err.no Git - linux-2.6/blob - fs/gfs2/log.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/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/lm_interface.h>
  18 #include <linux/delay.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "glock.h"
  24 #include "log.h"
  25 #include "lops.h"
  26 #include "meta_io.h"
  27 #include "util.h"
  28 #include "dir.h"
  29
  30 #define PULL 1
  31
  32 /**
  33  * gfs2_struct2blk - compute stuff
  34  * @sdp: the filesystem
  35  * @nstruct: the number of structures
  36  * @ssize: the size of the structures
  37  *
  38  * Compute the number of log descriptor blocks needed to hold a certain number
  39  * of structures of a certain size.
  40  *
  41  * Returns: the number of blocks needed (minimum is always 1)
  42  */
  43
  44 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  45                              unsigned int ssize)
  46 {
  47         unsigned int blks;
  48         unsigned int first, second;
  49
  50         blks = 1;
  51         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  52
  53         if (nstruct > first) {
  54                 second = (sdp->sd_sb.sb_bsize -
  55                           sizeof(struct gfs2_meta_header)) / ssize;
  56                 blks += DIV_ROUND_UP(nstruct - first, second);
  57         }
  58
  59         return blks;
  60 }
  61
  62 /**
  63  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  64  * @mapping: The associated mapping (maybe NULL)
  65  * @bd: The gfs2_bufdata to remove
  66  *
  67  * The log lock _must_ be held when calling this function
  68  *
  69  */
  70
  71 void gfs2_remove_from_ail(struct address_space *mapping, struct gfs2_bufdata *bd)
  72 {
  73         bd->bd_ail = NULL;
  74         list_del_init(&bd->bd_ail_st_list);
  75         list_del_init(&bd->bd_ail_gl_list);
  76         atomic_dec(&bd->bd_gl->gl_ail_count);
  77         if (mapping)
  78                 gfs2_meta_cache_flush(GFS2_I(mapping->host));
  79         brelse(bd->bd_bh);
  80 }
  81
  82 /**
  83  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  84  * @sdp: the filesystem
  85  * @tr: the part of the AIL
  86  *
  87  */
  88
  89 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  90 {
  91         struct gfs2_bufdata *bd, *s;
  92         struct buffer_head *bh;
  93         int retry;
  94
  95         BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
  96
  97         do {
  98                 retry = 0;
  99
 100                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 101                                                  bd_ail_st_list) {
 102                         bh = bd->bd_bh;
 103
 104                         gfs2_assert(sdp, bd->bd_ail == ai);
 105
 106                         if (!buffer_busy(bh)) {
 107                                 if (!buffer_uptodate(bh)) {
 108                                         gfs2_log_unlock(sdp);
 109                                         gfs2_io_error_bh(sdp, bh);
 110                                         gfs2_log_lock(sdp);
 111                                 }
 112                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 113                                 continue;
 114                         }
 115
 116                         if (!buffer_dirty(bh))
 117                                 continue;
 118
 119                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 120
 121                         gfs2_log_unlock(sdp);
 122                         wait_on_buffer(bh);
 123                         ll_rw_block(WRITE, 1, &bh);
 124                         gfs2_log_lock(sdp);
 125
 126                         retry = 1;
 127                         break;
 128                 }
 129         } while (retry);
 130 }
 131
 132 /**
 133  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 134  * @sdp: the filesystem
 135  * @ai: the AIL entry
 136  *
 137  */
 138
 139 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 140 {
 141         struct gfs2_bufdata *bd, *s;
 142         struct buffer_head *bh;
 143
 144         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 145                                          bd_ail_st_list) {
 146                 bh = bd->bd_bh;
 147
 148                 gfs2_assert(sdp, bd->bd_ail == ai);
 149
 150                 if (buffer_busy(bh)) {
 151                         if (flags & DIO_ALL)
 152                                 continue;
 153                         else
