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_jindex_refresh_secs = 60;
  58         gt->gt_recoverd_secs = 60;
  59         gt->gt_logd_secs = 1;
  60         gt->gt_quotad_secs = 5;
  61         gt->gt_quota_simul_sync = 64;
  62         gt->gt_quota_warn_period = 10;
  63         gt->gt_quota_scale_num = 1;
  64         gt->gt_quota_scale_den = 1;
  65         gt->gt_quota_cache_secs = 300;
  66         gt->gt_quota_quantum = 60;
  67         gt->gt_atime_quantum = 3600;
  68         gt->gt_new_files_jdata = 0;
  69         gt->gt_new_files_directio = 0;
  70         gt->gt_max_readahead = 1 << 18;
  71         gt->gt_stall_secs = 600;
  72         gt->gt_complain_secs = 10;
  73         gt->gt_reclaim_limit = 5000;
  74         gt->gt_statfs_quantum = 30;
  75         gt->gt_statfs_slow = 0;
  76 }
  77
  78 /**
  79  * gfs2_check_sb - Check superblock
  80  * @sdp: the filesystem
  81  * @sb: The superblock
  82  * @silent: Don't print a message if the check fails
  83  *
  84  * Checks the version code of the FS is one that we understand how to
  85  * read and that the sizes of the various on-disk structures have not
  86  * changed.
  87  */
  88
  89 int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
  90 {
  91         unsigned int x;
  92
  93         if (sb->sb_magic != GFS2_MAGIC ||
  94             sb->sb_type != GFS2_METATYPE_SB) {
  95                 if (!silent)
  96                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
  97                 return -EINVAL;
  98         }
  99
 100         /*  If format numbers match exactly, we're done.  */
 101
 102         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
 103             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
 104                 return 0;
 105
 106         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
 107                 for (x = 0; gfs2_old_fs_formats[x]; x++)
 108                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
 109                                 break;
 110
 111                 if (!gfs2_old_fs_formats[x]) {
 112                         printk(KERN_WARNING
 113                                "GFS2: code version (%u, %u) is incompatible "
 114                                "with ondisk format (%u, %u)\n",
 115                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 116                                sb->sb_fs_format, sb->sb_multihost_format);
 117                         printk(KERN_WARNING
 118                                "GFS2: I don't know how to upgrade this FS\n");
 119                         return -EINVAL;
 120                 }
 121         }
 122
 123         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
 124                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
 125                         if (gfs2_old_multihost_formats[x] ==
 126                             sb->sb_multihost_format)
 127                                 break;
 128
 129                 if (!gfs2_old_multihost_formats[x]) {
 130                         printk(KERN_WARNING
 131                                "GFS2: code version (%u, %u) is incompatible "
 132                                "with ondisk format (%u, %u)\n",
 133                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 134                                sb->sb_fs_format, sb->sb_multihost_format);
 135                         printk(KERN_WARNING
 136                                "GFS2: I don't know how to upgrade this FS\n");
 137                         return -EINVAL;
 138                 }
 139         }
 140
 141         if (!sdp->sd_args.ar_upgrade) {
 142                 printk(KERN_WARNING
 143                        "GFS2: code version (%u, %u) is incompatible "
 144                        "with ondisk format (%u, %u)\n",
 145                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
 146                        sb->sb_fs_format, sb->sb_multihost_format);
 147                 printk(KERN_INFO
 148                        "GFS2: Use the \"upgrade\" mount option to upgrade "
 149                        "the FS\n");
 150                 printk(KERN_INFO "GFS2: See the manual for more details\n");
 151                 return -EINVAL;
 152         }
 153
 154         return 0;
 155 }
 156
 157
 158 static void end_bio_io_page(struct bio *bio, int error)
 159 {
 160         struct page *page = bio->bi_private;
 161
 162         if (!error)
 163                 SetPageUptodate(page);
 164         else
 165                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
 166         unlock_page(page);
 167 }
 168
 169 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 170 {
 171         const struct gfs2_sb *str = buf;
 172
 173         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
 174         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
 175         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
 176         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
 177         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
 178         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
 179         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
 180         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
 181         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
 182         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
 183         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
 184
 185         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
 186         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
 187 }
 188
 189 /**
 190  * gfs2_read_super - Read the gfs2 super block from disk
 191  * @sdp: The GFS2 super block
 192  * @sector: The location of the super block
 193  * @error: The error code to return
 194  *
 195  * This uses the bio functions to read the super block from disk
 196  * because we want to be 100% sure that we never read cached data.
