gfs2_sbd *sdp, struct gfs2_glock *sb_gl)
 256 {
 257         return 0;
 258 }
 259
 260 /**
 261  * gfs2_jindex_hold - Grab a lock on the jindex
 262  * @sdp: The GFS2 superblock
 263  * @ji_gh: the holder for the jindex glock
 264  *
 265  * This is very similar to the gfs2_rindex_hold() function, except that
 266  * in general we hold the jindex lock for longer periods of time and
 267  * we grab it far less frequently (in general) then the rgrp lock.
 268  *
 269  * Returns: errno
 270  */
 271
 272 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 273 {
 274         struct gfs2_inode *dip = sdp->sd_jindex;
 275         struct qstr name;
 276         char buf[20];
 277         struct gfs2_jdesc *jd;
 278         int error;
 279
 280         name.name = buf;
 281
 282         down(&sdp->sd_jindex_mutex);
 283
 284         for (;;) {
 285                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED,
 286                                            GL_LOCAL_EXCL, ji_gh);
 287                 if (error)
 288                         break;
 289
 290                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 291
 292                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 293                 if (error == -ENOENT) {
 294                         error = 0;
 295                         break;
 296                 }
 297
 298                 gfs2_glock_dq_uninit(ji_gh);
 299
 300                 if (error)
 301                         break;
 302
 303                 error = -ENOMEM;
 304                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 305                 if (!jd)
 306                         break;
 307
 308                 error = gfs2_lookupi(dip, &name, 1, &jd->jd_inode);
 309                 if (error) {
 310                         kfree(jd);
 311                         break;
 312                 }
 313
 314                 spin_lock(&sdp->sd_jindex_spin);
 315                 jd->jd_jid = sdp->sd_journals++;
 316                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 317                 spin_unlock(&sdp->sd_jindex_spin);
 318         }
 319
 320         up(&sdp->sd_jindex_mutex);
 321
 322         return error;
 323 }
 324
 325 /**
 326  * gfs2_jindex_free - Clear all the journal index information
 327  * @sdp: The GFS2 superblock
 328  *
 329  */
 330
 331 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 332 {
 333         struct list_head list;
 334         struct gfs2_jdesc *jd;
 335
 336         spin_lock(&sdp->sd_jindex_spin);
 337         list_add(&list, &sdp->sd_jindex_list);
 338         list_del_init(&sdp->sd_jindex_list);
 339         sdp->sd_journals = 0;
 340         spin_unlock(&sdp->sd_jindex_spin);
 341
 342         while (!list_empty(&list)) {
 343                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 344                 list_del(&jd->jd_list);
 345                 gfs2_inode_put(jd->jd_inode);
 346                 kfree(jd);
 347         }
 348 }
 349
 350 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 351 {
 352         struct gfs2_jdesc *jd;
 353         int found = 0;
 354
 355         list_for_each_entry(jd, head, jd_list) {
 356                 if (jd->jd_jid == jid) {
 357                         found = 1;
 358                         break;
 359                 }
 360         }
 361
 362         if (!found)
 363                 jd = NULL;
 364
 365         return jd;
 366 }
 367
 368 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 369 {
 370         struct gfs2_jdesc *jd;
 371
 372         spin_lock(&sdp->sd_jindex_spin);
 373         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 374         spin_unlock(&sdp->sd_jindex_spin);
 375
 376         return jd;
 377 }
 378
 379 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 380 {
 381         struct gfs2_jdesc *jd;
 382
 383         spin_lock(&sdp->sd_jindex_spin);
 384         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 385         if (jd)
 386                 jd->jd_dirty = 1;
 387         spin_unlock(&sdp->sd_jindex_spin);
 388 }
 389
 390 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 391 {
 392         struct gfs2_jdesc *jd;
 393         int found = 0;
 394
 395         spin_lock(&sdp->sd_jindex_spin);
 396
 397         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 398                 if (jd->jd_dirty) {
 399                         jd->jd_dirty = 0;
 400                         found = 1;
