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_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_host *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 /**
 184  * gfs2_read_super - Read the gfs2 super block from disk
 185  * @sb: The VFS super block
 186  * @sector: The location of the super block
 187  *
 188  * This uses the bio functions to read the super block from disk
 189  * because we want to be 100% sure that we never read cached data.
 190  * A super block is read twice only during each GFS2 mount and is
 191  * never written to by the filesystem. The first time its read no
 192  * locks are held, and the only details which are looked at are those
 193  * relating to the locking protocol. Once locking is up and working,
 194  * the sb is read again under the lock to establish the location of
 195  * the master directory (contains pointers to journals etc) and the
 196  * root directory.
 197  *
 198  * Returns: A page containing the sb or NULL
 199  */
 200
 201 struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
 202 {
 203         struct page *page;
 204         struct bio *bio;
 205
 206         page = alloc_page(GFP_KERNEL);
 207         if (unlikely(!page))
 208                 return NULL;
 209
 210         ClearPageUptodate(page);
 211         ClearPageDirty(page);
 212         lock_page(page);
 213
 214         bio = bio_alloc(GFP_KERNEL, 1);
 215         if (unlikely(!bio)) {
 216                 __free_page(page);
 217                 return NULL;
 218         }
 219
 220         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 221         bio->bi_bdev = sb->s_bdev;
 222         bio_add_page(bio, page, PAGE_SIZE, 0);
 223
 224         bio->bi_end_io = end_bio_io_page;
 225         bio->bi_private = page;
 226         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 227         wait_on_page_locked(page);
 228         bio_put(bio);
 229         if (!PageUptodate(page)) {
 230                 __free_page(page);
 231                 return NULL;
 232         }
 233         return page;
 234 }
 235
 236 /**
 237  * gfs2_read_sb - Read super block
 238  * @sdp: The GFS2 superblock
 239  * @gl: the glock for the superblock (assumed to be held)
 240  * @silent: Don't print message if mount fails
 241  *
 242  */
 243
 244 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 245 {
 246         u32 hash_blocks, ind_blocks, leaf_blocks;
 247         u32 tmp_blocks;
 248         unsigned int x;
 249         int error;
 250         struct page *page;
 251         char *sb;
 252
 253         page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 254         if (!page) {
 255                 if (!silent)
 256                         fs_err(sdp, "can't read superblock\n");
 257                 return -EIO;
 258         }
 259         sb = kmap(page);
 260         gfs2_sb_in(&sdp->sd_sb, sb);
 261         kunmap(page);
 262         __free_page(page);
 263
 264         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 265         if (error)
 266                 return error;
 267
 268         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 269                                GFS2_BASIC_BLOCK_SHIFT;
 270         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 271         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 272                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 273         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 274                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 275         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 276         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 277         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 278         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 279         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 280                                 sizeof(struct gfs2_meta_header)) /
 281                                 sizeof(struct gfs2_quota_change);
 282
 283         /* Compute maximum reservation required to add a entry to a directory */
 284
 285         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 286                              sdp->sd_jbsize);
 287
 288         ind_blocks = 0;
 289         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 290                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 291                 ind_blocks += tmp_blocks;
 292         }
 293
 294         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 295
 296         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 297
 298         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 299                                 sizeof(struct gfs2_dinode);
 300         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 301         for (x = 2;; x++) {
 302                 u64 space, d;
 303                 u32 m;
 304
 305                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 306                 d = space;
 307                 m = do_div(d, sdp->sd_inptrs);
 308
 309                 if (d != sdp->sd_heightsize[x - 1] || m)
 310                         break;
 311                 sdp->sd_heightsize[x] = space;
 312         }
 313         sdp->sd_max_height = x;
 314         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 315
 316         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 317                                  sizeof(struct gfs2_dinode);
 318         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 319         for (x = 2;; x++) {
 320                 u64 space, d;
 321                 u32 m;
 322
 323                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 324                 d = space;
 325                 m = do_div(d, sdp->sd_inptrs);
 326
 327                 if (d != sdp->sd_jheightsize[x - 1] || m)
 328                         break;
 329                 sdp->sd_jheightsize[x] = space;
 330         }
 331         sdp->sd_max_jheight = x;
 332         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 333
 334         return 0;
 335 }
 336
 337 /**
 338  * gfs2_jindex_hold - Grab a lock on the jindex
 339  * @sdp: The GFS2 superblock
 340  * @ji_gh: the holder for the jindex glock
 341  *
 342  * This is very similar to the gfs2_rindex_hold() function, except that
 343  * in general we hold the jindex lock for longer periods of time and
 344  * we grab it far less frequently (in general) then the rgrp lock.
