fs/gfs2/super.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 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_demote_secs = 300;
  55         gt->gt_incore_log_blocks = 1024;
  56         gt->gt_log_flush_secs = 60;
  57         gt->gt_recoverd_secs = 60;
  58         gt->gt_logd_secs = 1;
  59         gt->gt_quotad_secs = 5;
  60         gt->gt_quota_simul_sync = 64;
  61         gt->gt_quota_warn_period = 10;
  62         gt->gt_quota_scale_num = 1;
  63         gt->gt_quota_scale_den = 1;
  64         gt->gt_quota_cache_secs = 300;
  65         gt->gt_quota_quantum = 60;
  66         gt->gt_atime_quantum = 3600;
  67         gt->gt_new_files_jdata = 0;
  68         gt->gt_new_files_directio = 0;
  69         gt->gt_max_readahead = 1 << 18;
  70         gt->gt_stall_secs = 600;
  71         gt->gt_complain_secs = 10;
  72         gt->gt_statfs_quantum = 30;
  73         gt->gt_statfs_slow = 0;
  74 }
  75
  76 /**
  77  * gfs2_check_sb - Check superblock
  78  * @sdp: the filesystem
  79  * @sb: The superblock
  80  * @silent: Don't print a message if the check fails
  81  *
  82  * Checks the version code of the FS is one that we understand how to
  83  * read and that the sizes of the various on-disk structures have not
  84  * changed.
  85  */
  86
  87 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  88 {
  89         unsigned int x;
  90
  91         if (sb->sb_magic != GFS2_MAGIC ||
  92             sb->sb_type != GFS2_METATYPE_SB) {
  93                 if (!silent)
  94                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
  95                 return -EINVAL;
  96         }
  97
  98         /*  If format numbers match exactly, we're done.  */
  99
 100         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 101             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 102                 return 0;
 103
 104         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 105                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 106                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 107                                 break;
 108
 109                 if (!gfs2_old_fs_formats[x]) {
 110                         printk(KERN_WARNING
 111                                "GFS2: code version (%u, %u) is incompatible "
 112                                "with ondisk format (%u, %u)\n",
 113                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 114                                sb->sb_fs_format, sb->sb_multihost_format);
 115                         printk(KERN_WARNING
 116                                "GFS2: I don't know how to upgrade this FS\n");
 117                         return -EINVAL;
 118                 }
 119         }
 120
 121         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 122                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 123                         if (gfs2_old_multihost_formats[x] ==
 124                             sb->sb_multihost_format)
 125                                 break;
 126
 127                 if (!gfs2_old_multihost_formats[x]) {
 128                         printk(KERN_WARNING
 129                                "GFS2: code version (%u, %u) is incompatible "
 130                                "with ondisk format (%u, %u)\n",
 131                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 132                                sb->sb_fs_format, sb->sb_multihost_format);
 133                         printk(KERN_WARNING
 134                                "GFS2: I don't know how to upgrade this FS\n");
 135                         return -EINVAL;
 136                 }
 137         }
 138
 139         if (!sdp->sd_args.ar_upgrade) {
 140                 printk(KERN_WARNING
 141                        "GFS2: code version (%u, %u) is incompatible "
 142                        "with ondisk format (%u, %u)\n",
 143                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 144                        sb->sb_fs_format, sb->sb_multihost_format);
 145                 printk(KERN_INFO
 146                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 147                        "the FS\n");
 148                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 149                 return -EINVAL;
 150         }
 151
 152         return 0;
 153 }
 154
 155
 156 static void end_bio_io_page(struct bio *bio, int error)
 157 {
 158         struct page *page = bio->bi_private;
 159
 160         if (!error)
 161                 SetPageUptodate(page);
 162         else
 163                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 164         unlock_page(page);
 165 }
 166
 167 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 168 {
 169         const struct gfs2_sb *str = buf;
 170
 171         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 172         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 173         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 174         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 175         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 176         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 177         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 178         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 179         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 180         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 181         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 182
 183         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 184         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 185 }
 186
 187 /**
 188  * gfs2_read_super - Read the gfs2 super block from disk
 189  * @sdp: The GFS2 super block
 190  * @sector: The location of the super block
 191  * @error: The error code to return
 192  *
 193  * This uses the bio functions to read the super block from disk
 194  * because we want to be 100% sure that we never read cached data.
