fs/gfs2/log.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/lm_interface.h>
  18 #include <linux/delay.h>
  19 #include <linux/kthread.h>
  20 #include <linux/freezer.h>
  21
  22 #include "gfs2.h"
  23 #include "incore.h"
  24 #include "bmap.h"
  25 #include "glock.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "util.h"
  30 #include "dir.h"
  31
  32 #define PULL 1
  33
  34 /**
  35  * gfs2_struct2blk - compute stuff
  36  * @sdp: the filesystem
  37  * @nstruct: the number of structures
  38  * @ssize: the size of the structures
  39  *
  40  * Compute the number of log descriptor blocks needed to hold a certain number
  41  * of structures of a certain size.
  42  *
  43  * Returns: the number of blocks needed (minimum is always 1)
  44  */
  45
  46 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  47                              unsigned int ssize)
  48 {
  49         unsigned int blks;
  50         unsigned int first, second;
  51
  52         blks = 1;
  53         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  54
  55         if (nstruct > first) {
  56                 second = (sdp->sd_sb.sb_bsize -
  57                           sizeof(struct gfs2_meta_header)) / ssize;
  58                 blks += DIV_ROUND_UP(nstruct - first, second);
  59         }
  60
  61         return blks;
  62 }
  63
  64 /**
  65  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  66  * @mapping: The associated mapping (maybe NULL)
  67  * @bd: The gfs2_bufdata to remove
  68  *
  69  * The log lock _must_ be held when calling this function
  70  *
  71  */
  72
  73 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  74 {
  75         bd->bd_ail = NULL;
  76         list_del_init(&bd->bd_ail_st_list);
  77         list_del_init(&bd->bd_ail_gl_list);
  78         atomic_dec(&bd->bd_gl->gl_ail_count);
  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         do {
  96                 retry = 0;
  97
  98                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
  99                                                  bd_ail_st_list) {
 100                         bh = bd->bd_bh;
 101
 102                         gfs2_assert(sdp, bd->bd_ail == ai);
 103
 104                         if (!buffer_busy(bh)) {
 105                                 if (!buffer_uptodate(bh))
 106                                         gfs2_io_error_bh(sdp, bh);
 107                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 108                                 continue;
 109                         }
 110
 111                         if (!buffer_dirty(bh))
 112                                 continue;
 113
 114                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 115
 116                         get_bh(bh);
 117                         gfs2_log_unlock(sdp);
 118                         lock_buffer(bh);
 119                         if (test_clear_buffer_dirty(bh)) {
 120                                 bh->b_end_io = end_buffer_write_sync;
 121                                 submit_bh(WRITE, bh);
 122                         } else {
 123                                 unlock_buffer(bh);
 124                                 brelse(bh);
 125                         }
 126                         gfs2_log_lock(sdp);
 127
 128                         retry = 1;
 129                         break;
 130                 }
 131         } while (retry);
 132 }
 133
 134 /**
 135  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 136  * @sdp: the filesystem
 137  * @ai: the AIL entry
 138  *
 139  */
 140
 141 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 142 {
 143         struct gfs2_bufdata *bd, *s;
 144         struct buffer_head *bh;
 145
 146         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 147                                          bd_ail_st_list) {
 148                 bh = bd->bd_bh;
 149
 150                 gfs2_assert(sdp, bd->bd_ail == ai);
 151
 152                 if (buffer_busy(bh)) {
 153                         if (flags & DIO_ALL)
 154                                 continue;
 155                         else
 156                                 break;
 157                 }
 158
 159                 if (!buffer_uptodate(bh))
 160                         gfs2_io_error_bh(sdp, bh);
 161
 162                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 163         }
 164
 165         return list_empty(&ai->ai_ail1_list);
 166 }
 167
 168 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
 169 {
 170         struct list_head *head;
 171         u64 sync_gen;
 172         struct list_head *first;
 173         struct gfs2_ail *first_ai, *ai, *tmp;
 174         int done = 0;
 175
 176         gfs2_log_lock(sdp);
 177         head = &sdp->sd_ail1_list;
 178         if (list_empty(head)) {
 179                 gfs2_log_unlock(sdp);
 180                 return;
 181         }
 182         sync_gen = sdp->sd_ail_sync_gen++;
 183
 184         first = head->prev;
 185         first_ai = list_entry(first, struct gfs2_ail, ai_list);
 186         first_ai->ai_sync_gen = sync_gen;
 187         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
 188
 189         if (flags & DIO_ALL)
 190                 first = NULL;
 191
 192         while(!done) {
 193                 if (first && (head->prev != first ||
 194                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
 195                         break;
 196
 197                 done = 1;
 198                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
 199                         if (ai->ai_sync_gen >= sync_gen)
 200                                 continue;
 201                         ai->ai_sync_gen = sync_gen;
