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_statfs_quantum = 30;
  80         gt->gt_statfs_slow = 0;
  81 }
  82
  83 /**
  84  * gfs2_check_sb - Check superblock
  85  * @sdp: the filesystem
  86  * @sb: The superblock
  87  * @silent: Don't print a message if the check fails
  88  *
  89  * Checks the version code of the FS is one that we understand how to
  90  * read and that the sizes of the various on-disk structures have not
  91  * changed.
  92  */
  93
  94 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  95 {
  96         unsigned int x;
  97
  98         if (sb->sb_magic != GFS2_MAGIC ||
  99             sb->sb_type != GFS2_METATYPE_SB) {
 100                 if (!silent)
 101                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
 102                 return -EINVAL;
 103         }
 104
 105         /*  If format numbers match exactly, we're done.  */
 106
 107         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 108             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 109                 return 0;
 110
 111         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 112                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 113                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 114                                 break;
 115
 116                 if (!gfs2_old_fs_formats[x]) {
 117                         printk(KERN_WARNING
 118                                "GFS2: code version (%u, %u) is incompatible "
 119                                "with ondisk format (%u, %u)\n",
 120                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 121                                sb->sb_fs_format, sb->sb_multihost_format);
 122                         printk(KERN_WARNING
 123                                "GFS2: I don't know how to upgrade this FS\n");
 124                         return -EINVAL;
 125                 }
 126         }
 127
 128         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 129                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 130                         if (gfs2_old_multihost_formats[x] ==
 131                             sb->sb_multihost_format)
 132                                 break;
 133
 134                 if (!gfs2_old_multihost_formats[x]) {
 135                         printk(KERN_WARNING
 136                                "GFS2: code version (%u, %u) is incompatible "
 137                                "with ondisk format (%u, %u)\n",
 138                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 139                                sb->sb_fs_format, sb->sb_multihost_format);
 140                         printk(KERN_WARNING
 141                                "GFS2: I don't know how to upgrade this FS\n");
 142                         return -EINVAL;
 143                 }
 144         }
 145
 146         if (!sdp->sd_args.ar_upgrade) {
 147                 printk(KERN_WARNING
 148                        "GFS2: code version (%u, %u) is incompatible "
 149                        "with ondisk format (%u, %u)\n",
 150                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 151                        sb->sb_fs_format, sb->sb_multihost_format);
 152                 printk(KERN_INFO
 153                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 154                        "the FS\n");
 155                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 156                 return -EINVAL;
 157         }
 158
 159         return 0;
 160 }
 161
 162
 163 static void end_bio_io_page(struct bio *bio, int error)
 164 {
 165         struct page *page = bio->bi_private;
 166
 167         if (!error)
 168                 SetPageUptodate(page);
 169         else
 170                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 171         unlock_page(page);
 172         return 0;
 173 }
 174
 175 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 176 {
 177         const struct gfs2_sb *str = buf;
 178
 179         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 180         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 181         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 182         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 183         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 184         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 185         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 186         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 187         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 188         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 189         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 190
 191         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 192         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 193 }
 194
 195 /**
 196  * gfs2_read_super - Read the gfs2 super block from disk
 197  * @sdp: The GFS2 super block
 198  * @sector: The location of the super block
 199  * @error: The error code to return
 200  *
 201  * This uses the bio functions to read the super block from disk
 202  * because we want to be 100% sure that we never read cached data.
 203  * A super block is read twice only during each GFS2 mount and is
 204  * never written to by the filesystem. The first time its read no
 205  * locks are held, and the only details which are looked at are those
 206  * relating to the locking protocol. Once locking is up and working,
 207  * the sb is read again under the lock to establish the location of
 208  * the master directory (contains pointers to journals etc) and the
 209  * root directory.