 195  * A super block is read twice only during each GFS2 mount and is
 196  * never written to by the filesystem. The first time its read no
 197  * locks are held, and the only details which are looked at are those
 198  * relating to the locking protocol. Once locking is up and working,
 199  * the sb is read again under the lock to establish the location of
 200  * the master directory (contains pointers to journals etc) and the
 201  * root directory.
 202  *
 203  * Returns: 0 on success or error
 204  */
 205
 206 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 207 {
 208         struct super_block *sb = sdp->sd_vfs;
 209         struct gfs2_sb *p;
 210         struct page *page;
 211         struct bio *bio;
 212
 213         page = alloc_page(GFP_KERNEL);
 214         if (unlikely(!page))
 215                 return -ENOBUFS;
 216
 217         ClearPageUptodate(page);
 218         ClearPageDirty(page);
 219         lock_page(page);
 220
 221         bio = bio_alloc(GFP_KERNEL, 1);
 222         if (unlikely(!bio)) {
 223                 __free_page(page);
 224                 return -ENOBUFS;
 225         }
 226
 227         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 228         bio->bi_bdev = sb->s_bdev;
 229         bio_add_page(bio, page, PAGE_SIZE, 0);
 230
 231         bio->bi_end_io = end_bio_io_page;
 232         bio->bi_private = page;
 233         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 234         wait_on_page_locked(page);
 235         bio_put(bio);
 236         if (!PageUptodate(page)) {
 237                 __free_page(page);
 238                 return -EIO;
 239         }
 240         p = kmap(page);
 241         gfs2_sb_in(&sdp->sd_sb, p);
 242         kunmap(page);
 243         __free_page(page);
 244         return 0;
 245 }
 246
 247 /**
 248  * gfs2_read_sb - Read super block
 249  * @sdp: The GFS2 superblock
 250  * @gl: the glock for the superblock (assumed to be held)
 251  * @silent: Don't print message if mount fails
 252  *
 253  */
 254
 255 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 256 {
 257         u32 hash_blocks, ind_blocks, leaf_blocks;
 258         u32 tmp_blocks;
 259         unsigned int x;
 260         int error;
 261
 262         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 263         if (error) {
 264                 if (!silent)
 265                         fs_err(sdp, "can't read superblock\n");
 266                 return error;
 267         }
 268
 269         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 270         if (error)
 271                 return error;
 272
 273         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 274                                GFS2_BASIC_BLOCK_SHIFT;
 275         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 276         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 277                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 278         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 279                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 280         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 281         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 282         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 283         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 284         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 285                                 sizeof(struct gfs2_meta_header)) /
 286                                 sizeof(struct gfs2_quota_change);
 287
 288         /* Compute maximum reservation required to add a entry to a directory */
 289
 290         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 291                              sdp->sd_jbsize);
 292
 293         ind_blocks = 0;
 294         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 295                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 296                 ind_blocks += tmp_blocks;
 297         }
 298
 299         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 300
 301         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 302
 303         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 304                                 sizeof(struct gfs2_dinode);
 305         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 306         for (x = 2;; x++) {
 307                 u64 space, d;
 308                 u32 m;
 309
 310                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 311                 d = space;
 312                 m = do_div(d, sdp->sd_inptrs);
 313
 314                 if (d != sdp->sd_heightsize[x - 1] || m)
 315                         break;
 316                 sdp->sd_heightsize[x] = space;
 317         }
 318         sdp->sd_max_height = x;
 319         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 320
 321         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 322                                  sizeof(struct gfs2_dinode);
 323         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 324         for (x = 2;; x++) {
 325                 u64 space, d;
 326                 u32 m;
 327
 328                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 329                 d = space;
 330                 m = do_div(d, sdp->sd_inptrs);
 331
 332                 if (d != sdp->sd_jheightsize[x - 1] || m)
 333                         break;
 334                 sdp->sd_jheightsize[x] = space;
 335         }
 336         sdp->sd_max_jheight = x;
 337         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 338
 339         return 0;
 340 }
 341
 342 /**
 343  * gfs2_jindex_hold - Grab a lock on the jindex
 344  * @sdp: The GFS2 superblock
 345  * @ji_gh: the holder for the jindex glock
 346  *
 347  * This is very similar to the gfs2_rindex_hold() function, except that
 348  * in general we hold the jindex lock for longer periods of time and
 349  * we grab it far less frequently (in general) then the rgrp lock.
