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_greedy_default = HZ / 10;
  81         gt->gt_greedy_quantum = HZ / 40;
  82         gt->gt_greedy_max = HZ / 4;
  83         gt->gt_statfs_quantum = 30;
  84         gt->gt_statfs_slow = 0;
  85 }
  86
  87 /**
  88  * gfs2_check_sb - Check superblock
  89  * @sdp: the filesystem
  90  * @sb: The superblock
  91  * @silent: Don't print a message if the check fails
  92  *
  93  * Checks the version code of the FS is one that we understand how to
  94  * read and that the sizes of the various on-disk structures have not
  95  * changed.
  96  */
  97
  98 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  99 {
 100         unsigned int x;
 101
 102         if (sb->sb_header.mh_magic != GFS2_MAGIC ||
 103             sb->sb_header.mh_type != GFS2_METATYPE_SB) {
 104                 if (!silent)
 105                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 106                 return -EINVAL;
 107         }
 108
 109         /*  If format numbers match exactly, we're done.  */
 110
 111         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 112             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 113                 return 0;
 114
 115         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 116                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 117                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 118                                 break;
 119
 120                 if (!gfs2_old_fs_formats[x]) {
 121                         printk(KERN_WARNING
 122                                "GFS2: code version (%u, %u) is incompatible "
 123                                "with ondisk format (%u, %u)\n",
 124                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 125                                sb->sb_fs_format, sb->sb_multihost_format);
 126                         printk(KERN_WARNING
 127                                "GFS2: I don't know how to upgrade this FS\n");
 128                         return -EINVAL;
 129                 }
 130         }
 131
 132         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 133                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 134                         if (gfs2_old_multihost_formats[x] ==
 135                             sb->sb_multihost_format)
 136                                 break;
 137
 138                 if (!gfs2_old_multihost_formats[x]) {
 139                         printk(KERN_WARNING
 140                                "GFS2: code version (%u, %u) is incompatible "
 141                                "with ondisk format (%u, %u)\n",
 142                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 143                                sb->sb_fs_format, sb->sb_multihost_format);
 144                         printk(KERN_WARNING
 145                                "GFS2: I don't know how to upgrade this FS\n");
 146                         return -EINVAL;
 147                 }
 148         }
 149
 150         if (!sdp->sd_args.ar_upgrade) {
 151                 printk(KERN_WARNING
 152                        "GFS2: code version (%u, %u) is incompatible "
 153                        "with ondisk format (%u, %u)\n",
 154                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 155                        sb->sb_fs_format, sb->sb_multihost_format);
 156                 printk(KERN_INFO
 157                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 158                        "the FS\n");
 159                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 160                 return -EINVAL;
 161         }
 162
 163         return 0;
 164 }
 165
 166
 167 static int end_bio_io_page(struct bio *bio, unsigned int bytes_done, int error)
 168 {
 169         struct page *page = bio->bi_private;
 170         if (bio->bi_size)
 171                 return 1;
 172
 173         if (!error)
 174                 SetPageUptodate(page);
 175         else
 176                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 177         unlock_page(page);
 178         return 0;
 179 }
 180
 181 /**
 182  * gfs2_read_super - Read the gfs2 super block from disk
 183  * @sb: The VFS super block
 184  * @sector: The location of the super block
 185  *
 186  * This uses the bio functions to read the super block from disk
 187  * because we want to be 100% sure that we never read cached data.
 188  * A super block is read twice only during each GFS2 mount and is
 189  * never written to by the filesystem. The first time its read no
 190  * locks are held, and the only details which are looked at are those
 191  * relating to the locking protocol. Once locking is up and working,
 192  * the sb is read again under the lock to establish the location of
 193  * the master directory (contains pointers to journals etc) and the
 194  * root directory.