 197  * A super block is read twice only during each GFS2 mount and is
 198  * never written to by the filesystem. The first time its read no
 199  * locks are held, and the only details which are looked at are those
 200  * relating to the locking protocol. Once locking is up and working,
 201  * the sb is read again under the lock to establish the location of
 202  * the master directory (contains pointers to journals etc) and the
 203  * root directory.
 204  *
 205  * Returns: 0 on success or error
 206  */
 207
 208 int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
 209 {
 210         struct super_block *sb = sdp->sd_vfs;
 211         struct gfs2_sb *p;
 212         struct page *page;
 213         struct bio *bio;
 214
 215         page = alloc_page(GFP_KERNEL);
 216         if (unlikely(!page))
 217                 return -ENOBUFS;
 218
 219         ClearPageUptodate(page);
 220         ClearPageDirty(page);
 221         lock_page(page);
 222
 223         bio = bio_alloc(GFP_KERNEL, 1);
 224         if (unlikely(!bio)) {
 225                 __free_page(page);
 226                 return -ENOBUFS;
 227         }
 228
 229         bio->bi_sector = sector * (sb->s_blocksize >> 9);
 230         bio->bi_bdev = sb->s_bdev;
 231         bio_add_page(bio, page, PAGE_SIZE, 0);
 232
 233         bio->bi_end_io = end_bio_io_page;
 234         bio->bi_private = page;
 235         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
 236         wait_on_page_locked(page);
 237         bio_put(bio);
 238         if (!PageUptodate(page)) {
 239                 __free_page(page);
 240                 return -EIO;
 241         }
 242         p = kmap(page);
 243         gfs2_sb_in(&sdp->sd_sb, p);
 244         kunmap(page);
 245         __free_page(page);
 246         return 0;
 247 }
 248
 249 /**
 250  * gfs2_read_sb - Read super block
 251  * @sdp: The GFS2 superblock
 252  * @gl: the glock for the superblock (assumed to be held)
 253  * @silent: Don't print message if mount fails
 254  *
 255  */
 256
 257 int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
 258 {
 259         u32 hash_blocks, ind_blocks, leaf_blocks;
 260         u32 tmp_blocks;
 261         unsigned int x;
 262         int error;
 263
 264         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
 265         if (error) {
 266                 if (!silent)
 267                         fs_err(sdp, "can't read superblock\n");
 268                 return error;
 269         }
 270
 271         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
 272         if (error)
 273                 return error;
 274
 275         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
 276                                GFS2_BASIC_BLOCK_SHIFT;
 277         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 278         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
 279                           sizeof(struct gfs2_dinode)) / sizeof(u64);
 280         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
 281                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
 282         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
 283         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
 284         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
 285         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
 286         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
 287                                 sizeof(struct gfs2_meta_header)) /
 288                                 sizeof(struct gfs2_quota_change);
 289
 290         /* Compute maximum reservation required to add a entry to a directory */
 291
 292         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
 293                              sdp->sd_jbsize);
 294
 295         ind_blocks = 0;
 296         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
 297                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
 298                 ind_blocks += tmp_blocks;
 299         }
 300
 301         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
 302
 303         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
 304
 305         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
 306                                 sizeof(struct gfs2_dinode);
 307         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
 308         for (x = 2;; x++) {
 309                 u64 space, d;
 310                 u32 m;
 311
 312                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
 313                 d = space;
 314                 m = do_div(d, sdp->sd_inptrs);
