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) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/crc32.h>
  16 #include <linux/gfs2_ondisk.h>
  17 #include <linux/bio.h>
  18 #include <linux/lm_interface.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "dir.h"
  24 #include "glock.h"
  25 #include "glops.h"
  26 #include "inode.h"
  27 #include "log.h"
  28 #include "meta_io.h"
  29 #include "quota.h"
  30 #include "recovery.h"
  31 #include "rgrp.h"
  32 #include "super.h"
  33 #include "trans.h"
  34 #include "util.h"
  35
  36 static const u32 gfs2_old_fs_formats[] = {
  37         0
  38 };
  39
  40 static const u32 gfs2_old_multihost_formats[] = {
  41         0
  42 };
  43
  44 /**
  45  * gfs2_tune_init - Fill a gfs2_tune structure with default values
  46  * @gt: tune
  47  *
  48  */
  49
  50 void gfs2_tune_init(struct gfs2_tune *gt)
  51 {
  52         spin_lock_init(&gt->gt_spin);
  53
  54         gt->gt_ilimit = 100;
  55         gt->gt_ilimit_tries = 3;
  56         gt->gt_ilimit_min = 1;
  57         gt->gt_demote_secs = 300;
  58         gt->gt_incore_log_blocks = 1024;
  59         gt->gt_log_flush_secs = 60;
  60         gt->gt_jindex_refresh_secs = 60;
  61         gt->gt_scand_secs = 15;
  62         gt->gt_recoverd_secs = 60;
  63         gt->gt_logd_secs = 1;
  64         gt->gt_quotad_secs = 5;
  65         gt->gt_quota_simul_sync = 64;
  66         gt->gt_quota_warn_period = 10;
  67         gt->gt_quota_scale_num = 1;
  68         gt->gt_quota_scale_den = 1;
  69         gt->gt_quota_cache_secs = 300;
  70         gt->gt_quota_quantum = 60;
  71         gt->gt_atime_quantum = 3600;
  72         gt->gt_new_files_jdata = 0;
  73         gt->gt_new_files_directio = 0;
  74         gt->gt_max_readahead = 1 << 18;
  75         gt->gt_lockdump_size = 131072;
  76         gt->gt_stall_secs = 600;
  77         gt->gt_complain_secs = 10;
  78         gt->gt_reclaim_limit = 5000;
  79         gt->gt_entries_per_readdir = 32;
  80         gt->gt_statfs_quantum = 30;
  81         gt->gt_statfs_slow = 0;
  82 }
  83
  84 /**
  85  * gfs2_check_sb - Check superblock
  86  * @sdp: the filesystem
  87  * @sb: The superblock
  88  * @silent: Don't print a message if the check fails
  89  *
  90  * Checks the version code of the FS is one that we understand how to
  91  * read and that the sizes of the various on-disk structures have not
  92  * changed.
  93  */
  94
  95 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  96 {
  97         unsigned int x;
  98
  99         if (sb->sb_header.mh_magic != GFS2_MAGIC ||
 100             sb->sb_header.mh_type != GFS2_METATYPE_SB) {
 101                 if (!silent)
 102                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 103                 return -EINVAL;
 104         }
 105
 106         /*  If format numbers match exactly, we're done.  */
 107
 108         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 109             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 110                 return 0;
 111
 112         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 113                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 114                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 115                                 break;
 116
 117                 if (!gfs2_old_fs_formats[x]) {
 118                         printk(KERN_WARNING
 119                                "GFS2: code version (%u, %u) is incompatible "
 120                                "with ondisk format (%u, %u)\n",
 121                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 122                                sb->sb_fs_format, sb->sb_multihost_format);
 123                         printk(KERN_WARNING
 124                                "GFS2: I don't know how to upgrade this FS\n");
 125                         return -EINVAL;
 126                 }
 127         }
 128
 129         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 130                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 131                         if (gfs2_old_multihost_formats[x] ==
 132                             sb->sb_multihost_format)
 133                                 break;
 134
 135                 if (!gfs2_old_multihost_formats[x]) {
 136                         printk(KERN_WARNING
 137                                "GFS2: code version (%u, %u) is incompatible "
 138                                "with ondisk format (%u, %u)\n",
 139                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 140                                sb->sb_fs_format, sb->sb_multihost_format);
 141                         printk(KERN_WARNING
 142                                "GFS2: I don't know how to upgrade this FS\n");
 143                         return -EINVAL;
 144                 }
 145         }
 146
 147         if (!sdp->sd_args.ar_upgrade) {
 148                 printk(KERN_WARNING
 149                        "GFS2: code version (%u, %u) is incompatible "
 150                        "with ondisk format (%u, %u)\n",
 151                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 152                        sb->sb_fs_format, sb->sb_multihost_format);
 153                 printk(KERN_INFO
 154                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 155                        "the FS\n");
 156                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 157                 return -EINVAL;
 158         }
 159
 160         return 0;
 161 }
 162
 163
 164 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
 165 {
 166         struct page *page = bio->bi_private;
 167         if (bio->bi_size)
 168                 return 1;
 169
 170         if (!error)
 171                 SetPageUptodate(page);
 172         else
 173                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 174         unlock_page(page);
 175         return 0;
 176 }
 177
 178 /**
 179  * gfs2_read_super - Read the gfs2 super block from disk
 180  * @sb: The VFS super block
 181  * @sector: The location of the super block
 182  *
 183  * This uses the bio functions to read the super block from disk
 184  * because we want to be 100% sure that we never read cached data.