 345  *
 346  * Returns: errno
 347  */
 348
 349 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 350 {
 351         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 352         struct qstr name;
 353         char buf[20];
 354         struct gfs2_jdesc *jd;
 355         int error;
 356
 357         name.name = buf;
 358
 359         mutex_lock(&sdp->sd_jindex_mutex);
 360
 361         for (;;) {
 362                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED,
 363                                            GL_LOCAL_EXCL, ji_gh);
 364                 if (error)
 365                         break;
 366
 367                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 368                 name.hash = gfs2_disk_hash(name.name, name.len);
 369
 370                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 371                 if (error == -ENOENT) {
 372                         error = 0;
 373                         break;
 374                 }
 375
 376                 gfs2_glock_dq_uninit(ji_gh);
 377
 378                 if (error)
 379                         break;
 380
 381                 error = -ENOMEM;
 382                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 383                 if (!jd)
 384                         break;
 385
 386                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 387                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 388                         if (!jd->jd_inode)
 389                                 error = -ENOENT;
 390                         else
 391                                 error = PTR_ERR(jd->jd_inode);
 392                         kfree(jd);
 393                         break;
 394                 }
 395
 396                 spin_lock(&sdp->sd_jindex_spin);
 397                 jd->jd_jid = sdp->sd_journals++;
 398                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 399                 spin_unlock(&sdp->sd_jindex_spin);
 400         }
 401
 402         mutex_unlock(&sdp->sd_jindex_mutex);
 403
 404         return error;
 405 }
 406
 407 /**
 408  * gfs2_jindex_free - Clear all the journal index information
 409  * @sdp: The GFS2 superblock
 410  *
 411  */
 412
 413 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 414 {
 415         struct list_head list;
 416         struct gfs2_jdesc *jd;
 417
 418         spin_lock(&sdp->sd_jindex_spin);
 419         list_add(&list, &sdp->sd_jindex_list);
 420         list_del_init(&sdp->sd_jindex_list);
 421         sdp->sd_journals = 0;
 422         spin_unlock(&sdp->sd_jindex_spin);
 423
 424         while (!list_empty(&list)) {
 425                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 426                 list_del(&jd->jd_list);
 427                 iput(jd->jd_inode);
 428                 kfree(jd);
 429         }
 430 }
 431
 432 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 433 {
 434         struct gfs2_jdesc *jd;
 435         int found = 0;
 436
 437         list_for_each_entry(jd, head, jd_list) {
 438                 if (jd->jd_jid == jid) {
 439                         found = 1;
 440                         break;
 441                 }
 442         }
 443
 444         if (!found)
 445                 jd = NULL;
 446
 447         return jd;
 448 }
 449
 450 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 451 {
 452         struct gfs2_jdesc *jd;
 453
 454         spin_lock(&sdp->sd_jindex_spin);
 455         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 456         spin_unlock(&sdp->sd_jindex_spin);
 457
 458         return jd;
 459 }
 460
 461 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 462 {
 463         struct gfs2_jdesc *jd;
 464
 465         spin_lock(&sdp->sd_jindex_spin);
 466         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 467         if (jd)
 468                 jd->jd_dirty = 1;
 469         spin_unlock(&sdp->sd_jindex_spin);
 470 }
 471
 472 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 473 {
 474         struct gfs2_jdesc *jd;
 475         int found = 0;
 476
 477         spin_lock(&sdp->sd_jindex_spin);
 478
 479         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 480                 if (jd->jd_dirty) {
 481                         jd->jd_dirty = 0;
 482                         found = 1;
 483                         break;
 484                 }
 485         }
 486         spin_unlock(&sdp->sd_jindex_spin);
 487
 488         if (!found)
 489                 jd = NULL;
 490
 491         return jd;
 492 }
 493
 494 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 495 {
 496         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 497         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 498         int ar;
 499         int error;
 500
 501         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 502             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 503                 gfs2_consist_inode(ip);