 154                                 break;
 155                 }
 156
 157                 if (!buffer_uptodate(bh))
 158                         gfs2_io_error_bh(sdp, bh);
 159
 160                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 161         }
 162
 163         return list_empty(&ai->ai_ail1_list);
 164 }
 165
 166 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
 167 {
 168         struct list_head *head;
 169         u64 sync_gen;
 170         struct list_head *first;
 171         struct gfs2_ail *first_ai, *ai, *tmp;
 172         int done = 0;
 173
 174         gfs2_log_lock(sdp);
 175         head = &sdp->sd_ail1_list;
 176         if (list_empty(head)) {
 177                 gfs2_log_unlock(sdp);
 178                 return;
 179         }
 180         sync_gen = sdp->sd_ail_sync_gen++;
 181
 182         first = head->prev;
 183         first_ai = list_entry(first, struct gfs2_ail, ai_list);
 184         first_ai->ai_sync_gen = sync_gen;
 185         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
 186
 187         if (flags & DIO_ALL)
 188                 first = NULL;
 189
 190         while(!done) {
 191                 if (first && (head->prev != first ||
 192                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
 193                         break;
 194
 195                 done = 1;
 196                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
 197                         if (ai->ai_sync_gen >= sync_gen)
 198                                 continue;
 199                         ai->ai_sync_gen = sync_gen;
 200                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
 201                         done = 0;
 202                         break;
 203                 }
 204         }
 205
 206         gfs2_log_unlock(sdp);
 207 }
 208
 209 int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 210 {
 211         struct gfs2_ail *ai, *s;
 212         int ret;
 213
 214         gfs2_log_lock(sdp);
 215
 216         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 217                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 218                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 219                 else if (!(flags & DIO_ALL))
 220                         break;
 221         }
 222
 223         ret = list_empty(&sdp->sd_ail1_list);
 224
 225         gfs2_log_unlock(sdp);
 226
 227         return ret;
 228 }
 229
 230
 231 /**
 232  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 233  * @sdp: the filesystem
 234  * @ai: the AIL entry
 235  *
 236  */
 237
 238 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 239 {
 240         struct list_head *head = &ai->ai_ail2_list;
 241         struct gfs2_bufdata *bd;
 242
 243         while (!list_empty(head)) {
 244                 bd = list_entry(head->prev, struct gfs2_bufdata,
 245                                 bd_ail_st_list);
 246                 gfs2_assert(sdp, bd->bd_ail == ai);
 247                 gfs2_remove_from_ail(bd->bd_bh->b_page->mapping, bd);
 248         }
 249 }
 250
 251 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 252 {
 253         struct gfs2_ail *ai, *safe;
 254         unsigned int old_tail = sdp->sd_log_tail;
 255         int wrap = (new_tail < old_tail);
 256         int a, b, rm;
 257
 258         gfs2_log_lock(sdp);
 259
 260         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 261                 a = (old_tail <= ai->ai_first);
 262                 b = (ai->ai_first < new_tail);
 263                 rm = (wrap) ? (a || b) : (a && b);
 264                 if (!rm)
 265                         continue;
 266
 267                 gfs2_ail2_empty_one(sdp, ai);
 268                 list_del(&ai->ai_list);
 269                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 270                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 271                 kfree(ai);
 272         }
 273
 274         gfs2_log_unlock(sdp);
 275 }
 276
 277 /**
 278  * gfs2_log_reserve - Make a log reservation
 279  * @sdp: The GFS2 superblock
 280  * @blks: The number of blocks to reserve
 281  *
 282  * Note that we never give out the last few blocks of the journal. Thats
 283  * due to the fact that there is a small number of header blocks
 284  * associated with each log flush. The exact number can't be known until
 285  * flush time, so we ensure that we have just enough free blocks at all
 286  * times to avoid running out during a log flush.