 185  * A super block is read twice only during each GFS2 mount and is
 186  * never written to by the filesystem. The first time its read no
 187  * locks are held, and the only details which are looked at are those
 188  * relating to the locking protocol. Once locking is up and working,
 189  * the sb is read again under the lock to establish the location of
 190  * the master directory (contains pointers to journals etc) and the
 191  * root directory.
 192  *
 193  * Returns: A page containing the sb or NULL
 194  */
 195
 196 struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
 197 {
 198         struct page *page;
 199         struct bio *bio;
 200
 201         page = alloc_page(GFP_KERNEL);
 202         if (unlikely(!page))
 203                 return NULL;
 204
 205         ClearPageUptodate(page);
 206         ClearPageDirty(page);
 207         lock_page(page);
 208
 209         bio = bio_alloc(GFP_KERNEL, 1);
 210         if (unlikely(!bio)) {
 211                 __free_page(page);
 212                 return NULL;
 213         }
 214
 215         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 216         bio->bi_bdev = sb->s_bdev;
 217         bio_add_page(bio, page, PAGE_SIZE, 0);
 218
 219         bio->bi_end_io = end_bio_io_page;
 220         bio->bi_private = page;
 221         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 222         wait_on_page_locked(page);
 223         bio_put(bio);
 224         if (!PageUptodate(page)) {
 225                 __free_page(page);
 226                 return NULL;
 227         }
 228         return page;
 229 }
 230
 231 /**
 232  * gfs2_read_sb - Read super block
 233  * @sdp: The GFS2 superblock
 234  * @gl: the glock for the superblock (assumed to be held)
 235  * @silent: Don't print message if mount fails
 236  *
 237  */
 238
 239 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 240 {
 241         u32 hash_blocks, ind_blocks, leaf_blocks;
 242         u32 tmp_blocks;
 243         unsigned int x;
 244         int error;
 245         struct page *page;
 246         char *sb;
 247
 248         page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 249         if (!page) {
 250                 if (!silent)
 251                         fs_err(sdp, "can't read superblock\n");
 252                 return -EIO;
 253         }
 254         sb = kmap(page);
 255         gfs2_sb_in(&sdp->sd_sb, sb);
 256         kunmap(page);
 257         __free_page(page);
 258
 259         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 260         if (error)
 261                 return error;
 262
 263         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 264                                GFS2_BASIC_BLOCK_SHIFT;
 265         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 266         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 267                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 268         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 269                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 270         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 271         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 272         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 273         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 274         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 275                                 sizeof(struct gfs2_meta_header)) /
 276                                 sizeof(struct gfs2_quota_change);
 277
 278         /* Compute maximum reservation required to add a entry to a directory */
 279
 280         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 281                              sdp->sd_jbsize);
 282
 283         ind_blocks = 0;
 284         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 285                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 286                 ind_blocks += tmp_blocks;
 287         }
 288
 289         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 290
 291         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 292
 293         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 294                                 sizeof(struct gfs2_dinode);
 295         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 296         for (x = 2;; x++) {
 297                 u64 space, d;
 298                 u32 m;
 299
 300                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 301                 d = space;
 302                 m = do_div(d, sdp->sd_inptrs);
 303
 304                 if (d != sdp->sd_heightsize[x - 1] || m)
 305                         break;
 306                 sdp->sd_heightsize[x] = space;
 307         }
 308         sdp->sd_max_height = x;
 309         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 310
 311         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 312                                  sizeof(struct gfs2_dinode);
 313         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 314         for (x = 2;; x++) {
 315                 u64 space, d;
 316                 u32 m;
 317
 318                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 319                 d = space;
 320                 m = do_div(d, sdp->sd_inptrs);
 321
 322                 if (d != sdp->sd_jheightsize[x - 1] || m)
 323                         break;
 324                 sdp->sd_jheightsize[x] = space;
 325         }
 326         sdp->sd_max_jheight = x;
 327         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 328
 329         return 0;
 330 }
 331
 332 /**
 333  * gfs2_jindex_hold - Grab a lock on the jindex
 334  * @sdp: The GFS2 superblock
 335  * @ji_gh: the holder for the jindex glock
 336  *
 337  * This is very similar to the gfs2_rindex_hold() function, except that
 338  * in general we hold the jindex lock for longer periods of time and
 339  * we grab it far less frequently (in general) then the rgrp lock.