 315
 316                 if (d != sdp->sd_heightsize[x - 1] || m)
 317                         break;
 318                 sdp->sd_heightsize[x] = space;
 319         }
 320         sdp->sd_max_height = x;
 321         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
 322
 323         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
 324                                  sizeof(struct gfs2_dinode);
 325         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
 326         for (x = 2;; x++) {
 327                 u64 space, d;
 328                 u32 m;
 329
 330                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
 331                 d = space;
 332                 m = do_div(d, sdp->sd_inptrs);
 333
 334                 if (d != sdp->sd_jheightsize[x - 1] || m)
 335                         break;
 336                 sdp->sd_jheightsize[x] = space;
 337         }
 338         sdp->sd_max_jheight = x;
 339         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
 340
 341         return 0;
 342 }
 343
 344 /**
 345  * gfs2_jindex_hold - Grab a lock on the jindex
 346  * @sdp: The GFS2 superblock
 347  * @ji_gh: the holder for the jindex glock
 348  *
 349  * This is very similar to the gfs2_rindex_hold() function, except that
 350  * in general we hold the jindex lock for longer periods of time and
 351  * we grab it far less frequently (in general) then the rgrp lock.
 352  *
 353  * Returns: errno
 354  */
 355
 356 int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
 357 {
 358         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
 359         struct qstr name;
 360         char buf[20];
 361         struct gfs2_jdesc *jd;
 362         int error;
 363
 364         name.name = buf;
 365
 366         mutex_lock(&sdp->sd_jindex_mutex);
 367
 368         for (;;) {
 369                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
 370                 if (error)
 371                         break;
 372
 373                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
 374                 name.hash = gfs2_disk_hash(name.name, name.len);
 375
 376                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
 377                 if (error == -ENOENT) {
 378                         error = 0;
 379                         break;
 380                 }
 381
 382                 gfs2_glock_dq_uninit(ji_gh);
 383
 384                 if (error)
 385                         break;
 386
 387                 error = -ENOMEM;
 388                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
 389                 if (!jd)
 390                         break;
 391
 392                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1, NULL);
 393                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
 394                         if (!jd->jd_inode)
 395                                 error = -ENOENT;
 396                         else
 397                                 error = PTR_ERR(jd->jd_inode);
 398                         kfree(jd);
 399                         break;
 400                 }
 401
 402                 spin_lock(&sdp->sd_jindex_spin);
 403                 jd->jd_jid = sdp->sd_journals++;
 404                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
 405                 spin_unlock(&sdp->sd_jindex_spin);
 406         }
 407
 408         mutex_unlock(&sdp->sd_jindex_mutex);
 409
 410         return error;
 411 }
 412
 413 /**
 414  * gfs2_jindex_free - Clear all the journal index information
 415  * @sdp: The GFS2 superblock
 416  *
 417  */
 418
 419 void gfs2_jindex_free(struct gfs2_sbd *sdp)
 420 {
 421         struct list_head list;
 422         struct gfs2_jdesc *jd;
 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                 list_del(&jd->jd_list);
 433                 iput(jd->jd_inode);
 434                 kfree(jd);
 435         }
 436 }
 437
 438 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
 439 {
 440         struct gfs2_jdesc *jd;
 441         int found = 0;
 442
 443         list_for_each_entry(jd, head, jd_list) {
 444                 if (jd->jd_jid == jid) {
 445                         found = 1;
 446                         break;
 447                 }
 448         }
 449
 450         if (!found)
 451                 jd = NULL;
 452
 453         return jd;
 454 }
 455
 456 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
 457 {
 458         struct gfs2_jdesc *jd;
 459
 460         spin_lock(&sdp->sd_jindex_spin);
 461         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 462         spin_unlock(&sdp->sd_jindex_spin);