 195  *
 196  * Returns: A page containing the sb or NULL
 197  */
 198
 199 struct page *gfs2_read_super(struct super_block *sb, sector_t sector)
 200 {
 201         struct page *page;
 202         struct bio *bio;
 203
 204         page = alloc_page(GFP_KERNEL);
 205         if (unlikely(!page))
 206                 return NULL;
 207
 208         ClearPageUptodate(page);
 209         ClearPageDirty(page);
 210         lock_page(page);
 211
 212         bio = bio_alloc(GFP_KERNEL, 1);
 213         if (unlikely(!bio)) {
 214                 __free_page(page);
 215                 return NULL;
 216         }
 217
 218         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 219         bio->bi_bdev = sb->s_bdev;
 220         bio_add_page(bio, page, PAGE_SIZE, 0);
 221
 222         bio->bi_end_io = end_bio_io_page;
 223         bio->bi_private = page;
 224         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 225         wait_on_page_locked(page);
 226         bio_put(bio);
 227         if (!PageUptodate(page)) {
 228                 __free_page(page);
 229                 return NULL;
 230         }
 231         return page;
 232 }
 233
 234 /**
 235  * gfs2_read_sb - Read super block
 236  * @sdp: The GFS2 superblock
 237  * @gl: the glock for the superblock (assumed to be held)
 238  * @silent: Don't print message if mount fails
 239  *
 240  */
 241
 242 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 243 {
 244         u32 hash_blocks, ind_blocks, leaf_blocks;
 245         u32 tmp_blocks;
 246         unsigned int x;
 247         int error;
 248         struct page *page;
 249         char *sb;
 250
 251         page = gfs2_read_super(sdp->sd_vfs, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 252         if (!page) {
 253                 if (!silent)
 254                         fs_err(sdp, "can't read superblock\n");
 255                 return -EIO;
 256         }
 257         sb = kmap(page);
 258         gfs2_sb_in(&sdp->sd_sb, sb);
 259         kunmap(page);
 260         __free_page(page);
 261
 262         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 263         if (error)
 264                 return error;
 265
 266         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 267                                GFS2_BASIC_BLOCK_SHIFT;
 268         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 269         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 270                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 271         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 272                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 273         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 274         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 275         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 276         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 277         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 278                                 sizeof(struct gfs2_meta_header)) /
 279                                 sizeof(struct gfs2_quota_change);
 280
 281         /* Compute maximum reservation required to add a entry to a directory */
 282
 283         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 284                              sdp->sd_jbsize);
 285
 286         ind_blocks = 0;
 287         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 288                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 289                 ind_blocks += tmp_blocks;
 290         }
 291
 292         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 293
 294         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 295
 296         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 297                                 sizeof(struct gfs2_dinode);
 298         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 299         for (x = 2;; x++) {
 300                 u64 space, d;
 301                 u32 m;
 302
 303                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 304                 d = space;
 305                 m = do_div(d, sdp->sd_inptrs);
 306
 307                 if (d != sdp->sd_heightsize[x - 1] || m)
 308                         break;
 309                 sdp->sd_heightsize[x] = space;
 310         }
 311         sdp->sd_max_height = x;
 312         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 313
 314         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 315                                  sizeof(struct gfs2_dinode);
 316         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 317         for (x = 2;; x++) {
 318                 u64 space, d;
 319                 u32 m;
 320
 321                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 322                 d = space;
 323                 m = do_div(d, sdp->sd_inptrs);
 324
 325                 if (d != sdp->sd_jheightsize[x - 1] || m)
 326                         break;
 327                 sdp->sd_jheightsize[x] = space;
 328         }
 329         sdp->sd_max_jheight = x;
 330         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 331
 332         return 0;
 333 }
 334
 335 /**
 336  * gfs2_jindex_hold - Grab a lock on the jindex
 337  * @sdp: The GFS2 superblock
 338  * @ji_gh: the holder for the jindex glock
 339  *
 340  * This is very similar to the gfs2_rindex_hold() function, except that
 341  * in general we hold the jindex lock for longer periods of time and
 342  * we grab it far less frequently (in general) then the rgrp lock.