 504                 return -EIO;
 505         }
 506         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 507
 508         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 509         if (!error && ar) {
 510                 gfs2_consist_inode(ip);
 511                 error = -EIO;
 512         }
 513
 514         return error;
 515 }
 516
 517 /**
 518  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 519  * @sdp: the filesystem
 520  *
 521  * Returns: errno
 522  */
 523
 524 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 525 {
 526         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 527         struct gfs2_glock *j_gl = ip->i_gl;
 528         struct gfs2_holder t_gh;
 529         struct gfs2_log_header_host head;
 530         int error;
 531
 532         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 533                                    GL_LOCAL_EXCL, &t_gh);
 534         if (error)
 535                 return error;
 536
 537         gfs2_meta_cache_flush(ip);
 538         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 539
 540         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 541         if (error)
 542                 goto fail;
 543
 544         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 545                 gfs2_consist(sdp);
 546                 error = -EIO;
 547                 goto fail;
 548         }
 549
 550         /*  Initialize some head of the log stuff  */
 551         sdp->sd_log_sequence = head.lh_sequence + 1;
 552         gfs2_log_pointers_init(sdp, head.lh_blkno);
 553
 554         error = gfs2_quota_init(sdp);
 555         if (error)
 556                 goto fail;
 557
 558         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 559
 560         gfs2_glock_dq_uninit(&t_gh);
 561
 562         return 0;
 563
 564 fail:
 565         t_gh.gh_flags |= GL_NOCACHE;
 566         gfs2_glock_dq_uninit(&t_gh);
 567
 568         return error;
 569 }
 570
 571 /**
 572  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 573  * @sdp: the filesystem
 574  *
 575  * Returns: errno
 576  */
 577
 578 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 579 {
 580         struct gfs2_holder t_gh;
 581         int error;
 582
 583         gfs2_quota_sync(sdp);
 584         gfs2_statfs_sync(sdp);
 585
 586         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 587                                 GL_LOCAL_EXCL | GL_NOCACHE,
 588                                 &t_gh);
 589         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 590                 return error;
 591
 592         gfs2_meta_syncfs(sdp);
 593         gfs2_log_shutdown(sdp);
 594
 595         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 596
 597         if (t_gh.gh_gl)
 598                 gfs2_glock_dq_uninit(&t_gh);
 599
 600         gfs2_quota_cleanup(sdp);
 601
 602         return error;
 603 }
 604
 605 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 606 {
 607         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 608         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 609         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 610         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 611         struct buffer_head *m_bh, *l_bh;
 612         struct gfs2_holder gh;
 613         int error;
 614
 615         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 616                                    &gh);
 617         if (error)
 618                 return error;
 619
 620         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 621         if (error)
 622                 goto out;
 623
 624         if (sdp->sd_args.ar_spectator) {
 625                 spin_lock(&sdp->sd_statfs_spin);
 626                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 627                                       sizeof(struct gfs2_dinode));
 628                 spin_unlock(&sdp->sd_statfs_spin);
 629         } else {
 630                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 631                 if (error)
 632                         goto out_m_bh;
 633
 634                 spin_lock(&sdp->sd_statfs_spin);
 635                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 636                                       sizeof(struct gfs2_dinode));
 637                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 638                                       sizeof(struct gfs2_dinode));
 639                 spin_unlock(&sdp->sd_statfs_spin);
 640
 641                 brelse(l_bh);
 642         }
 643
 644 out_m_bh:
 645         brelse(m_bh);
 646 out:
 647         gfs2_glock_dq_uninit(&gh);
 648         return 0;
 649 }
 650
 651 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 652                         s64 dinodes)
 653 {