 350  *
 351  * Returns: errno
 352  */
 353
 354 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 355 {
 356         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 357         struct qstr name;
 358         char buf[20];
 359         struct gfs2_jdesc *jd;
 360         int error;
 361
 362         name.name = buf;
 363
 364         mutex_lock(&sdp->sd_jindex_mutex);
 365
 366         for (;;) {
 367                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 368                 if (error)
 369                         break;
 370
 371                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 372                 name.hash = gfs2_disk_hash(name.name, name.len);
 373
 374                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 375                 if (error == -ENOENT) {
 376                         error = 0;
 377                         break;
 378                 }
 379
 380                 gfs2_glock_dq_uninit(ji_gh);
 381
 382                 if (error)
 383                         break;
 384
 385                 error = -ENOMEM;
 386                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 387                 if (!jd)
 388                         break;
 389
 390                 INIT_LIST_HEAD(&jd->extent_list);
 391                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 392                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 393                         if (!jd->jd_inode)
 394                                 error = -ENOENT;
 395                         else
 396                                 error = PTR_ERR(jd->jd_inode);
 397                         kfree(jd);
 398                         break;
 399                 }
 400
 401                 spin_lock(&sdp->sd_jindex_spin);
 402                 jd->jd_jid = sdp->sd_journals++;
 403                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 404                 spin_unlock(&sdp->sd_jindex_spin);
 405         }
 406
 407         mutex_unlock(&sdp->sd_jindex_mutex);
 408
 409         return error;
 410 }
 411
 412 /**
 413  * gfs2_jindex_free - Clear all the journal index information
 414  * @sdp: The GFS2 superblock
 415  *
 416  */
 417
 418 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 419 {
 420         struct list_head list, *head;
 421         struct gfs2_jdesc *jd;
 422         struct gfs2_journal_extent *jext;
 423
 424         spin_lock(&sdp->sd_jindex_spin);
 425         list_add(&list, &sdp->sd_jindex_list);
 426         list_del_init(&sdp->sd_jindex_list);
 427         sdp->sd_journals = 0;
 428         spin_unlock(&sdp->sd_jindex_spin);
 429
 430         while (!list_empty(&list)) {
 431                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 432                 head = &jd->extent_list;
 433                 while (!list_empty(head)) {
 434                         jext = list_entry(head->next,
 435                                           struct gfs2_journal_extent,
 436                                           extent_list);
 437                         list_del(&jext->extent_list);
 438                         kfree(jext);
 439                 }
 440                 list_del(&jd->jd_list);
 441                 iput(jd->jd_inode);
 442                 kfree(jd);
 443         }
 444 }
 445
 446 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 447 {
 448         struct gfs2_jdesc *jd;
 449         int found = 0;
 450
 451         list_for_each_entry(jd, head, jd_list) {
 452                 if (jd->jd_jid == jid) {
 453                         found = 1;
 454                         break;
 455                 }
 456         }
 457
 458         if (!found)
 459                 jd = NULL;
 460
 461         return jd;
 462 }
 463
 464 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 465 {
 466         struct gfs2_jdesc *jd;
 467
 468         spin_lock(&sdp->sd_jindex_spin);
 469         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 470         spin_unlock(&sdp->sd_jindex_spin);
 471
 472         return jd;
 473 }
 474
 475 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 476 {
 477         struct gfs2_jdesc *jd;
 478
 479         spin_lock(&sdp->sd_jindex_spin);
 480         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 481         if (jd)
 482                 jd->jd_dirty = 1;
 483         spin_unlock(&sdp->sd_jindex_spin);
 484 }
 485
 486 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 487 {
 488         struct gfs2_jdesc *jd;
 489         int found = 0;
 490
 491         spin_lock(&sdp->sd_jindex_spin);
 492
 493         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 494                 if (jd->jd_dirty) {
 495                         jd->jd_dirty = 0;
 496                         found = 1;
 497                         break;
 498                 }
 499         }
 500         spin_unlock(&sdp->sd_jindex_spin);
 501
 502         if (!found)
 503                 jd = NULL;
 504
 505         return jd;
 506 }
 507
 508 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 509 {
