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_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                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 391                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 392                         if (!jd->jd_inode)
 393                                 error = -ENOENT;
 394                         else
 395                                 error = PTR_ERR(jd->jd_inode);
 396                         kfree(jd);
 397                         break;
 398                 }
 399
 400                 spin_lock(&sdp->sd_jindex_spin);
 401                 jd->jd_jid = sdp->sd_journals++;
 402                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 403                 spin_unlock(&sdp->sd_jindex_spin);
 404         }
 405
 406         mutex_unlock(&sdp->sd_jindex_mutex);
 407
 408         return error;
 409 }
 410
 411 /**
 412  * gfs2_jindex_free - Clear all the journal index information
 413  * @sdp: The GFS2 superblock
 414  *
 415  */
 416
 417 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 418 {
 419         struct list_head list;
 420         struct gfs2_jdesc *jd;
 421
 422         spin_lock(&sdp->sd_jindex_spin);
 423         list_add(&list, &sdp->sd_jindex_list);
 424         list_del_init(&sdp->sd_jindex_list);
 425         sdp->sd_journals = 0;
 426         spin_unlock(&sdp->sd_jindex_spin);
 427
 428         while (!list_empty(&list)) {
 429                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
 430                 list_del(&jd->jd_list);
 431                 iput(jd->jd_inode);
 432                 kfree(jd);
 433         }
 434 }
 435
 436 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 437 {
 438         struct gfs2_jdesc *jd;
 439         int found = 0;
 440
 441         list_for_each_entry(jd, head, jd_list) {
 442                 if (jd->jd_jid == jid) {
 443                         found = 1;
 444                         break;
 445                 }
 446         }
 447
 448         if (!found)
 449                 jd = NULL;
 450
 451         return jd;
 452 }
 453
 454 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 455 {
 456         struct gfs2_jdesc *jd;
 457
 458         spin_lock(&sdp->sd_jindex_spin);
 459         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 460         spin_unlock(&sdp->sd_jindex_spin);
 461
 462         return jd;
 463 }
 464
 465 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 466 {
 467         struct gfs2_jdesc *jd;
 468
 469         spin_lock(&sdp->sd_jindex_spin);
 470         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 471         if (jd)
 472                 jd->jd_dirty = 1;
 473         spin_unlock(&sdp->sd_jindex_spin);
 474 }
 475
 476 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 477 {
 478         struct gfs2_jdesc *jd;
 479         int found = 0;
 480
 481         spin_lock(&sdp->sd_jindex_spin);
 482
 483         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 484                 if (jd->jd_dirty) {
 485                         jd->jd_dirty = 0;
 486                         found = 1;
 487                         break;
 488                 }
 489         }
 490         spin_unlock(&sdp->sd_jindex_spin);
 491
 492         if (!found)
 493                 jd = NULL;
 494
 495         return jd;
 496 }
 497
 498 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 499 {
 500         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 501         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 502         int ar;
 503         int error;
 504
 505         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 506             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 507                 gfs2_consist_inode(ip);
 508                 return -EIO;
 509         }
 510         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 511
 512         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 513         if (!error && ar) {
 514                 gfs2_consist_inode(ip);
 515                 error = -EIO;
 516         }
 517
 518         return error;
 519 }
 520
 521 /**
 522  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 523  * @sdp: the filesystem
 524  *
 525  * Returns: errno
 526  */
 527
 528 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 529 {
 530         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 531         struct gfs2_glock *j_gl = ip->i_gl;
 532         struct gfs2_holder t_gh;
 533         struct gfs2_log_header_host head;
 534         int error;
 535
 536         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 537         if (error)
