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
  19 #include "gfs2.h"
  20 #include "lm_interface.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 uint32_t gfs2_old_fs_formats[] = {
  37         0
  38 };
  39
  40 static const uint32_t 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_atomic_write = 4 << 20;
  75         gt->gt_max_readahead = 1 << 18;
  76         gt->gt_lockdump_size = 131072;
  77         gt->gt_stall_secs = 600;
  78         gt->gt_complain_secs = 10;
  79         gt->gt_reclaim_limit = 5000;
  80         gt->gt_entries_per_readdir = 32;
  81         gt->gt_prefetch_secs = 10;
  82         gt->gt_greedy_default = HZ / 10;
  83         gt->gt_greedy_quantum = HZ / 40;
  84         gt->gt_greedy_max = HZ / 4;
  85         gt->gt_statfs_quantum = 30;
  86         gt->gt_statfs_slow = 0;
  87 }
  88
  89 /**
  90  * gfs2_check_sb - Check superblock
  91  * @sdp: the filesystem
  92  * @sb: The superblock
  93  * @silent: Don't print a message if the check fails
  94  *
  95  * Checks the version code of the FS is one that we understand how to
  96  * read and that the sizes of the various on-disk structures have not
  97  * changed.
  98  */
  99
 100 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb *sb, int silent)
 101 {
 102         unsigned int x;
 103
 104         if (sb->sb_header.mh_magic != GFS2_MAGIC ||
 105             sb->sb_header.mh_type != GFS2_METATYPE_SB) {
 106                 if (!silent)
 107                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 108                 return -EINVAL;
 109         }
 110
 111         /*  If format numbers match exactly, we're done.  */
 112
 113         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 114             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 115                 return 0;
 116
 117         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 118                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 119                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 120                                 break;
 121
 122                 if (!gfs2_old_fs_formats[x]) {
 123                         printk(KERN_WARNING
 124                                "GFS2: code version (%u, %u) is incompatible "
 125                                "with ondisk format (%u, %u)\n",
 126                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 127                                sb->sb_fs_format, sb->sb_multihost_format);
 128                         printk(KERN_WARNING
 129                                "GFS2: I don't know how to upgrade this FS\n");
 130                         return -EINVAL;
 131                 }
 132         }
 133
 134         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 135                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 136                         if (gfs2_old_multihost_formats[x] ==
 137                             sb->sb_multihost_format)
 138                                 break;
 139
 140                 if (!gfs2_old_multihost_formats[x]) {
 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_WARNING
 147                                "GFS2: I don't know how to upgrade this FS\n");
 148                         return -EINVAL;
 149                 }
 150         }
 151
 152         if (!sdp->sd_args.ar_upgrade) {
 153                 printk(KERN_WARNING
 154                        "GFS2: code version (%u, %u) is incompatible "
 155                        "with ondisk format (%u, %u)\n",
 156                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 157                        sb->sb_fs_format, sb->sb_multihost_format);
 158                 printk(KERN_INFO
 159                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 160                        "the FS\n");
 161                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 162                 return -EINVAL;
 163         }
 164
 165         return 0;
 166 }
 167
 168
 169 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
 170 {
 171         struct page *page = bio->bi_private;
 172         if (bio->bi_size)
 173                 return 1;
 174
 175         if (!error)
 176                 SetPageUptodate(page);
 177         else
 178                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 179         unlock_page(page);
 180         return 0;
 181 }
 182
 183 static struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
 184 {
 185         struct page *page;
 186         struct bio *bio;
 187
 188         page = alloc_page(GFP_KERNEL);
 189         if (unlikely(!page))
 190                 return NULL;
 191
 192         ClearPageUptodate(page);
 193         ClearPageDirty(page);
 194         lock_page(page);
 195
 196         bio = bio_alloc(GFP_KERNEL, 1);
 197         if (unlikely(!bio)) {
 198                 __free_page(page);
 199                 return NULL;
 200         }
 201