 510         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 511         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 512         int ar;
 513         int error;
 514
 515         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 516             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 517                 gfs2_consist_inode(ip);
 518                 return -EIO;
 519         }
 520         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 521
 522         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 523         if (!error && ar) {
 524                 gfs2_consist_inode(ip);
 525                 error = -EIO;
 526         }
 527
 528         return error;
 529 }
 530
 531 /**
 532  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 533  * @sdp: the filesystem
 534  *
 535  * Returns: errno
 536  */
 537
 538 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 539 {
 540         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 541         struct gfs2_glock *j_gl = ip->i_gl;
 542         struct gfs2_holder t_gh;
 543         struct gfs2_log_header_host head;
 544         int error;
 545
 546         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 547         if (error)
 548                 return error;
 549
 550         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 551
 552         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 553         if (error)
 554                 goto fail;
 555
 556         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 557                 gfs2_consist(sdp);
 558                 error = -EIO;
 559                 goto fail;
 560         }
 561
 562         /*  Initialize some head of the log stuff  */
 563         sdp->sd_log_sequence = head.lh_sequence + 1;
 564         gfs2_log_pointers_init(sdp, head.lh_blkno);
 565
 566         error = gfs2_quota_init(sdp);
 567         if (error)
 568                 goto fail;
 569
 570         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 571
 572         gfs2_glock_dq_uninit(&t_gh);
 573
 574         return 0;
 575
 576 fail:
 577         t_gh.gh_flags |= GL_NOCACHE;
 578         gfs2_glock_dq_uninit(&t_gh);
 579
 580         return error;
 581 }
 582
 583 /**
 584  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 585  * @sdp: the filesystem
 586  *
 587  * Returns: errno
 588  */
 589
 590 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 591 {
 592         struct gfs2_holder t_gh;
 593         int error;
 594
 595         gfs2_quota_sync(sdp);
 596         gfs2_statfs_sync(sdp);
 597
 598         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 599                                    &t_gh);
 600         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 601                 return error;
 602
 603         gfs2_meta_syncfs(sdp);
 604         gfs2_log_shutdown(sdp);
 605
 606         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 607
 608         if (t_gh.gh_gl)
 609                 gfs2_glock_dq_uninit(&t_gh);
 610
 611         gfs2_quota_cleanup(sdp);
 612
 613         return error;
 614 }
 615
 616 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 617 {
 618         const struct gfs2_statfs_change *str = buf;
 619
 620         sc->sc_total = be64_to_cpu(str->sc_total);
 621         sc->sc_free = be64_to_cpu(str->sc_free);
 622         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 623 }
 624
 625 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 626 {
 627         struct gfs2_statfs_change *str = buf;
 628
 629         str->sc_total = cpu_to_be64(sc->sc_total);
 630         str->sc_free = cpu_to_be64(sc->sc_free);
 631         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 632 }
 633
 634 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 635 {
 636         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 637         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 638         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 639         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 640         struct buffer_head *m_bh, *l_bh;
 641         struct gfs2_holder gh;
 642         int error;
 643
 644         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 645                                    &gh);
 646         if (error)
 647                 return error;
 648
 649         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 650         if (error)
 651                 goto out;
 652
 653         if (sdp->sd_args.ar_spectator) {
 654                 spin_lock(&sdp->sd_statfs_spin);
 655                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 656                                       sizeof(struct gfs2_dinode));
 657                 spin_unlock(&sdp->sd_statfs_spin);
 658         } else {
 659                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 660                 if (error)
 661                         goto out_m_bh;
 662
 663                 spin_lock(&sdp->sd_statfs_spin);
 664                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 665                                       sizeof(struct gfs2_dinode));
 666                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 667                                       sizeof(struct gfs2_dinode));
