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 #include "trace_gfs2.h"
  32
  33 #define PULL 1
  34
  35 /**
  36  * gfs2_struct2blk - compute stuff
  37  * @sdp: the filesystem
  38  * @nstruct: the number of structures
  39  * @ssize: the size of the structures
  40  *
  41  * Compute the number of log descriptor blocks needed to hold a certain number
  42  * of structures of a certain size.
  43  *
  44  * Returns: the number of blocks needed (minimum is always 1)
  45  */
  46
  47 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  48                              unsigned int ssize)
  49 {
  50         unsigned int blks;
  51         unsigned int first, second;
  52
  53         blks = 1;
  54         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  55
  56         if (nstruct > first) {
  57                 second = (sdp->sd_sb.sb_bsize -
  58                           sizeof(struct gfs2_meta_header)) / ssize;
  59                 blks += DIV_ROUND_UP(nstruct - first, second);
  60         }
  61
  62         return blks;
  63 }
  64
  65 /**
  66  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  67  * @mapping: The associated mapping (maybe NULL)
  68  * @bd: The gfs2_bufdata to remove
  69  *
  70  * The log lock _must_ be held when calling this function
  71  *
  72  */
  73
  74 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  75 {
  76         bd->bd_ail = NULL;
  77         list_del_init(&bd->bd_ail_st_list);
  78         list_del_init(&bd->bd_ail_gl_list);
  79         atomic_dec(&bd->bd_gl->gl_ail_count);
  80         brelse(bd->bd_bh);
  81 }
  82
  83 /**
  84  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  85  * @sdp: the filesystem
  86  * @tr: the part of the AIL
  87  *
  88  */
  89
  90 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  91 __releases(&sdp->sd_log_lock)
  92 __acquires(&sdp->sd_log_lock)
  93 {
  94         struct gfs2_bufdata *bd, *s;
  95         struct buffer_head *bh;
  96         int retry;
  97
  98         do {
  99                 retry = 0;
 100
 101                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 102                                                  bd_ail_st_list) {
 103                         bh = bd->bd_bh;
 104
 105                         gfs2_assert(sdp, bd->bd_ail == ai);
 106
 107                         if (!buffer_busy(bh)) {
 108                                 if (!buffer_uptodate(bh))
 109                                         gfs2_io_error_bh(sdp, bh);
 110                                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 111                                 continue;
 112                         }
 113
 114                         if (!buffer_dirty(bh))
 115                                 continue;
 116
 117                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 118
 119                         get_bh(bh);
 120                         gfs2_log_unlock(sdp);
 121                         lock_buffer(bh);
 122                         if (test_clear_buffer_dirty(bh)) {
 123                                 bh->b_end_io = end_buffer_write_sync;
 124                                 submit_bh(WRITE_SYNC_PLUG, bh);
 125                         } else {
 126                                 unlock_buffer(bh);
 127                                 brelse(bh);
 128                         }
 129                         gfs2_log_lock(sdp);
 130
 131                         retry = 1;
 132                         break;
 133                 }
 134         } while (retry);
 135 }
 136
 137 /**
 138  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 139  * @sdp: the filesystem
 140  * @ai: the AIL entry
 141  *
 142  */
 143
 144 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 145 {
 146         struct gfs2_bufdata *bd, *s;
 147         struct buffer_head *bh;
 148
 149         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 150                                          bd_ail_st_list) {
 151                 bh = bd->bd_bh;
 152
 153                 gfs2_assert(sdp, bd->bd_ail == ai);
 154
 155                 if (buffer_busy(bh)) {
 156                         if (flags & DIO_ALL)
 157                                 continue;
 158                         else
 159                                 break;
 160                 }
 161
 162                 if (!buffer_uptodate(bh))