 202                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
 203                         done = 0;
 204                         break;
 205                 }
 206         }
 207
 208         gfs2_log_unlock(sdp);
 209 }
 210
 211 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 212 {
 213         struct gfs2_ail *ai, *s;
 214         int ret;
 215
 216         gfs2_log_lock(sdp);
 217
 218         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 219                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 220                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 221                 else if (!(flags & DIO_ALL))
 222                         break;
 223         }
 224
 225         ret = list_empty(&sdp->sd_ail1_list);
 226
 227         gfs2_log_unlock(sdp);
 228
 229         return ret;
 230 }
 231
 232
 233 /**
 234  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 235  * @sdp: the filesystem
 236  * @ai: the AIL entry
 237  *
 238  */
 239
 240 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 241 {
 242         struct list_head *head = &ai->ai_ail2_list;
 243         struct gfs2_bufdata *bd;
 244
 245         while (!list_empty(head)) {
 246                 bd = list_entry(head->prev, struct gfs2_bufdata,
 247                                 bd_ail_st_list);
 248                 gfs2_assert(sdp, bd->bd_ail == ai);
 249                 gfs2_remove_from_ail(bd);
 250         }
 251 }
 252
 253 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 254 {
 255         struct gfs2_ail *ai, *safe;
 256         unsigned int old_tail = sdp->sd_log_tail;
 257         int wrap = (new_tail < old_tail);
 258         int a, b, rm;
 259
 260         gfs2_log_lock(sdp);
 261
 262         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 263                 a = (old_tail <= ai->ai_first);
 264                 b = (ai->ai_first < new_tail);
 265                 rm = (wrap) ? (a || b) : (a && b);
 266                 if (!rm)
 267                         continue;
 268
 269                 gfs2_ail2_empty_one(sdp, ai);
 270                 list_del(&ai->ai_list);
 271                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 272                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 273                 kfree(ai);
 274         }
 275
 276         gfs2_log_unlock(sdp);
 277 }
 278
 279 /**
 280  * gfs2_log_reserve - Make a log reservation
 281  * @sdp: The GFS2 superblock
 282  * @blks: The number of blocks to reserve
 283  *
 284  * Note that we never give out the last few blocks of the journal. Thats
 285  * due to the fact that there is a small number of header blocks
 286  * associated with each log flush. The exact number can't be known until
 287  * flush time, so we ensure that we have just enough free blocks at all
 288  * times to avoid running out during a log flush.
 289  *
 290  * Returns: errno
 291  */
 292
 293 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 294 {
 295         unsigned int try = 0;
 296         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
 297
 298         if (gfs2_assert_warn(sdp, blks) ||
 299             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 300                 return -EINVAL;
 301
 302         mutex_lock(&sdp->sd_log_reserve_mutex);
 303         gfs2_log_lock(sdp);
 304         while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) {
 305                 gfs2_log_unlock(sdp);
 306                 gfs2_ail1_empty(sdp, 0);
 307                 gfs2_log_flush(sdp, NULL);
 308
 309                 if (try++)
 310                         gfs2_ail1_start(sdp, 0);
 311                 gfs2_log_lock(sdp);
 312         }
 313         atomic_sub(blks, &sdp->sd_log_blks_free);
 314         gfs2_log_unlock(sdp);
 315         mutex_unlock(&sdp->sd_log_reserve_mutex);
 316
 317         down_read(&sdp->sd_log_flush_lock);
 318
 319         return 0;
 320 }
 321
 322 /**
 323  * gfs2_log_release - Release a given number of log blocks
 324  * @sdp: The GFS2 superblock
 325  * @blks: The number of blocks
 326  *
 327  */
 328
 329 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 330 {
 331
 332         gfs2_log_lock(sdp);
 333         atomic_add(blks, &sdp->sd_log_blks_free);
 334         gfs2_assert_withdraw(sdp,
 335                              atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 336         gfs2_log_unlock(sdp);
 337         up_read(&sdp->sd_log_flush_lock);
 338 }
 339
 340 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 341 {
 342         struct gfs2_journal_extent *je;
 343
 344         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
 345                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks)
 346                         return je->dblock + lbn - je->lblock;
 347         }
 348
 349         return -1;
 350 }
 351
 352 /**
 353  * log_distance - Compute distance between two journal blocks
 354  * @sdp: The GFS2 superblock
 355  * @newer: The most recent journal block of the pair
 356  * @older: The older journal block of the pair
 357  *
 358  *   Compute the distance (in the journal direction) between two
 359  *   blocks in the journal
 360  *
 361  * Returns: the distance in blocks
 362  */
 363
 364 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 365                                         unsigned int older)
 366 {
 367         int dist;
 368
 369         dist = newer - older;
 370         if (dist < 0)
 371                 dist += sdp->sd_jdesc->jd_blocks;
 372
 373         return dist;
 374 }
 375
 376 /**
 377  * calc_reserved - Calculate the number of blocks to reserve when
 378  *                 refunding a transaction's unused buffers.