 343  *
 344  * Returns: errno
 345  */
 346
 347 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 348 {
 349         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 350         struct qstr name;
 351         char buf[20];
 352         struct gfs2_jdesc *jd;
 353         int error;
 354
 355         name.name = buf;
 356
 357         mutex_lock(&sdp->sd_jindex_mutex);
 358
 359         for (;;) {
 360                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED,
 361                                            GL_LOCAL_EXCL, ji_gh);
 362                 if (error)
 363                         break;
 364
 365                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 366                 name.hash = gfs2_disk_hash(name.name, name.len);
 367
 368                 error = gfs2_dir_search(sdp->sd_jindex, &name, NULL, NULL);
 369                 if (error == -ENOENT) {
 370                         error = 0;
 371                         break;
 372                 }
 373
 374                 gfs2_glock_dq_uninit(ji_gh);
 375
 376                 if (error)
 377                         break;
 378
 379                 error = -ENOMEM;
 380                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 381                 if (!jd)
 382                         break;
 383
 384                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 385                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 386                         if (!jd->jd_inode)
 387                                 error = -ENOENT;
 388                         else
 389                                 error = PTR_ERR(jd->jd_inode);
 390                         kfree(jd);
 391                         break;
 392                 }
 393
 394                 spin_lock(&sdp->sd_jindex_spin);
 395                 jd->jd_jid = sdp->sd_journals++;
 396                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 397                 spin_unlock(&sdp->sd_jindex_spin);
 398         }
 399
 400         mutex_unlock(&sdp->sd_jindex_mutex);
 401
 402         return error;
 403 }
 404
 405 /**
 406  * gfs2_jindex_free - Clear all the journal index information
 407  * @sdp: The GFS2 superblock
 408  *
 409  */
 410
 411 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 412 {
 413         struct list_head list;
 414         struct gfs2_jdesc *jd;
 415
 416         spin_lock(&sdp->sd_jindex_spin);
 417         list_add(&list, &sdp->sd_jindex_list);
 418         list_del_init(&sdp->sd_jindex_list);
 419         sdp->sd_journals = 0;
 420         spin_unlock(&sdp->sd_jindex_spin);
 421
 422         while (!list_empty(&list)) {
 423                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 424                 list_del(&jd->jd_list);
 425                 iput(jd->jd_inode);
 426                 kfree(jd);
 427         }
 428 }
 429
 430 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 431 {
 432         struct gfs2_jdesc *jd;
 433         int found = 0;
 434
 435         list_for_each_entry(jd, head, jd_list) {
 436                 if (jd->jd_jid == jid) {
 437                         found = 1;
 438                         break;
 439                 }
 440         }
 441
 442         if (!found)
 443                 jd = NULL;
 444
 445         return jd;
 446 }
 447
 448 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 449 {
 450         struct gfs2_jdesc *jd;
 451
 452         spin_lock(&sdp->sd_jindex_spin);
 453         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 454         spin_unlock(&sdp->sd_jindex_spin);
 455
 456         return jd;
 457 }
 458
 459 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 460 {
 461         struct gfs2_jdesc *jd;
 462
 463         spin_lock(&sdp->sd_jindex_spin);
 464         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 465         if (jd)
 466                 jd->jd_dirty = 1;
 467         spin_unlock(&sdp->sd_jindex_spin);
 468 }
 469
 470 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 471 {
 472         struct gfs2_jdesc *jd;
 473         int found = 0;
 474
 475         spin_lock(&sdp->sd_jindex_spin);
 476
 477         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 478                 if (jd->jd_dirty) {
 479                         jd->jd_dirty = 0;
 480                         found = 1;
 481                         break;
 482                 }
 483         }
 484         spin_unlock(&sdp->sd_jindex_spin);
 485
 486         if (!found)
 487                 jd = NULL;
 488
 489         return jd;
 490 }
 491
 492 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 493 {
 494         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 495         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 496         int ar;
 497         int error;
 498
 499         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 500             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 501                 gfs2_consist_inode(ip);