 287  *
 288  * Returns: errno
 289  */
 290
 291 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 292 {
 293         unsigned int try = 0;
 294         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
 295
 296         if (gfs2_assert_warn(sdp, blks) ||
 297             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 298                 return -EINVAL;
 299
 300         mutex_lock(&sdp->sd_log_reserve_mutex);
 301         gfs2_log_lock(sdp);
 302         while(sdp->sd_log_blks_free <= (blks + reserved_blks)) {
 303                 gfs2_log_unlock(sdp);
 304                 gfs2_ail1_empty(sdp, 0);
 305                 gfs2_log_flush(sdp, NULL);
 306
 307                 if (try++)
 308                         gfs2_ail1_start(sdp, 0);
 309                 gfs2_log_lock(sdp);
 310         }
 311         sdp->sd_log_blks_free -= blks;
 312         gfs2_log_unlock(sdp);
 313         mutex_unlock(&sdp->sd_log_reserve_mutex);
 314
 315         down_read(&sdp->sd_log_flush_lock);
 316
 317         return 0;
 318 }
 319
 320 /**
 321  * gfs2_log_release - Release a given number of log blocks
 322  * @sdp: The GFS2 superblock
 323  * @blks: The number of blocks
 324  *
 325  */
 326
 327 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 328 {
 329
 330         gfs2_log_lock(sdp);
 331         sdp->sd_log_blks_free += blks;
 332         gfs2_assert_withdraw(sdp,
 333                              sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 334         gfs2_log_unlock(sdp);
 335         up_read(&sdp->sd_log_flush_lock);
 336 }
 337
 338 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 339 {
 340         struct inode *inode = sdp->sd_jdesc->jd_inode;
 341         int error;
 342         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 343
 344         bh_map.b_size = 1 << inode->i_blkbits;
 345         error = gfs2_block_map(inode, lbn, 0, &bh_map);
 346         if (error || !bh_map.b_blocknr)
 347                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
 348                        (unsigned long long)bh_map.b_blocknr, lbn);
 349         gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
 350
 351         return bh_map.b_blocknr;
 352 }
 353
 354 /**
 355  * log_distance - Compute distance between two journal blocks
 356  * @sdp: The GFS2 superblock
 357  * @newer: The most recent journal block of the pair
 358  * @older: The older journal block of the pair
 359  *
 360  *   Compute the distance (in the journal direction) between two
 361  *   blocks in the journal
 362  *
 363  * Returns: the distance in blocks
 364  */
 365
 366 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 367                                         unsigned int older)
 368 {
 369         int dist;
 370
 371         dist = newer - older;
 372         if (dist < 0)
 373                 dist += sdp->sd_jdesc->jd_blocks;
 374
 375         return dist;
 376 }
 377
 378 /**
 379  * calc_reserved - Calculate the number of blocks to reserve when
 380  *                 refunding a transaction's unused buffers.
 381  * @sdp: The GFS2 superblock
 382  *
 383  * This is complex.  We need to reserve room for all our currently used
 384  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 385  * all our journaled data buffers for journaled files (e.g. files in the
 386  * meta_fs like rindex, or files for which chattr +j was done.)
 387  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 388  * will count it as free space (sd_log_blks_free) and corruption will follow.
 389  *
 390  * We can have metadata bufs and jdata bufs in the same journal.  So each
 391  * type gets its own log header, for which we need to reserve a block.
 392  * In fact, each type has the potential for needing more than one header
 393  * in cases where we have more buffers than will fit on a journal page.
 394  * Metadata journal entries take up half the space of journaled buffer entries.
 395  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 396  * databuf_limit (251) before they cause a wrap around.
 397  *
 398  * Also, we need to reserve blocks for revoke journal entries and one for an
 399  * overall header for the lot.
 400  *
 401  * Returns: the number of blocks reserved
 402  */
 403 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 404 {
 405         unsigned int reserved = 0;
 406         unsigned int mbuf_limit, metabufhdrs_needed;
 407         unsigned int dbuf_limit, databufhdrs_needed;
 408         unsigned int revokes = 0;
 409
 410         mbuf_limit = buf_limit(sdp);
 411         metabufhdrs_needed = (sdp->sd_log_commited_buf +
 412                               (mbuf_limit - 1)) / mbuf_limit;
 413         dbuf_limit = databuf_limit(sdp);
 414         databufhdrs_needed = (sdp->sd_log_commited_databuf +
 415                               (dbuf_limit - 1)) / dbuf_limit;
 416
 417         if (sdp->sd_log_commited_revoke)
 418                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 419                                           sizeof(u64));