 379  * @sdp: The GFS2 superblock
 380  *
 381  * This is complex.  We need to reserve room for all our currently used
 382  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 383  * all our journaled data buffers for journaled files (e.g. files in the
 384  * meta_fs like rindex, or files for which chattr +j was done.)
 385  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 386  * will count it as free space (sd_log_blks_free) and corruption will follow.
 387  *
 388  * We can have metadata bufs and jdata bufs in the same journal.  So each
 389  * type gets its own log header, for which we need to reserve a block.
 390  * In fact, each type has the potential for needing more than one header
 391  * in cases where we have more buffers than will fit on a journal page.
 392  * Metadata journal entries take up half the space of journaled buffer entries.
 393  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 394  * databuf_limit (251) before they cause a wrap around.
 395  *
 396  * Also, we need to reserve blocks for revoke journal entries and one for an
 397  * overall header for the lot.
 398  *
 399  * Returns: the number of blocks reserved
 400  */
 401 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 402 {
 403         unsigned int reserved = 0;
 404         unsigned int mbuf_limit, metabufhdrs_needed;
 405         unsigned int dbuf_limit, databufhdrs_needed;
 406         unsigned int revokes = 0;
 407
 408         mbuf_limit = buf_limit(sdp);
 409         metabufhdrs_needed = (sdp->sd_log_commited_buf +
 410                               (mbuf_limit - 1)) / mbuf_limit;
 411         dbuf_limit = databuf_limit(sdp);
 412         databufhdrs_needed = (sdp->sd_log_commited_databuf +
 413                               (dbuf_limit - 1)) / dbuf_limit;
 414
 415         if (sdp->sd_log_commited_revoke)
 416                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 417                                           sizeof(u64));
 418
 419         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
 420                 sdp->sd_log_commited_databuf + databufhdrs_needed +
 421                 revokes;
 422         /* One for the overall header */
 423         if (reserved)
 424                 reserved++;
 425         return reserved;
 426 }
 427
 428 static unsigned int current_tail(struct gfs2_sbd *sdp)
 429 {
 430         struct gfs2_ail *ai;
 431         unsigned int tail;
 432
 433         gfs2_log_lock(sdp);
 434
 435         if (list_empty(&sdp->sd_ail1_list)) {
 436                 tail = sdp->sd_log_head;
 437         } else {
 438                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 439                 tail = ai->ai_first;
 440         }
 441
 442         gfs2_log_unlock(sdp);
 443
 444         return tail;
 445 }
 446
 447 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 448 {
 449         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 450                 BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head);
 451
 452         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 453                 sdp->sd_log_flush_head = 0;
 454                 sdp->sd_log_flush_wrapped = 1;
 455         }
 456 }
 457
 458 /**
 459  * gfs2_log_write_endio - End of I/O for a log buffer
 460  * @bh: The buffer head
 461  * @uptodate: I/O Status
 462  *
 463  */
 464
 465 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 466 {
 467         struct gfs2_sbd *sdp = bh->b_private;
 468         bh->b_private = NULL;
 469
 470         end_buffer_write_sync(bh, uptodate);
 471         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 472                 wake_up(&sdp->sd_log_flush_wait);
 473 }
 474
 475 /**
 476  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 477  * @sdp: The GFS2 superblock
 478  *
 479  * Returns: the buffer_head
 480  */
 481
 482 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 483 {
 484         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 485         struct buffer_head *bh;
 486
 487         bh = sb_getblk(sdp->sd_vfs, blkno);
 488         lock_buffer(bh);
 489         memset(bh->b_data, 0, bh->b_size);
 490         set_buffer_uptodate(bh);
 491         clear_buffer_dirty(bh);
 492         gfs2_log_incr_head(sdp);
 493         atomic_inc(&sdp->sd_log_in_flight);
 494         bh->b_private = sdp;
 495         bh->b_end_io = gfs2_log_write_endio;
 496
 497         return bh;
 498 }
 499
 500 /**
 501  * gfs2_fake_write_endio -
 502  * @bh: The buffer head
 503  * @uptodate: The I/O Status
 504  *
 505  */
 506
 507 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 508 {
