git.oblomov.eu Git - linux-2.6/blob - fs/gfs2/super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 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_demote_secs = 300;
  55         gt->gt_incore_log_blocks = 1024;
  56         gt->gt_log_flush_secs = 60;
  57         gt->gt_recoverd_secs = 60;
  58         gt->gt_logd_secs = 1;
  59         gt->gt_quotad_secs = 5;
  60         gt->gt_quota_simul_sync = 64;
  61         gt->gt_quota_warn_period = 10;
  62         gt->gt_quota_scale_num = 1;
  63         gt->gt_quota_scale_den = 1;
  64         gt->gt_quota_cache_secs = 300;
  65         gt->gt_quota_quantum = 60;
  66         gt->gt_atime_quantum = 3600;
  67         gt->gt_new_files_jdata = 0;
  68         gt->gt_new_files_directio = 0;
  69         gt->gt_max_readahead = 1 << 18;
  70         gt->gt_stall_secs = 600;
  71         gt->gt_complain_secs = 10;
  72         gt->gt_statfs_quantum = 30;
  73         gt->gt_statfs_slow = 0;
  74 }
  75
  76 /**
  77  * gfs2_check_sb - Check superblock
  78  * @sdp: the filesystem
  79  * @sb: The superblock
  80  * @silent: Don't print a message if the check fails
  81  *
  82  * Checks the version code of the FS is one that we understand how to
  83  * read and that the sizes of the various on-disk structures have not
  84  * changed.
  85  */
  86
  87 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  88 {
  89         unsigned int x;
  90
  91         if (sb->sb_magic != GFS2_MAGIC ||
  92             sb->sb_type != GFS2_METATYPE_SB) {
  93                 if (!silent)
  94                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
  95                 return -EINVAL;
  96         }
  97
  98         /*  If format numbers match exactly, we're done.  */
  99
 100         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 101             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 102                 return 0;
 103
 104         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 105                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 106                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 107                                 break;
 108
 109                 if (!gfs2_old_fs_formats[x]) {
 110                         printk(KERN_WARNING
 111                                "GFS2: code version (%u, %u) is incompatible "
 112                                "with ondisk format (%u, %u)\n",
 113                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 114                                sb->sb_fs_format, sb->sb_multihost_format);
 115                         printk(KERN_WARNING
 116                                "GFS2: I don't know how to upgrade this FS\n");
 117                         return -EINVAL;
 118                 }
 119         }
 120
 121         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 122                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 123                         if (gfs2_old_multihost_formats[x] ==
 124                             sb->sb_multihost_format)
 125                                 break;
 126
 127                 if (!gfs2_old_multihost_formats[x]) {
 128                         printk(KERN_WARNING
 129                                "GFS2: code version (%u, %u) is incompatible "
 130                                "with ondisk format (%u, %u)\n",
 131                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 132                                sb->sb_fs_format, sb->sb_multihost_format);
 133                         printk(KERN_WARNING
 134                                "GFS2: I don't know how to upgrade this FS\n");
 135                         return -EINVAL;
 136                 }
 137         }
 138
 139         if (!sdp->sd_args.ar_upgrade) {
 140                 printk(KERN_WARNING
 141                        "GFS2: code version (%u, %u) is incompatible "
 142                        "with ondisk format (%u, %u)\n",
 143                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 144                        sb->sb_fs_format, sb->sb_multihost_format);
 145                 printk(KERN_INFO
 146                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 147                        "the FS\n");
 148                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 149                 return -EINVAL;
 150         }
 151
 152         return 0;
 153 }
 154
 155
 156 static void end_bio_io_page(struct bio *bio, int error)
 157 {
 158         struct page *page = bio->bi_private;
 159
 160         if (!error)
 161                 SetPageUptodate(page);
 162         else
 163                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 164         unlock_page(page);
 165 }
 166
 167 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 168 {
 169         const struct gfs2_sb *str = buf;
 170
 171         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 172         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 173         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 174         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 175         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 176         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 177         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 178         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 179         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 180         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 181         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 182
 183         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 184         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 185 }
 186
 187 /**
 188  * gfs2_read_super - Read the gfs2 super block from disk
 189  * @sdp: The GFS2 super block
 190  * @sector: The location of the super block
 191  * @error: The error code to return
 192  *
 193  * This uses the bio functions to read the super block from disk
 194  * because we want to be 100% sure that we never read cached data.
