SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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_demote_secs = 300;
  55         gt->gt_incore_log_blocks = 1024;
  56         gt->gt_log_flush_secs = 60;
  57         gt->gt_jindex_refresh_secs = 60;
  58         gt->gt_recoverd_secs = 60;
  59         gt->gt_logd_secs = 1;
  60         gt->gt_quotad_secs = 5;
  61         gt->gt_quota_simul_sync = 64;
  62         gt->gt_quota_warn_period = 10;
  63         gt->gt_quota_scale_num = 1;
  64         gt->gt_quota_scale_den = 1;
  65         gt->gt_quota_cache_secs = 300;
  66         gt->gt_quota_quantum = 60;
  67         gt->gt_atime_quantum = 3600;
  68         gt->gt_new_files_jdata = 0;
  69         gt->gt_new_files_directio = 0;
  70         gt->gt_max_readahead = 1 << 18;
  71         gt->gt_stall_secs = 600;
  72         gt->gt_complain_secs = 10;
  73         gt->gt_statfs_quantum = 30;
  74         gt->gt_statfs_slow = 0;
  75 }
  76
  77 /**
  78  * gfs2_check_sb - Check superblock
  79  * @sdp: the filesystem
  80  * @sb: The superblock
  81  * @silent: Don't print a message if the check fails
  82  *
  83  * Checks the version code of the FS is one that we understand how to
  84  * read and that the sizes of the various on-disk structures have not
  85  * changed.
  86  */
  87
  88 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  89 {
  90         unsigned int x;
  91
  92         if (sb->sb_magic != GFS2_MAGIC ||
  93             sb->sb_type != GFS2_METATYPE_SB) {
  94                 if (!silent)
  95                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
  96                 return -EINVAL;
  97         }
  98
  99         /*  If format numbers match exactly, we're done.  */
 100
 101         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 102             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 103                 return 0;
 104
 105         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 106                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 107                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 108                                 break;
 109
 110                 if (!gfs2_old_fs_formats[x]) {
 111                         printk(KERN_WARNING
 112                                "GFS2: code version (%u, %u) is incompatible "
 113                                "with ondisk format (%u, %u)\n",
 114                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 115                                sb->sb_fs_format, sb->sb_multihost_format);
 116                         printk(KERN_WARNING
 117                                "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] ==
 125                             sb->sb_multihost_format)
 126                                 break;
 127
 128                 if (!gfs2_old_multihost_formats[x]) {
 129                         printk(KERN_WARNING
 130                                "GFS2: code version (%u, %u) is incompatible "
 131                                "with ondisk format (%u, %u)\n",
 132                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 133                                sb->sb_fs_format, sb->sb_multihost_format);
 134                         printk(KERN_WARNING
 135                                "GFS2: I don't know how to upgrade this FS\n");
 136                         return -EINVAL;
 137                 }
 138         }
 139
 140         if (!sdp->sd_args.ar_upgrade) {
 141                 printk(KERN_WARNING
 142                        "GFS2: code version (%u, %u) is incompatible "
 143                        "with ondisk format (%u, %u)\n",
 144                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 145                        sb->sb_fs_format, sb->sb_multihost_format);
 146                 printk(KERN_INFO
 147                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 148                        "the FS\n");
 149                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 150                 return -EINVAL;
 151         }
 152
 153         return 0;
 154 }
 155
 156
 157 static void end_bio_io_page(struct bio *bio, int error)
 158 {
 159         struct page *page = bio->bi_private;
 160
 161         if (!error)
 162                 SetPageUptodate(page);
 163         else
 164                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 165         unlock_page(page);
 166 }
 167
 168 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 169 {
 170         const struct gfs2_sb *str = buf;
 171
 172         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 173         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 174         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 175         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 176         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 177         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 178         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 179         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 180         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 181         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 182         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 183
 184         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 185         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 186 }
 187
 188 /**
 189  * gfs2_read_super - Read the gfs2 super block from disk
 190  * @sdp: The GFS2 super block
 191  * @sector: The location of the super block
 192  * @error: The error code to return
 193  *
 194  * This uses the bio functions to read the super block from disk
 195  * because we want to be 100% sure that we never read cached data.