 654         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 655         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 656         struct buffer_head *l_bh;
 657         int error;
 658
 659         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 660         if (error)
 661                 return;
 662
 663         mutex_lock(&sdp->sd_statfs_mutex);
 664         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 665         mutex_unlock(&sdp->sd_statfs_mutex);
 666
 667         spin_lock(&sdp->sd_statfs_spin);
 668         l_sc->sc_total += total;
 669         l_sc->sc_free += free;
 670         l_sc->sc_dinodes += dinodes;
 671         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 672         spin_unlock(&sdp->sd_statfs_spin);
 673
 674         brelse(l_bh);
 675 }
 676
 677 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 678 {
 679         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 680         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 681         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 682         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 683         struct gfs2_holder gh;
 684         struct buffer_head *m_bh, *l_bh;
 685         int error;
 686
 687         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 688                                    &gh);
 689         if (error)
 690                 return error;
 691
 692         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 693         if (error)
 694                 goto out;
 695
 696         spin_lock(&sdp->sd_statfs_spin);
 697         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 698                               sizeof(struct gfs2_dinode));
 699         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 700                 spin_unlock(&sdp->sd_statfs_spin);
 701                 goto out_bh;
 702         }
 703         spin_unlock(&sdp->sd_statfs_spin);
 704
 705         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 706         if (error)
 707                 goto out_bh;
 708
 709         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 710         if (error)
 711                 goto out_bh2;
 712
 713         mutex_lock(&sdp->sd_statfs_mutex);
 714         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 715         mutex_unlock(&sdp->sd_statfs_mutex);
 716
 717         spin_lock(&sdp->sd_statfs_spin);
 718         m_sc->sc_total += l_sc->sc_total;
 719         m_sc->sc_free += l_sc->sc_free;
 720         m_sc->sc_dinodes += l_sc->sc_dinodes;
 721         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 722         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 723                0, sizeof(struct gfs2_statfs_change));
 724         spin_unlock(&sdp->sd_statfs_spin);
 725
 726         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 727         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 728
 729         gfs2_trans_end(sdp);
 730
 731 out_bh2:
 732         brelse(l_bh);
 733 out_bh:
 734         brelse(m_bh);
 735 out:
 736         gfs2_glock_dq_uninit(&gh);
 737         return error;
 738 }
 739
 740 /**
 741  * gfs2_statfs_i - Do a statfs
 742  * @sdp: the filesystem
 743  * @sg: the sg structure
 744  *
 745  * Returns: errno
 746  */
 747
 748 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 749 {
 750         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 751         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 752
 753         spin_lock(&sdp->sd_statfs_spin);
 754
 755         *sc = *m_sc;
 756         sc->sc_total += l_sc->sc_total;
 757         sc->sc_free += l_sc->sc_free;
 758         sc->sc_dinodes += l_sc->sc_dinodes;
 759
 760         spin_unlock(&sdp->sd_statfs_spin);
 761
 762         if (sc->sc_free < 0)
 763                 sc->sc_free = 0;
 764         if (sc->sc_free > sc->sc_total)
 765                 sc->sc_free = sc->sc_total;
 766         if (sc->sc_dinodes < 0)
 767                 sc->sc_dinodes = 0;
 768
 769         return 0;
 770 }
 771
 772 /**
 773  * statfs_fill - fill in the sg for a given RG
 774  * @rgd: the RG
 775  * @sc: the sc structure
 776  *
 777  * Returns: 0 on success, -ESTALE if the LVB is invalid
 778  */
 779
 780 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 781                             struct gfs2_statfs_change_host *sc)
 782 {
 783         gfs2_rgrp_verify(rgd);
 784         sc->sc_total += rgd->rd_ri.ri_data;
 785         sc->sc_free += rgd->rd_rg.rg_free;
 786         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 787         return 0;
 788 }
 789
 790 /**
 791  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 792  * @sdp: the filesystem
 793  * @sc: the sc info that will be returned
 794  *
 795  * Any error (other than a signal) will cause this routine to fall back
 796  * to the synchronous version.
 797  *
 798  * FIXME: This really shouldn't busy wait like this.