 463
 464         return jd;
 465 }
 466
 467 void gfs2_jdesc_make_dirty(struct gfs2_sbd *sdp, unsigned int jid)
 468 {
 469         struct gfs2_jdesc *jd;
 470
 471         spin_lock(&sdp->sd_jindex_spin);
 472         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
 473         if (jd)
 474                 jd->jd_dirty = 1;
 475         spin_unlock(&sdp->sd_jindex_spin);
 476 }
 477
 478 struct gfs2_jdesc *gfs2_jdesc_find_dirty(struct gfs2_sbd *sdp)
 479 {
 480         struct gfs2_jdesc *jd;
 481         int found = 0;
 482
 483         spin_lock(&sdp->sd_jindex_spin);
 484
 485         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 486                 if (jd->jd_dirty) {
 487                         jd->jd_dirty = 0;
 488                         found = 1;
 489                         break;
 490                 }
 491         }
 492         spin_unlock(&sdp->sd_jindex_spin);
 493
 494         if (!found)
 495                 jd = NULL;
 496
 497         return jd;
 498 }
 499
 500 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
 501 {
 502         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 503         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 504         int ar;
 505         int error;
 506
 507         if (ip->i_di.di_size < (8 << 20) || ip->i_di.di_size > (1 << 30) ||
 508             (ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1))) {
 509                 gfs2_consist_inode(ip);
 510                 return -EIO;
 511         }
 512         jd->jd_blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
 513
 514         error = gfs2_write_alloc_required(ip, 0, ip->i_di.di_size, &ar);
 515         if (!error && ar) {
 516                 gfs2_consist_inode(ip);
 517                 error = -EIO;
 518         }
 519
 520         return error;
 521 }
 522
 523 /**
 524  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
 525  * @sdp: the filesystem
 526  *
 527  * Returns: errno
 528  */
 529
 530 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
 531 {
 532         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
 533         struct gfs2_glock *j_gl = ip->i_gl;
 534         struct gfs2_holder t_gh;
 535         struct gfs2_log_header_host head;
 536         int error;
 537
 538         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
 539         if (error)
 540                 return error;
 541
 542         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
 543
 544         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
 545         if (error)
 546                 goto fail;
 547
 548         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 549                 gfs2_consist(sdp);
 550                 error = -EIO;
 551                 goto fail;
 552         }
 553
 554         /*  Initialize some head of the log stuff  */
 555         sdp->sd_log_sequence = head.lh_sequence + 1;
 556         gfs2_log_pointers_init(sdp, head.lh_blkno);
 557
 558         error = gfs2_quota_init(sdp);
 559         if (error)
 560                 goto fail;
 561
 562         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 563
 564         gfs2_glock_dq_uninit(&t_gh);
 565
 566         return 0;
 567
 568 fail:
 569         t_gh.gh_flags |= GL_NOCACHE;
 570         gfs2_glock_dq_uninit(&t_gh);
 571
 572         return error;
 573 }
 574
 575 /**
 576  * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
 577  * @sdp: the filesystem
 578  *
 579  * Returns: errno
 580  */
 581
 582 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
 583 {
 584         struct gfs2_holder t_gh;
 585         int error;
 586
 587         gfs2_quota_sync(sdp);
 588         gfs2_statfs_sync(sdp);
 589
 590         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE,
 591                                    &t_gh);
 592         if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
 593                 return error;
 594
 595         gfs2_meta_syncfs(sdp);
 596         gfs2_log_shutdown(sdp);
 597
 598         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
 599
 600         if (t_gh.gh_gl)
 601                 gfs2_glock_dq_uninit(&t_gh);
 602
 603         gfs2_quota_cleanup(sdp);
 604
 605         return error;
 606 }
 607
 608 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
 609 {
 610         const struct gfs2_statfs_change *str = buf;
 611
 612         sc->sc_total = be64_to_cpu(str->sc_total);
 613         sc->sc_free = be64_to_cpu(str->sc_free);
 614         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
 615 }