 196  * A super block is read twice only during each GFS2 mount and is
 197  * never written to by the filesystem. The first time its read no
 198  * locks are held, and the only details which are looked at are those
 199  * relating to the locking protocol. Once locking is up and working,
 200  * the sb is read again under the lock to establish the location of
 201  * the master directory (contains pointers to journals etc) and the
 202  * root directory.
 203  *
 204  * Returns: 0 on success or error
 205  */
 206
 207 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 208 {
 209         struct super_block *sb = sdp->sd_vfs;
 210         struct gfs2_sb *p;
 211         struct page *page;
 212         struct bio *bio;
 213
 214         page = alloc_page(GFP_KERNEL);
 215         if (unlikely(!page))
 216                 return -ENOBUFS;
 217
 218         ClearPageUptodate(page);
 219         ClearPageDirty(page);
 220         lock_page(page);
 221
 222         bio = bio_alloc(GFP_KERNEL, 1);
 223         if (unlikely(!bio)) {
 224                 __free_page(page);
 225                 return -ENOBUFS;
 226         }
 227
 228         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 229         bio->bi_bdev = sb->s_bdev;
 230         bio_add_page(bio, page, PAGE_SIZE, 0);
 231
 232         bio->bi_end_io = end_bio_io_page;
 233         bio->bi_private = page;
 234         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 235         wait_on_page_locked(page);
 236         bio_put(bio);
 237         if (!PageUptodate(page)) {
 238                 __free_page(page);
 239                 return -EIO;
 240         }
 241         p = kmap(page);
 242         gfs2_sb_in(&sdp->sd_sb, p);
 243         kunmap(page);
 244         __free_page(page);
 245         return 0;
 246 }
 247
 248 /**
 249  * gfs2_read_sb - Read super block
 250  * @sdp: The GFS2 superblock
 251  * @gl: the glock for the superblock (assumed to be held)
 252  * @silent: Don't print message if mount fails
 253  *
 254  */
 255
 256 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 257 {
 258         u32 hash_blocks, ind_blocks, leaf_blocks;
 259         u32 tmp_blocks;
 260         unsigned int x;
 261         int error;
 262
 263         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 264         if (error) {
 265                 if (!silent)
 266                         fs_err(sdp, "can't read superblock\n");
 267                 return error;
 268         }
 269
 270         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 271         if (error)
 272                 return error;
 273
 274         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 275                                GFS2_BASIC_BLOCK_SHIFT;
 276         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 277         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 278                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 279         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 280                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 281         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 282         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 283         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 284         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 285         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 286                                 sizeof(struct gfs2_meta_header)) /
 287                                 sizeof(struct gfs2_quota_change);
 288
 289         /* Compute maximum reservation required to add a entry to a directory */
 290
 291         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 292                              sdp->sd_jbsize);
 293
 294         ind_blocks = 0;
 295         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 296                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 297                 ind_blocks += tmp_blocks;
 298         }
 299
 300         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 301
 302         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 303
 304         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 305                                 sizeof(struct gfs2_dinode);
 306         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 307         for (x = 2;; x++) {
 308                 u64 space, d;
 309                 u32 m;
 310
 311                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 312                 d = space;
 313                 m = do_div(d, sdp->sd_inptrs);
 314
 315                 if (d != sdp->sd_heightsize[x - 1] || m)
 316                         break;
 317                 sdp->sd_heightsize[x] = space;
 318         }
 319         sdp->sd_max_height = x;
 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         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 339
 340         return 0;
 341 }
 342
 343 /**
 344  * gfs2_jindex_hold - Grab a lock on the jindex
 345  * @sdp: The GFS2 superblock
 346  * @ji_gh: the holder for the jindex glock
 347  *
 348  * This is very similar to the gfs2_rindex_hold() function, except that
 349  * in general we hold the jindex lock for longer periods of time and
 350  * we grab it far less frequently (in general) then the rgrp lock.