 195  * A super block is read twice only during each GFS2 mount and is
 196  * never written to by the filesystem. The first time its read no
 197  * locks are held, and the only details which are looked at are those
 198  * relating to the locking protocol. Once locking is up and working,
 199  * the sb is read again under the lock to establish the location of
 200  * the master directory (contains pointers to journals etc) and the
 201  * root directory.
 202  *
 203  * Returns: 0 on success or error
 204  */
 205
 206 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 207 {
 208         struct super_block *sb = sdp->sd_vfs;
 209         struct gfs2_sb *p;
 210         struct page *page;
 211         struct bio *bio;
 212
 213         page = alloc_page(GFP_NOFS);
 214         if (unlikely(!page))
 215                 return -ENOBUFS;
 216
 217         ClearPageUptodate(page);
 218         ClearPageDirty(page);
 219         lock_page(page);
 220
 221         bio = bio_alloc(GFP_NOFS, 1);
 222         if (unlikely(!bio)) {
 223                 __free_page(page);
 224                 return -ENOBUFS;
 225         }
 226
 227         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 228         bio->bi_bdev = sb->s_bdev;
 229         bio_add_page(bio, page, PAGE_SIZE, 0);
 230
 231         bio->bi_end_io = end_bio_io_page;
 232         bio->bi_private = page;
 233         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 234         wait_on_page_locked(page);
 235         bio_put(bio);
 236         if (!PageUptodate(page)) {
 237                 __free_page(page);
 238                 return -EIO;
 239         }
 240         p = kmap(page);
 241         gfs2_sb_in(&sdp->sd_sb, p);
 242         kunmap(page);
 243         __free_page(page);
 244         return 0;
 245 }
 246
 247 /**
 248  * gfs2_read_sb - Read super block
 249  * @sdp: The GFS2 superblock
 250  * @gl: the glock for the superblock (assumed to be held)
 251  * @silent: Don't print message if mount fails
 252  *
 253  */
 254
 255 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 256 {
 257         u32 hash_blocks, ind_blocks, leaf_blocks;
 258         u32 tmp_blocks;
 259         unsigned int x;
 260         int error;
 261
 262         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 263         if (error) {
 264                 if (!silent)
 265                         fs_err(sdp, "can't read superblock\n");
 266                 return error;
 267         }
 268
 269         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 270         if (error)
 271                 return error;
 272
 273         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 274                                GFS2_BASIC_BLOCK_SHIFT;
 275         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 276         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 277                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 278         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 279                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 280         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 281         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 282         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 283         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 284         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 285                                 sizeof(struct gfs2_meta_header)) /
 286                                 sizeof(struct gfs2_quota_change);
 287
 288         /* Compute maximum reservation required to add a entry to a directory */
 289
 290         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 291                              sdp->sd_jbsize);
 292
 293         ind_blocks = 0;
 294         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 295                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 296                 ind_blocks += tmp_blocks;
 297         }
 298
 299         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 300
 301         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 302
 303         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 304                                 sizeof(struct gfs2_dinode);
 305         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 306         for (x = 2;; x++) {
 307                 u64 space, d;
 308                 u32 m;
 309
 310                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 311                 d = space;
 312                 m = do_div(d, sdp->sd_inptrs);
 313
 314                 if (d != sdp->sd_heightsize[x - 1] || m)
 315                         break;
 316                 sdp->sd_heightsize[x] = space;
 317         }
 318         sdp->sd_max_height = x;
 319         sdp->sd_heightsize[x] = ~0;
 320         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 321
 322         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 323                                  sizeof(struct gfs2_dinode);
 324         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 325         for (x = 2;; x++) {
 326                 u64 space, d;
 327                 u32 m;
 328
 329                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 330                 d = space;
 331                 m = do_div(d, sdp->sd_inptrs);
 332
 333                 if (d != sdp->sd_jheightsize[x - 1] || m)
 334                         break;
 335                 sdp->sd_jheightsize[x] = space;
 336         }
 337         sdp->sd_max_jheight = x;
 338         sdp->sd_jheightsize[x] = ~0;
 339         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 340
 341         return 0;
 342 }
 343
 344 /**
 345  * gfs2_jindex_hold - Grab a lock on the jindex
 346  * @sdp: The GFS2 superblock
 347  * @ji_gh: the holder for the jindex glock
 348  *
 349  * This is very similar to the gfs2_rindex_hold() function, except that
 350  * in general we hold the jindex lock for longer periods of time and
 351  * we grab it far less frequently (in general) then the rgrp lock.