 538                 return error;
 539
 540         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 541
 542         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 543         if (error)
 544                 goto fail;
 545
 546         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 547                 gfs2_consist(sdp);
 548                 error = -EIO;
 549                 goto fail;
 550         }
 551
 552         /*  Initialize some head of the log stuff  */
 553         sdp->sd_log_sequence = head.lh_sequence + 1;
 554         gfs2_log_pointers_init(sdp, head.lh_blkno);
 555
 556         error = gfs2_quota_init(sdp);
 557         if (error)
 558                 goto fail;
 559
 560         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 561
 562         gfs2_glock_dq_uninit(&t_gh);
 563
 564         return 0;
 565
 566 fail:
 567         t_gh.gh_flags |= GL_NOCACHE;
 568         gfs2_glock_dq_uninit(&t_gh);
 569
 570         return error;
 571 }
 572
 573 /**
 574  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 575  * @sdp: the filesystem
 576  *
 577  * Returns: errno
 578  */
 579
 580 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 581 {
 582         struct gfs2_holder t_gh;
 583         int error;
 584
 585         gfs2_quota_sync(sdp);
 586         gfs2_statfs_sync(sdp);
 587
 588         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 589                                    &t_gh);
 590         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 591                 return error;
 592
 593         gfs2_meta_syncfs(sdp);
 594         gfs2_log_shutdown(sdp);
 595
 596         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 597
 598         if (t_gh.gh_gl)
 599                 gfs2_glock_dq_uninit(&t_gh);
 600
 601         gfs2_quota_cleanup(sdp);
 602
 603         return error;
 604 }
 605
 606 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 607 {
 608         const struct gfs2_statfs_change *str = buf;
 609
 610         sc->sc_total = be64_to_cpu(str->sc_total);
 611         sc->sc_free = be64_to_cpu(str->sc_free);
 612         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 613 }
 614
 615 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 616 {
 617         struct gfs2_statfs_change *str = buf;
 618
 619         str->sc_total = cpu_to_be64(sc->sc_total);
 620         str->sc_free = cpu_to_be64(sc->sc_free);
 621         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 622 }
 623
 624 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 625 {
 626         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 627         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 628         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 629         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 630         struct buffer_head *m_bh, *l_bh;
 631         struct gfs2_holder gh;
 632         int error;
 633
 634         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 635                                    &gh);
 636         if (error)
 637                 return error;
 638
 639         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 640         if (error)
 641                 goto out;
 642
 643         if (sdp->sd_args.ar_spectator) {
 644                 spin_lock(&sdp->sd_statfs_spin);
 645                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 646                                       sizeof(struct gfs2_dinode));
 647                 spin_unlock(&sdp->sd_statfs_spin);
 648         } else {
 649                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 650                 if (error)
 651                         goto out_m_bh;
 652
 653                 spin_lock(&sdp->sd_statfs_spin);
 654                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 655                                       sizeof(struct gfs2_dinode));
 656                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 657                                       sizeof(struct gfs2_dinode));
 658                 spin_unlock(&sdp->sd_statfs_spin);
 659
 660                 brelse(l_bh);
 661         }
 662
 663 out_m_bh:
 664         brelse(m_bh);
 665 out:
 666         gfs2_glock_dq_uninit(&gh);
 667         return 0;
 668 }
 669
 670 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 671                         s64 dinodes)
 672 {
 673         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 674         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 675         struct buffer_head *l_bh;
 676         int error;
 677
 678         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 679         if (error)