 351  *
 352  * Returns: errno
 353  */
 354
 355 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 356 {
 357         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 358         struct qstr name;
 359         char buf[20];
 360         struct gfs2_jdesc *jd;
 361         int error;
 362
 363         name.name = buf;
 364
 365         mutex_lock(&sdp->sd_jindex_mutex);
 366
 367         for (;;) {
 368                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 369                 if (error)
 370                         break;
 371
 372                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 373                 name.hash = gfs2_disk_hash(name.name, name.len);
 374
 375                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 376                 if (error == -ENOENT) {
 377                         error = 0;
 378                         break;
 379                 }
 380
 381                 gfs2_glock_dq_uninit(ji_gh);
 382
 383                 if (error)
 384                         break;
 385
 386                 error = -ENOMEM;
 387                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 388                 if (!jd)
 389                         break;
 390
 391                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 392                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 393                         if (!jd->jd_inode)
 394                                 error = -ENOENT;
 395                         else
 396                                 error = PTR_ERR(jd->jd_inode);
 397                         kfree(jd);
 398                         break;
 399                 }
 400
 401                 spin_lock(&sdp->sd_jindex_spin);
 402                 jd->jd_jid = sdp->sd_journals++;
 403                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 404                 spin_unlock(&sdp->sd_jindex_spin);
 405         }
 406
 407         mutex_unlock(&sdp->sd_jindex_mutex);
 408
 409         return error;
 410 }
 411
 412 /**
 413  * gfs2_jindex_free - Clear all the journal index information
 414  * @sdp: The GFS2 superblock
 415  *
 416  */
 417
 418 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 419 {
 420         struct list_head list;
 421         struct gfs2_jdesc *jd;
 422
 423         spin_lock(&sdp->sd_jindex_spin);
 424         list_add(&list, &sdp->sd_jindex_list);
 425         list_del_init(&sdp->sd_jindex_list);
 426         sdp->sd_journals = 0;
 427         spin_unlock(&sdp->sd_jindex_spin);
 428
 429         while (!list_empty(&list)) {
 430                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 431                 list_del(&jd->jd_list);
 432                 iput(jd->jd_inode);
 433                 kfree(jd);
 434         }
 435 }
 436
 437 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 438 {
 439         struct gfs2_jdesc *jd;
 440         int found = 0;
 441
 442         list_for_each_entry(jd, head, jd_list) {
 443                 if (jd->jd_jid == jid) {
 444                         found = 1;
 445                         break;
 446                 }
 447         }
 448
 449         if (!found)
 450                 jd = NULL;
 451
 452         return jd;
 453 }
 454
 455 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 456 {
 457         struct gfs2_jdesc *jd;
 458
 459         spin_lock(&sdp->sd_jindex_spin);
 460         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 461         spin_unlock(&sdp->sd_jindex_spin);
 462
 463         return jd;
 464 }
 465
 466 void gfs2_jdesc_make_dirty(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         if (jd)
 473                 jd->jd_dirty = 1;
 474         spin_unlock(&sdp->sd_jindex_spin);
 475 }
 476
 477 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 478 {
 479         struct gfs2_jdesc *jd;
 480         int found = 0;
 481
 482         spin_lock(&sdp->sd_jindex_spin);
 483
 484         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 485                 if (jd->jd_dirty) {
 486                         jd->jd_dirty = 0;
 487                         found = 1;
 488                         break;
 489                 }
 490         }
 491         spin_unlock(&sdp->sd_jindex_spin);
 492
 493         if (!found)
 494                 jd = NULL;
 495
 496         return jd;
 497 }
 498
 499 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 500 {
 501         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 502         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 503         int ar;
 504         int error;
 505
 506         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 507             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 508                 gfs2_consist_inode(ip);
 509                 return -EIO;
 510         }
 511         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 512
