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/delay.h>
  18 #include <linux/kthread.h>
  19 #include <linux/freezer.h>
  20 #include <linux/bio.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 __releases(&sdp->sd_log_lock)
  91 __acquires(&sdp->sd_log_lock)
  92 {
  93         struct gfs2_bufdata *bd, *s;
  94         struct buffer_head *bh;
  95         int retry;
  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_io_error_bh(sdp, bh);
 109                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 110                                 continue;
 111                         }
 112
 113                         if (!buffer_dirty(bh))
 114                                 continue;
 115
 116                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 117
 118                         get_bh(bh);
 119                         gfs2_log_unlock(sdp);
 120                         lock_buffer(bh);
 121                         if (test_clear_buffer_dirty(bh)) {
 122                                 bh->b_end_io = end_buffer_write_sync;
 123                                 submit_bh(WRITE, bh);
 124                         } else {
 125                                 unlock_buffer(bh);
 126                                 brelse(bh);
 127                         }
 128                         gfs2_log_lock(sdp);
 129
 130                         retry = 1;
 131                         break;
 132                 }
 133         } while (retry);
 134 }
 135
 136 /**
 137  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 138  * @sdp: the filesystem
 139  * @ai: the AIL entry
 140  *
 141  */
 142
 143 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 144 {
 145         struct gfs2_bufdata *bd, *s;
 146         struct buffer_head *bh;
 147
 148         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 149                                          bd_ail_st_list) {
 150                 bh = bd->bd_bh;
 151
 152                 gfs2_assert(sdp, bd->bd_ail == ai);
 153
 154                 if (buffer_busy(bh)) {
 155                         if (flags & DIO_ALL)
 156                                 continue;
 157                         else
 158                                 break;
 159                 }
 160
 161                 if (!buffer_uptodate(bh))
 162                         gfs2_io_error_bh(sdp, bh);
 163
 164                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 165         }
 166
 167         return list_empty(&ai->ai_ail1_list);
 168 }
 169
 170 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
 171 {
 172         struct list_head *head;
 173         u64 sync_gen;
 174         struct list_head *first;
 175         struct gfs2_ail *first_ai, *ai, *tmp;
 176         int done = 0;
 177
 178         gfs2_log_lock(sdp);
 179         head = &sdp->sd_ail1_list;
 180         if (list_empty(head)) {
 181                 gfs2_log_unlock(sdp);
 182                 return;
 183         }
 184         sync_gen = sdp->sd_ail_sync_gen++;
 185
 186         first = head->prev;
 187         first_ai = list_entry(first, struct gfs2_ail, ai_list);
 188         first_ai->ai_sync_gen = sync_gen;
 189         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
 190
 191         if (flags & DIO_ALL)
 192                 first = NULL;
 193
 194         while(!done) {
 195                 if (first && (head->prev != first ||
 196                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
 197                         break;
 198
 199                 done = 1;
 200                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
 201                         if (ai->ai_sync_gen >= sync_gen)
 202                                 continue;
 203                         ai->ai_sync_gen = sync_gen;
 204                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
 205                         done = 0;
 206                         break;
 207                 }
 208         }
 209
 210         gfs2_log_unlock(sdp);
 211 }
 212
 213 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 214 {
 215         struct gfs2_ail *ai, *s;
 216         int ret;
 217
 218         gfs2_log_lock(sdp);
 219
 220         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 221                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 222                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 223                 else if (!(flags & DIO_ALL))
 224                         break;
 225         }
 226
 227         ret = list_empty(&sdp->sd_ail1_list);
 228
 229         gfs2_log_unlock(sdp);
 230
 231         return ret;
 232 }
 233
 234
 235 /**
 236  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 237  * @sdp: the filesystem
 238  * @ai: the AIL entry
 239  *
 240  */
 241
 242 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 243 {
 244         struct list_head *head = &ai->ai_ail2_list;
 245         struct gfs2_bufdata *bd;
 246
 247         while (!list_empty(head)) {
 248                 bd = list_entry(head->prev, struct gfs2_bufdata,
 249                                 bd_ail_st_list);
 250                 gfs2_assert(sdp, bd->bd_ail == ai);
 251                 gfs2_remove_from_ail(bd);
 252         }
 253 }
 254
 255 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 256 {
 257         struct gfs2_ail *ai, *safe;
 258         unsigned int old_tail = sdp->sd_log_tail;
 259         int wrap = (new_tail < old_tail);
 260         int a, b, rm;
 261
 262         gfs2_log_lock(sdp);
 263
 264         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 265                 a = (old_tail <= ai->ai_first);
 266                 b = (ai->ai_first < new_tail);
 267                 rm = (wrap) ? (a || b) : (a && b);
 268                 if (!rm)
 269                         continue;
 270
 271                 gfs2_ail2_empty_one(sdp, ai);
 272                 list_del(&ai->ai_list);
 273                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 274                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 275                 kfree(ai);
 276         }
 277
 278         gfs2_log_unlock(sdp);
 279 }
 280
 281 /**
 282  * gfs2_log_reserve - Make a log reservation
 283  * @sdp: The GFS2 superblock
 284  * @blks: The number of blocks to reserve
 285  *
 286  * Note that we never give out the last few blocks of the journal. Thats
 287  * due to the fact that there is a small number of header blocks
 288  * associated with each log flush. The exact number can't be known until
 289  * flush time, so we ensure that we have just enough free blocks at all
 290  * times to avoid running out during a log flush.