 340  *
 341  * Returns: errno
 342  */
 343
 344 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 345 {
 346         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 347         struct qstr name;
 348         char buf[20];
 349         struct gfs2_jdesc *jd;
 350         int error;
 351
 352         name.name = buf;
 353
 354         mutex_lock(&sdp->sd_jindex_mutex);
 355
 356         for (;;) {
 357                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 358                 if (error)
 359                         break;
 360
 361                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 362                 name.hash = gfs2_disk_hash(name.name, name.len);
 363
 364                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 365                 if (error == -ENOENT) {
 366                         error = 0;
 367                         break;
 368                 }
 369
 370                 gfs2_glock_dq_uninit(ji_gh);
 371
 372                 if (error)
 373                         break;
 374
 375                 error = -ENOMEM;
 376                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 377                 if (!jd)
 378                         break;
 379
 380                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 381                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 382                         if (!jd->jd_inode)
 383                                 error = -ENOENT;
 384                         else
 385                                 error = PTR_ERR(jd->jd_inode);
 386                         kfree(jd);
 387                         break;
 388                 }
 389
 390                 spin_lock(&sdp->sd_jindex_spin);
 391                 jd->jd_jid = sdp->sd_journals++;
 392                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 393                 spin_unlock(&sdp->sd_jindex_spin);
 394         }
 395
 396         mutex_unlock(&sdp->sd_jindex_mutex);
 397
 398         return error;
 399 }
 400
 401 /**
 402  * gfs2_jindex_free - Clear all the journal index information
 403  * @sdp: The GFS2 superblock
 404  *
 405  */
 406
 407 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 408 {
 409         struct list_head list;
 410         struct gfs2_jdesc *jd;
 411
 412         spin_lock(&sdp->sd_jindex_spin);
 413         list_add(&list, &sdp->sd_jindex_list);
 414         list_del_init(&sdp->sd_jindex_list);
 415         sdp->sd_journals = 0;
 416         spin_unlock(&sdp->sd_jindex_spin);
 417
 418         while (!list_empty(&list)) {
 419                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 420                 list_del(&jd->jd_list);
 421                 iput(jd->jd_inode);
 422                 kfree(jd);
 423         }
 424 }
 425
 426 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 427 {
 428         struct gfs2_jdesc *jd;
 429         int found = 0;
 430
 431         list_for_each_entry(jd, head, jd_list) {
 432                 if (jd->jd_jid == jid) {
 433                         found = 1;
 434                         break;
 435                 }
 436         }
 437
 438         if (!found)
 439                 jd = NULL;
 440
 441         return jd;
 442 }
 443
 444 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 445 {
 446         struct gfs2_jdesc *jd;
 447
 448         spin_lock(&sdp->sd_jindex_spin);
 449         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 450         spin_unlock(&sdp->sd_jindex_spin);
 451
 452         return jd;
 453 }
 454
 455 void gfs2_jdesc_make_dirty(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         if (jd)
 462                 jd->jd_dirty = 1;
 463         spin_unlock(&sdp->sd_jindex_spin);
 464 }
 465
 466 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 467 {
 468         struct gfs2_jdesc *jd;
 469         int found = 0;
 470
 471         spin_lock(&sdp->sd_jindex_spin);
 472
 473         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 474                 if (jd->jd_dirty) {
 475                         jd->jd_dirty = 0;
 476                         found = 1;
 477                         break;
 478                 }
 479         }
 480         spin_unlock(&sdp->sd_jindex_spin);
 481
 482         if (!found)
 483                 jd = NULL;
 484
 485         return jd;
 486 }
 487
 488 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 489 {
 490         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 491         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 492         int ar;
 493         int error;
 494
 495         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 496             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 497                 gfs2_consist_inode(ip);
 498                 return -EIO;