 202         bio->bi_sector = sector;
 203         bio->bi_bdev = sb->s_bdev;
 204         bio_add_page(bio, page, PAGE_SIZE, 0);
 205
 206         bio->bi_end_io = end_bio_io_page;
 207         bio->bi_private = page;
 208         submit_bio(READ_SYNC, bio);
 209         wait_on_page_locked(page);
 210         bio_put(bio);
 211         if (!PageUptodate(page)) {
 212                 __free_page(page);
 213                 return NULL;
 214         }
 215         return page;
 216 }
 217
 218 /**
 219  * gfs2_read_sb - Read super block
 220  * @sdp: The GFS2 superblock
 221  * @gl: the glock for the superblock (assumed to be held)
 222  * @silent: Don't print message if mount fails
 223  *
 224  */
 225
 226 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 227 {
 228         uint32_t hash_blocks, ind_blocks, leaf_blocks;
 229         uint32_t tmp_blocks;
 230         unsigned int x;
 231         int error;
 232         struct page *page;
 233         char *sb;
 234
 235         page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 236         if (!page) {
 237                 if (!silent)
 238                         fs_err(sdp, "can't read superblock\n");
 239                 return -EIO;
 240         }
 241         sb = kmap(page);
 242         gfs2_sb_in(&sdp->sd_sb, sb);
 243         kunmap(page);
 244         __free_page(page);
 245
 246         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 247         if (error)
 248                 return error;
 249
 250         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 251                                GFS2_BASIC_BLOCK_SHIFT;
 252         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 253         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 254                           sizeof(struct gfs2_dinode)) / sizeof(uint64_t);
 255         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 256                           sizeof(struct gfs2_meta_header)) / sizeof(uint64_t);
 257         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 258         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 259         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 260         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(uint64_t);
 261         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 262                                 sizeof(struct gfs2_meta_header)) /
 263                                 sizeof(struct gfs2_quota_change);
 264
 265         /* Compute maximum reservation required to add a entry to a directory */
 266
 267         hash_blocks = DIV_ROUND_UP(sizeof(uint64_t) * (1 << GFS2_DIR_MAX_DEPTH),
 268                              sdp->sd_jbsize);
 269
 270         ind_blocks = 0;
 271         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 272                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 273                 ind_blocks += tmp_blocks;
 274         }
 275
 276         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 277
 278         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 279
 280         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 281                                 sizeof(struct gfs2_dinode);
 282         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 283         for (x = 2;; x++) {
 284                 uint64_t space, d;
 285                 uint32_t m;
 286
 287                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 288                 d = space;
 289                 m = do_div(d, sdp->sd_inptrs);
 290
 291                 if (d != sdp->sd_heightsize[x - 1] || m)
 292                         break;
 293                 sdp->sd_heightsize[x] = space;
 294         }
 295         sdp->sd_max_height = x;
 296         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 297
 298         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 299                                  sizeof(struct gfs2_dinode);
 300         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 301         for (x = 2;; x++) {
 302                 uint64_t space, d;
 303                 uint32_t m;
 304
 305                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 306                 d = space;
 307                 m = do_div(d, sdp->sd_inptrs);
 308
 309                 if (d != sdp->sd_jheightsize[x - 1] || m)
 310                         break;
 311                 sdp->sd_jheightsize[x] = space;
 312         }
 313         sdp->sd_max_jheight = x;
 314         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 315
 316         return 0;
 317 }
 318
 319 /**
 320  * gfs2_jindex_hold - Grab a lock on the jindex
 321  * @sdp: The GFS2 superblock
 322  * @ji_gh: the holder for the jindex glock
 323  *
 324  * This is very similar to the gfs2_rindex_hold() function, except that
 325  * in general we hold the jindex lock for longer periods of time and
 326  * we grab it far less frequently (in general) then the rgrp lock.