 799  *
 800  * Returns: errno
 801  */
 802
 803 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 804 {
 805         struct gfs2_holder ri_gh;
 806         struct gfs2_rgrpd *rgd_next;
 807         struct gfs2_holder *gha, *gh;
 808         unsigned int slots = 64;
 809         unsigned int x;
 810         int done;
 811         int error = 0, err;
 812
 813         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 814         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 815         if (!gha)
 816                 return -ENOMEM;
 817
 818         error = gfs2_rindex_hold(sdp, &ri_gh);
 819         if (error)
 820                 goto out;
 821
 822         rgd_next = gfs2_rgrpd_get_first(sdp);
 823
 824         for (;;) {
 825                 done = 1;
 826
 827                 for (x = 0; x < slots; x++) {
 828                         gh = gha + x;
 829
 830                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 831                                 err = gfs2_glock_wait(gh);
 832                                 if (err) {
 833                                         gfs2_holder_uninit(gh);
 834                                         error = err;
 835                                 } else {
 836                                         if (!error)
 837                                                 error = statfs_slow_fill(
 838                                                         gh->gh_gl->gl_object, sc);
 839                                         gfs2_glock_dq_uninit(gh);
 840                                 }
 841                         }
 842
 843                         if (gh->gh_gl)
 844                                 done = 0;
 845                         else if (rgd_next && !error) {
 846                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 847                                                            LM_ST_SHARED,
 848                                                            GL_ASYNC,
 849                                                            gh);
 850                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 851                                 done = 0;
 852                         }
 853
 854                         if (signal_pending(current))
 855                                 error = -ERESTARTSYS;
 856                 }
 857
 858                 if (done)
 859                         break;
 860
 861                 yield();
 862         }
 863
 864         gfs2_glock_dq_uninit(&ri_gh);
 865
 866 out:
 867         kfree(gha);
 868         return error;
 869 }
 870
 871 struct lfcc {
 872         struct list_head list;
 873         struct gfs2_holder gh;
 874 };
 875
 876 /**
 877  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 878  *                            journals are clean
 879  * @sdp: the file system
 880  * @state: the state to put the transaction lock into
 881  * @t_gh: the hold on the transaction lock
 882  *
 883  * Returns: errno
 884  */
 885
 886 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 887                                     struct gfs2_holder *t_gh)
 888 {
 889         struct gfs2_inode *ip;
 890         struct gfs2_holder ji_gh;
 891         struct gfs2_jdesc *jd;
 892         struct lfcc *lfcc;
 893         LIST_HEAD(list);
 894         struct gfs2_log_header_host lh;
 895         int error;
 896
 897         error = gfs2_jindex_hold(sdp, &ji_gh);
 898         if (error)
 899                 return error;
 900
 901         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 902                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 903                 if (!lfcc) {
 904                         error = -ENOMEM;
 905                         goto out;
 906                 }
 907                 ip = GFS2_I(jd->jd_inode);
 908                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 909                 if (error) {
 910                         kfree(lfcc);
 911                         goto out;
 912                 }
 913                 list_add(&lfcc->list, &list);
 914         }
 915
 916         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 917                                LM_FLAG_PRIORITY | GL_NOCACHE,
 918                                t_gh);
 919
 920         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 921                 error = gfs2_jdesc_check(jd);
 922                 if (error)
 923                         break;
 924                 error = gfs2_find_jhead(jd, &lh);
 925                 if (error)
 926                         break;
 927                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 928                         error = -EBUSY;
 929                         break;
 930                 }
 931         }
 932
 933         if (error)
 934                 gfs2_glock_dq_uninit(t_gh);
 935
 936 out:
 937         while (!list_empty(&list)) {
 938                 lfcc = list_entry(list.next, struct lfcc, list);
 939                 list_del(&lfcc->list);
 940                 gfs2_glock_dq_uninit(&lfcc->gh);
 941                 kfree(lfcc);
 942         }
 943         gfs2_glock_dq_uninit(&ji_gh);
 944         return error;
 945 }
 946
 947 /**
 948  * gfs2_freeze_fs - freezes the file system
 949  * @sdp: the file system
 950  *
 951  * This function flushes data and meta data for all machines by
 952  * aquiring the transaction log exclusively.  All journals are
 953  * ensured to be in a clean state as well.
 954  *
 955  * Returns: errno
 956  */
 957
 958 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 959 {
 960         int error = 0;
 961
 962         mutex_lock(&sdp->sd_freeze_lock);
 963
 964         if (!sdp->sd_freeze_count++) {
 965                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 966                 if (error)
 967                         sdp->sd_freeze_count--;
 968         }
 969
 970         mutex_unlock(&sdp->sd_freeze_lock);
 971
 972         return error;
 973 }
 974
 975 /**
 976  * gfs2_unfreeze_fs - unfreezes the file system
 977  * @sdp: the file system
 978  *
 979  * This function allows the file system to proceed by unlocking
 980  * the exclusively held transaction lock.  Other GFS2 nodes are
 981  * now free to acquire the lock shared and go on with their lives.
 982  *
 983  */
 984
 985 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 986 {
 987         mutex_lock(&sdp->sd_freeze_lock);
 988
 989         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 990                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 991
 992         mutex_unlock(&sdp->sd_freeze_lock);
 993 }
 994