 499         }
 500         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 501
 502         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 503         if (!error && ar) {
 504                 gfs2_consist_inode(ip);
 505                 error = -EIO;
 506         }
 507
 508         return error;
 509 }
 510
 511 /**
 512  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 513  * @sdp: the filesystem
 514  *
 515  * Returns: errno
 516  */
 517
 518 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 519 {
 520         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 521         struct gfs2_glock *j_gl = ip->i_gl;
 522         struct gfs2_holder t_gh;
 523         struct gfs2_log_header_host head;
 524         int error;
 525
 526         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 527         if (error)
 528                 return error;
 529
 530         gfs2_meta_cache_flush(ip);
 531         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 532
 533         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 534         if (error)
 535                 goto fail;
 536
 537         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 538                 gfs2_consist(sdp);
 539                 error = -EIO;
 540                 goto fail;
 541         }
 542
 543         /*  Initialize some head of the log stuff  */
 544         sdp->sd_log_sequence = head.lh_sequence + 1;
 545         gfs2_log_pointers_init(sdp, head.lh_blkno);
 546
 547         error = gfs2_quota_init(sdp);
 548         if (error)
 549                 goto fail;
 550
 551         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 552
 553         gfs2_glock_dq_uninit(&t_gh);
 554
 555         return 0;
 556
 557 fail:
 558         t_gh.gh_flags |= GL_NOCACHE;
 559         gfs2_glock_dq_uninit(&t_gh);
 560
 561         return error;
 562 }
 563
 564 /**
 565  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 566  * @sdp: the filesystem
 567  *
 568  * Returns: errno
 569  */
 570
 571 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 572 {
 573         struct gfs2_holder t_gh;
 574         int error;
 575
 576         gfs2_quota_sync(sdp);
 577         gfs2_statfs_sync(sdp);
 578
 579         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 580                                    &t_gh);
 581         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 582                 return error;
 583
 584         gfs2_meta_syncfs(sdp);
 585         gfs2_log_shutdown(sdp);
 586
 587         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 588
 589         if (t_gh.gh_gl)
 590                 gfs2_glock_dq_uninit(&t_gh);
 591
 592         gfs2_quota_cleanup(sdp);
 593
 594         return error;
 595 }
 596
 597 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 598 {
 599         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 600         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 601         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 602         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 603         struct buffer_head *m_bh, *l_bh;
 604         struct gfs2_holder gh;
 605         int error;
 606
 607         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 608                                    &gh);
 609         if (error)
 610                 return error;
 611
 612         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 613         if (error)
 614                 goto out;
 615
 616         if (sdp->sd_args.ar_spectator) {
 617                 spin_lock(&sdp->sd_statfs_spin);
 618                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 619                                       sizeof(struct gfs2_dinode));
 620                 spin_unlock(&sdp->sd_statfs_spin);
 621         } else {
 622                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 623                 if (error)
 624                         goto out_m_bh;
 625
 626                 spin_lock(&sdp->sd_statfs_spin);
 627                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 628                                       sizeof(struct gfs2_dinode));
 629                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 630                                       sizeof(struct gfs2_dinode));
 631                 spin_unlock(&sdp->sd_statfs_spin);
 632
 633                 brelse(l_bh);
 634         }
 635
 636 out_m_bh:
 637         brelse(m_bh);
 638 out:
 639         gfs2_glock_dq_uninit(&gh);
 640         return 0;
 641 }
 642
 643 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 644                         s64 dinodes)
 645 {
 646         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 647         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 648         struct buffer_head *l_bh;