 616
 617 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
 618 {
 619         struct gfs2_statfs_change *str = buf;
 620
 621         str->sc_total = cpu_to_be64(sc->sc_total);
 622         str->sc_free = cpu_to_be64(sc->sc_free);
 623         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
 624 }
 625
 626 int gfs2_statfs_init(struct gfs2_sbd *sdp)
 627 {
 628         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 629         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 630         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 631         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 632         struct buffer_head *m_bh, *l_bh;
 633         struct gfs2_holder gh;
 634         int error;
 635
 636         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 637                                    &gh);
 638         if (error)
 639                 return error;
 640
 641         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 642         if (error)
 643                 goto out;
 644
 645         if (sdp->sd_args.ar_spectator) {
 646                 spin_lock(&sdp->sd_statfs_spin);
 647                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 648                                       sizeof(struct gfs2_dinode));
 649                 spin_unlock(&sdp->sd_statfs_spin);
 650         } else {
 651                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 652                 if (error)
 653                         goto out_m_bh;
 654
 655                 spin_lock(&sdp->sd_statfs_spin);
 656                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
 657                                       sizeof(struct gfs2_dinode));
 658                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
 659                                       sizeof(struct gfs2_dinode));
 660                 spin_unlock(&sdp->sd_statfs_spin);
 661
 662                 brelse(l_bh);
 663         }
 664
 665 out_m_bh:
 666         brelse(m_bh);
 667 out:
 668         gfs2_glock_dq_uninit(&gh);
 669         return 0;
 670 }
 671
 672 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
 673                         s64 dinodes)
 674 {
 675         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 676         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 677         struct buffer_head *l_bh;
 678         int error;
 679
 680         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 681         if (error)
 682                 return;
 683
 684         mutex_lock(&sdp->sd_statfs_mutex);
 685         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 686         mutex_unlock(&sdp->sd_statfs_mutex);
 687
 688         spin_lock(&sdp->sd_statfs_spin);
 689         l_sc->sc_total += total;
 690         l_sc->sc_free += free;
 691         l_sc->sc_dinodes += dinodes;
 692         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
 693         spin_unlock(&sdp->sd_statfs_spin);
 694
 695         brelse(l_bh);
 696 }
 697
 698 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
 699 {
 700         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
 701         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
 702         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 703         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 704         struct gfs2_holder gh;
 705         struct buffer_head *m_bh, *l_bh;
 706         int error;
 707
 708         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
 709                                    &gh);
 710         if (error)
 711                 return error;
 712
 713         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
 714         if (error)
 715                 goto out;
 716
 717         spin_lock(&sdp->sd_statfs_spin);
 718         gfs2_statfs_change_in(m_sc, m_bh->b_data +
 719                               sizeof(struct gfs2_dinode));
 720         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
 721                 spin_unlock(&sdp->sd_statfs_spin);
 722                 goto out_bh;
 723         }
 724         spin_unlock(&sdp->sd_statfs_spin);
 725
 726         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
 727         if (error)
 728                 goto out_bh;
 729
 730         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
 731         if (error)
 732                 goto out_bh2;
 733
 734         mutex_lock(&sdp->sd_statfs_mutex);
 735         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
 736         mutex_unlock(&sdp->sd_statfs_mutex);
 737
 738         spin_lock(&sdp->sd_statfs_spin);