 509         struct buffer_head *real_bh = bh->b_private;
 510         struct gfs2_bufdata *bd = real_bh->b_private;
 511         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 512
 513         end_buffer_write_sync(bh, uptodate);
 514         free_buffer_head(bh);
 515         unlock_buffer(real_bh);
 516         brelse(real_bh);
 517         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 518                 wake_up(&sdp->sd_log_flush_wait);
 519 }
 520
 521 /**
 522  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 523  * @sdp: the filesystem
 524  * @data: the data the buffer_head should point to
 525  *
 526  * Returns: the log buffer descriptor
 527  */
 528
 529 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 530                                       struct buffer_head *real)
 531 {
 532         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 533         struct buffer_head *bh;
 534
 535         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 536         atomic_set(&bh->b_count, 1);
 537         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 538         set_bh_page(bh, real->b_page, bh_offset(real));
 539         bh->b_blocknr = blkno;
 540         bh->b_size = sdp->sd_sb.sb_bsize;
 541         bh->b_bdev = sdp->sd_vfs->s_bdev;
 542         bh->b_private = real;
 543         bh->b_end_io = gfs2_fake_write_endio;
 544
 545         gfs2_log_incr_head(sdp);
 546         atomic_inc(&sdp->sd_log_in_flight);
 547
 548         return bh;
 549 }
 550
 551 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 552 {
 553         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 554
 555         ail2_empty(sdp, new_tail);
 556
 557         gfs2_log_lock(sdp);
 558         atomic_add(dist, &sdp->sd_log_blks_free);
 559         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 560         gfs2_log_unlock(sdp);
 561
 562         sdp->sd_log_tail = new_tail;
 563 }
 564
 565 /**
 566  * log_write_header - Get and initialize a journal header buffer
 567  * @sdp: The GFS2 superblock
 568  *
 569  * Returns: the initialized log buffer descriptor
 570  */
 571
 572 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 573 {
 574         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 575         struct buffer_head *bh;
 576         struct gfs2_log_header *lh;
 577         unsigned int tail;
 578         u32 hash;
 579
 580         bh = sb_getblk(sdp->sd_vfs, blkno);
 581         lock_buffer(bh);
 582         memset(bh->b_data, 0, bh->b_size);
 583         set_buffer_uptodate(bh);
 584         clear_buffer_dirty(bh);
 585         unlock_buffer(bh);
 586
 587         gfs2_ail1_empty(sdp, 0);
 588         tail = current_tail(sdp);
 589
 590         lh = (struct gfs2_log_header *)bh->b_data;
 591         memset(lh, 0, sizeof(struct gfs2_log_header));
 592         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 593         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 594         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 595         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 596         lh->lh_flags = cpu_to_be32(flags);
 597         lh->lh_tail = cpu_to_be32(tail);
 598         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 599         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 600         lh->lh_hash = cpu_to_be32(hash);
 601
 602         set_buffer_dirty(bh);
 603         if (sync_dirty_buffer(bh))
 604                 gfs2_io_error_bh(sdp, bh);
 605         brelse(bh);
 606
 607         if (sdp->sd_log_tail != tail)
 608                 log_pull_tail(sdp, tail);
 609         else
 610                 gfs2_assert_withdraw(sdp, !pull);
 611
 612         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 613         gfs2_log_incr_head(sdp);
 614 }
 615
 616 static void log_flush_commit(struct gfs2_sbd *sdp)
 617 {
 618         DEFINE_WAIT(wait);
 619
 620         if (atomic_read(&sdp->sd_log_in_flight)) {
 621                 do {
 622                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 623                                         TASK_UNINTERRUPTIBLE);
 624                         if (atomic_read(&sdp->sd_log_in_flight))
 625                                 io_schedule();
 626                 } while(atomic_read(&sdp->sd_log_in_flight));
 627                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 628         }
 629
 630         log_write_header(sdp, 0, 0);
 631 }
 632
 633 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 634 {
 635         struct gfs2_bufdata *bd;
 636         struct buffer_head *bh;