 668                 spin_unlock(&sdp->sd_statfs_spin);
 669
 670                 brelse(l_bh);
 671         }
 672
 673 out_m_bh:
 674         brelse(m_bh);
 675 out:
 676         gfs2_glock_dq_uninit(&gh);
 677         return 0;
 678 }
 679
 680 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 681                         s64 dinodes)
 682 {
 683         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 684         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 685         struct buffer_head *l_bh;
 686         int error;
 687
 688         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 689         if (error)
 690                 return;
 691
 692         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 693
 694         spin_lock(&sdp->sd_statfs_spin);
 695         l_sc->sc_total += total;
 696         l_sc->sc_free += free;
 697         l_sc->sc_dinodes += dinodes;
 698         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 699         spin_unlock(&sdp->sd_statfs_spin);
 700
 701         brelse(l_bh);
 702 }
 703
 704 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 705 {
 706         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 707         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 708         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 709         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 710         struct gfs2_holder gh;
 711         struct buffer_head *m_bh, *l_bh;
 712         int error;
 713
 714         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 715                                    &gh);
 716         if (error)
 717                 return error;
 718
 719         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 720         if (error)
 721                 goto out;
 722
 723         spin_lock(&sdp->sd_statfs_spin);
 724         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 725                               sizeof(struct gfs2_dinode));
 726         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 727                 spin_unlock(&sdp->sd_statfs_spin);
 728                 goto out_bh;
 729         }
 730         spin_unlock(&sdp->sd_statfs_spin);
 731
 732         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 733         if (error)
 734                 goto out_bh;
 735
 736         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 737         if (error)
 738                 goto out_bh2;
 739
 740         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 741
 742         spin_lock(&sdp->sd_statfs_spin);
 743         m_sc->sc_total += l_sc->sc_total;
 744         m_sc->sc_free += l_sc->sc_free;
 745         m_sc->sc_dinodes += l_sc->sc_dinodes;
 746         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 747         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 748                0, sizeof(struct gfs2_statfs_change));
 749         spin_unlock(&sdp->sd_statfs_spin);
 750
 751         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 752         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 753
 754         gfs2_trans_end(sdp);
 755
 756 out_bh2:
 757         brelse(l_bh);
 758 out_bh:
 759         brelse(m_bh);
 760 out:
 761         gfs2_glock_dq_uninit(&gh);
 762         return error;
 763 }
 764
 765 /**
 766  * gfs2_statfs_i - Do a statfs
 767  * @sdp: the filesystem
 768  * @sg: the sg structure
 769  *
 770  * Returns: errno
 771  */
 772
 773 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 774 {
 775         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 776         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 777
 778         spin_lock(&sdp->sd_statfs_spin);
 779
 780         *sc = *m_sc;
 781         sc->sc_total += l_sc->sc_total;
 782         sc->sc_free += l_sc->sc_free;
 783         sc->sc_dinodes += l_sc->sc_dinodes;
 784
 785         spin_unlock(&sdp->sd_statfs_spin);
 786
 787         if (sc->sc_free < 0)
 788                 sc->sc_free = 0;
 789         if (sc->sc_free > sc->sc_total)
 790                 sc->sc_free = sc->sc_total;
 791         if (sc->sc_dinodes < 0)
 792                 sc->sc_dinodes = 0;
 793
 794         return 0;
 795 }
 796
 797 /**
 798  * statfs_fill - fill in the sg for a given RG
 799  * @rgd: the RG
 800  * @sc: the sc structure
 801  *
 802  * Returns: 0 on success, -ESTALE if the LVB is invalid
 803  */
 804
 805 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 806                             struct gfs2_statfs_change_host *sc)
 807 {
 808         gfs2_rgrp_verify(rgd);
 809         sc->sc_total += rgd->rd_data;
 810         sc->sc_free += rgd->rd_rg.rg_free;
 811         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 812         return 0;
 813 }
 814
 815 /**
 816  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 817  * @sdp: the filesystem
 818  * @sc: the sc info that will be returned
 819  *
 820  * Any error (other than a signal) will cause this routine to fall back
 821  * to the synchronous version.
 822  *
 823  * FIXME: This really shouldn't busy wait like this.