 163                         gfs2_io_error_bh(sdp, bh);
 164
 165                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 166         }
 167
 168         return list_empty(&ai->ai_ail1_list);
 169 }
 170
 171 static void gfs2_ail1_start(struct gfs2_sbd *sdp, int flags)
 172 {
 173         struct list_head *head;
 174         u64 sync_gen;
 175         struct list_head *first;
 176         struct gfs2_ail *first_ai, *ai, *tmp;
 177         int done = 0;
 178
 179         gfs2_log_lock(sdp);
 180         head = &sdp->sd_ail1_list;
 181         if (list_empty(head)) {
 182                 gfs2_log_unlock(sdp);
 183                 return;
 184         }
 185         sync_gen = sdp->sd_ail_sync_gen++;
 186
 187         first = head->prev;
 188         first_ai = list_entry(first, struct gfs2_ail, ai_list);
 189         first_ai->ai_sync_gen = sync_gen;
 190         gfs2_ail1_start_one(sdp, first_ai); /* This may drop log lock */
 191
 192         if (flags & DIO_ALL)
 193                 first = NULL;
 194
 195         while(!done) {
 196                 if (first && (head->prev != first ||
 197                               gfs2_ail1_empty_one(sdp, first_ai, 0)))
 198                         break;
 199
 200                 done = 1;
 201                 list_for_each_entry_safe_reverse(ai, tmp, head, ai_list) {
 202                         if (ai->ai_sync_gen >= sync_gen)
 203                                 continue;
 204                         ai->ai_sync_gen = sync_gen;
 205                         gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
 206                         done = 0;
 207                         break;
 208                 }
 209         }
 210
 211         gfs2_log_unlock(sdp);
 212 }
 213
 214 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 215 {
 216         struct gfs2_ail *ai, *s;
 217         int ret;
 218
 219         gfs2_log_lock(sdp);
 220
 221         list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 222                 if (gfs2_ail1_empty_one(sdp, ai, flags))
 223                         list_move(&ai->ai_list, &sdp->sd_ail2_list);
 224                 else if (!(flags & DIO_ALL))
 225                         break;
 226         }
 227
 228         ret = list_empty(&sdp->sd_ail1_list);
 229
 230         gfs2_log_unlock(sdp);
 231
 232         return ret;
 233 }
 234
 235
 236 /**
 237  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 238  * @sdp: the filesystem
 239  * @ai: the AIL entry
 240  *
 241  */
 242
 243 static void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 244 {
 245         struct list_head *head = &ai->ai_ail2_list;
 246         struct gfs2_bufdata *bd;
 247
 248         while (!list_empty(head)) {
 249                 bd = list_entry(head->prev, struct gfs2_bufdata,
 250                                 bd_ail_st_list);
 251                 gfs2_assert(sdp, bd->bd_ail == ai);
 252                 gfs2_remove_from_ail(bd);
 253         }
 254 }
 255
 256 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 257 {
 258         struct gfs2_ail *ai, *safe;
 259         unsigned int old_tail = sdp->sd_log_tail;
 260         int wrap = (new_tail < old_tail);
 261         int a, b, rm;
 262
 263         gfs2_log_lock(sdp);
 264
 265         list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 266                 a = (old_tail <= ai->ai_first);
 267                 b = (ai->ai_first < new_tail);
 268                 rm = (wrap) ? (a || b) : (a && b);
 269                 if (!rm)
 270                         continue;
 271
 272                 gfs2_ail2_empty_one(sdp, ai);
 273                 list_del(&ai->ai_list);
 274                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail1_list));
 275                 gfs2_assert_warn(sdp, list_empty(&ai->ai_ail2_list));
 276                 kfree(ai);
 277         }
 278
 279         gfs2_log_unlock(sdp);
 280 }
 281
 282 /**
 283  * gfs2_log_reserve - Make a log reservation
 284  * @sdp: The GFS2 superblock
 285  * @blks: The number of blocks to reserve
 286  *
 287  * Note that we never give out the last few blocks of the journal. Thats
 288  * due to the fact that there is a small number of header blocks
 289  * associated with each log flush. The exact number can't be known until
 290  * flush time, so we ensure that we have just enough free blocks at all
 291  * times to avoid running out during a log flush.