 680                 return;
 681
 682         mutex_lock(&sdp->sd_statfs_mutex);
 683         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 684         mutex_unlock(&sdp->sd_statfs_mutex);
 685
 686         spin_lock(&sdp->sd_statfs_spin);
 687         l_sc->sc_total += total;
 688         l_sc->sc_free += free;
 689         l_sc->sc_dinodes += dinodes;
 690         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 691         spin_unlock(&sdp->sd_statfs_spin);
 692
 693         brelse(l_bh);
 694 }
 695
 696 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 697 {
 698         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 699         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 700         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 701         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 702         struct gfs2_holder gh;
 703         struct buffer_head *m_bh, *l_bh;
 704         int error;
 705
 706         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 707                                    &gh);
 708         if (error)
 709                 return error;
 710
 711         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 712         if (error)
 713                 goto out;
 714
 715         spin_lock(&sdp->sd_statfs_spin);
 716         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 717                               sizeof(struct gfs2_dinode));
 718         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 719                 spin_unlock(&sdp->sd_statfs_spin);
 720                 goto out_bh;
 721         }
 722         spin_unlock(&sdp->sd_statfs_spin);
 723
 724         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 725         if (error)
 726                 goto out_bh;
 727
 728         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 729         if (error)
 730                 goto out_bh2;
 731
 732         mutex_lock(&sdp->sd_statfs_mutex);
 733         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 734         mutex_unlock(&sdp->sd_statfs_mutex);
 735
 736         spin_lock(&sdp->sd_statfs_spin);
 737         m_sc->sc_total += l_sc->sc_total;
 738         m_sc->sc_free += l_sc->sc_free;
 739         m_sc->sc_dinodes += l_sc->sc_dinodes;
 740         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 741         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 742                0, sizeof(struct gfs2_statfs_change));
 743         spin_unlock(&sdp->sd_statfs_spin);
 744
 745         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 746         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 747
 748         gfs2_trans_end(sdp);
 749
 750 out_bh2:
 751         brelse(l_bh);
 752 out_bh:
 753         brelse(m_bh);
 754 out:
 755         gfs2_glock_dq_uninit(&gh);
 756         return error;
 757 }
 758
 759 /**
 760  * gfs2_statfs_i - Do a statfs
 761  * @sdp: the filesystem
 762  * @sg: the sg structure
 763  *
 764  * Returns: errno
 765  */
 766
 767 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 768 {
 769         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 770         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 771
 772         spin_lock(&sdp->sd_statfs_spin);
 773
 774         *sc = *m_sc;
 775         sc->sc_total += l_sc->sc_total;
 776         sc->sc_free += l_sc->sc_free;
 777         sc->sc_dinodes += l_sc->sc_dinodes;
 778
 779         spin_unlock(&sdp->sd_statfs_spin);
 780
 781         if (sc->sc_free < 0)
 782                 sc->sc_free = 0;
 783         if (sc->sc_free > sc->sc_total)
 784                 sc->sc_free = sc->sc_total;
 785         if (sc->sc_dinodes < 0)
 786                 sc->sc_dinodes = 0;
 787
 788         return 0;
 789 }
 790
 791 /**
 792  * statfs_fill - fill in the sg for a given RG
 793  * @rgd: the RG
 794  * @sc: the sc structure
 795  *
 796  * Returns: 0 on success, -ESTALE if the LVB is invalid
 797  */
 798
 799 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 800                             struct gfs2_statfs_change_host *sc)
 801 {
 802         gfs2_rgrp_verify(rgd);
 803         sc->sc_total += rgd->rd_data;
 804         sc->sc_free += rgd->rd_rg.rg_free;
 805         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 806         return 0;
 807 }
 808
 809 /**
 810  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 811  * @sdp: the filesystem
 812  * @sc: the sc info that will be returned
 813  *
 814  * Any error (other than a signal) will cause this routine to fall back
 815  * to the synchronous version.
 816  *
 817  * FIXME: This really shouldn't busy wait like this.