 291  *
 292  * Returns: errno
 293  */
 294
 295 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 296 {
 297         unsigned int try = 0;
 298         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
 299
 300         if (gfs2_assert_warn(sdp, blks) ||
 301             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 302                 return -EINVAL;
 303
 304         mutex_lock(&sdp->sd_log_reserve_mutex);
 305         gfs2_log_lock(sdp);
 306         while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) {
 307                 gfs2_log_unlock(sdp);
 308                 gfs2_ail1_empty(sdp, 0);
 309                 gfs2_log_flush(sdp, NULL);
 310
 311                 if (try++)
 312                         gfs2_ail1_start(sdp, 0);
 313                 gfs2_log_lock(sdp);
 314         }
 315         atomic_sub(blks, &sdp->sd_log_blks_free);
 316         gfs2_log_unlock(sdp);
 317         mutex_unlock(&sdp->sd_log_reserve_mutex);
 318
 319         down_read(&sdp->sd_log_flush_lock);
 320
 321         return 0;
 322 }
 323
 324 /**
 325  * gfs2_log_release - Release a given number of log blocks
 326  * @sdp: The GFS2 superblock
 327  * @blks: The number of blocks
 328  *
 329  */
 330
 331 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 332 {
 333
 334         gfs2_log_lock(sdp);
 335         atomic_add(blks, &sdp->sd_log_blks_free);
 336         gfs2_assert_withdraw(sdp,
 337                              atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 338         gfs2_log_unlock(sdp);
 339         up_read(&sdp->sd_log_flush_lock);
 340 }
 341
 342 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 343 {
 344         struct gfs2_journal_extent *je;
 345
 346         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
 347                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks)
 348                         return je->dblock + lbn - je->lblock;
 349         }
 350
 351         return -1;
 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 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 450 {
 451         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 452                 BUG_ON(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_write_endio - End of I/O for a log buffer
 462  * @bh: The buffer head
 463  * @uptodate: I/O Status
 464  *
 465  */
 466
 467 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 468 {
 469         struct gfs2_sbd *sdp = bh->b_private;
 470         bh->b_private = NULL;
 471
 472         end_buffer_write_sync(bh, uptodate);
 473         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 474                 wake_up(&sdp->sd_log_flush_wait);
 475 }
 476
 477 /**
 478  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 479  * @sdp: The GFS2 superblock
 480  *
 481  * Returns: the buffer_head
 482  */
 483
 484 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 485 {
 486         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 487         struct buffer_head *bh;
 488
 489         bh = sb_getblk(sdp->sd_vfs, blkno);
 490         lock_buffer(bh);
 491         memset(bh->b_data, 0, bh->b_size);
 492         set_buffer_uptodate(bh);
 493         clear_buffer_dirty(bh);
 494         gfs2_log_incr_head(sdp);
 495         atomic_inc(&sdp->sd_log_in_flight);
 496         bh->b_private = sdp;
 497         bh->b_end_io = gfs2_log_write_endio;
 498
 499         return bh;
 500 }
 501
 502 /**
 503  * gfs2_fake_write_endio -
 504  * @bh: The buffer head
 505  * @uptodate: The I/O Status
 506  *
 507  */
 508
 509 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 510 {
 511         struct buffer_head *real_bh = bh->b_private;
 512         struct gfs2_bufdata *bd = real_bh->b_private;
 513         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 514
 515         end_buffer_write_sync(bh, uptodate);
 516         free_buffer_head(bh);
 517         unlock_buffer(real_bh);
 518         brelse(real_bh);
 519         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 520                 wake_up(&sdp->sd_log_flush_wait);
 521 }
 522
 523 /**
 524  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 525  * @sdp: the filesystem
 526  * @data: the data the buffer_head should point to
 527  *
 528  * Returns: the log buffer descriptor
 529  */
 530
 531 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 532                                       struct buffer_head *real)
 533 {
 534         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 535         struct buffer_head *bh;
 536
 537         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 538         atomic_set(&bh->b_count, 1);