 513         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 514         if (!error && ar) {
 515                 gfs2_consist_inode(ip);
 516                 error = -EIO;
 517         }
 518
 519         return error;
 520 }
 521
 522 /**
 523  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 524  * @sdp: the filesystem
 525  *
 526  * Returns: errno
 527  */
 528
 529 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 530 {
 531         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 532         struct gfs2_glock *j_gl = ip->i_gl;
 533         struct gfs2_holder t_gh;
 534         struct gfs2_log_header_host head;
 535         int error;
 536
 537         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 538         if (error)
 539                 return error;
 540
 541         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 542
 543         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 544         if (error)
 545                 goto fail;
 546
 547         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 548                 gfs2_consist(sdp);
 549                 error = -EIO;
 550                 goto fail;
 551         }
 552
 553         /*  Initialize some head of the log stuff  */
 554         sdp->sd_log_sequence = head.lh_sequence + 1;
 555         gfs2_log_pointers_init(sdp, head.lh_blkno);
 556
 557         error = gfs2_quota_init(sdp);
 558         if (error)
 559                 goto fail;
 560
 561         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 562
 563         gfs2_glock_dq_uninit(&t_gh);
 564
 565         return 0;
 566
 567 fail:
 568         t_gh.gh_flags |= GL_NOCACHE;
 569         gfs2_glock_dq_uninit(&t_gh);
 570
 571         return error;
 572 }
 573
 574 /**
 575  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 576  * @sdp: the filesystem
 577  *
 578  * Returns: errno
 579  */
 580
 581 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 582 {
 583         struct gfs2_holder t_gh;
 584         int error;
 585
 586         gfs2_quota_sync(sdp);
 587         gfs2_statfs_sync(sdp);
 588
 589         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 590                                    &t_gh);
 591         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 592                 return error;
 593
 594         gfs2_meta_syncfs(sdp);
 595         gfs2_log_shutdown(sdp);
 596
 597         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 598
 599         if (t_gh.gh_gl)
 600                 gfs2_glock_dq_uninit(&t_gh);
 601
 602         gfs2_quota_cleanup(sdp);
 603
 604         return error;
 605 }
 606
 607 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 608 {
 609         const struct gfs2_statfs_change *str = buf;
 610
 611         sc->sc_total = be64_to_cpu(str->sc_total);
 612         sc->sc_free = be64_to_cpu(str->sc_free);
 613         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 614 }
 615
 616 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 617 {
 618         struct gfs2_statfs_change *str = buf;
 619
 620         str->sc_total = cpu_to_be64(sc->sc_total);
 621         str->sc_free = cpu_to_be64(sc->sc_free);
 622         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 623 }
 624
 625 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 626 {
 627         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 628         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 629         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 630         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 631         struct buffer_head *m_bh, *l_bh;
 632         struct gfs2_holder gh;
 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         if (sdp->sd_args.ar_spectator) {
 645                 spin_lock(&sdp->sd_statfs_spin);
 646                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 647                                       sizeof(struct gfs2_dinode));
 648                 spin_unlock(&sdp->sd_statfs_spin);
 649         } else {
 650                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 651                 if (error)
 652                         goto out_m_bh;
 653
 654                 spin_lock(&sdp->sd_statfs_spin);
 655                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 656                                       sizeof(struct gfs2_dinode));
 657                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 658                                       sizeof(struct gfs2_dinode));
 659                 spin_unlock(&sdp->sd_statfs_spin);
 660
 661                 brelse(l_bh);
 662         }
 663
 664 out_m_bh:
 665         brelse(m_bh);
 666 out:
 667         gfs2_glock_dq_uninit(&gh);
 668         return 0;
 669 }
 670
 671 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 672                         s64 dinodes)
 673 {