 818  *
 819  * Returns: errno
 820  */
 821
 822 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 823 {
 824         struct gfs2_holder ri_gh;
 825         struct gfs2_rgrpd *rgd_next;
 826         struct gfs2_holder *gha, *gh;
 827         unsigned int slots = 64;
 828         unsigned int x;
 829         int done;
 830         int error = 0, err;
 831
 832         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 833         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 834         if (!gha)
 835                 return -ENOMEM;
 836
 837         error = gfs2_rindex_hold(sdp, &ri_gh);
 838         if (error)
 839                 goto out;
 840
 841         rgd_next = gfs2_rgrpd_get_first(sdp);
 842
 843         for (;;) {
 844                 done = 1;
 845
 846                 for (x = 0; x < slots; x++) {
 847                         gh = gha + x;
 848
 849                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 850                                 err = gfs2_glock_wait(gh);
 851                                 if (err) {
 852                                         gfs2_holder_uninit(gh);
 853                                         error = err;
 854                                 } else {
 855                                         if (!error)
 856                                                 error = statfs_slow_fill(
 857                                                         gh->gh_gl->gl_object, sc);
 858                                         gfs2_glock_dq_uninit(gh);
 859                                 }
 860                         }
 861
 862                         if (gh->gh_gl)
 863                                 done = 0;
 864                         else if (rgd_next && !error) {
 865                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 866                                                            LM_ST_SHARED,
 867                                                            GL_ASYNC,
 868                                                            gh);
 869                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 870                                 done = 0;
 871                         }
 872
 873                         if (signal_pending(current))
 874                                 error = -ERESTARTSYS;
 875                 }
 876
 877                 if (done)
 878                         break;
 879
 880                 yield();
 881         }
 882
 883         gfs2_glock_dq_uninit(&ri_gh);
 884
 885 out:
 886         kfree(gha);
 887         return error;
 888 }
 889
 890 struct lfcc {
 891         struct list_head list;
 892         struct gfs2_holder gh;
 893 };
 894
 895 /**
 896  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 897  *                            journals are clean
 898  * @sdp: the file system
 899  * @state: the state to put the transaction lock into
 900  * @t_gh: the hold on the transaction lock
 901  *
 902  * Returns: errno
 903  */
 904
 905 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 906                                     struct gfs2_holder *t_gh)
 907 {
 908         struct gfs2_inode *ip;
 909         struct gfs2_holder ji_gh;
 910         struct gfs2_jdesc *jd;
 911         struct lfcc *lfcc;
 912         LIST_HEAD(list);
 913         struct gfs2_log_header_host lh;
 914         int error;
 915
 916         error = gfs2_jindex_hold(sdp, &ji_gh);
 917         if (error)
 918                 return error;
 919
 920         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 921                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 922                 if (!lfcc) {
 923                         error = -ENOMEM;
 924                         goto out;
 925                 }
 926                 ip = GFS2_I(jd->jd_inode);
 927                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 928                 if (error) {
 929                         kfree(lfcc);
 930                         goto out;
 931                 }
 932                 list_add(&lfcc->list, &list);
 933         }
 934
 935         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 936                                LM_FLAG_PRIORITY | GL_NOCACHE,
 937                                t_gh);
 938
 939         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 940                 error = gfs2_jdesc_check(jd);
 941                 if (error)
 942                         break;
 943                 error = gfs2_find_jhead(jd, &lh);
 944                 if (error)
 945                         break;
 946                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 947                         error = -EBUSY;
 948                         break;
 949                 }
 950         }
 951
 952         if (error)
 953                 gfs2_glock_dq_uninit(t_gh);
 954
 955 out:
 956         while (!list_empty(&list)) {
 957                 lfcc = list_entry(list.next, struct lfcc, list);
 958                 list_del(&lfcc->list);
 959                 gfs2_glock_dq_uninit(&lfcc->gh);
 960                 kfree(lfcc);
 961         }
 962         gfs2_glock_dq_uninit(&ji_gh);
 963         return error;
 964 }
 965
 966 /**
 967  * gfs2_freeze_fs - freezes the file system
 968  * @sdp: the file system
 969  *
 970  * This function flushes data and meta data for all machines by
 971  * aquiring the transaction log exclusively.  All journals are
 972  * ensured to be in a clean state as well.
 973  *
 974  * Returns: errno
 975  */
 976
 977 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 978 {
 979         int error = 0;
 980
 981         mutex_lock(&sdp->sd_freeze_lock);
 982
 983         if (!sdp->sd_freeze_count++) {
 984                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 985                 if (error)
 986                         sdp->sd_freeze_count--;
 987         }
 988
 989         mutex_unlock(&sdp->sd_freeze_lock);
 990
 991         return error;
 992 }
 993
 994 /**
 995  * gfs2_unfreeze_fs - unfreezes the file system
 996  * @sdp: the file system
 997  *
 998  * This function allows the file system to proceed by unlocking
 999  * the exclusively held transaction lock.  Other GFS2 nodes are
1000  * now free to acquire the lock shared and go on with their lives.
1001  *
1002  */
1003
1004 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1005 {
1006         mutex_lock(&sdp->sd_freeze_lock);
1007
1008         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1009                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1010
1011         mutex_unlock(&sdp->sd_freeze_lock);
1012 }
1013