 539         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 540         set_bh_page(bh, real->b_page, bh_offset(real));
 541         bh->b_blocknr = blkno;
 542         bh->b_size = sdp->sd_sb.sb_bsize;
 543         bh->b_bdev = sdp->sd_vfs->s_bdev;
 544         bh->b_private = real;
 545         bh->b_end_io = gfs2_fake_write_endio;
 546
 547         gfs2_log_incr_head(sdp);
 548         atomic_inc(&sdp->sd_log_in_flight);
 549
 550         return bh;
 551 }
 552
 553 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 554 {
 555         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 556
 557         ail2_empty(sdp, new_tail);
 558
 559         gfs2_log_lock(sdp);
 560         atomic_add(dist, &sdp->sd_log_blks_free);
 561         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 562         gfs2_log_unlock(sdp);
 563
 564         sdp->sd_log_tail = new_tail;
 565 }
 566
 567 /**
 568  * log_write_header - Get and initialize a journal header buffer
 569  * @sdp: The GFS2 superblock
 570  *
 571  * Returns: the initialized log buffer descriptor
 572  */
 573
 574 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 575 {
 576         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 577         struct buffer_head *bh;
 578         struct gfs2_log_header *lh;
 579         unsigned int tail;
 580         u32 hash;
 581
 582         bh = sb_getblk(sdp->sd_vfs, blkno);
 583         lock_buffer(bh);
 584         memset(bh->b_data, 0, bh->b_size);
 585         set_buffer_uptodate(bh);
 586         clear_buffer_dirty(bh);
 587
 588         gfs2_ail1_empty(sdp, 0);
 589         tail = current_tail(sdp);
 590
 591         lh = (struct gfs2_log_header *)bh->b_data;
 592         memset(lh, 0, sizeof(struct gfs2_log_header));
 593         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 594         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 595         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 596         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 597         lh->lh_flags = cpu_to_be32(flags);
 598         lh->lh_tail = cpu_to_be32(tail);
 599         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 600         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 601         lh->lh_hash = cpu_to_be32(hash);
 602
 603         bh->b_end_io = end_buffer_write_sync;
 604         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
 605                 goto skip_barrier;
 606         get_bh(bh);
 607         submit_bh(WRITE_BARRIER | (1 << BIO_RW_META), bh);
 608         wait_on_buffer(bh);
 609         if (buffer_eopnotsupp(bh)) {
 610                 clear_buffer_eopnotsupp(bh);
 611                 set_buffer_uptodate(bh);
 612                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
 613                 lock_buffer(bh);
 614 skip_barrier:
 615                 get_bh(bh);
 616                 submit_bh(WRITE_SYNC | (1 << BIO_RW_META), bh);
 617                 wait_on_buffer(bh);
 618         }
 619         if (!buffer_uptodate(bh))
 620                 gfs2_io_error_bh(sdp, bh);
 621         brelse(bh);
 622
 623         if (sdp->sd_log_tail != tail)
 624                 log_pull_tail(sdp, tail);
 625         else
 626                 gfs2_assert_withdraw(sdp, !pull);
 627
 628         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 629         gfs2_log_incr_head(sdp);
 630 }
 631
 632 static void log_flush_commit(struct gfs2_sbd *sdp)
 633 {
 634         DEFINE_WAIT(wait);
 635
 636         if (atomic_read(&sdp->sd_log_in_flight)) {
 637                 do {
 638                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 639                                         TASK_UNINTERRUPTIBLE);
 640                         if (atomic_read(&sdp->sd_log_in_flight))
 641                                 io_schedule();
 642                 } while(atomic_read(&sdp->sd_log_in_flight));
 643                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 644         }
 645
 646         log_write_header(sdp, 0, 0);
 647 }
 648
 649 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 650 {
 651         struct gfs2_bufdata *bd;
 652         struct buffer_head *bh;
 653         LIST_HEAD(written);
 654
 655         gfs2_log_lock(sdp);
 656         while (!list_empty(&sdp->sd_log_le_ordered)) {
 657                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 658                 list_move(&bd->bd_le.le_list, &written);
 659                 bh = bd->bd_bh;
 660                 if (!buffer_dirty(bh))
 661                         continue;
 662                 get_bh(bh);
 663                 gfs2_log_unlock(sdp);