 502                 return -EIO;
 503         }
 504         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 505
 506         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 507         if (!error && ar) {
 508                 gfs2_consist_inode(ip);
 509                 error = -EIO;
 510         }
 511
 512         return error;
 513 }
 514
 515 /**
 516  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 517  * @sdp: the filesystem
 518  *
 519  * Returns: errno
 520  */
 521
 522 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 523 {
 524         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 525         struct gfs2_glock *j_gl = ip->i_gl;
 526         struct gfs2_holder t_gh;
 527         struct gfs2_log_header_host head;
 528         int error;
 529
 530         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 531                                    GL_LOCAL_EXCL, &t_gh);
 532         if (error)
 533                 return error;
 534
 535         gfs2_meta_cache_flush(ip);
 536         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 537
 538         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 539         if (error)
 540                 goto fail;
 541
 542         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 543                 gfs2_consist(sdp);
 544                 error = -EIO;
 545                 goto fail;
 546         }
 547
 548         /*  Initialize some head of the log stuff  */
 549         sdp->sd_log_sequence = head.lh_sequence + 1;
 550         gfs2_log_pointers_init(sdp, head.lh_blkno);
 551
 552         error = gfs2_quota_init(sdp);
 553         if (error)
 554                 goto fail;
 555
 556         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 557
 558         gfs2_glock_dq_uninit(&t_gh);
 559
 560         return 0;
 561
 562 fail:
 563         t_gh.gh_flags |= GL_NOCACHE;
 564         gfs2_glock_dq_uninit(&t_gh);
 565
 566         return error;
 567 }
 568
 569 /**
 570  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 571  * @sdp: the filesystem
 572  *
 573  * Returns: errno
 574  */
 575
 576 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 577 {
 578         struct gfs2_holder t_gh;
 579         int error;
 580
 581         gfs2_quota_sync(sdp);
 582         gfs2_statfs_sync(sdp);
 583
 584         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED,
 585                                 GL_LOCAL_EXCL | GL_NOCACHE,
 586                                 &t_gh);
 587         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 588                 return error;
 589
 590         gfs2_meta_syncfs(sdp);
 591         gfs2_log_shutdown(sdp);
 592
 593         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 594
 595         if (t_gh.gh_gl)
 596                 gfs2_glock_dq_uninit(&t_gh);
 597
 598         gfs2_quota_cleanup(sdp);
 599
 600         return error;
 601 }
 602
 603 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 604 {
 605         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 606         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 607         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 608         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 609         struct buffer_head *m_bh, *l_bh;
 610         struct gfs2_holder gh;
 611         int error;
 612
 613         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 614                                    &gh);
 615         if (error)
 616                 return error;
 617
 618         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 619         if (error)
 620                 goto out;
 621
 622         if (sdp->sd_args.ar_spectator) {
 623                 spin_lock(&sdp->sd_statfs_spin);
 624                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 625                                       sizeof(struct gfs2_dinode));
 626                 spin_unlock(&sdp->sd_statfs_spin);
 627         } else {
 628                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 629                 if (error)
 630                         goto out_m_bh;
 631
 632                 spin_lock(&sdp->sd_statfs_spin);
 633                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 634                                       sizeof(struct gfs2_dinode));
 635                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 636                                       sizeof(struct gfs2_dinode));
 637                 spin_unlock(&sdp->sd_statfs_spin);
 638
 639                 brelse(l_bh);
 640         }
 641
 642 out_m_bh:
 643         brelse(m_bh);
 644 out:
 645         gfs2_glock_dq_uninit(&gh);
 646         return 0;
 647 }
 648
 649 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 650                         s64 dinodes)
 651 {