 637         LIST_HEAD(written);
 638
 639         gfs2_log_lock(sdp);
 640         while (!list_empty(&sdp->sd_log_le_ordered)) {
 641                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 642                 list_move(&bd->bd_le.le_list, &written);
 643                 bh = bd->bd_bh;
 644                 if (!buffer_dirty(bh))
 645                         continue;
 646                 get_bh(bh);
 647                 gfs2_log_unlock(sdp);
 648                 lock_buffer(bh);
 649                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
 650                         bh->b_end_io = end_buffer_write_sync;
 651                         submit_bh(WRITE, bh);
 652                 } else {
 653                         unlock_buffer(bh);
 654                         brelse(bh);
 655                 }
 656                 gfs2_log_lock(sdp);
 657         }
 658         list_splice(&written, &sdp->sd_log_le_ordered);
 659         gfs2_log_unlock(sdp);
 660 }
 661
 662 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 663 {
 664         struct gfs2_bufdata *bd;
 665         struct buffer_head *bh;
 666
 667         gfs2_log_lock(sdp);
 668         while (!list_empty(&sdp->sd_log_le_ordered)) {
 669                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 670                 bh = bd->bd_bh;
 671                 if (buffer_locked(bh)) {
 672                         get_bh(bh);
 673                         gfs2_log_unlock(sdp);
 674                         wait_on_buffer(bh);
 675                         brelse(bh);
 676                         gfs2_log_lock(sdp);
 677                         continue;
 678                 }
 679                 list_del_init(&bd->bd_le.le_list);
 680         }
 681         gfs2_log_unlock(sdp);
 682 }
 683
 684 /**
 685  * gfs2_log_flush - flush incore transaction(s)
 686  * @sdp: the filesystem
 687  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 688  *
 689  */
 690
 691 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 692 {
 693         struct gfs2_ail *ai;
 694
 695         down_write(&sdp->sd_log_flush_lock);
 696
 697         /* Log might have been flushed while we waited for the flush lock */
 698         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 699                 up_write(&sdp->sd_log_flush_lock);
 700                 return;
 701         }
 702
 703         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 704         INIT_LIST_HEAD(&ai->ai_ail1_list);
 705         INIT_LIST_HEAD(&ai->ai_ail2_list);
 706
 707         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
 708                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
 709                        sdp->sd_log_commited_buf);
 710                 gfs2_assert_withdraw(sdp, 0);
 711         }
 712         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
 713                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
 714                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
 715                 gfs2_assert_withdraw(sdp, 0);
 716         }
 717         gfs2_assert_withdraw(sdp,
 718                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 719
 720         sdp->sd_log_flush_head = sdp->sd_log_head;
 721         sdp->sd_log_flush_wrapped = 0;
 722         ai->ai_first = sdp->sd_log_flush_head;
 723
 724         gfs2_ordered_write(sdp);
 725         lops_before_commit(sdp);
 726         gfs2_ordered_wait(sdp);
 727
 728         if (sdp->sd_log_head != sdp->sd_log_flush_head)
 729                 log_flush_commit(sdp);
 730         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 731                 gfs2_log_lock(sdp);
 732                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 733                 gfs2_log_unlock(sdp);
 734                 log_write_header(sdp, 0, PULL);
 735         }
 736         lops_after_commit(sdp, ai);
 737
 738         gfs2_log_lock(sdp);
 739         sdp->sd_log_head = sdp->sd_log_flush_head;
 740         sdp->sd_log_blks_reserved = 0;
 741         sdp->sd_log_commited_buf = 0;
 742         sdp->sd_log_commited_databuf = 0;
 743         sdp->sd_log_commited_revoke = 0;
 744
 745         if (!list_empty(&ai->ai_ail1_list)) {
 746                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 747                 ai = NULL;
 748         }
 749         gfs2_log_unlock(sdp);
 750
 751         sdp->sd_vfs->s_dirt = 0;
 752         up_write(&sdp->sd_log_flush_lock);
 753
 754         kfree(ai);
 755 }
 756
 757 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 758 {
 759         unsigned int reserved;