 210  *
 211  * Returns: 0 on success or error
 212  */
 213
 214 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 215 {
 216         struct super_block *sb = sdp->sd_vfs;
 217         struct gfs2_sb *p;
 218         struct page *page;
 219         struct bio *bio;
 220
 221         page = alloc_page(GFP_KERNEL);
 222         if (unlikely(!page))
 223                 return -ENOBUFS;
 224
 225         ClearPageUptodate(page);
 226         ClearPageDirty(page);
 227         lock_page(page);
 228
 229         bio = bio_alloc(GFP_KERNEL, 1);
 230         if (unlikely(!bio)) {
 231                 __free_page(page);
 232                 return -ENOBUFS;
 233         }
 234
 235         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 236         bio->bi_bdev = sb->s_bdev;
 237         bio_add_page(bio, page, PAGE_SIZE, 0);
 238
 239         bio->bi_end_io = end_bio_io_page;
 240         bio->bi_private = page;
 241         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 242         wait_on_page_locked(page);
 243         bio_put(bio);
 244         if (!PageUptodate(page)) {
 245                 __free_page(page);
 246                 return -EIO;
 247         }
 248         p = kmap(page);
 249         gfs2_sb_in(&sdp->sd_sb, p);
 250         kunmap(page);
 251         __free_page(page);
 252         return 0;
 253 }
 254
 255 /**
 256  * gfs2_read_sb - Read super block
 257  * @sdp: The GFS2 superblock
 258  * @gl: the glock for the superblock (assumed to be held)
 259  * @silent: Don't print message if mount fails
 260  *
 261  */
 262
 263 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 264 {
 265         u32 hash_blocks, ind_blocks, leaf_blocks;
 266         u32 tmp_blocks;
 267         unsigned int x;
 268         int error;
 269
 270         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 271         if (error) {
 272                 if (!silent)
 273                         fs_err(sdp, "can't read superblock\n");
 274                 return error;
 275         }
 276
 277         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 278         if (error)
 279                 return error;
 280
 281         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 282                                GFS2_BASIC_BLOCK_SHIFT;
 283         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 284         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 285                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 286         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 287                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 288         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 289         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 290         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 291         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 292         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 293                                 sizeof(struct gfs2_meta_header)) /
 294                                 sizeof(struct gfs2_quota_change);
 295
 296         /* Compute maximum reservation required to add a entry to a directory */
 297
 298         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 299                              sdp->sd_jbsize);
 300
 301         ind_blocks = 0;
 302         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 303                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 304                 ind_blocks += tmp_blocks;
 305         }
 306
 307         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 308
 309         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 310
 311         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 312                                 sizeof(struct gfs2_dinode);
 313         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 314         for (x = 2;; x++) {
 315                 u64 space, d;
 316                 u32 m;
 317
 318                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 319                 d = space;
 320                 m = do_div(d, sdp->sd_inptrs);
 321
 322                 if (d != sdp->sd_heightsize[x - 1] || m)
 323                         break;
 324                 sdp->sd_heightsize[x] = space;
 325         }
 326         sdp->sd_max_height = x;
 327         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 328
 329         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 330                                  sizeof(struct gfs2_dinode);
 331         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 332         for (x = 2;; x++) {
 333                 u64 space, d;
 334                 u32 m;
 335
 336                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 337                 d = space;
 338                 m = do_div(d, sdp->sd_inptrs);
 339
 340                 if (d != sdp->sd_jheightsize[x - 1] || m)
 341                         break;
 342                 sdp->sd_jheightsize[x] = space;
 343         }
 344         sdp->sd_max_jheight = x;
 345         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 346
 347         return 0;
 348 }
 349
 350 /**
 351  * gfs2_jindex_hold - Grab a lock on the jindex
 352  * @sdp: The GFS2 superblock
 353  * @ji_gh: the holder for the jindex glock
 354  *
 355  * This is very similar to the gfs2_rindex_hold() function, except that
 356  * in general we hold the jindex lock for longer periods of time and
 357  * we grab it far less frequently (in general) then the rgrp lock.