 664                 lock_buffer(bh);
 665                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
 666                         bh->b_end_io = end_buffer_write_sync;
 667                         submit_bh(WRITE, bh);
 668                 } else {
 669                         unlock_buffer(bh);
 670                         brelse(bh);
 671                 }
 672                 gfs2_log_lock(sdp);
 673         }
 674         list_splice(&written, &sdp->sd_log_le_ordered);
 675         gfs2_log_unlock(sdp);
 676 }
 677
 678 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 679 {
 680         struct gfs2_bufdata *bd;
 681         struct buffer_head *bh;
 682
 683         gfs2_log_lock(sdp);
 684         while (!list_empty(&sdp->sd_log_le_ordered)) {
 685                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 686                 bh = bd->bd_bh;
 687                 if (buffer_locked(bh)) {
 688                         get_bh(bh);
 689                         gfs2_log_unlock(sdp);
 690                         wait_on_buffer(bh);
 691                         brelse(bh);
 692                         gfs2_log_lock(sdp);
 693                         continue;
 694                 }
 695                 list_del_init(&bd->bd_le.le_list);
 696         }
 697         gfs2_log_unlock(sdp);
 698 }
 699
 700 /**
 701  * gfs2_log_flush - flush incore transaction(s)
 702  * @sdp: the filesystem
 703  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 704  *
 705  */
 706
 707 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 708 {
 709         struct gfs2_ail *ai;
 710
 711         down_write(&sdp->sd_log_flush_lock);
 712
 713         /* Log might have been flushed while we waited for the flush lock */
 714         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 715                 up_write(&sdp->sd_log_flush_lock);
 716                 return;
 717         }
 718
 719         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 720         INIT_LIST_HEAD(&ai->ai_ail1_list);
 721         INIT_LIST_HEAD(&ai->ai_ail2_list);
 722
 723         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
 724                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
 725                        sdp->sd_log_commited_buf);
 726                 gfs2_assert_withdraw(sdp, 0);
 727         }
 728         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
 729                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
 730                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
 731                 gfs2_assert_withdraw(sdp, 0);
 732         }
 733         gfs2_assert_withdraw(sdp,
 734                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 735
 736         sdp->sd_log_flush_head = sdp->sd_log_head;
 737         sdp->sd_log_flush_wrapped = 0;
 738         ai->ai_first = sdp->sd_log_flush_head;
 739
 740         gfs2_ordered_write(sdp);
 741         lops_before_commit(sdp);
 742         gfs2_ordered_wait(sdp);
 743
 744         if (sdp->sd_log_head != sdp->sd_log_flush_head)
 745                 log_flush_commit(sdp);
 746         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 747                 gfs2_log_lock(sdp);
 748                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 749                 gfs2_log_unlock(sdp);
 750                 log_write_header(sdp, 0, PULL);
 751         }
 752         lops_after_commit(sdp, ai);
 753
 754         gfs2_log_lock(sdp);
 755         sdp->sd_log_head = sdp->sd_log_flush_head;
 756         sdp->sd_log_blks_reserved = 0;
 757         sdp->sd_log_commited_buf = 0;
 758         sdp->sd_log_commited_databuf = 0;
 759         sdp->sd_log_commited_revoke = 0;
 760
 761         if (!list_empty(&ai->ai_ail1_list)) {
 762                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 763                 ai = NULL;
 764         }
 765         gfs2_log_unlock(sdp);
 766
 767         sdp->sd_vfs->s_dirt = 0;
 768         up_write(&sdp->sd_log_flush_lock);
 769
 770         kfree(ai);
 771 }
 772
 773 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 774 {
 775         unsigned int reserved;
 776         unsigned int unused;
 777
 778         gfs2_log_lock(sdp);
 779
 780         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 781         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 782                 tr->tr_num_databuf_rm;
 783         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 784                              (((int)sdp->sd_log_commited_databuf) >= 0));
 785         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 786         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 787         reserved = calc_reserved(sdp);