 760         unsigned int unused;
 761
 762         gfs2_log_lock(sdp);
 763
 764         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 765         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 766                 tr->tr_num_databuf_rm;
 767         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 768                              (((int)sdp->sd_log_commited_databuf) >= 0));
 769         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 770         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 771         reserved = calc_reserved(sdp);
 772         unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
 773         gfs2_assert_withdraw(sdp, unused >= 0);
 774         atomic_add(unused, &sdp->sd_log_blks_free);
 775         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 776                              sdp->sd_jdesc->jd_blocks);
 777         sdp->sd_log_blks_reserved = reserved;
 778
 779         gfs2_log_unlock(sdp);
 780 }
 781
 782 /**
 783  * gfs2_log_commit - Commit a transaction to the log
 784  * @sdp: the filesystem
 785  * @tr: the transaction
 786  *
 787  * Returns: errno
 788  */
 789
 790 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 791 {
 792         log_refund(sdp, tr);
 793         lops_incore_commit(sdp, tr);
 794
 795         sdp->sd_vfs->s_dirt = 1;
 796         up_read(&sdp->sd_log_flush_lock);
 797
 798         gfs2_log_lock(sdp);
 799         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 800                 wake_up_process(sdp->sd_logd_process);
 801         gfs2_log_unlock(sdp);
 802 }
 803
 804 /**
 805  * gfs2_log_shutdown - write a shutdown header into a journal
 806  * @sdp: the filesystem
 807  *
 808  */
 809
 810 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 811 {
 812         down_write(&sdp->sd_log_flush_lock);
 813
 814         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 815         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 816         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 817         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 818         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 819         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 820
 821         sdp->sd_log_flush_head = sdp->sd_log_head;
 822         sdp->sd_log_flush_wrapped = 0;
 823
 824         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 825                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 826
 827         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 828         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 829         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 830
 831         sdp->sd_log_head = sdp->sd_log_flush_head;
 832         sdp->sd_log_tail = sdp->sd_log_head;
 833
 834         up_write(&sdp->sd_log_flush_lock);
 835 }
 836
 837
 838 /**
 839  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 840  * @sdp: the filesystem
 841  *
 842  */
 843
 844 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 845 {
 846         gfs2_log_flush(sdp, NULL);
 847         for (;;) {
 848                 gfs2_ail1_start(sdp, DIO_ALL);
 849                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 850                         break;
 851                 msleep(10);
 852         }
 853 }
 854
 855
 856 /**
 857  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 858  * @sdp: Pointer to GFS2 superblock
 859  *
 860  * Also, periodically check to make sure that we're using the most recent
 861  * journal index.
 862  */
 863
 864 int gfs2_logd(void *data)
 865 {
 866         struct gfs2_sbd *sdp = data;
 867         unsigned long t;
 868         int need_flush;
 869
 870         while (!kthread_should_stop()) {
 871                 /* Advance the log tail */
 872
 873                 t = sdp->sd_log_flush_time +
 874                     gfs2_tune_get(sdp, gt_log_flush_secs) * HZ;
 875
 876                 gfs2_ail1_empty(sdp, DIO_ALL);
 877                 gfs2_log_lock(sdp);
 878                 need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks);
 879                 gfs2_log_unlock(sdp);
 880                 if (need_flush || time_after_eq(jiffies, t)) {
 881                         gfs2_log_flush(sdp, NULL);
 882                         sdp->sd_log_flush_time = jiffies;
 883                 }
 884
 885                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
 886                 if (freezing(current))
 887                         refrigerator();
 888                 schedule_timeout_interruptible(t);
 889         }
 890
 891         return 0;
 892 }
 893