 327  *
 328  * Returns: errno
 329  */
 330
 331 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 332 {
 333         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 334         struct qstr name;
 335         char buf[20];
 336         struct gfs2_jdesc *jd;
 337         int error;
 338
 339         name.name = buf;
 340
 341         mutex_lock(&sdp->sd_jindex_mutex);
 342
 343         for (;;) {
 344                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED,
 345                                            GL_LOCAL_EXCL, ji_gh);
 346                 if (error)
 347                         break;
 348
 349                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 350                 name.hash = gfs2_disk_hash(name.name, name.len);
 351
 352                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 353                 if (error == -ENOENT) {
 354                         error = 0;
 355                         break;
 356                 }
 357
 358                 gfs2_glock_dq_uninit(ji_gh);
 359
 360                 if (error)
 361                         break;
 362
 363                 error = -ENOMEM;
 364                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 365                 if (!jd)
 366                         break;
 367
 368                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 369                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 370                         if (!jd->jd_inode)
 371                                 error = -ENOENT;
 372                         else
 373                                 error = PTR_ERR(jd->jd_inode);
 374                         kfree(jd);
 375                         break;
 376                 }
 377
 378                 spin_lock(&sdp->sd_jindex_spin);
 379                 jd->jd_jid = sdp->sd_journals++;
 380                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 381                 spin_unlock(&sdp->sd_jindex_spin);
 382         }
 383
 384         mutex_unlock(&sdp->sd_jindex_mutex);
 385
 386         return error;
 387 }
 388
 389 /**
 390  * gfs2_jindex_free - Clear all the journal index information
 391  * @sdp: The GFS2 superblock
 392  *
 393  */
 394
 395 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 396 {
 397         struct list_head list;
 398         struct gfs2_jdesc *jd;
 399
 400         spin_lock(&sdp->sd_jindex_spin);
 401         list_add(&list, &sdp->sd_jindex_list);
 402         list_del_init(&sdp->sd_jindex_list);
 403         sdp->sd_journals = 0;
 404         spin_unlock(&sdp->sd_jindex_spin);
 405
 406         while (!list_empty(&list)) {
 407                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 408                 list_del(&jd->jd_list);
 409                 iput(jd->jd_inode);
 410                 kfree(jd);
 411         }
 412 }
 413
 414 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 415 {
 416         struct gfs2_jdesc *jd;
 417         int found = 0;
 418
 419         list_for_each_entry(jd, head, jd_list) {
 420                 if (jd->jd_jid == jid) {
 421                         found = 1;
 422                         break;
 423                 }
 424         }
 425
 426         if (!found)
 427                 jd = NULL;
 428
 429         return jd;
 430 }
 431
 432 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 433 {
 434         struct gfs2_jdesc *jd;
 435
 436         spin_lock(&sdp->sd_jindex_spin);
 437         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 438         spin_unlock(&sdp->sd_jindex_spin);
 439
 440         return jd;
 441 }
 442
 443 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 444 {
 445         struct gfs2_jdesc *jd;
 446
 447         spin_lock(&sdp->sd_jindex_spin);
 448         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 449         if (jd)
 450                 jd->jd_dirty = 1;
 451         spin_unlock(&sdp->sd_jindex_spin);
 452 }
 453
 454 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 455 {
 456         struct gfs2_jdesc *jd;
 457         int found = 0;
 458
 459         spin_lock(&sdp->sd_jindex_spin);
 460
 461         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 462                 if (jd->jd_dirty) {
 463                         jd->jd_dirty = 0;
 464                         found = 1;
 465                         break;
 466                 }
 467         }
 468         spin_unlock(&sdp->sd_jindex_spin);
 469
 470         if (!found)
 471                 jd = NULL;
 472
 473         return jd;
 474 }
 475
 476 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 477 {
 478         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 479         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 480         int ar;
 481         int error;
 482
 483         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 484             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 485                 gfs2_consist_inode(ip);