 352  *
 353  * Returns: errno
 354  */
 355
 356 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 357 {
 358         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 359         struct qstr name;
 360         char buf[20];
 361         struct gfs2_jdesc *jd;
 362         int error;
 363
 364         name.name = buf;
 365
 366         mutex_lock(&sdp->sd_jindex_mutex);
 367
 368         for (;;) {
 369                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 370                 if (error)
 371                         break;
 372
 373                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 374                 name.hash = gfs2_disk_hash(name.name, name.len);
 375
 376                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 377                 if (error == -ENOENT) {
 378                         error = 0;
 379                         break;
 380                 }
 381
 382                 gfs2_glock_dq_uninit(ji_gh);
 383
 384                 if (error)
 385                         break;
 386
 387                 error = -ENOMEM;
 388                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 389                 if (!jd)
 390                         break;
 391
 392                 INIT_LIST_HEAD(&jd->extent_list);
 393                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 394                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 395                         if (!jd->jd_inode)
 396                                 error = -ENOENT;
 397                         else
 398                                 error = PTR_ERR(jd->jd_inode);
 399                         kfree(jd);
 400                         break;
 401                 }
 402
 403                 spin_lock(&sdp->sd_jindex_spin);
 404                 jd->jd_jid = sdp->sd_journals++;
 405                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 406                 spin_unlock(&sdp->sd_jindex_spin);
 407         }
 408
 409         mutex_unlock(&sdp->sd_jindex_mutex);
 410
 411         return error;
 412 }
 413
 414 /**
 415  * gfs2_jindex_free - Clear all the journal index information
 416  * @sdp: The GFS2 superblock
 417  *
 418  */
 419
 420 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 421 {
 422         struct list_head list, *head;
 423         struct gfs2_jdesc *jd;
 424         struct gfs2_journal_extent *jext;
 425
 426         spin_lock(&sdp->sd_jindex_spin);
 427         list_add(&list, &sdp->sd_jindex_list);
 428         list_del_init(&sdp->sd_jindex_list);
 429         sdp->sd_journals = 0;
 430         spin_unlock(&sdp->sd_jindex_spin);
 431
 432         while (!list_empty(&list)) {
 433                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 434                 head = &jd->extent_list;
 435                 while (!list_empty(head)) {
 436                         jext = list_entry(head->next,
 437                                           struct gfs2_journal_extent,
 438                                           extent_list);
 439                         list_del(&jext->extent_list);
 440                         kfree(jext);
 441                 }
 442                 list_del(&jd->jd_list);
 443                 iput(jd->jd_inode);
 444                 kfree(jd);
 445         }
 446 }
 447
 448 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 449 {
 450         struct gfs2_jdesc *jd;
 451         int found = 0;
 452
 453         list_for_each_entry(jd, head, jd_list) {
 454                 if (jd->jd_jid == jid) {
 455                         found = 1;
 456                         break;
 457                 }
 458         }
 459
 460         if (!found)
 461                 jd = NULL;
 462
 463         return jd;
 464 }
 465
 466 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 467 {
 468         struct gfs2_jdesc *jd;
 469
 470         spin_lock(&sdp->sd_jindex_spin);
 471         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 472         spin_unlock(&sdp->sd_jindex_spin);
 473
 474         return jd;
 475 }
 476
 477 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 478 {
 479         struct gfs2_jdesc *jd;
 480
 481         spin_lock(&sdp->sd_jindex_spin);
 482         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 483         if (jd)
 484                 jd->jd_dirty = 1;
 485         spin_unlock(&sdp->sd_jindex_spin);
 486 }
 487
 488 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 489 {
 490         struct gfs2_jdesc *jd;
 491         int found = 0;
 492
 493         spin_lock(&sdp->sd_jindex_spin);
 494
 495         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 496                 if (jd->jd_dirty) {
 497                         jd->jd_dirty = 0;
 498                         found = 1;
 499                         break;
 500                 }
 501         }
 502         spin_unlock(&sdp->sd_jindex_spin);
 503
 504         if (!found)
 505                 jd = NULL;
 506
 507         return jd;
 508 }
 509
 510 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 511 {