 649         int error;
 650
 651         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 652         if (error)
 653                 return;
 654
 655         mutex_lock(&sdp->sd_statfs_mutex);
 656         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 657         mutex_unlock(&sdp->sd_statfs_mutex);
 658
 659         spin_lock(&sdp->sd_statfs_spin);
 660         l_sc->sc_total += total;
 661         l_sc->sc_free += free;
 662         l_sc->sc_dinodes += dinodes;
 663         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 664         spin_unlock(&sdp->sd_statfs_spin);
 665
 666         brelse(l_bh);
 667 }
 668
 669 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 670 {
 671         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 672         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 673         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 674         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 675         struct gfs2_holder gh;
 676         struct buffer_head *m_bh, *l_bh;
 677         int error;
 678
 679         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 680                                    &gh);
 681         if (error)
 682                 return error;
 683
 684         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 685         if (error)
 686                 goto out;
 687
 688         spin_lock(&sdp->sd_statfs_spin);
 689         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 690                               sizeof(struct gfs2_dinode));
 691         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 692                 spin_unlock(&sdp->sd_statfs_spin);
 693                 goto out_bh;
 694         }
 695         spin_unlock(&sdp->sd_statfs_spin);
 696
 697         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 698         if (error)
 699                 goto out_bh;
 700
 701         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 702         if (error)
 703                 goto out_bh2;
 704
 705         mutex_lock(&sdp->sd_statfs_mutex);
 706         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 707         mutex_unlock(&sdp->sd_statfs_mutex);
 708
 709         spin_lock(&sdp->sd_statfs_spin);
 710         m_sc->sc_total += l_sc->sc_total;
 711         m_sc->sc_free += l_sc->sc_free;
 712         m_sc->sc_dinodes += l_sc->sc_dinodes;
 713         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 714         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 715                0, sizeof(struct gfs2_statfs_change));
 716         spin_unlock(&sdp->sd_statfs_spin);
 717
 718         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 719         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 720
 721         gfs2_trans_end(sdp);
 722
 723 out_bh2:
 724         brelse(l_bh);
 725 out_bh:
 726         brelse(m_bh);
 727 out:
 728         gfs2_glock_dq_uninit(&gh);
 729         return error;
 730 }
 731
 732 /**
 733  * gfs2_statfs_i - Do a statfs
 734  * @sdp: the filesystem
 735  * @sg: the sg structure
 736  *
 737  * Returns: errno
 738  */
 739
 740 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 741 {
 742         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 743         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 744
 745         spin_lock(&sdp->sd_statfs_spin);
 746
 747         *sc = *m_sc;
 748         sc->sc_total += l_sc->sc_total;
 749         sc->sc_free += l_sc->sc_free;
 750         sc->sc_dinodes += l_sc->sc_dinodes;
 751
 752         spin_unlock(&sdp->sd_statfs_spin);
 753
 754         if (sc->sc_free < 0)
 755                 sc->sc_free = 0;
 756         if (sc->sc_free > sc->sc_total)
 757                 sc->sc_free = sc->sc_total;
 758         if (sc->sc_dinodes < 0)
 759                 sc->sc_dinodes = 0;
 760
 761         return 0;
 762 }
 763
 764 /**
 765  * statfs_fill - fill in the sg for a given RG
 766  * @rgd: the RG
 767  * @sc: the sc structure
 768  *
 769  * Returns: 0 on success, -ESTALE if the LVB is invalid
 770  */
 771
 772 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 773                             struct gfs2_statfs_change_host *sc)
 774 {
 775         gfs2_rgrp_verify(rgd);
 776         sc->sc_total += rgd->rd_ri.ri_data;
 777         sc->sc_free += rgd->rd_rg.rg_free;
 778         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 779         return 0;
 780 }
 781
 782 /**
 783  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 784  * @sdp: the filesystem
 785  * @sc: the sc info that will be returned
 786  *
 787  * Any error (other than a signal) will cause this routine to fall back
 788  * to the synchronous version.
 789  *
 790  * FIXME: This really shouldn't busy wait like this.