 292  *
 293  * Returns: errno
 294  */
 295
 296 int gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks)
 297 {
 298         unsigned int try = 0;
 299         unsigned reserved_blks = 6 * (4096 / sdp->sd_vfs->s_blocksize);
 300
 301         if (gfs2_assert_warn(sdp, blks) ||
 302             gfs2_assert_warn(sdp, blks <= sdp->sd_jdesc->jd_blocks))
 303                 return -EINVAL;
 304
 305         mutex_lock(&sdp->sd_log_reserve_mutex);
 306         gfs2_log_lock(sdp);
 307         while(atomic_read(&sdp->sd_log_blks_free) <= (blks + reserved_blks)) {
 308                 gfs2_log_unlock(sdp);
 309                 gfs2_ail1_empty(sdp, 0);
 310                 gfs2_log_flush(sdp, NULL);
 311
 312                 if (try++)
 313                         gfs2_ail1_start(sdp, 0);
 314                 gfs2_log_lock(sdp);
 315         }
 316         atomic_sub(blks, &sdp->sd_log_blks_free);
 317         trace_gfs2_log_blocks(sdp, -blks);
 318         gfs2_log_unlock(sdp);
 319         mutex_unlock(&sdp->sd_log_reserve_mutex);
 320
 321         down_read(&sdp->sd_log_flush_lock);
 322
 323         return 0;
 324 }
 325
 326 /**
 327  * gfs2_log_release - Release a given number of log blocks
 328  * @sdp: The GFS2 superblock
 329  * @blks: The number of blocks
 330  *
 331  */
 332
 333 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
 334 {
 335
 336         gfs2_log_lock(sdp);
 337         atomic_add(blks, &sdp->sd_log_blks_free);
 338         trace_gfs2_log_blocks(sdp, blks);
 339         gfs2_assert_withdraw(sdp,
 340                              atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 341         gfs2_log_unlock(sdp);
 342         up_read(&sdp->sd_log_flush_lock);
 343 }
 344
 345 static u64 log_bmap(struct gfs2_sbd *sdp, unsigned int lbn)
 346 {
 347         struct gfs2_journal_extent *je;
 348
 349         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
 350                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks)
 351                         return je->dblock + lbn - je->lblock;
 352         }
 353
 354         return -1;
 355 }
 356
 357 /**
 358  * log_distance - Compute distance between two journal blocks
 359  * @sdp: The GFS2 superblock
 360  * @newer: The most recent journal block of the pair
 361  * @older: The older journal block of the pair
 362  *
 363  *   Compute the distance (in the journal direction) between two
 364  *   blocks in the journal
 365  *
 366  * Returns: the distance in blocks
 367  */
 368
 369 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 370                                         unsigned int older)
 371 {
 372         int dist;
 373
 374         dist = newer - older;
 375         if (dist < 0)
 376                 dist += sdp->sd_jdesc->jd_blocks;
 377
 378         return dist;
 379 }
 380
 381 /**
 382  * calc_reserved - Calculate the number of blocks to reserve when
 383  *                 refunding a transaction's unused buffers.
 384  * @sdp: The GFS2 superblock
 385  *
 386  * This is complex.  We need to reserve room for all our currently used
 387  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 388  * all our journaled data buffers for journaled files (e.g. files in the
 389  * meta_fs like rindex, or files for which chattr +j was done.)
 390  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 391  * will count it as free space (sd_log_blks_free) and corruption will follow.
 392  *
 393  * We can have metadata bufs and jdata bufs in the same journal.  So each
 394  * type gets its own log header, for which we need to reserve a block.
 395  * In fact, each type has the potential for needing more than one header
 396  * in cases where we have more buffers than will fit on a journal page.
 397  * Metadata journal entries take up half the space of journaled buffer entries.
 398  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 399  * databuf_limit (251) before they cause a wrap around.