 420
 421         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
 422                 sdp->sd_log_commited_databuf + databufhdrs_needed +
 423                 revokes;
 424         /* One for the overall header */
 425         if (reserved)
 426                 reserved++;
 427         return reserved;
 428 }
 429
 430 static unsigned int current_tail(struct gfs2_sbd *sdp)
 431 {
 432         struct gfs2_ail *ai;
 433         unsigned int tail;
 434
 435         gfs2_log_lock(sdp);
 436
 437         if (list_empty(&sdp->sd_ail1_list)) {
 438                 tail = sdp->sd_log_head;
 439         } else {
 440                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 441                 tail = ai->ai_first;
 442         }
 443
 444         gfs2_log_unlock(sdp);
 445
 446         return tail;
 447 }
 448
 449 static inline void log_incr_head(struct gfs2_sbd *sdp)
 450 {
 451         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 452                 gfs2_assert_withdraw(sdp, sdp->sd_log_flush_head == sdp->sd_log_head);
 453
 454         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 455                 sdp->sd_log_flush_head = 0;
 456                 sdp->sd_log_flush_wrapped = 1;
 457         }
 458 }
 459
 460 /**
 461  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 462  * @sdp: The GFS2 superblock
 463  *
 464  * Returns: the buffer_head
 465  */
 466
 467 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 468 {
 469         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 470         struct gfs2_log_buf *lb;
 471         struct buffer_head *bh;
 472
 473         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 474         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 475
 476         bh = lb->lb_bh = sb_getblk(sdp->sd_vfs, blkno);
 477         lock_buffer(bh);
 478         memset(bh->b_data, 0, bh->b_size);
 479         set_buffer_uptodate(bh);
 480         clear_buffer_dirty(bh);
 481         unlock_buffer(bh);
 482
 483         log_incr_head(sdp);
 484
 485         return bh;
 486 }
 487
 488 /**
 489  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 490  * @sdp: the filesystem
 491  * @data: the data the buffer_head should point to
 492  *
 493  * Returns: the log buffer descriptor
 494  */
 495
 496 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 497                                       struct buffer_head *real)
 498 {
 499         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 500         struct gfs2_log_buf *lb;
 501         struct buffer_head *bh;
 502
 503         lb = kzalloc(sizeof(struct gfs2_log_buf), GFP_NOFS | __GFP_NOFAIL);
 504         list_add(&lb->lb_list, &sdp->sd_log_flush_list);
 505         lb->lb_real = real;
 506
 507         bh = lb->lb_bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 508         atomic_set(&bh->b_count, 1);
 509         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate);
 510         set_bh_page(bh, real->b_page, bh_offset(real));
 511         bh->b_blocknr = blkno;
 512         bh->b_size = sdp->sd_sb.sb_bsize;
 513         bh->b_bdev = sdp->sd_vfs->s_bdev;
 514
 515         log_incr_head(sdp);
 516
 517         return bh;
 518 }
 519
 520 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 521 {
 522         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 523
 524         ail2_empty(sdp, new_tail);
 525
 526         gfs2_log_lock(sdp);
 527         sdp->sd_log_blks_free += dist;
 528         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <= sdp->sd_jdesc->jd_blocks);
 529         gfs2_log_unlock(sdp);
 530
 531         sdp->sd_log_tail = new_tail;
 532 }
 533
 534 /**
 535  * log_write_header - Get and initialize a journal header buffer
 536  * @sdp: The GFS2 superblock
 537  *
 538  * Returns: the initialized log buffer descriptor
 539  */
 540
 541 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 542 {
 543         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 544         struct buffer_head *bh;
 545         struct gfs2_log_header *lh;
 546         unsigned int tail;
 547         u32 hash;
 548
 549         bh = sb_getblk(sdp->sd_vfs, blkno);
 550         lock_buffer(bh);
 551         memset(bh->b_data, 0, bh->b_size);
 552         set_buffer_uptodate(bh);
 553         clear_buffer_dirty(bh);
 554         unlock_buffer(bh);
 555
 556         gfs2_ail1_empty(sdp, 0);
 557         tail = current_tail(sdp);
 558
 559         lh = (struct gfs2_log_header *)bh->b_data;
 560         memset(lh, 0, sizeof(struct gfs2_log_header));
 561         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 562         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 563         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 564         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 565         lh->lh_flags = cpu_to_be32(flags);
 566         lh->lh_tail = cpu_to_be32(tail);