 788         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved);
 789         unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
 790         atomic_add(unused, &sdp->sd_log_blks_free);
 791         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 792                              sdp->sd_jdesc->jd_blocks);
 793         sdp->sd_log_blks_reserved = reserved;
 794
 795         gfs2_log_unlock(sdp);
 796 }
 797
 798 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 799 {
 800         struct list_head *head = &tr->tr_list_buf;
 801         struct gfs2_bufdata *bd;
 802
 803         gfs2_log_lock(sdp);
 804         while (!list_empty(head)) {
 805                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
 806                 list_del_init(&bd->bd_list_tr);
 807                 tr->tr_num_buf--;
 808         }
 809         gfs2_log_unlock(sdp);
 810         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 811 }
 812
 813 /**
 814  * gfs2_log_commit - Commit a transaction to the log
 815  * @sdp: the filesystem
 816  * @tr: the transaction
 817  *
 818  * Returns: errno
 819  */
 820
 821 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 822 {
 823         log_refund(sdp, tr);
 824         buf_lo_incore_commit(sdp, tr);
 825
 826         sdp->sd_vfs->s_dirt = 1;
 827         up_read(&sdp->sd_log_flush_lock);
 828
 829         gfs2_log_lock(sdp);
 830         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 831                 wake_up_process(sdp->sd_logd_process);
 832         gfs2_log_unlock(sdp);
 833 }
 834
 835 /**
 836  * gfs2_log_shutdown - write a shutdown header into a journal
 837  * @sdp: the filesystem
 838  *
 839  */
 840
 841 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 842 {
 843         down_write(&sdp->sd_log_flush_lock);
 844
 845         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 846         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 847         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 848         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 849         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 850         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 851
 852         sdp->sd_log_flush_head = sdp->sd_log_head;
 853         sdp->sd_log_flush_wrapped = 0;
 854
 855         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 856                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 857
 858         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 859         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 860         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 861
 862         sdp->sd_log_head = sdp->sd_log_flush_head;
 863         sdp->sd_log_tail = sdp->sd_log_head;
 864
 865         up_write(&sdp->sd_log_flush_lock);
 866 }
 867
 868
 869 /**
 870  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 871  * @sdp: the filesystem
 872  *
 873  */
 874
 875 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 876 {
 877         gfs2_log_flush(sdp, NULL);
 878         for (;;) {
 879                 gfs2_ail1_start(sdp, DIO_ALL);
 880                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 881                         break;
 882                 msleep(10);
 883         }
 884 }
 885
 886
 887 /**
 888  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 889  * @sdp: Pointer to GFS2 superblock
 890  *
 891  * Also, periodically check to make sure that we're using the most recent
 892  * journal index.
 893  */
 894
 895 int gfs2_logd(void *data)
 896 {
 897         struct gfs2_sbd *sdp = data;
 898         unsigned long t;
 899         int need_flush;
 900
 901         while (!kthread_should_stop()) {
 902                 /* Advance the log tail */
 903
 904                 t = sdp->sd_log_flush_time +
 905                     gfs2_tune_get(sdp, gt_log_flush_secs) * HZ;
 906
 907                 gfs2_ail1_empty(sdp, DIO_ALL);
 908                 gfs2_log_lock(sdp);
 909                 need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks);
 910                 gfs2_log_unlock(sdp);
 911                 if (need_flush || time_after_eq(jiffies, t)) {
 912                         gfs2_log_flush(sdp, NULL);
 913                         sdp->sd_log_flush_time = jiffies;
 914                 }
 915
 916                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
 917                 if (freezing(current))
 918                         refrigerator();
 919                 schedule_timeout_interruptible(t);
 920         }
 921
 922         return 0;
 923 }
 924