 674         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 675         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 676         struct buffer_head *l_bh;
 677         int error;
 678
 679         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 680         if (error)
 681                 return;
 682
 683         mutex_lock(&sdp->sd_statfs_mutex);
 684         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 685         mutex_unlock(&sdp->sd_statfs_mutex);
 686
 687         spin_lock(&sdp->sd_statfs_spin);
 688         l_sc->sc_total += total;
 689         l_sc->sc_free += free;
 690         l_sc->sc_dinodes += dinodes;
 691         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 692         spin_unlock(&sdp->sd_statfs_spin);
 693
 694         brelse(l_bh);
 695 }
 696
 697 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 698 {
 699         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 700         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 701         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 702         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 703         struct gfs2_holder gh;
 704         struct buffer_head *m_bh, *l_bh;
 705         int error;
 706
 707         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 708                                    &gh);
 709         if (error)
 710                 return error;
 711
 712         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 713         if (error)
 714                 goto out;
 715
 716         spin_lock(&sdp->sd_statfs_spin);
 717         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 718                               sizeof(struct gfs2_dinode));
 719         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 720                 spin_unlock(&sdp->sd_statfs_spin);
 721                 goto out_bh;
 722         }
 723         spin_unlock(&sdp->sd_statfs_spin);
 724
 725         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 726         if (error)
 727                 goto out_bh;
 728
 729         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 730         if (error)
 731                 goto out_bh2;
 732
 733         mutex_lock(&sdp->sd_statfs_mutex);
 734         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 735         mutex_unlock(&sdp->sd_statfs_mutex);
 736
 737         spin_lock(&sdp->sd_statfs_spin);
 738         m_sc->sc_total += l_sc->sc_total;
 739         m_sc->sc_free += l_sc->sc_free;
 740         m_sc->sc_dinodes += l_sc->sc_dinodes;
 741         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 742         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 743                0, sizeof(struct gfs2_statfs_change));
 744         spin_unlock(&sdp->sd_statfs_spin);
 745
 746         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 747         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 748
 749         gfs2_trans_end(sdp);
 750
 751 out_bh2:
 752         brelse(l_bh);
 753 out_bh:
 754         brelse(m_bh);
 755 out:
 756         gfs2_glock_dq_uninit(&gh);
 757         return error;
 758 }
 759
 760 /**
 761  * gfs2_statfs_i - Do a statfs
 762  * @sdp: the filesystem
 763  * @sg: the sg structure
 764  *
 765  * Returns: errno
 766  */
 767
 768 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 769 {
 770         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 771         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 772
 773         spin_lock(&sdp->sd_statfs_spin);
 774
 775         *sc = *m_sc;
 776         sc->sc_total += l_sc->sc_total;
 777         sc->sc_free += l_sc->sc_free;
 778         sc->sc_dinodes += l_sc->sc_dinodes;
 779
 780         spin_unlock(&sdp->sd_statfs_spin);
 781
 782         if (sc->sc_free < 0)
 783                 sc->sc_free = 0;
 784         if (sc->sc_free > sc->sc_total)
 785                 sc->sc_free = sc->sc_total;
 786         if (sc->sc_dinodes < 0)
 787                 sc->sc_dinodes = 0;
 788
 789         return 0;
 790 }
 791
 792 /**
 793  * statfs_fill - fill in the sg for a given RG
 794  * @rgd: the RG
 795  * @sc: the sc structure
 796  *
 797  * Returns: 0 on success, -ESTALE if the LVB is invalid
 798  */
 799
 800 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 801                             struct gfs2_statfs_change_host *sc)
 802 {
 803         gfs2_rgrp_verify(rgd);
 804         sc->sc_total += rgd->rd_data;
 805         sc->sc_free += rgd->rd_rg.rg_free;
 806         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 807         return 0;
 808 }
 809
 810 /**
 811  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 812  * @sdp: the filesystem
 813  * @sc: the sc info that will be returned
 814  *
 815  * Any error (other than a signal) will cause this routine to fall back
 816  * to the synchronous version.
 817  *
 818  * FIXME: This really shouldn't busy wait like this.