 567         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 568         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 569         lh->lh_hash = cpu_to_be32(hash);
 570
 571         set_buffer_dirty(bh);
 572         if (sync_dirty_buffer(bh))
 573                 gfs2_io_error_bh(sdp, bh);
 574         brelse(bh);
 575
 576         if (sdp->sd_log_tail != tail)
 577                 log_pull_tail(sdp, tail);
 578         else
 579                 gfs2_assert_withdraw(sdp, !pull);
 580
 581         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 582         log_incr_head(sdp);
 583 }
 584
 585 static void log_flush_commit(struct gfs2_sbd *sdp)
 586 {
 587         struct list_head *head = &sdp->sd_log_flush_list;
 588         struct gfs2_log_buf *lb;
 589         struct buffer_head *bh;
 590         int flushcount = 0;
 591
 592         while (!list_empty(head)) {
 593                 lb = list_entry(head->next, struct gfs2_log_buf, lb_list);
 594                 list_del(&lb->lb_list);
 595                 bh = lb->lb_bh;
 596
 597                 wait_on_buffer(bh);
 598                 if (!buffer_uptodate(bh))
 599                         gfs2_io_error_bh(sdp, bh);
 600                 if (lb->lb_real) {
 601                         while (atomic_read(&bh->b_count) != 1)  /* Grrrr... */
 602                                 schedule();
 603                         free_buffer_head(bh);
 604                 } else
 605                         brelse(bh);
 606                 kfree(lb);
 607                 flushcount++;
 608         }
 609
 610         /* If nothing was journaled, the header is unplanned and unwanted. */
 611         if (flushcount) {
 612                 log_write_header(sdp, 0, 0);
 613         } else {
 614                 unsigned int tail;
 615                 tail = current_tail(sdp);
 616
 617                 gfs2_ail1_empty(sdp, 0);
 618                 if (sdp->sd_log_tail != tail)
 619                         log_pull_tail(sdp, tail);
 620         }
 621 }
 622
 623 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 624 {
 625         struct gfs2_bufdata *bd;
 626         struct buffer_head *bh;
 627         LIST_HEAD(written);
 628
 629         gfs2_log_lock(sdp);
 630         while (!list_empty(&sdp->sd_log_le_ordered)) {
 631                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 632                 list_move(&bd->bd_le.le_list, &written);
 633                 bh = bd->bd_bh;
 634                 if (!buffer_dirty(bh))
 635                         continue;
 636                 get_bh(bh);
 637                 gfs2_log_unlock(sdp);
 638                 lock_buffer(bh);
 639                 if (test_clear_buffer_dirty(bh)) {
 640                         bh->b_end_io = end_buffer_write_sync;
 641                         submit_bh(WRITE, bh);
 642                 } else {
 643                         unlock_buffer(bh);
 644                         brelse(bh);
 645                 }
 646                 gfs2_log_lock(sdp);
 647         }
 648         list_splice(&written, &sdp->sd_log_le_ordered);
 649         gfs2_log_unlock(sdp);
 650 }
 651
 652 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 653 {
 654         struct gfs2_bufdata *bd;
 655         struct buffer_head *bh;
 656
 657         gfs2_log_lock(sdp);
 658         while (!list_empty(&sdp->sd_log_le_ordered)) {
 659                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 660                 bh = bd->bd_bh;
 661                 if (buffer_locked(bh)) {
 662                         get_bh(bh);
 663                         gfs2_log_unlock(sdp);
 664                         wait_on_buffer(bh);
 665                         brelse(bh);
 666                         gfs2_log_lock(sdp);
 667                         continue;
 668                 }
 669                 list_del_init(&bd->bd_le.le_list);
 670         }
 671         gfs2_log_unlock(sdp);
 672 }
 673
 674 /**
 675  * gfs2_log_flush - flush incore transaction(s)
 676  * @sdp: the filesystem
 677  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 678  *
 679  */
 680
 681 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 682 {
 683         struct gfs2_ail *ai;
 684
 685         down_write(&sdp->sd_log_flush_lock);
 686
 687         if (gl) {
 688                 gfs2_log_lock(sdp);
 689                 if (list_empty(&gl->gl_le.le_list)) {
 690                         gfs2_log_unlock(sdp);
 691                         up_write(&sdp->sd_log_flush_lock);
 692                         return;
 693                 }
 694                 gfs2_log_unlock(sdp);
 695         }
 696
 697         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 698         INIT_LIST_HEAD(&ai->ai_ail1_list);
 699         INIT_LIST_HEAD(&ai->ai_ail2_list);
 700
 701         gfs2_assert_withdraw(sdp,
 702                              sdp->sd_log_num_buf + sdp->sd_log_num_databuf ==
 703                              sdp->sd_log_commited_buf +
 704                              sdp->sd_log_commited_databuf);
 705         gfs2_assert_withdraw(sdp,
