fs/gfs2/super.c

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