 401                         break;
 402                 }
 403         }
 404         spin_unlock(&sdp->sd_jindex_spin);
 405
 406         if (!found)
 407                 jd = NULL;
 408
 409         return jd;
 410 }
 411
 412 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 413 {
 414         struct gfs2_inode *ip = jd->jd_inode;
 415         struct gfs2_sbd *sdp = ip->i_sbd;
 416         int ar;
 417         int error;
 418
 419         if (ip->i_di.di_size < (8 << 20) ||
 420             ip->i_di.di_size > (1 << 30) ||
 421             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 422                 gfs2_consist_inode(ip);
 423                 return -EIO;
 424         }
 425         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 426
 427         error = gfs2_write_alloc_required(ip,
 428                                           0, ip->i_di.di_size,
 429                                           &ar);
 430         if (!error && ar) {
 431                 gfs2_consist_inode(ip);
 432                 error = -EIO;
 433         }
 434
 435         return error;
 436 }
 437
 438 int gfs2_lookup_master_dir(struct gfs2_sbd *sdp)
 439 {
 440         struct gfs2_glock *gl;
 441         int error;
 442
 443         error = gfs2_glock_get(sdp,
 444                                sdp->sd_sb.sb_master_dir.no_addr,
 445                                &gfs2_inode_glops, CREATE, &gl);
 446         if (!error) {
 447                 error = gfs2_inode_get(gl, &sdp->sd_sb.sb_master_dir, CREATE,
 448                                        &sdp->sd_master_dir);
 449                 gfs2_glock_put(gl);
 450         }
 451
 452         return error;
 453 }
 454
 455 /**
 456  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 457  * @sdp: the filesystem
 458  *
 459  * Returns: errno
 460  */
 461
 462 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 463 {
 464         struct gfs2_glock *j_gl = sdp->sd_jdesc->jd_inode->i_gl;
 465         struct gfs2_holder t_gh;
 466         struct gfs2_log_header head;
 467         int error;
 468
 469         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 470                                    GL_LOCAL_EXCL | GL_NEVER_RECURSE, &t_gh);
 471         if (error)
 472                 return error;
 473
 474         gfs2_meta_cache_flush(sdp->sd_jdesc->jd_inode);
 475         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA | DIO_DATA);
 476
 477         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 478         if (error)
 479                 goto fail;
 480
 481         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 482                 gfs2_consist(sdp);
 483                 error = -EIO;
 484                 goto fail;
 485         }
 486
 487         /*  Initialize some head of the log stuff  */
 488         sdp->sd_log_sequence = head.lh_sequence + 1;
 489         gfs2_log_pointers_init(sdp, head.lh_blkno);
 490
 491         error = gfs2_unlinked_init(sdp);
 492         if (error)
 493                 goto fail;
 494         error = gfs2_quota_init(sdp);
 495         if (error)
 496                 goto fail_unlinked;
 497
 498         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 499
 500         gfs2_glock_dq_uninit(&t_gh);
 501
 502         return 0;
 503
 504  fail_unlinked:
 505         gfs2_unlinked_cleanup(sdp);
 506
 507  fail:
 508         t_gh.gh_flags |= GL_NOCACHE;
 509         gfs2_glock_dq_uninit(&t_gh);
 510
 511         return error;
 512 }
 513
 514 /**
 515  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 516  * @sdp: the filesystem
 517  *
 518  * Returns: errno
 519  */
 520
 521 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 522 {
 523         struct gfs2_holder t_gh;
 524         int error;
 525
 526         gfs2_unlinked_dealloc(sdp);
 527         gfs2_quota_sync(sdp);
 528         gfs2_statfs_sync(sdp);
 529
 530         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 531                                 GL_LOCAL_EXCL | GL_NEVER_RECURSE | GL_NOCACHE,
 532                                 &t_gh);
 533         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 534                 return error;
 535
 536         gfs2_meta_syncfs(sdp);
 537         gfs2_log_shutdown(sdp);
 538
 539         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 540
 541         if (t_gh.gh_gl)
 542                 gfs2_glock_dq_uninit(&t_gh);
 543
 544         gfs2_unlinked_cleanup(sdp);
 545         gfs2_quota_cleanup(sdp);