 706                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 707
 708         sdp->sd_log_flush_head = sdp->sd_log_head;
 709         sdp->sd_log_flush_wrapped = 0;
 710         ai->ai_first = sdp->sd_log_flush_head;
 711
 712         gfs2_ordered_write(sdp);
 713         lops_before_commit(sdp);
 714         gfs2_ordered_wait(sdp);
 715
 716         if (!list_empty(&sdp->sd_log_flush_list))
 717                 log_flush_commit(sdp);
 718         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 719                 gfs2_log_lock(sdp);
 720                 sdp->sd_log_blks_free--; /* Adjust for unreserved buffer */
 721                 gfs2_log_unlock(sdp);
 722                 log_write_header(sdp, 0, PULL);
 723         }
 724         lops_after_commit(sdp, ai);
 725
 726         gfs2_log_lock(sdp);
 727         sdp->sd_log_head = sdp->sd_log_flush_head;
 728         sdp->sd_log_blks_reserved = 0;
 729         sdp->sd_log_commited_buf = 0;
 730         sdp->sd_log_commited_databuf = 0;
 731         sdp->sd_log_commited_revoke = 0;
 732
 733         if (!list_empty(&ai->ai_ail1_list)) {
 734                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 735                 ai = NULL;
 736         }
 737         gfs2_log_unlock(sdp);
 738
 739         sdp->sd_vfs->s_dirt = 0;
 740         up_write(&sdp->sd_log_flush_lock);
 741
 742         kfree(ai);
 743 }
 744
 745 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 746 {
 747         unsigned int reserved;
 748         unsigned int old;
 749
 750         gfs2_log_lock(sdp);
 751
 752         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 753         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 754                 tr->tr_num_databuf_rm;
 755         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 756                              (((int)sdp->sd_log_commited_databuf) >= 0));
 757         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 758         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 759         reserved = calc_reserved(sdp);
 760         old = sdp->sd_log_blks_free;
 761         sdp->sd_log_blks_free += tr->tr_reserved -
 762                                  (reserved - sdp->sd_log_blks_reserved);
 763
 764         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free >= old);
 765         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_free <=
 766                              sdp->sd_jdesc->jd_blocks);
 767
 768         sdp->sd_log_blks_reserved = reserved;
 769
 770         gfs2_log_unlock(sdp);
 771 }
 772
 773 /**
 774  * gfs2_log_commit - Commit a transaction to the log
 775  * @sdp: the filesystem
 776  * @tr: the transaction
 777  *
 778  * Returns: errno
 779  */
 780
 781 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 782 {
 783         log_refund(sdp, tr);
 784         lops_incore_commit(sdp, tr);
 785
 786         sdp->sd_vfs->s_dirt = 1;
 787         up_read(&sdp->sd_log_flush_lock);
 788
 789         gfs2_log_lock(sdp);
 790         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 791                 wake_up_process(sdp->sd_logd_process);
 792         gfs2_log_unlock(sdp);
 793 }
 794
 795 /**
 796  * gfs2_log_shutdown - write a shutdown header into a journal
 797  * @sdp: the filesystem
 798  *
 799  */
 800
 801 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 802 {
 803         down_write(&sdp->sd_log_flush_lock);
 804
 805         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 806         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_gl);
 807         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 808         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 809         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 810         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 811         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 812
 813         sdp->sd_log_flush_head = sdp->sd_log_head;
 814         sdp->sd_log_flush_wrapped = 0;
 815
 816         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 817                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 818
 819         gfs2_assert_warn(sdp, sdp->sd_log_blks_free == sdp->sd_jdesc->jd_blocks);
 820         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 821         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 822
 823         sdp->sd_log_head = sdp->sd_log_flush_head;
 824         sdp->sd_log_tail = sdp->sd_log_head;
 825
 826         up_write(&sdp->sd_log_flush_lock);
 827 }
 828
 829
 830 /**
 831  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 832  * @sdp: the filesystem
 833  *
 834  */
 835
 836 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 837 {
 838         gfs2_log_flush(sdp, NULL);
 839         for (;;) {
 840                 gfs2_ail1_start(sdp, DIO_ALL);
 841                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 842                         break;
 843                 msleep(10);
 844         }
 845 }
 846