 512         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 513         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 514         int ar;
 515         int error;
 516
 517         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 518             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 519                 gfs2_consist_inode(ip);
 520                 return -EIO;
 521         }
 522         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 523
 524         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 525         if (!error && ar) {
 526                 gfs2_consist_inode(ip);
 527                 error = -EIO;
 528         }
 529
 530         return error;
 531 }
 532
 533 /**
 534  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 535  * @sdp: the filesystem
 536  *
 537  * Returns: errno
 538  */
 539
 540 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 541 {
 542         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 543         struct gfs2_glock *j_gl = ip->i_gl;
 544         struct gfs2_holder t_gh;
 545         struct gfs2_log_header_host head;
 546         int error;
 547
 548         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 549         if (error)
 550                 return error;
 551
 552         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 553
 554         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 555         if (error)
 556                 goto fail;
 557
 558         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 559                 gfs2_consist(sdp);
 560                 error = -EIO;
 561                 goto fail;
 562         }
 563
 564         /*  Initialize some head of the log stuff  */
 565         sdp->sd_log_sequence = head.lh_sequence + 1;
 566         gfs2_log_pointers_init(sdp, head.lh_blkno);
 567
 568         error = gfs2_quota_init(sdp);
 569         if (error)
 570                 goto fail;
 571
 572         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 573
 574         gfs2_glock_dq_uninit(&t_gh);
 575
 576         return 0;
 577
 578 fail:
 579         t_gh.gh_flags |= GL_NOCACHE;
 580         gfs2_glock_dq_uninit(&t_gh);
 581
 582         return error;
 583 }
 584
 585 /**
 586  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 587  * @sdp: the filesystem
 588  *
 589  * Returns: errno
 590  */
 591
 592 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 593 {
 594         struct gfs2_holder t_gh;
 595         int error;
 596
 597         gfs2_quota_sync(sdp);
 598         gfs2_statfs_sync(sdp);
 599
 600         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 601                                    &t_gh);
 602         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 603                 return error;
 604
 605         gfs2_meta_syncfs(sdp);
 606         gfs2_log_shutdown(sdp);
 607
 608         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 609
 610         if (t_gh.gh_gl)
 611                 gfs2_glock_dq_uninit(&t_gh);
 612
 613         gfs2_quota_cleanup(sdp);
 614
 615         return error;
 616 }
 617
 618 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 619 {
 620         const struct gfs2_statfs_change *str = buf;
 621
 622         sc->sc_total = be64_to_cpu(str->sc_total);
 623         sc->sc_free = be64_to_cpu(str->sc_free);
 624         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 625 }
 626
 627 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 628 {
 629         struct gfs2_statfs_change *str = buf;
 630
 631         str->sc_total = cpu_to_be64(sc->sc_total);
 632         str->sc_free = cpu_to_be64(sc->sc_free);
 633         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 634 }
 635
 636 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 637 {
 638         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 639         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 640         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 641         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 642         struct buffer_head *m_bh, *l_bh;
 643         struct gfs2_holder gh;
 644         int error;
 645
 646         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 647                                    &gh);
 648         if (error)
 649                 return error;
 650
 651         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 652         if (error)
 653                 goto out;
 654
 655         if (sdp->sd_args.ar_spectator) {
 656                 spin_lock(&sdp->sd_statfs_spin);
 657                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 658                                       sizeof(struct gfs2_dinode));
 659                 spin_unlock(&sdp->sd_statfs_spin);
 660         } else {
 661                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 662                 if (error)
 663                         goto out_m_bh;
 664
 665                 spin_lock(&sdp->sd_statfs_spin);
 666                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 667                                       sizeof(struct gfs2_dinode));
 668                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 669                                       sizeof(struct gfs2_dinode));