 652         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 653         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 654         struct buffer_head *l_bh;
 655         int error;
 656
 657         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 658         if (error)
 659                 return;
 660
 661         mutex_lock(&sdp->sd_statfs_mutex);
 662         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 663         mutex_unlock(&sdp->sd_statfs_mutex);
 664
 665         spin_lock(&sdp->sd_statfs_spin);
 666         l_sc->sc_total += total;
 667         l_sc->sc_free += free;
 668         l_sc->sc_dinodes += dinodes;
 669         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 670         spin_unlock(&sdp->sd_statfs_spin);
 671
 672         brelse(l_bh);
 673 }
 674
 675 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 676 {
 677         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 678         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 679         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 680         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 681         struct gfs2_holder gh;
 682         struct buffer_head *m_bh, *l_bh;
 683         int error;
 684
 685         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 686                                    &gh);
 687         if (error)
 688                 return error;
 689
 690         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 691         if (error)
 692                 goto out;
 693
 694         spin_lock(&sdp->sd_statfs_spin);
 695         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 696                               sizeof(struct gfs2_dinode));
 697         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 698                 spin_unlock(&sdp->sd_statfs_spin);
 699                 goto out_bh;
 700         }
 701         spin_unlock(&sdp->sd_statfs_spin);
 702
 703         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 704         if (error)
 705                 goto out_bh;
 706
 707         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 708         if (error)
 709                 goto out_bh2;
 710
 711         mutex_lock(&sdp->sd_statfs_mutex);
 712         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 713         mutex_unlock(&sdp->sd_statfs_mutex);
 714
 715         spin_lock(&sdp->sd_statfs_spin);
 716         m_sc->sc_total += l_sc->sc_total;
 717         m_sc->sc_free += l_sc->sc_free;
 718         m_sc->sc_dinodes += l_sc->sc_dinodes;
 719         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 720         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 721                0, sizeof(struct gfs2_statfs_change));
 722         spin_unlock(&sdp->sd_statfs_spin);
 723
 724         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 725         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 726
 727         gfs2_trans_end(sdp);
 728
 729 out_bh2:
 730         brelse(l_bh);
 731 out_bh:
 732         brelse(m_bh);
 733 out:
 734         gfs2_glock_dq_uninit(&gh);
 735         return error;
 736 }
 737
 738 /**
 739  * gfs2_statfs_i - Do a statfs
 740  * @sdp: the filesystem
 741  * @sg: the sg structure
 742  *
 743  * Returns: errno
 744  */
 745
 746 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 747 {
 748         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 749         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 750
 751         spin_lock(&sdp->sd_statfs_spin);
 752
 753         *sc = *m_sc;
 754         sc->sc_total += l_sc->sc_total;
 755         sc->sc_free += l_sc->sc_free;
 756         sc->sc_dinodes += l_sc->sc_dinodes;
 757
 758         spin_unlock(&sdp->sd_statfs_spin);
 759
 760         if (sc->sc_free < 0)
 761                 sc->sc_free = 0;
 762         if (sc->sc_free > sc->sc_total)
 763                 sc->sc_free = sc->sc_total;
 764         if (sc->sc_dinodes < 0)
 765                 sc->sc_dinodes = 0;
 766
 767         return 0;
 768 }
 769
 770 /**
 771  * statfs_fill - fill in the sg for a given RG
 772  * @rgd: the RG
 773  * @sc: the sc structure
 774  *
 775  * Returns: 0 on success, -ESTALE if the LVB is invalid
 776  */
 777
 778 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 779                             struct gfs2_statfs_change_host *sc)
 780 {
 781         gfs2_rgrp_verify(rgd);
 782         sc->sc_total += rgd->rd_ri.ri_data;
 783         sc->sc_free += rgd->rd_rg.rg_free;
 784         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 785         return 0;
 786 }
 787
 788 /**
 789  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 790  * @sdp: the filesystem
 791  * @sc: the sc info that will be returned
 792  *
 793  * Any error (other than a signal) will cause this routine to fall back
 794  * to the synchronous version.
 795  *
 796  * FIXME: This really shouldn't busy wait like this.