 400  *
 401  * Also, we need to reserve blocks for revoke journal entries and one for an
 402  * overall header for the lot.
 403  *
 404  * Returns: the number of blocks reserved
 405  */
 406 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 407 {
 408         unsigned int reserved = 0;
 409         unsigned int mbuf_limit, metabufhdrs_needed;
 410         unsigned int dbuf_limit, databufhdrs_needed;
 411         unsigned int revokes = 0;
 412
 413         mbuf_limit = buf_limit(sdp);
 414         metabufhdrs_needed = (sdp->sd_log_commited_buf +
 415                               (mbuf_limit - 1)) / mbuf_limit;
 416         dbuf_limit = databuf_limit(sdp);
 417         databufhdrs_needed = (sdp->sd_log_commited_databuf +
 418                               (dbuf_limit - 1)) / dbuf_limit;
 419
 420         if (sdp->sd_log_commited_revoke > 0)
 421                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 422                                           sizeof(u64));
 423
 424         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
 425                 sdp->sd_log_commited_databuf + databufhdrs_needed +
 426                 revokes;
 427         /* One for the overall header */
 428         if (reserved)
 429                 reserved++;
 430         return reserved;
 431 }
 432
 433 static unsigned int current_tail(struct gfs2_sbd *sdp)
 434 {
 435         struct gfs2_ail *ai;
 436         unsigned int tail;
 437
 438         gfs2_log_lock(sdp);
 439
 440         if (list_empty(&sdp->sd_ail1_list)) {
 441                 tail = sdp->sd_log_head;
 442         } else {
 443                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 444                 tail = ai->ai_first;
 445         }
 446
 447         gfs2_log_unlock(sdp);
 448
 449         return tail;
 450 }
 451
 452 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 453 {
 454         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 455                 BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head);
 456
 457         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 458                 sdp->sd_log_flush_head = 0;
 459                 sdp->sd_log_flush_wrapped = 1;
 460         }
 461 }
 462
 463 /**
 464  * gfs2_log_write_endio - End of I/O for a log buffer
 465  * @bh: The buffer head
 466  * @uptodate: I/O Status
 467  *
 468  */
 469
 470 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 471 {
 472         struct gfs2_sbd *sdp = bh->b_private;
 473         bh->b_private = NULL;
 474
 475         end_buffer_write_sync(bh, uptodate);
 476         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 477                 wake_up(&sdp->sd_log_flush_wait);
 478 }
 479
 480 /**
 481  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 482  * @sdp: The GFS2 superblock
 483  *
 484  * Returns: the buffer_head
 485  */
 486
 487 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 488 {
 489         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 490         struct buffer_head *bh;
 491
 492         bh = sb_getblk(sdp->sd_vfs, blkno);
 493         lock_buffer(bh);
 494         memset(bh->b_data, 0, bh->b_size);
 495         set_buffer_uptodate(bh);
 496         clear_buffer_dirty(bh);
 497         gfs2_log_incr_head(sdp);
 498         atomic_inc(&sdp->sd_log_in_flight);
 499         bh->b_private = sdp;
 500         bh->b_end_io = gfs2_log_write_endio;
 501
 502         return bh;
 503 }
 504
 505 /**
 506  * gfs2_fake_write_endio -
 507  * @bh: The buffer head
 508  * @uptodate: The I/O Status
 509  *
 510  */
 511
 512 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 513 {
 514         struct buffer_head *real_bh = bh->b_private;
 515         struct gfs2_bufdata *bd = real_bh->b_private;
 516         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 517
 518         end_buffer_write_sync(bh, uptodate);
 519         free_buffer_head(bh);
 520         unlock_buffer(real_bh);
 521         brelse(real_bh);
 522         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 523                 wake_up(&sdp->sd_log_flush_wait);
 524 }
 525
 526 /**
 527  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 528  * @sdp: the filesystem
 529  * @data: the data the buffer_head should point to
 530  *
 531  * Returns: the log buffer descriptor
 532  */
 533