 546
 547         return error;
 548 }
 549
 550 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 551 {
 552         struct gfs2_inode *m_ip = sdp->sd_statfs_inode;
 553         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 554         struct gfs2_inode *l_ip = sdp->sd_sc_inode;
 555         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 556         struct buffer_head *m_bh, *l_bh;
 557         struct gfs2_holder gh;
 558         int error;
 559
 560         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 561                                    &gh);
 562         if (error)
 563                 return error;
 564
 565         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 566         if (error)
 567                 goto out;
 568
 569         if (sdp->sd_args.ar_spectator) {
 570                 spin_lock(&sdp->sd_statfs_spin);
 571                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 572                                       sizeof(struct gfs2_dinode));
 573                 spin_unlock(&sdp->sd_statfs_spin);
 574         } else {
 575                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 576                 if (error)
 577                         goto out_m_bh;
 578
 579                 spin_lock(&sdp->sd_statfs_spin);
 580                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 581                                       sizeof(struct gfs2_dinode));
 582                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 583                                       sizeof(struct gfs2_dinode));
 584                 spin_unlock(&sdp->sd_statfs_spin);
 585
 586                 brelse(l_bh);
 587         }
 588
 589  out_m_bh:
 590         brelse(m_bh);
 591
 592  out:
 593         gfs2_glock_dq_uninit(&gh);
 594
 595         return 0;
 596 }
 597
 598 void gfs2_statfs_change(struct gfs2_sbd *sdp, int64_t total, int64_t free,
 599                         int64_t dinodes)
 600 {
 601         struct gfs2_inode *l_ip = sdp->sd_sc_inode;
 602         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 603         struct buffer_head *l_bh;
 604         int error;
 605
 606         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 607         if (error)
 608                 return;
 609
 610         down(&sdp->sd_statfs_mutex);
 611         gfs2_trans_add_bh(l_ip->i_gl, l_bh);
 612         up(&sdp->sd_statfs_mutex);
 613
 614         spin_lock(&sdp->sd_statfs_spin);
 615         l_sc->sc_total += total;
 616         l_sc->sc_free += free;
 617         l_sc->sc_dinodes += dinodes;
 618         gfs2_statfs_change_out(l_sc, l_bh->b_data +
 619                                sizeof(struct gfs2_dinode));
 620         spin_unlock(&sdp->sd_statfs_spin);
 621
 622         brelse(l_bh);
 623 }
 624
 625 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 626 {
 627         struct gfs2_inode *m_ip = sdp->sd_statfs_inode;
 628         struct gfs2_inode *l_ip = sdp->sd_sc_inode;
 629         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 630         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 631         struct gfs2_holder gh;
 632         struct buffer_head *m_bh, *l_bh;
 633         int error;
 634
 635         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 636                                    &gh);
 637         if (error)
 638                 return error;
 639
 640         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 641         if (error)
 642                 goto out;
 643
 644         spin_lock(&sdp->sd_statfs_spin);
 645         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 646                               sizeof(struct gfs2_dinode));
 647         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 648                 spin_unlock(&sdp->sd_statfs_spin);
 649                 goto out_bh;
 650         }
 651         spin_unlock(&sdp->sd_statfs_spin);
 652
 653         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 654         if (error)
 655                 goto out_bh;
 656
 657         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 658         if (error)
 659                 goto out_bh2;
 660
 661         down(&sdp->sd_statfs_mutex);
 662         gfs2_trans_add_bh(l_ip->i_gl, l_bh);
 663         up(&sdp->sd_statfs_mutex);
 664
 665         spin_lock(&sdp->sd_statfs_spin);
 666         m_sc->sc_total += l_sc->sc_total;
 667         m_sc->sc_free += l_sc->sc_free;