 358  *
 359  * Returns: errno
 360  */
 361
 362 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 363 {
 364         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 365         struct qstr name;
 366         char buf[20];
 367         struct gfs2_jdesc *jd;
 368         int error;
 369
 370         name.name = buf;
 371
 372         mutex_lock(&sdp->sd_jindex_mutex);
 373
 374         for (;;) {
 375                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 376                 if (error)
 377                         break;
 378
 379                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 380                 name.hash = gfs2_disk_hash(name.name, name.len);
 381
 382                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 383                 if (error == -ENOENT) {
 384                         error = 0;
 385                         break;
 386                 }
 387
 388                 gfs2_glock_dq_uninit(ji_gh);
 389
 390                 if (error)
 391                         break;
 392
 393                 error = -ENOMEM;
 394                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 395                 if (!jd)
 396                         break;
 397
 398                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 399                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 400                         if (!jd->jd_inode)
 401                                 error = -ENOENT;
 402                         else
 403                                 error = PTR_ERR(jd->jd_inode);
 404                         kfree(jd);
 405                         break;
 406                 }
 407
 408                 spin_lock(&sdp->sd_jindex_spin);
 409                 jd->jd_jid = sdp->sd_journals++;
 410                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 411                 spin_unlock(&sdp->sd_jindex_spin);
 412         }
 413
 414         mutex_unlock(&sdp->sd_jindex_mutex);
 415
 416         return error;
 417 }
 418
 419 /**
 420  * gfs2_jindex_free - Clear all the journal index information
 421  * @sdp: The GFS2 superblock
 422  *
 423  */
 424
 425 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 426 {
 427         struct list_head list;
 428         struct gfs2_jdesc *jd;
 429
 430         spin_lock(&sdp->sd_jindex_spin);
 431         list_add(&list, &sdp->sd_jindex_list);
 432         list_del_init(&sdp->sd_jindex_list);
 433         sdp->sd_journals = 0;
 434         spin_unlock(&sdp->sd_jindex_spin);
 435
 436         while (!list_empty(&list)) {
 437                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 438                 list_del(&jd->jd_list);
 439                 iput(jd->jd_inode);
 440                 kfree(jd);
 441         }
 442 }
 443
 444 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 445 {
 446         struct gfs2_jdesc *jd;
 447         int found = 0;
 448
 449         list_for_each_entry(jd, head, jd_list) {
 450                 if (jd->jd_jid == jid) {
 451                         found = 1;
 452                         break;
 453                 }
 454         }
 455
 456         if (!found)
 457                 jd = NULL;
 458
 459         return jd;
 460 }
 461
 462 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 463 {
 464         struct gfs2_jdesc *jd;
 465
 466         spin_lock(&sdp->sd_jindex_spin);
 467         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 468         spin_unlock(&sdp->sd_jindex_spin);
 469
 470         return jd;
 471 }
 472
 473 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 474 {
 475         struct gfs2_jdesc *jd;
 476
 477         spin_lock(&sdp->sd_jindex_spin);
 478         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 479         if (jd)
 480                 jd->jd_dirty = 1;
 481         spin_unlock(&sdp->sd_jindex_spin);
 482 }
 483
 484 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 485 {
 486         struct gfs2_jdesc *jd;
 487         int found = 0;
 488
 489         spin_lock(&sdp->sd_jindex_spin);
 490
 491         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 492                 if (jd->jd_dirty) {
 493                         jd->jd_dirty = 0;
 494                         found = 1;
 495                         break;
 496                 }
 497         }
 498         spin_unlock(&sdp->sd_jindex_spin);
 499
 500         if (!found)
 501                 jd = NULL;
 502
 503         return jd;
 504 }
 505
 506 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 507 {
 508         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 509         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 510         int ar;
 511         int error;
 512
 513         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 514             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 515                 gfs2_consist_inode(ip);
 516                 return -EIO;
 517         }
 518         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 519