 739         m_sc->sc_total += l_sc->sc_total;
 740         m_sc->sc_free += l_sc->sc_free;
 741         m_sc->sc_dinodes += l_sc->sc_dinodes;
 742         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
 743         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
 744                0, sizeof(struct gfs2_statfs_change));
 745         spin_unlock(&sdp->sd_statfs_spin);
 746
 747         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
 748         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
 749
 750         gfs2_trans_end(sdp);
 751
 752 out_bh2:
 753         brelse(l_bh);
 754 out_bh:
 755         brelse(m_bh);
 756 out:
 757         gfs2_glock_dq_uninit(&gh);
 758         return error;
 759 }
 760
 761 /**
 762  * gfs2_statfs_i - Do a statfs
 763  * @sdp: the filesystem
 764  * @sg: the sg structure
 765  *
 766  * Returns: errno
 767  */
 768
 769 int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 770 {
 771         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
 772         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
 773
 774         spin_lock(&sdp->sd_statfs_spin);
 775
 776         *sc = *m_sc;
 777         sc->sc_total += l_sc->sc_total;
 778         sc->sc_free += l_sc->sc_free;
 779         sc->sc_dinodes += l_sc->sc_dinodes;
 780
 781         spin_unlock(&sdp->sd_statfs_spin);
 782
 783         if (sc->sc_free < 0)
 784                 sc->sc_free = 0;
 785         if (sc->sc_free > sc->sc_total)
 786                 sc->sc_free = sc->sc_total;
 787         if (sc->sc_dinodes < 0)
 788                 sc->sc_dinodes = 0;
 789
 790         return 0;
 791 }
 792
 793 /**
 794  * statfs_fill - fill in the sg for a given RG
 795  * @rgd: the RG
 796  * @sc: the sc structure
 797  *
 798  * Returns: 0 on success, -ESTALE if the LVB is invalid
 799  */
 800
 801 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
 802                             struct gfs2_statfs_change_host *sc)
 803 {
 804         gfs2_rgrp_verify(rgd);
 805         sc->sc_total += rgd->rd_data;
 806         sc->sc_free += rgd->rd_rg.rg_free;
 807         sc->sc_dinodes += rgd->rd_rg.rg_dinodes;
 808         return 0;
 809 }
 810
 811 /**
 812  * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
 813  * @sdp: the filesystem
 814  * @sc: the sc info that will be returned
 815  *
 816  * Any error (other than a signal) will cause this routine to fall back
 817  * to the synchronous version.
 818  *
 819  * FIXME: This really shouldn't busy wait like this.
 820  *
 821  * Returns: errno
 822  */
 823
 824 int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
 825 {
 826         struct gfs2_holder ri_gh;
 827         struct gfs2_rgrpd *rgd_next;
 828         struct gfs2_holder *gha, *gh;
 829         unsigned int slots = 64;
 830         unsigned int x;
 831         int done;
 832         int error = 0, err;
 833
 834         memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
 835         gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
 836         if (!gha)
 837                 return -ENOMEM;
 838
 839         error = gfs2_rindex_hold(sdp, &ri_gh);
 840         if (error)
 841                 goto out;
 842
 843         rgd_next = gfs2_rgrpd_get_first(sdp);
 844
 845         for (;;) {
 846                 done = 1;
 847
 848                 for (x = 0; x < slots; x++) {
 849                         gh = gha + x;
 850
 851                         if (gh->gh_gl && gfs2_glock_poll(gh)) {
 852                                 err = gfs2_glock_wait(gh);
 853                                 if (err) {
 854                                         gfs2_holder_uninit(gh);
 855                                         error = err;
 856                                 } else {
 857                                         if (!error)
 858                                                 error = statfs_slow_fill(
 859                                                         gh->gh_gl->gl_object, sc);
 860                                         gfs2_glock_dq_uninit(gh);
 861                                 }
 862                         }
 863
 864                         if (gh->gh_gl)
 865                                 done = 0;
 866                         else if (rgd_next && !error) {
 867                                 error = gfs2_glock_nq_init(rgd_next->rd_gl,
 868                                                            LM_ST_SHARED,
 869                                                            GL_ASYNC,
 870                                                            gh);