 668         m_sc->sc_dinodes += l_sc->sc_dinodes;
 669         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 670         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 671                0, sizeof(struct gfs2_statfs_change));
 672         spin_unlock(&sdp->sd_statfs_spin);
 673
 674         gfs2_trans_add_bh(m_ip->i_gl, m_bh);
 675         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 676
 677         gfs2_trans_end(sdp);
 678
 679  out_bh2:
 680         brelse(l_bh);
 681
 682  out_bh:
 683         brelse(m_bh);
 684
 685  out:
 686         gfs2_glock_dq_uninit(&gh);
 687
 688         return error;
 689 }
 690
 691 /**
 692  * gfs2_statfs_i - Do a statfs
 693  * @sdp: the filesystem
 694  * @sg: the sg structure
 695  *
 696  * Returns: errno
 697  */
 698
 699 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change *sc)
 700 {
 701         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 702         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 703
 704         spin_lock(&sdp->sd_statfs_spin);
 705
 706         *sc = *m_sc;
 707         sc->sc_total += l_sc->sc_total;
 708         sc->sc_free += l_sc->sc_free;
 709         sc->sc_dinodes += l_sc->sc_dinodes;
 710
 711         spin_unlock(&sdp->sd_statfs_spin);
 712
 713         if (sc->sc_free < 0)
 714                 sc->sc_free = 0;
 715         if (sc->sc_free > sc->sc_total)
 716                 sc->sc_free = sc->sc_total;
 717         if (sc->sc_dinodes < 0)
 718                 sc->sc_dinodes = 0;
 719
 720         return 0;
 721 }
 722
 723 /**
 724  * statfs_fill - fill in the sg for a given RG
 725  * @rgd: the RG
 726  * @sc: the sc structure
 727  *
 728  * Returns: 0 on success, -ESTALE if the LVB is invalid
 729  */
 730
 731 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 732                             struct gfs2_statfs_change *sc)
 733 {
 734         gfs2_rgrp_verify(rgd);
 735         sc->sc_total += rgd->rd_ri.ri_data;
 736         sc->sc_free += rgd->rd_rg.rg_free;
 737         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 738         return 0;
 739 }
 740
 741 /**
 742  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 743  * @sdp: the filesystem
 744  * @sc: the sc info that will be returned
 745  *
 746  * Any error (other than a signal) will cause this routine to fall back
 747  * to the synchronous version.
 748  *
 749  * FIXME: This really shouldn't busy wait like this.
 750  *
 751  * Returns: errno
 752  */
 753
 754 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change *sc)
 755 {
 756         struct gfs2_holder ri_gh;
 757         struct gfs2_rgrpd *rgd_next;
 758         struct gfs2_holder *gha, *gh;
 759         unsigned int slots = 64;
 760         unsigned int x;
 761         int done;
 762         int error = 0, err;
 763
 764         memset(sc, 0, sizeof(struct gfs2_statfs_change));
 765         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 766         if (!gha)
 767                 return -ENOMEM;
 768
 769         error = gfs2_rindex_hold(sdp, &ri_gh);
 770         if (error)
 771                 goto out;
 772
 773         rgd_next = gfs2_rgrpd_get_first(sdp);
 774
 775         for (;;) {
 776                 done = 1;
 777
 778                 for (x = 0; x < slots; x++) {
 779                         gh = gha + x;
 780
 781                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 782                                 err = gfs2_glock_wait(gh);
 783                                 if (err) {
 784                                         gfs2_holder_uninit(gh);
 785                                         error = err;
 786                                 } else {
 787                                         if (!error)
 788                                                 error = statfs_slow_fill(get_gl2rgd(gh->gh_gl), sc);
 789                                         gfs2_glock_dq_uninit(gh);
 790                                 }
 791                         }
 792
 793                         if (gh->gh_gl)
 794                                 done = 0;
 795                         else if (rgd_next && !error) {
 796                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 797                                                            LM_ST_SHARED,
 798                                                            GL_ASYNC,
 799                                                            gh);
 800                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 801                                 done = 0;