 670                 spin_unlock(&sdp->sd_statfs_spin);
 671
 672                 brelse(l_bh);
 673         }
 674
 675 out_m_bh:
 676         brelse(m_bh);
 677 out:
 678         gfs2_glock_dq_uninit(&gh);
 679         return 0;
 680 }
 681
 682 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 683                         s64 dinodes)
 684 {
 685         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 686         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 687         struct buffer_head *l_bh;
 688         int error;
 689
 690         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 691         if (error)
 692                 return;
 693
 694         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 695
 696         spin_lock(&sdp->sd_statfs_spin);
 697         l_sc->sc_total += total;
 698         l_sc->sc_free += free;
 699         l_sc->sc_dinodes += dinodes;
 700         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 701         spin_unlock(&sdp->sd_statfs_spin);
 702
 703         brelse(l_bh);
 704 }
 705
 706 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 707 {
 708         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 709         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 710         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 711         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 712         struct gfs2_holder gh;
 713         struct buffer_head *m_bh, *l_bh;
 714         int error;
 715
 716         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 717                                    &gh);
 718         if (error)
 719                 return error;
 720
 721         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 722         if (error)
 723                 goto out;
 724
 725         spin_lock(&sdp->sd_statfs_spin);
 726         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 727                               sizeof(struct gfs2_dinode));
 728         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 729                 spin_unlock(&sdp->sd_statfs_spin);
 730                 goto out_bh;
 731         }
 732         spin_unlock(&sdp->sd_statfs_spin);
 733
 734         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 735         if (error)
 736                 goto out_bh;
 737
 738         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 739         if (error)
 740                 goto out_bh2;
 741
 742         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 743
 744         spin_lock(&sdp->sd_statfs_spin);
 745         m_sc->sc_total += l_sc->sc_total;
 746         m_sc->sc_free += l_sc->sc_free;
 747         m_sc->sc_dinodes += l_sc->sc_dinodes;
 748         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 749         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 750                0, sizeof(struct gfs2_statfs_change));
 751         spin_unlock(&sdp->sd_statfs_spin);
 752
 753         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 754         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 755
 756         gfs2_trans_end(sdp);
 757
 758 out_bh2:
 759         brelse(l_bh);
 760 out_bh:
 761         brelse(m_bh);
 762 out:
 763         gfs2_glock_dq_uninit(&gh);
 764         return error;
 765 }
 766
 767 /**
 768  * gfs2_statfs_i - Do a statfs
 769  * @sdp: the filesystem
 770  * @sg: the sg structure
 771  *
 772  * Returns: errno
 773  */
 774
 775 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 776 {
 777         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 778         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 779
 780         spin_lock(&sdp->sd_statfs_spin);
 781
 782         *sc = *m_sc;
 783         sc->sc_total += l_sc->sc_total;
 784         sc->sc_free += l_sc->sc_free;
 785         sc->sc_dinodes += l_sc->sc_dinodes;
 786
 787         spin_unlock(&sdp->sd_statfs_spin);
 788
 789         if (sc->sc_free < 0)
 790                 sc->sc_free = 0;
 791         if (sc->sc_free > sc->sc_total)
 792                 sc->sc_free = sc->sc_total;
 793         if (sc->sc_dinodes < 0)
 794                 sc->sc_dinodes = 0;
 795
 796         return 0;
 797 }
 798
 799 /**
 800  * statfs_fill - fill in the sg for a given RG
 801  * @rgd: the RG
 802  * @sc: the sc structure
 803  *
 804  * Returns: 0 on success, -ESTALE if the LVB is invalid
 805  */
 806
 807 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 808                             struct gfs2_statfs_change_host *sc)
 809 {
 810         gfs2_rgrp_verify(rgd);
 811         sc->sc_total += rgd->rd_data;
 812         sc->sc_free += rgd->rd_rg.rg_free;
 813         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 814         return 0;
 815 }
 816
 817 /**
 818  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 819  * @sdp: the filesystem
 820  * @sc: the sc info that will be returned
 821  *
 822  * Any error (other than a signal) will cause this routine to fall back
 823  * to the synchronous version.
 824  *
 825  * FIXME: This really shouldn't busy wait like this.