 520         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 521         if (!error && ar) {
 522                 gfs2_consist_inode(ip);
 523                 error = -EIO;
 524         }
 525
 526         return error;
 527 }
 528
 529 /**
 530  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 531  * @sdp: the filesystem
 532  *
 533  * Returns: errno
 534  */
 535
 536 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 537 {
 538         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 539         struct gfs2_glock *j_gl = ip->i_gl;
 540         struct gfs2_holder t_gh;
 541         struct gfs2_log_header_host head;
 542         int error;
 543
 544         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 545         if (error)
 546                 return error;
 547
 548         gfs2_meta_cache_flush(ip);
 549         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 550
 551         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 552         if (error)
 553                 goto fail;
 554
 555         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 556                 gfs2_consist(sdp);
 557                 error = -EIO;
 558                 goto fail;
 559         }
 560
 561         /*  Initialize some head of the log stuff  */
 562         sdp->sd_log_sequence = head.lh_sequence + 1;
 563         gfs2_log_pointers_init(sdp, head.lh_blkno);
 564
 565         error = gfs2_quota_init(sdp);
 566         if (error)
 567                 goto fail;
 568
 569         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 570
 571         gfs2_glock_dq_uninit(&t_gh);
 572
 573         return 0;
 574
 575 fail:
 576         t_gh.gh_flags |= GL_NOCACHE;
 577         gfs2_glock_dq_uninit(&t_gh);
 578
 579         return error;
 580 }
 581
 582 /**
 583  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 584  * @sdp: the filesystem
 585  *
 586  * Returns: errno
 587  */
 588
 589 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 590 {
 591         struct gfs2_holder t_gh;
 592         int error;
 593
 594         gfs2_quota_sync(sdp);
 595         gfs2_statfs_sync(sdp);
 596
 597         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 598                                    &t_gh);
 599         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 600                 return error;
 601
 602         gfs2_meta_syncfs(sdp);
 603         gfs2_log_shutdown(sdp);
 604
 605         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 606
 607         if (t_gh.gh_gl)
 608                 gfs2_glock_dq_uninit(&t_gh);
 609
 610         gfs2_quota_cleanup(sdp);
 611
 612         return error;
 613 }
 614
 615 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 616 {
 617         const struct gfs2_statfs_change *str = buf;
 618
 619         sc->sc_total = be64_to_cpu(str->sc_total);
 620         sc->sc_free = be64_to_cpu(str->sc_free);
 621         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 622 }
 623
 624 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 625 {
 626         struct gfs2_statfs_change *str = buf;
 627
 628         str->sc_total = cpu_to_be64(sc->sc_total);
 629         str->sc_free = cpu_to_be64(sc->sc_free);
 630         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 631 }
 632
 633 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 634 {
 635         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 636         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 637         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 638         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 639         struct buffer_head *m_bh, *l_bh;
 640         struct gfs2_holder gh;
 641         int error;
 642
 643         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 644                                    &gh);
 645         if (error)
 646                 return error;
 647
 648         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 649         if (error)
 650                 goto out;
 651
 652         if (sdp->sd_args.ar_spectator) {
 653                 spin_lock(&sdp->sd_statfs_spin);
 654                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 655                                       sizeof(struct gfs2_dinode));
 656                 spin_unlock(&sdp->sd_statfs_spin);
 657         } else {
 658                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 659                 if (error)
 660                         goto out_m_bh;
 661
 662                 spin_lock(&sdp->sd_statfs_spin);
 663                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 664                                       sizeof(struct gfs2_dinode));
 665                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 666                                       sizeof(struct gfs2_dinode));
 667                 spin_unlock(&sdp->sd_statfs_spin);
 668
 669                 brelse(l_bh);
 670         }
 671
 672 out_m_bh:
 673         brelse(m_bh);
 674 out:
 675         gfs2_glock_dq_uninit(&gh);
 676         return 0;