 797  *
 798  * Returns: errno
 799  */
 800
 801 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 802 {
 803         struct gfs2_holder ri_gh;
 804         struct gfs2_rgrpd *rgd_next;
 805         struct gfs2_holder *gha, *gh;
 806         unsigned int slots = 64;
 807         unsigned int x;
 808         int done;
 809         int error = 0, err;
 810
 811         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 812         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 813         if (!gha)
 814                 return -ENOMEM;
 815
 816         error = gfs2_rindex_hold(sdp, &ri_gh);
 817         if (error)
 818                 goto out;
 819
 820         rgd_next = gfs2_rgrpd_get_first(sdp);
 821
 822         for (;;) {
 823                 done = 1;
 824
 825                 for (x = 0; x < slots; x++) {
 826                         gh = gha + x;
 827
 828                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 829                                 err = gfs2_glock_wait(gh);
 830                                 if (err) {
 831                                         gfs2_holder_uninit(gh);
 832                                         error = err;
 833                                 } else {
 834                                         if (!error)
 835                                                 error = statfs_slow_fill(
 836                                                         gh->gh_gl->gl_object, sc);
 837                                         gfs2_glock_dq_uninit(gh);
 838                                 }
 839                         }
 840
 841                         if (gh->gh_gl)
 842                                 done = 0;
 843                         else if (rgd_next && !error) {
 844                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 845                                                            LM_ST_SHARED,
 846                                                            GL_ASYNC,
 847                                                            gh);
 848                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 849                                 done = 0;
 850                         }
 851
 852                         if (signal_pending(current))
 853                                 error = -ERESTARTSYS;
 854                 }
 855
 856                 if (done)
 857                         break;
 858
 859                 yield();
 860         }
 861
 862         gfs2_glock_dq_uninit(&ri_gh);
 863
 864 out:
 865         kfree(gha);
 866         return error;
 867 }
 868
 869 struct lfcc {
 870         struct list_head list;
 871         struct gfs2_holder gh;
 872 };
 873
 874 /**
 875  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 876  *                            journals are clean
 877  * @sdp: the file system
 878  * @state: the state to put the transaction lock into
 879  * @t_gh: the hold on the transaction lock
 880  *
 881  * Returns: errno
 882  */
 883
 884 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 885                                     struct gfs2_holder *t_gh)
 886 {
 887         struct gfs2_inode *ip;
 888         struct gfs2_holder ji_gh;
 889         struct gfs2_jdesc *jd;
 890         struct lfcc *lfcc;
 891         LIST_HEAD(list);
 892         struct gfs2_log_header_host lh;
 893         int error;
 894
 895         error = gfs2_jindex_hold(sdp, &ji_gh);
 896         if (error)
 897                 return error;
 898
 899         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 900                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 901                 if (!lfcc) {
 902                         error = -ENOMEM;
 903                         goto out;
 904                 }
 905                 ip = GFS2_I(jd->jd_inode);
 906                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 907                 if (error) {
 908                         kfree(lfcc);
 909                         goto out;
 910                 }
 911                 list_add(&lfcc->list, &list);
 912         }
 913
 914         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 915                                LM_FLAG_PRIORITY | GL_NOCACHE,
 916                                t_gh);
 917
 918         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 919                 error = gfs2_jdesc_check(jd);
 920                 if (error)
 921                         break;
 922                 error = gfs2_find_jhead(jd, &lh);
 923                 if (error)
 924                         break;
 925                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 926                         error = -EBUSY;
 927                         break;
 928                 }
 929         }
 930
 931         if (error)
 932                 gfs2_glock_dq_uninit(t_gh);
 933
 934 out:
 935         while (!list_empty(&list)) {
 936                 lfcc = list_entry(list.next, struct lfcc, list);
 937                 list_del(&lfcc->list);
 938                 gfs2_glock_dq_uninit(&lfcc->gh);
 939                 kfree(lfcc);
 940         }
 941         gfs2_glock_dq_uninit(&ji_gh);
 942         return error;
 943 }
 944
 945 /**
 946  * gfs2_freeze_fs - freezes the file system
 947  * @sdp: the file system
 948  *
 949  * This function flushes data and meta data for all machines by
 950  * aquiring the transaction log exclusively.  All journals are
 951  * ensured to be in a clean state as well.
 952  *
 953  * Returns: errno
 954  */
 955
 956 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 957 {
 958         int error = 0;
 959
 960         mutex_lock(&sdp->sd_freeze_lock);
 961
 962         if (!sdp->sd_freeze_count++) {
 963                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 964                 if (error)
 965                         sdp->sd_freeze_count--;
 966         }
 967
 968         mutex_unlock(&sdp->sd_freeze_lock);
 969
 970         return error;
 971 }
 972
 973 /**
 974  * gfs2_unfreeze_fs - unfreezes the file system
 975  * @sdp: the file system
 976  *
 977  * This function allows the file system to proceed by unlocking
 978  * the exclusively held transaction lock.  Other GFS2 nodes are
 979  * now free to acquire the lock shared and go on with their lives.
 980  *
 981  */
 982
 983 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
 984 {
 985         mutex_lock(&sdp->sd_freeze_lock);
 986
 987         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
 988                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
 989
 990         mutex_unlock(&sdp->sd_freeze_lock);
 991 }
 992