 791  *
 792  * Returns: errno
 793  */
 794
 795 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 796 {
 797         struct gfs2_holder ri_gh;
 798         struct gfs2_rgrpd *rgd_next;
 799         struct gfs2_holder *gha, *gh;
 800         unsigned int slots = 64;
 801         unsigned int x;
 802         int done;
 803         int error = 0, err;
 804
 805         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 806         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 807         if (!gha)
 808                 return -ENOMEM;
 809
 810         error = gfs2_rindex_hold(sdp, &ri_gh);
 811         if (error)
 812                 goto out;
 813
 814         rgd_next = gfs2_rgrpd_get_first(sdp);
 815
 816         for (;;) {
 817                 done = 1;
 818
 819                 for (x = 0; x < slots; x++) {
 820                         gh = gha + x;
 821
 822                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 823                                 err = gfs2_glock_wait(gh);
 824                                 if (err) {
 825                                         gfs2_holder_uninit(gh);
 826                                         error = err;
 827                                 } else {
 828                                         if (!error)
 829                                                 error = statfs_slow_fill(
 830                                                         gh->gh_gl->gl_object, sc);
 831                                         gfs2_glock_dq_uninit(gh);
 832                                 }
 833                         }
 834
 835                         if (gh->gh_gl)
 836                                 done = 0;
 837                         else if (rgd_next && !error) {
 838                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 839                                                            LM_ST_SHARED,
 840                                                            GL_ASYNC,
 841                                                            gh);
 842                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 843                                 done = 0;
 844                         }
 845
 846                         if (signal_pending(current))
 847                                 error = -ERESTARTSYS;
 848                 }
 849
 850                 if (done)
 851                         break;
 852
 853                 yield();
 854         }
 855
 856         gfs2_glock_dq_uninit(&ri_gh);
 857
 858 out:
 859         kfree(gha);
 860         return error;
 861 }
 862
 863 struct lfcc {
 864         struct list_head list;
 865         struct gfs2_holder gh;
 866 };
 867
 868 /**
 869  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 870  *                            journals are clean
 871  * @sdp: the file system
 872  * @state: the state to put the transaction lock into
 873  * @t_gh: the hold on the transaction lock
 874  *
 875  * Returns: errno
 876  */
 877
 878 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 879                                     struct gfs2_holder *t_gh)
 880 {
 881         struct gfs2_inode *ip;
 882         struct gfs2_holder ji_gh;
 883         struct gfs2_jdesc *jd;
 884         struct lfcc *lfcc;
 885         LIST_HEAD(list);
 886         struct gfs2_log_header_host lh;
 887         int error;
 888
 889         error = gfs2_jindex_hold(sdp, &ji_gh);
 890         if (error)
 891                 return error;
 892
 893         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 894                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 895                 if (!lfcc) {
 896                         error = -ENOMEM;
 897                         goto out;
 898                 }
 899                 ip = GFS2_I(jd->jd_inode);
 900                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 901                 if (error) {
 902                         kfree(lfcc);
 903                         goto out;
 904                 }
 905                 list_add(&lfcc->list, &list);
 906         }
 907
 908         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 909                                LM_FLAG_PRIORITY | GL_NOCACHE,
 910                                t_gh);
 911
 912         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 913                 error = gfs2_jdesc_check(jd);
 914                 if (error)
 915                         break;
 916                 error = gfs2_find_jhead(jd, &lh);
 917                 if (error)
 918                         break;
 919                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 920                         error = -EBUSY;
 921                         break;
 922                 }
 923         }
 924
 925         if (error)
 926                 gfs2_glock_dq_uninit(t_gh);
 927
 928 out:
 929         while (!list_empty(&list)) {
 930                 lfcc = list_entry(list.next, struct lfcc, list);
 931                 list_del(&lfcc->list);
 932                 gfs2_glock_dq_uninit(&lfcc->gh);
 933                 kfree(lfcc);
 934         }
 935         gfs2_glock_dq_uninit(&ji_gh);
 936         return error;
 937 }
 938
 939 /**
 940  * gfs2_freeze_fs - freezes the file system
 941  * @sdp: the file system
 942  *
 943  * This function flushes data and meta data for all machines by
 944  * aquiring the transaction log exclusively.  All journals are
 945  * ensured to be in a clean state as well.
 946  *
 947  * Returns: errno
 948  */
 949
 950 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 951 {
 952         int error = 0;
 953
 954         mutex_lock(&sdp->sd_freeze_lock);
 955
 956         if (!sdp->sd_freeze_count++) {
 957                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 958                 if (error)
 959                         sdp->sd_freeze_count--;
 960         }
 961
 962         mutex_unlock(&sdp->sd_freeze_lock);
 963
 964         return error;
 965 }
 966
 967 /**
 968  * gfs2_unfreeze_fs - unfreezes the file system
 969  * @sdp: the file system
 970  *
 971  * This function allows the file system to proceed by unlocking
 972  * the exclusively held transaction lock.  Other GFS2 nodes are
 973  * now free to acquire the lock shared and go on with their lives.
 974  *
 975  */
 976
 977 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 978 {
 979         mutex_lock(&sdp->sd_freeze_lock);
 980
 981         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 982                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 983
 984         mutex_unlock(&sdp->sd_freeze_lock);
 985 }
 986