 826  *
 827  * Returns: errno
 828  */
 829
 830 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 831 {
 832         struct gfs2_holder ri_gh;
 833         struct gfs2_rgrpd *rgd_next;
 834         struct gfs2_holder *gha, *gh;
 835         unsigned int slots = 64;
 836         unsigned int x;
 837         int done;
 838         int error = 0, err;
 839
 840         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 841         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 842         if (!gha)
 843                 return -ENOMEM;
 844
 845         error = gfs2_rindex_hold(sdp, &ri_gh);
 846         if (error)
 847                 goto out;
 848
 849         rgd_next = gfs2_rgrpd_get_first(sdp);
 850
 851         for (;;) {
 852                 done = 1;
 853
 854                 for (x = 0; x < slots; x++) {
 855                         gh = gha + x;
 856
 857                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 858                                 err = gfs2_glock_wait(gh);
 859                                 if (err) {
 860                                         gfs2_holder_uninit(gh);
 861                                         error = err;
 862                                 } else {
 863                                         if (!error)
 864                                                 error = statfs_slow_fill(
 865                                                         gh->gh_gl->gl_object, sc);
 866                                         gfs2_glock_dq_uninit(gh);
 867                                 }
 868                         }
 869
 870                         if (gh->gh_gl)
 871                                 done = 0;
 872                         else if (rgd_next && !error) {
 873                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 874                                                            LM_ST_SHARED,
 875                                                            GL_ASYNC,
 876                                                            gh);
 877                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 878                                 done = 0;
 879                         }
 880
 881                         if (signal_pending(current))
 882                                 error = -ERESTARTSYS;
 883                 }
 884
 885                 if (done)
 886                         break;
 887
 888                 yield();
 889         }
 890
 891         gfs2_glock_dq_uninit(&ri_gh);
 892
 893 out:
 894         kfree(gha);
 895         return error;
 896 }
 897
 898 struct lfcc {
 899         struct list_head list;
 900         struct gfs2_holder gh;
 901 };
 902
 903 /**
 904  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 905  *                            journals are clean
 906  * @sdp: the file system
 907  * @state: the state to put the transaction lock into
 908  * @t_gh: the hold on the transaction lock
 909  *
 910  * Returns: errno
 911  */
 912
 913 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 914                                     struct gfs2_holder *t_gh)
 915 {
 916         struct gfs2_inode *ip;
 917         struct gfs2_holder ji_gh;
 918         struct gfs2_jdesc *jd;
 919         struct lfcc *lfcc;
 920         LIST_HEAD(list);
 921         struct gfs2_log_header_host lh;
 922         int error;
 923
 924         error = gfs2_jindex_hold(sdp, &ji_gh);
 925         if (error)
 926                 return error;
 927
 928         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 929                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 930                 if (!lfcc) {
 931                         error = -ENOMEM;
 932                         goto out;
 933                 }
 934                 ip = GFS2_I(jd->jd_inode);
 935                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 936                 if (error) {
 937                         kfree(lfcc);
 938                         goto out;
 939                 }
 940                 list_add(&lfcc->list, &list);
 941         }
 942
 943         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 944                                LM_FLAG_PRIORITY | GL_NOCACHE,
 945                                t_gh);
 946
 947         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 948                 error = gfs2_jdesc_check(jd);
 949                 if (error)
 950                         break;
 951                 error = gfs2_find_jhead(jd, &lh);
 952                 if (error)
 953                         break;
 954                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 955                         error = -EBUSY;
 956                         break;
 957                 }
 958         }
 959
 960         if (error)
 961                 gfs2_glock_dq_uninit(t_gh);
 962
 963 out:
 964         while (!list_empty(&list)) {
 965                 lfcc = list_entry(list.next, struct lfcc, list);
 966                 list_del(&lfcc->list);
 967                 gfs2_glock_dq_uninit(&lfcc->gh);
 968                 kfree(lfcc);
 969         }
 970         gfs2_glock_dq_uninit(&ji_gh);
 971         return error;
 972 }
 973
 974 /**
 975  * gfs2_freeze_fs - freezes the file system
 976  * @sdp: the file system
 977  *
 978  * This function flushes data and meta data for all machines by
 979  * aquiring the transaction log exclusively.  All journals are
 980  * ensured to be in a clean state as well.
 981  *
 982  * Returns: errno
 983  */
 984
 985 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 986 {
 987         int error = 0;
 988
 989         mutex_lock(&sdp->sd_freeze_lock);
 990
 991         if (!sdp->sd_freeze_count++) {
 992                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 993                 if (error)
 994                         sdp->sd_freeze_count--;
 995         }
 996
 997         mutex_unlock(&sdp->sd_freeze_lock);
 998
 999         return error;
1000 }
1001
1002 /**
1003  * gfs2_unfreeze_fs - unfreezes the file system
1004  * @sdp: the file system
1005  *
1006  * This function allows the file system to proceed by unlocking
1007  * the exclusively held transaction lock.  Other GFS2 nodes are
1008  * now free to acquire the lock shared and go on with their lives.
1009  *
1010  */
1011
1012 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1013 {
1014         mutex_lock(&sdp->sd_freeze_lock);
1015
1016         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1017                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1018
1019         mutex_unlock(&sdp->sd_freeze_lock);
1020 }
1021