 871                                 rgd_next = gfs2_rgrpd_get_next(rgd_next);
 872                                 done = 0;
 873                         }
 874
 875                         if (signal_pending(current))
 876                                 error = -ERESTARTSYS;
 877                 }
 878
 879                 if (done)
 880                         break;
 881
 882                 yield();
 883         }
 884
 885         gfs2_glock_dq_uninit(&ri_gh);
 886
 887 out:
 888         kfree(gha);
 889         return error;
 890 }
 891
 892 struct lfcc {
 893         struct list_head list;
 894         struct gfs2_holder gh;
 895 };
 896
 897 /**
 898  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
 899  *                            journals are clean
 900  * @sdp: the file system
 901  * @state: the state to put the transaction lock into
 902  * @t_gh: the hold on the transaction lock
 903  *
 904  * Returns: errno
 905  */
 906
 907 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
 908                                     struct gfs2_holder *t_gh)
 909 {
 910         struct gfs2_inode *ip;
 911         struct gfs2_holder ji_gh;
 912         struct gfs2_jdesc *jd;
 913         struct lfcc *lfcc;
 914         LIST_HEAD(list);
 915         struct gfs2_log_header_host lh;
 916         int error;
 917
 918         error = gfs2_jindex_hold(sdp, &ji_gh);
 919         if (error)
 920                 return error;
 921
 922         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 923                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
 924                 if (!lfcc) {
 925                         error = -ENOMEM;
 926                         goto out;
 927                 }
 928                 ip = GFS2_I(jd->jd_inode);
 929                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
 930                 if (error) {
 931                         kfree(lfcc);
 932                         goto out;
 933                 }
 934                 list_add(&lfcc->list, &list);
 935         }
 936
 937         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
 938                                LM_FLAG_PRIORITY | GL_NOCACHE,
 939                                t_gh);
 940
 941         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
 942                 error = gfs2_jdesc_check(jd);
 943                 if (error)
 944                         break;
 945                 error = gfs2_find_jhead(jd, &lh);
 946                 if (error)
 947                         break;
 948                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
 949                         error = -EBUSY;
 950                         break;
 951                 }
 952         }
 953
 954         if (error)
 955                 gfs2_glock_dq_uninit(t_gh);
 956
 957 out:
 958         while (!list_empty(&list)) {
 959                 lfcc = list_entry(list.next, struct lfcc, list);
 960                 list_del(&lfcc->list);
 961                 gfs2_glock_dq_uninit(&lfcc->gh);
 962                 kfree(lfcc);
 963         }
 964         gfs2_glock_dq_uninit(&ji_gh);
 965         return error;
 966 }
 967
 968 /**
 969  * gfs2_freeze_fs - freezes the file system
 970  * @sdp: the file system
 971  *
 972  * This function flushes data and meta data for all machines by
 973  * aquiring the transaction log exclusively.  All journals are
 974  * ensured to be in a clean state as well.
 975  *
 976  * Returns: errno
 977  */
 978
 979 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
 980 {
 981         int error = 0;
 982
 983         mutex_lock(&sdp->sd_freeze_lock);
 984
 985         if (!sdp->sd_freeze_count++) {
 986                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
 987                 if (error)
 988                         sdp->sd_freeze_count--;
 989         }
 990
 991         mutex_unlock(&sdp->sd_freeze_lock);
 992
 993         return error;
 994 }
 995
 996 /**
 997  * gfs2_unfreeze_fs - unfreezes the file system
 998  * @sdp: the file system
 999  *
1000  * This function allows the file system to proceed by unlocking
1001  * the exclusively held transaction lock.  Other GFS2 nodes are
1002  * now free to acquire the lock shared and go on with their lives.
1003  *
1004  */
1005
1006 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
1007 {
1008         mutex_lock(&sdp->sd_freeze_lock);
1009
1010         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
1011                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
1012
1013         mutex_unlock(&sdp->sd_freeze_lock);
1014 }
1015