 486                 return -EIO;
 487         }
 488         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 489
 490         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 491         if (!error && ar) {
 492                 gfs2_consist_inode(ip);
 493                 error = -EIO;
 494         }
 495
 496         return error;
 497 }
 498
 499 /**
 500  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 501  * @sdp: the filesystem
 502  *
 503  * Returns: errno
 504  */
 505
 506 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 507 {
 508         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 509         struct gfs2_glock *j_gl = ip->i_gl;
 510         struct gfs2_holder t_gh;
 511         struct gfs2_log_header head;
 512         int error;
 513
 514         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 515                                    GL_LOCAL_EXCL, &t_gh);
 516         if (error)
 517                 return error;
 518
 519         gfs2_meta_cache_flush(ip);
 520         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA | DIO_DATA);
 521
 522         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 523         if (error)
 524                 goto fail;
 525
 526         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 527                 gfs2_consist(sdp);
 528                 error = -EIO;
 529                 goto fail;
 530         }
 531
 532         /*  Initialize some head of the log stuff  */
 533         sdp->sd_log_sequence = head.lh_sequence + 1;
 534         gfs2_log_pointers_init(sdp, head.lh_blkno);
 535
 536         error = gfs2_quota_init(sdp);
 537         if (error)
 538                 goto fail;
 539
 540         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 541
 542         gfs2_glock_dq_uninit(&t_gh);
 543
 544         return 0;
 545
 546 fail:
 547         t_gh.gh_flags |= GL_NOCACHE;
 548         gfs2_glock_dq_uninit(&t_gh);
 549
 550         return error;
 551 }
 552
 553 /**
 554  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 555  * @sdp: the filesystem
 556  *
 557  * Returns: errno
 558  */
 559
 560 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 561 {
 562         struct gfs2_holder t_gh;
 563         int error;
 564
 565         gfs2_quota_sync(sdp);
 566         gfs2_statfs_sync(sdp);
 567
 568         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 569                                 GL_LOCAL_EXCL | GL_NOCACHE,
 570                                 &t_gh);
 571         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 572                 return error;
 573
 574         gfs2_meta_syncfs(sdp);
 575         gfs2_log_shutdown(sdp);
 576
 577         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 578
 579         if (t_gh.gh_gl)
 580                 gfs2_glock_dq_uninit(&t_gh);
 581
 582         gfs2_quota_cleanup(sdp);
 583
 584         return error;
 585 }
 586
 587 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 588 {
 589         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 590         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 591         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 592         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 593         struct buffer_head *m_bh, *l_bh;
 594         struct gfs2_holder gh;
 595         int error;
 596
 597         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 598                                    &gh);
 599         if (error)
 600                 return error;
 601
 602         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 603         if (error)
 604                 goto out;
 605
 606         if (sdp->sd_args.ar_spectator) {
 607                 spin_lock(&sdp->sd_statfs_spin);
 608                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 609                                       sizeof(struct gfs2_dinode));
 610                 spin_unlock(&sdp->sd_statfs_spin);
 611         } else {
 612                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 613                 if (error)
 614                         goto out_m_bh;
 615
 616                 spin_lock(&sdp->sd_statfs_spin);
 617                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 618                                       sizeof(struct gfs2_dinode));
 619                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 620                                       sizeof(struct gfs2_dinode));
 621                 spin_unlock(&sdp->sd_statfs_spin);
 622
 623                 brelse(l_bh);
 624         }
 625
 626 out_m_bh:
 627         brelse(m_bh);
 628 out:
 629         gfs2_glock_dq_uninit(&gh);
 630         return 0;
 631 }
 632
 633 void gfs2_statfs_change(struct gfs2_sbd *sdp, int64_t total, int64_t free,
 634                         int64_t dinodes)
 635 {