 677 }
 678
 679 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 680                         s64 dinodes)
 681 {
 682         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 683         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 684         struct buffer_head *l_bh;
 685         int error;
 686
 687         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 688         if (error)
 689                 return;
 690
 691         mutex_lock(&sdp->sd_statfs_mutex);
 692         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 693         mutex_unlock(&sdp->sd_statfs_mutex);
 694
 695         spin_lock(&sdp->sd_statfs_spin);
 696         l_sc->sc_total += total;
 697         l_sc->sc_free += free;
 698         l_sc->sc_dinodes += dinodes;
 699         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 700         spin_unlock(&sdp->sd_statfs_spin);
 701
 702         brelse(l_bh);
 703 }
 704
 705 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 706 {
 707         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 708         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 709         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 710         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 711         struct gfs2_holder gh;
 712         struct buffer_head *m_bh, *l_bh;
 713         int error;
 714
 715         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 716                                    &gh);
 717         if (error)
 718                 return error;
 719
 720         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 721         if (error)
 722                 goto out;
 723
 724         spin_lock(&sdp->sd_statfs_spin);
 725         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 726                               sizeof(struct gfs2_dinode));
 727         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 728                 spin_unlock(&sdp->sd_statfs_spin);
 729                 goto out_bh;
 730         }
 731         spin_unlock(&sdp->sd_statfs_spin);
 732
 733         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 734         if (error)
 735                 goto out_bh;
 736
 737         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 738         if (error)
 739                 goto out_bh2;
 740
 741         mutex_lock(&sdp->sd_statfs_mutex);
 742         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 743         mutex_unlock(&sdp->sd_statfs_mutex);
 744
 745         spin_lock(&sdp->sd_statfs_spin);
 746         m_sc->sc_total += l_sc->sc_total;
 747         m_sc->sc_free += l_sc->sc_free;
 748         m_sc->sc_dinodes += l_sc->sc_dinodes;
 749         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 750         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 751                0, sizeof(struct gfs2_statfs_change));
 752         spin_unlock(&sdp->sd_statfs_spin);
 753
 754         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 755         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 756
 757         gfs2_trans_end(sdp);
 758
 759 out_bh2:
 760         brelse(l_bh);
 761 out_bh:
 762         brelse(m_bh);
 763 out:
 764         gfs2_glock_dq_uninit(&gh);
 765         return error;
 766 }
 767
 768 /**
 769  * gfs2_statfs_i - Do a statfs
 770  * @sdp: the filesystem
 771  * @sg: the sg structure
 772  *
 773  * Returns: errno
 774  */
 775
 776 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 777 {
 778         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 779         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 780
 781         spin_lock(&sdp->sd_statfs_spin);
 782
 783         *sc = *m_sc;
 784         sc->sc_total += l_sc->sc_total;
 785         sc->sc_free += l_sc->sc_free;
 786         sc->sc_dinodes += l_sc->sc_dinodes;
 787
 788         spin_unlock(&sdp->sd_statfs_spin);
 789
 790         if (sc->sc_free < 0)
 791                 sc->sc_free = 0;
 792         if (sc->sc_free > sc->sc_total)
 793                 sc->sc_free = sc->sc_total;
 794         if (sc->sc_dinodes < 0)
 795                 sc->sc_dinodes = 0;
 796
 797         return 0;
 798 }
 799
 800 /**
 801  * statfs_fill - fill in the sg for a given RG
 802  * @rgd: the RG
 803  * @sc: the sc structure
 804  *
 805  * Returns: 0 on success, -ESTALE if the LVB is invalid
 806  */
 807
 808 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 809                             struct gfs2_statfs_change_host *sc)
 810 {
 811         gfs2_rgrp_verify(rgd);
 812         sc->sc_total += rgd->rd_data;
 813         sc->sc_free += rgd->rd_rg.rg_free;
 814         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 815         return 0;
 816 }
 817
 818 /**
 819  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 820  * @sdp: the filesystem
 821  * @sc: the sc info that will be returned
 822  *
 823  * Any error (other than a signal) will cause this routine to fall back
 824  * to the synchronous version.
 825  *
 826  * FIXME: This really shouldn't busy wait like this.