 824  *
 825  * Returns: errno
 826  */
 827
 828 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 829 {
 830         struct gfs2_holder ri_gh;
 831         struct gfs2_rgrpd *rgd_next;
 832         struct gfs2_holder *gha, *gh;
 833         unsigned int slots = 64;
 834         unsigned int x;
 835         int done;
 836         int error = 0, err;
 837
 838         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 839         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 840         if (!gha)
 841                 return -ENOMEM;
 842
 843         error = gfs2_rindex_hold(sdp, &ri_gh);
 844         if (error)
 845                 goto out;
 846
 847         rgd_next = gfs2_rgrpd_get_first(sdp);
 848
 849         for (;;) {
 850                 done = 1;
 851
 852                 for (x = 0; x < slots; x++) {
 853                         gh = gha + x;
 854
 855                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 856                                 err = gfs2_glock_wait(gh);
 857                                 if (err) {
 858                                         gfs2_holder_uninit(gh);
 859                                         error = err;
 860                                 } else {
 861                                         if (!error)
 862                                                 error = statfs_slow_fill(
 863                                                         gh->gh_gl->gl_object, sc);
 864                                         gfs2_glock_dq_uninit(gh);
 865                                 }
 866                         }
 867
 868                         if (gh->gh_gl)
 869                                 done = 0;
 870                         else if (rgd_next && !error) {
 871                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 872                                                            LM_ST_SHARED,
 873                                                            GL_ASYNC,
 874                                                            gh);
 875                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 876                                 done = 0;
 877                         }
 878
 879                         if (signal_pending(current))
 880                                 error = -ERESTARTSYS;
 881                 }
 882
 883                 if (done)
 884                         break;
 885
 886                 yield();
 887         }
 888
 889         gfs2_glock_dq_uninit(&ri_gh);
 890
 891 out:
 892         kfree(gha);
 893         return error;
 894 }
 895
 896 struct lfcc {
 897         struct list_head list;
 898         struct gfs2_holder gh;
 899 };
 900
 901 /**
 902  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 903  *                            journals are clean
 904  * @sdp: the file system
 905  * @state: the state to put the transaction lock into
 906  * @t_gh: the hold on the transaction lock
 907  *
 908  * Returns: errno
 909  */
 910
 911 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 912                                     struct gfs2_holder *t_gh)
 913 {
 914         struct gfs2_inode *ip;
 915         struct gfs2_holder ji_gh;
 916         struct gfs2_jdesc *jd;
 917         struct lfcc *lfcc;
 918         LIST_HEAD(list);
 919         struct gfs2_log_header_host lh;
 920         int error;
 921
 922         error = gfs2_jindex_hold(sdp, &ji_gh);
 923         if (error)
 924                 return error;
 925
 926         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 927                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 928                 if (!lfcc) {
 929                         error = -ENOMEM;
 930                         goto out;
 931                 }
 932                 ip = GFS2_I(jd->jd_inode);
 933                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 934                 if (error) {
 935                         kfree(lfcc);
 936                         goto out;
 937                 }
 938                 list_add(&lfcc->list, &list);
 939         }
 940
 941         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 942                                LM_FLAG_PRIORITY | GL_NOCACHE,
 943                                t_gh);
 944
 945         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 946                 error = gfs2_jdesc_check(jd);
 947                 if (error)
 948                         break;
 949                 error = gfs2_find_jhead(jd, &lh);
 950                 if (error)
 951                         break;
 952                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 953                         error = -EBUSY;
 954                         break;
 955                 }
 956         }
 957
 958         if (error)
 959                 gfs2_glock_dq_uninit(t_gh);
 960
 961 out:
 962         while (!list_empty(&list)) {
 963                 lfcc = list_entry(list.next, struct lfcc, list);
 964                 list_del(&lfcc->list);
 965                 gfs2_glock_dq_uninit(&lfcc->gh);
 966                 kfree(lfcc);
 967         }
 968         gfs2_glock_dq_uninit(&ji_gh);
 969         return error;
 970 }
 971
 972 /**
 973  * gfs2_freeze_fs - freezes the file system
 974  * @sdp: the file system
 975  *
 976  * This function flushes data and meta data for all machines by
 977  * aquiring the transaction log exclusively.  All journals are
 978  * ensured to be in a clean state as well.
 979  *
 980  * Returns: errno
 981  */
 982
 983 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 984 {
 985         int error = 0;
 986
 987         mutex_lock(&sdp->sd_freeze_lock);
 988
 989         if (!sdp->sd_freeze_count++) {
 990                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 991                 if (error)
 992                         sdp->sd_freeze_count--;
 993         }
 994
 995         mutex_unlock(&sdp->sd_freeze_lock);
 996
 997         return error;
 998 }
 999
1000 /**
1001  * gfs2_unfreeze_fs - unfreezes the file system
1002  * @sdp: the file system
1003  *
1004  * This function allows the file system to proceed by unlocking
1005  * the exclusively held transaction lock.  Other GFS2 nodes are
1006  * now free to acquire the lock shared and go on with their lives.
1007  *
1008  */
1009
1010 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1011 {
1012         mutex_lock(&sdp->sd_freeze_lock);
1013
1014         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1015                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1016
1017         mutex_unlock(&sdp->sd_freeze_lock);
1018 }
1019