 636         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 637         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 638         struct buffer_head *l_bh;
 639         int error;
 640
 641         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 642         if (error)
 643                 return;
 644
 645         mutex_lock(&sdp->sd_statfs_mutex);
 646         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 647         mutex_unlock(&sdp->sd_statfs_mutex);
 648
 649         spin_lock(&sdp->sd_statfs_spin);
 650         l_sc->sc_total += total;
 651         l_sc->sc_free += free;
 652         l_sc->sc_dinodes += dinodes;
 653         gfs2_statfs_change_out(l_sc, l_bh->b_data +
 654                                sizeof(struct gfs2_dinode));
 655         spin_unlock(&sdp->sd_statfs_spin);
 656
 657         brelse(l_bh);
 658 }
 659
 660 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 661 {
 662         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 663         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 664         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 665         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 666         struct gfs2_holder gh;
 667         struct buffer_head *m_bh, *l_bh;
 668         int error;
 669
 670         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 671                                    &gh);
 672         if (error)
 673                 return error;
 674
 675         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 676         if (error)
 677                 goto out;
 678
 679         spin_lock(&sdp->sd_statfs_spin);
 680         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 681                               sizeof(struct gfs2_dinode));
 682         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 683                 spin_unlock(&sdp->sd_statfs_spin);
 684                 goto out_bh;
 685         }
 686         spin_unlock(&sdp->sd_statfs_spin);
 687
 688         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 689         if (error)
 690                 goto out_bh;
 691
 692         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 693         if (error)
 694                 goto out_bh2;
 695
 696         mutex_lock(&sdp->sd_statfs_mutex);
 697         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 698         mutex_unlock(&sdp->sd_statfs_mutex);
 699
 700         spin_lock(&sdp->sd_statfs_spin);
 701         m_sc->sc_total += l_sc->sc_total;
 702         m_sc->sc_free += l_sc->sc_free;
 703         m_sc->sc_dinodes += l_sc->sc_dinodes;
 704         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 705         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 706                0, sizeof(struct gfs2_statfs_change));
 707         spin_unlock(&sdp->sd_statfs_spin);
 708
 709         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 710         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 711
 712         gfs2_trans_end(sdp);
 713
 714 out_bh2:
 715         brelse(l_bh);
 716 out_bh:
 717         brelse(m_bh);
 718 out:
 719         gfs2_glock_dq_uninit(&gh);
 720         return error;
 721 }
 722
 723 /**
 724  * gfs2_statfs_i - Do a statfs
 725  * @sdp: the filesystem
 726  * @sg: the sg structure
 727  *
 728  * Returns: errno
 729  */
 730
 731 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change *sc)
 732 {
 733         struct gfs2_statfs_change *m_sc = &sdp->sd_statfs_master;
 734         struct gfs2_statfs_change *l_sc = &sdp->sd_statfs_local;
 735
 736         spin_lock(&sdp->sd_statfs_spin);
 737
 738         *sc = *m_sc;
 739         sc->sc_total += l_sc->sc_total;
 740         sc->sc_free += l_sc->sc_free;
 741         sc->sc_dinodes += l_sc->sc_dinodes;
 742
 743         spin_unlock(&sdp->sd_statfs_spin);
 744
 745         if (sc->sc_free < 0)
 746                 sc->sc_free = 0;
 747         if (sc->sc_free > sc->sc_total)
 748                 sc->sc_free = sc->sc_total;
 749         if (sc->sc_dinodes < 0)
 750                 sc->sc_dinodes = 0;
 751
 752         return 0;
 753 }
 754
 755 /**
 756  * statfs_fill - fill in the sg for a given RG
 757  * @rgd: the RG
 758  * @sc: the sc structure
 759  *
 760  * Returns: 0 on success, -ESTALE if the LVB is invalid
 761  */
 762
 763 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 764                             struct gfs2_statfs_change *sc)
 765 {
 766         gfs2_rgrp_verify(rgd);
 767         sc->sc_total += rgd->rd_ri.ri_data;
 768         sc->sc_free += rgd->rd_rg.rg_free;
 769         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 770         return 0;
 771 }
 772
 773 /**
 774  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 775  * @sdp: the filesystem
 776  * @sc: the sc info that will be returned
 777  *
 778  * Any error (other than a signal) will cause this routine to fall back
 779  * to the synchronous version.
 780  *
 781  * FIXME: This really shouldn't busy wait like this.