 819  *
 820  * Returns: errno
 821  */
 822
 823 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 824 {
 825         struct gfs2_holder ri_gh;
 826         struct gfs2_rgrpd *rgd_next;
 827         struct gfs2_holder *gha, *gh;
 828         unsigned int slots = 64;
 829         unsigned int x;
 830         int done;
 831         int error = 0, err;
 832
 833         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 834         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 835         if (!gha)
 836                 return -ENOMEM;
 837
 838         error = gfs2_rindex_hold(sdp, &ri_gh);
 839         if (error)
 840                 goto out;
 841
 842         rgd_next = gfs2_rgrpd_get_first(sdp);
 843
 844         for (;;) {
 845                 done = 1;
 846
 847                 for (x = 0; x < slots; x++) {
 848                         gh = gha + x;
 849
 850                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 851                                 err = gfs2_glock_wait(gh);
 852                                 if (err) {
 853                                         gfs2_holder_uninit(gh);
 854                                         error = err;
 855                                 } else {
 856                                         if (!error)
 857                                                 error = statfs_slow_fill(
 858                                                         gh->gh_gl->gl_object, sc);
 859                                         gfs2_glock_dq_uninit(gh);
 860                                 }
 861                         }
 862
 863                         if (gh->gh_gl)
 864                                 done = 0;
 865                         else if (rgd_next && !error) {
 866                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 867                                                            LM_ST_SHARED,
 868                                                            GL_ASYNC,
 869                                                            gh);
 870                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 871                                 done = 0;
 872                         }
 873
 874                         if (signal_pending(current))
 875                                 error = -ERESTARTSYS;
 876                 }
 877
 878                 if (done)
 879                         break;
 880
 881                 yield();
 882         }
 883
 884         gfs2_glock_dq_uninit(&ri_gh);
 885
 886 out:
 887         kfree(gha);
 888         return error;
 889 }
 890
 891 struct lfcc {
 892         struct list_head list;
 893         struct gfs2_holder gh;
 894 };
 895
 896 /**
 897  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 898  *                            journals are clean
 899  * @sdp: the file system
 900  * @state: the state to put the transaction lock into
 901  * @t_gh: the hold on the transaction lock
 902  *
 903  * Returns: errno
 904  */
 905
 906 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 907                                     struct gfs2_holder *t_gh)
 908 {
 909         struct gfs2_inode *ip;
 910         struct gfs2_holder ji_gh;
 911         struct gfs2_jdesc *jd;
 912         struct lfcc *lfcc;
 913         LIST_HEAD(list);
 914         struct gfs2_log_header_host lh;
 915         int error;
 916
 917         error = gfs2_jindex_hold(sdp, &ji_gh);
 918         if (error)
 919                 return error;
 920
 921         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 922                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 923                 if (!lfcc) {
 924                         error = -ENOMEM;
 925                         goto out;
 926                 }
 927                 ip = GFS2_I(jd->jd_inode);
 928                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 929                 if (error) {
 930                         kfree(lfcc);
 931                         goto out;
 932                 }
 933                 list_add(&lfcc->list, &list);
 934         }
 935
 936         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 937                                LM_FLAG_PRIORITY | GL_NOCACHE,
 938                                t_gh);
 939
 940         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 941                 error = gfs2_jdesc_check(jd);
 942                 if (error)
 943                         break;
 944                 error = gfs2_find_jhead(jd, &lh);
 945                 if (error)
 946                         break;
 947                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 948                         error = -EBUSY;
 949                         break;
 950                 }
 951         }
 952
 953         if (error)
 954                 gfs2_glock_dq_uninit(t_gh);
 955
 956 out:
 957         while (!list_empty(&list)) {
 958                 lfcc = list_entry(list.next, struct lfcc, list);
 959                 list_del(&lfcc->list);
 960                 gfs2_glock_dq_uninit(&lfcc->gh);
 961                 kfree(lfcc);
 962         }
 963         gfs2_glock_dq_uninit(&ji_gh);
 964         return error;
 965 }
 966
 967 /**
 968  * gfs2_freeze_fs - freezes the file system
 969  * @sdp: the file system
 970  *
 971  * This function flushes data and meta data for all machines by
 972  * aquiring the transaction log exclusively.  All journals are
 973  * ensured to be in a clean state as well.
 974  *
 975  * Returns: errno
 976  */
 977
 978 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 979 {
 980         int error = 0;
 981
 982         mutex_lock(&sdp->sd_freeze_lock);
 983
 984         if (!sdp->sd_freeze_count++) {
 985                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 986                 if (error)
 987                         sdp->sd_freeze_count--;
 988         }
 989
 990         mutex_unlock(&sdp->sd_freeze_lock);
 991
 992         return error;
 993 }
 994
 995 /**
 996  * gfs2_unfreeze_fs - unfreezes the file system
 997  * @sdp: the file system
 998  *
 999  * This function allows the file system to proceed by unlocking
1000  * the exclusively held transaction lock.  Other GFS2 nodes are
1001  * now free to acquire the lock shared and go on with their lives.
1002  *
1003  */
1004
1005 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1006 {
1007         mutex_lock(&sdp->sd_freeze_lock);
1008
1009         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1010                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1011
1012         mutex_unlock(&sdp->sd_freeze_lock);
1013 }
1014