 827  *
 828  * Returns: errno
 829  */
 830
 831 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 832 {
 833         struct gfs2_holder ri_gh;
 834         struct gfs2_rgrpd *rgd_next;
 835         struct gfs2_holder *gha, *gh;
 836         unsigned int slots = 64;
 837         unsigned int x;
 838         int done;
 839         int error = 0, err;
 840
 841         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 842         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 843         if (!gha)
 844                 return -ENOMEM;
 845
 846         error = gfs2_rindex_hold(sdp, &ri_gh);
 847         if (error)
 848                 goto out;
 849
 850         rgd_next = gfs2_rgrpd_get_first(sdp);
 851
 852         for (;;) {
 853                 done = 1;
 854
 855                 for (x = 0; x < slots; x++) {
 856                         gh = gha + x;
 857
 858                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 859                                 err = gfs2_glock_wait(gh);
 860                                 if (err) {
 861                                         gfs2_holder_uninit(gh);
 862                                         error = err;
 863                                 } else {
 864                                         if (!error)
 865                                                 error = statfs_slow_fill(
 866                                                         gh->gh_gl->gl_object, sc);
 867                                         gfs2_glock_dq_uninit(gh);
 868                                 }
 869                         }
 870
 871                         if (gh->gh_gl)
 872                                 done = 0;
 873                         else if (rgd_next && !error) {
 874                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 875                                                            LM_ST_SHARED,
 876                                                            GL_ASYNC,
 877                                                            gh);
 878                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 879                                 done = 0;
 880                         }
 881
 882                         if (signal_pending(current))
 883                                 error = -ERESTARTSYS;
 884                 }
 885
 886                 if (done)
 887                         break;
 888
 889                 yield();
 890         }
 891
 892         gfs2_glock_dq_uninit(&ri_gh);
 893
 894 out:
 895         kfree(gha);
 896         return error;
 897 }
 898
 899 struct lfcc {
 900         struct list_head list;
 901         struct gfs2_holder gh;
 902 };
 903
 904 /**
 905  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 906  *                            journals are clean
 907  * @sdp: the file system
 908  * @state: the state to put the transaction lock into
 909  * @t_gh: the hold on the transaction lock
 910  *
 911  * Returns: errno
 912  */
 913
 914 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 915                                     struct gfs2_holder *t_gh)
 916 {
 917         struct gfs2_inode *ip;
 918         struct gfs2_holder ji_gh;
 919         struct gfs2_jdesc *jd;
 920         struct lfcc *lfcc;
 921         LIST_HEAD(list);
 922         struct gfs2_log_header_host lh;
 923         int error;
 924
 925         error = gfs2_jindex_hold(sdp, &ji_gh);
 926         if (error)
 927                 return error;
 928
 929         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 930                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 931                 if (!lfcc) {
 932                         error = -ENOMEM;
 933                         goto out;
 934                 }
 935                 ip = GFS2_I(jd->jd_inode);
 936                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 937                 if (error) {
 938                         kfree(lfcc);
 939                         goto out;
 940                 }
 941                 list_add(&lfcc->list, &list);
 942         }
 943
 944         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 945                                LM_FLAG_PRIORITY | GL_NOCACHE,
 946                                t_gh);
 947
 948         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 949                 error = gfs2_jdesc_check(jd);
 950                 if (error)
 951                         break;
 952                 error = gfs2_find_jhead(jd, &lh);
 953                 if (error)
 954                         break;
 955                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 956                         error = -EBUSY;
 957                         break;
 958                 }
 959         }
 960
 961         if (error)
 962                 gfs2_glock_dq_uninit(t_gh);
 963
 964 out:
 965         while (!list_empty(&list)) {
 966                 lfcc = list_entry(list.next, struct lfcc, list);
 967                 list_del(&lfcc->list);
 968                 gfs2_glock_dq_uninit(&lfcc->gh);
 969                 kfree(lfcc);
 970         }
 971         gfs2_glock_dq_uninit(&ji_gh);
 972         return error;
 973 }
 974
 975 /**
 976  * gfs2_freeze_fs - freezes the file system
 977  * @sdp: the file system
 978  *
 979  * This function flushes data and meta data for all machines by
 980  * aquiring the transaction log exclusively.  All journals are
 981  * ensured to be in a clean state as well.
 982  *
 983  * Returns: errno
 984  */
 985
 986 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 987 {
 988         int error = 0;
 989
 990         mutex_lock(&sdp->sd_freeze_lock);
 991
 992         if (!sdp->sd_freeze_count++) {
 993                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 994                 if (error)
 995                         sdp->sd_freeze_count--;
 996         }
 997
 998         mutex_unlock(&sdp->sd_freeze_lock);
 999
1000         return error;
1001 }
1002
1003 /**
1004  * gfs2_unfreeze_fs - unfreezes the file system
1005  * @sdp: the file system
1006  *
1007  * This function allows the file system to proceed by unlocking
1008  * the exclusively held transaction lock.  Other GFS2 nodes are
1009  * now free to acquire the lock shared and go on with their lives.
1010  *
1011  */
1012
1013 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1014 {
1015         mutex_lock(&sdp->sd_freeze_lock);
1016
1017         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1018                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1019
1020         mutex_unlock(&sdp->sd_freeze_lock);
1021 }
1022