 782  *
 783  * Returns: errno
 784  */
 785
 786 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change *sc)
 787 {
 788         struct gfs2_holder ri_gh;
 789         struct gfs2_rgrpd *rgd_next;
 790         struct gfs2_holder *gha, *gh;
 791         unsigned int slots = 64;
 792         unsigned int x;
 793         int done;
 794         int error = 0, err;
 795
 796         memset(sc, 0, sizeof(struct gfs2_statfs_change));
 797         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 798         if (!gha)
 799                 return -ENOMEM;
 800
 801         error = gfs2_rindex_hold(sdp, &ri_gh);
 802         if (error)
 803                 goto out;
 804
 805         rgd_next = gfs2_rgrpd_get_first(sdp);
 806
 807         for (;;) {
 808                 done = 1;
 809
 810                 for (x = 0; x < slots; x++) {
 811                         gh = gha + x;
 812
 813                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 814                                 err = gfs2_glock_wait(gh);
 815                                 if (err) {
 816                                         gfs2_holder_uninit(gh);
 817                                         error = err;
 818                                 } else {
 819                                         if (!error)
 820                                                 error = statfs_slow_fill(
 821                                                         gh->gh_gl->gl_object, sc);
 822                                         gfs2_glock_dq_uninit(gh);
 823                                 }
 824                         }
 825
 826                         if (gh->gh_gl)
 827                                 done = 0;
 828                         else if (rgd_next && !error) {
 829                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 830                                                            LM_ST_SHARED,
 831                                                            GL_ASYNC,
 832                                                            gh);
 833                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 834                                 done = 0;
 835                         }
 836
 837                         if (signal_pending(current))
 838                                 error = -ERESTARTSYS;
 839                 }
 840
 841                 if (done)
 842                         break;
 843
 844                 yield();
 845         }
 846
 847         gfs2_glock_dq_uninit(&ri_gh);
 848
 849 out:
 850         kfree(gha);
 851         return error;
 852 }
 853
 854 struct lfcc {
 855         struct list_head list;
 856         struct gfs2_holder gh;
 857 };
 858
 859 /**
 860  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 861  *                            journals are clean
 862  * @sdp: the file system
 863  * @state: the state to put the transaction lock into
 864  * @t_gh: the hold on the transaction lock
 865  *
 866  * Returns: errno
 867  */
 868
 869 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 870                                     struct gfs2_holder *t_gh)
 871 {
 872         struct gfs2_inode *ip;
 873         struct gfs2_holder ji_gh;
 874         struct gfs2_jdesc *jd;
 875         struct lfcc *lfcc;
 876         LIST_HEAD(list);
 877         struct gfs2_log_header lh;
 878         int error;
 879
 880         error = gfs2_jindex_hold(sdp, &ji_gh);
 881         if (error)
 882                 return error;
 883
 884         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 885                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 886                 if (!lfcc) {
 887                         error = -ENOMEM;
 888                         goto out;
 889                 }
 890                 ip = GFS2_I(jd->jd_inode);
 891                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 892                 if (error) {
 893                         kfree(lfcc);
 894                         goto out;
 895                 }
 896                 list_add(&lfcc->list, &list);
 897         }
 898
 899         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 900                                LM_FLAG_PRIORITY | GL_NOCACHE,
 901                                t_gh);
 902
 903         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 904                 error = gfs2_jdesc_check(jd);
 905                 if (error)
 906                         break;
 907                 error = gfs2_find_jhead(jd, &lh);
 908                 if (error)
 909                         break;
 910                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 911                         error = -EBUSY;
 912                         break;
 913                 }
 914         }
 915
 916         if (error)
 917                 gfs2_glock_dq_uninit(t_gh);
 918
 919 out:
 920         while (!list_empty(&list)) {
 921                 lfcc = list_entry(list.next, struct lfcc, list);
 922                 list_del(&lfcc->list);
 923                 gfs2_glock_dq_uninit(&lfcc->gh);
 924                 kfree(lfcc);
 925         }
 926         gfs2_glock_dq_uninit(&ji_gh);
 927         return error;
 928 }
 929
 930 /**
 931  * gfs2_freeze_fs - freezes the file system
 932  * @sdp: the file system
 933  *
 934  * This function flushes data and meta data for all machines by
 935  * aquiring the transaction log exclusively.  All journals are
 936  * ensured to be in a clean state as well.
 937  *
 938  * Returns: errno
 939  */
 940
 941 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 942 {
 943         int error = 0;
 944
 945         mutex_lock(&sdp->sd_freeze_lock);
 946
 947         if (!sdp->sd_freeze_count++) {
 948                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 949                 if (error)
 950                         sdp->sd_freeze_count--;
 951         }
 952
 953         mutex_unlock(&sdp->sd_freeze_lock);
 954
 955         return error;
 956 }
 957
 958 /**
 959  * gfs2_unfreeze_fs - unfreezes the file system
 960  * @sdp: the file system
 961  *
 962  * This function allows the file system to proceed by unlocking
 963  * the exclusively held transaction lock.  Other GFS2 nodes are
 964  * now free to acquire the lock shared and go on with their lives.
 965  *
 966  */
 967
 968 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 969 {
 970         mutex_lock(&sdp->sd_freeze_lock);
 971
 972         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 973                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 974
 975         mutex_unlock(&sdp->sd_freeze_lock);
 976 }
 977