 802                         }
 803
 804                         if (signal_pending(current))
 805                                 error = -ERESTARTSYS;
 806                 }
 807
 808                 if (done)
 809                         break;
 810
 811                 yield();
 812         }
 813
 814         gfs2_glock_dq_uninit(&ri_gh);
 815
 816  out:
 817         kfree(gha);
 818
 819         return error;
 820 }
 821
 822 struct lfcc {
 823         struct list_head list;
 824         struct gfs2_holder gh;
 825 };
 826
 827 /**
 828  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 829  *                            journals are clean
 830  * @sdp: the file system
 831  * @state: the state to put the transaction lock into
 832  * @t_gh: the hold on the transaction lock
 833  *
 834  * Returns: errno
 835  */
 836
 837 int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp, struct gfs2_holder *t_gh)
 838 {
 839         struct gfs2_holder ji_gh;
 840         struct gfs2_jdesc *jd;
 841         struct lfcc *lfcc;
 842         LIST_HEAD(list);
 843         struct gfs2_log_header lh;
 844         int error;
 845
 846         error = gfs2_jindex_hold(sdp, &ji_gh);
 847         if (error)
 848                 return error;
 849
 850         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 851                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 852                 if (!lfcc) {
 853                         error = -ENOMEM;
 854                         goto out;
 855                 }
 856                 error = gfs2_glock_nq_init(jd->jd_inode->i_gl, LM_ST_SHARED, 0,
 857                                            &lfcc->gh);
 858                 if (error) {
 859                         kfree(lfcc);
 860                         goto out;
 861                 }
 862                 list_add(&lfcc->list, &list);
 863         }
 864
 865         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 866                                LM_FLAG_PRIORITY | GL_NEVER_RECURSE | GL_NOCACHE,
 867                                t_gh);
 868
 869         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 870                 error = gfs2_jdesc_check(jd);
 871                 if (error)
 872                         break;
 873                 error = gfs2_find_jhead(jd, &lh);
 874                 if (error)
 875                         break;
 876                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 877                         error = -EBUSY;
 878                         break;
 879                 }
 880         }
 881
 882         if (error)
 883                 gfs2_glock_dq_uninit(t_gh);
 884
 885  out:
 886         while (!list_empty(&list)) {
 887                 lfcc = list_entry(list.next, struct lfcc, list);
 888                 list_del(&lfcc->list);
 889                 gfs2_glock_dq_uninit(&lfcc->gh);
 890                 kfree(lfcc);
 891         }
 892         gfs2_glock_dq_uninit(&ji_gh);
 893
 894         return error;
 895 }
 896
 897 /**
 898  * gfs2_freeze_fs - freezes the file system
 899  * @sdp: the file system
 900  *
 901  * This function flushes data and meta data for all machines by
 902  * aquiring the transaction log exclusively.  All journals are
 903  * ensured to be in a clean state as well.
 904  *
 905  * Returns: errno
 906  */
 907
 908 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 909 {
 910         int error = 0;
 911
 912         down(&sdp->sd_freeze_lock);
 913
 914         if (!sdp->sd_freeze_count++) {
 915                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 916                 if (error)
 917                         sdp->sd_freeze_count--;
 918         }
 919
 920         up(&sdp->sd_freeze_lock);
 921
 922         return error;
 923 }
 924
 925 /**
 926  * gfs2_unfreeze_fs - unfreezes the file system
 927  * @sdp: the file system
 928  *
 929  * This function allows the file system to proceed by unlocking
 930  * the exclusively held transaction lock.  Other GFS2 nodes are
 931  * now free to acquire the lock shared and go on with their lives.
 932  *
 933  */
 934
 935 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 936 {
 937         down(&sdp->sd_freeze_lock);
 938
 939         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 940                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 941
 942         up(&sdp->sd_freeze_lock);
 943 }
 944