 534 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 535                                       struct buffer_head *real)
 536 {
 537         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 538         struct buffer_head *bh;
 539
 540         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 541         atomic_set(&bh->b_count, 1);
 542         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 543         set_bh_page(bh, real->b_page, bh_offset(real));
 544         bh->b_blocknr = blkno;
 545         bh->b_size = sdp->sd_sb.sb_bsize;
 546         bh->b_bdev = sdp->sd_vfs->s_bdev;
 547         bh->b_private = real;
 548         bh->b_end_io = gfs2_fake_write_endio;
 549
 550         gfs2_log_incr_head(sdp);
 551         atomic_inc(&sdp->sd_log_in_flight);
 552
 553         return bh;
 554 }
 555
 556 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 557 {
 558         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 559
 560         ail2_empty(sdp, new_tail);
 561
 562         gfs2_log_lock(sdp);
 563         atomic_add(dist, &sdp->sd_log_blks_free);
 564         trace_gfs2_log_blocks(sdp, dist);
 565         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 566         gfs2_log_unlock(sdp);
 567
 568         sdp->sd_log_tail = new_tail;
 569 }
 570
 571 /**
 572  * log_write_header - Get and initialize a journal header buffer
 573  * @sdp: The GFS2 superblock
 574  *
 575  * Returns: the initialized log buffer descriptor
 576  */
 577
 578 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 579 {
 580         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 581         struct buffer_head *bh;
 582         struct gfs2_log_header *lh;
 583         unsigned int tail;
 584         u32 hash;
 585
 586         bh = sb_getblk(sdp->sd_vfs, blkno);
 587         lock_buffer(bh);
 588         memset(bh->b_data, 0, bh->b_size);
 589         set_buffer_uptodate(bh);
 590         clear_buffer_dirty(bh);
 591
 592         gfs2_ail1_empty(sdp, 0);
 593         tail = current_tail(sdp);
 594
 595         lh = (struct gfs2_log_header *)bh->b_data;
 596         memset(lh, 0, sizeof(struct gfs2_log_header));
 597         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 598         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 599         lh->lh_header.__pad0 = cpu_to_be64(0);
 600         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 601         lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 602         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 603         lh->lh_flags = cpu_to_be32(flags);
 604         lh->lh_tail = cpu_to_be32(tail);
 605         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 606         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 607         lh->lh_hash = cpu_to_be32(hash);
 608
 609         bh->b_end_io = end_buffer_write_sync;
 610         if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
 611                 goto skip_barrier;
 612         get_bh(bh);
 613         submit_bh(WRITE_SYNC | (1 << BIO_RW_BARRIER) | (1 << BIO_RW_META), bh);
 614         wait_on_buffer(bh);
 615         if (buffer_eopnotsupp(bh)) {
 616                 clear_buffer_eopnotsupp(bh);
 617                 set_buffer_uptodate(bh);
 618                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
 619                 lock_buffer(bh);
 620 skip_barrier:
 621                 get_bh(bh);
 622                 submit_bh(WRITE_SYNC | (1 << BIO_RW_META), bh);
 623                 wait_on_buffer(bh);
 624         }
 625         if (!buffer_uptodate(bh))
 626                 gfs2_io_error_bh(sdp, bh);
 627         brelse(bh);
 628
 629         if (sdp->sd_log_tail != tail)
 630                 log_pull_tail(sdp, tail);
 631         else
 632                 gfs2_assert_withdraw(sdp, !pull);
 633
 634         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 635         gfs2_log_incr_head(sdp);
 636 }
 637
 638 static void log_flush_commit(struct gfs2_sbd *sdp)
 639 {
 640         DEFINE_WAIT(wait);
 641
 642         if (atomic_read(&sdp->sd_log_in_flight)) {
 643                 do {
 644                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 645                                         TASK_UNINTERRUPTIBLE);
 646                         if (atomic_read(&sdp->sd_log_in_flight))
 647                                 io_schedule();
 648                 } while(atomic_read(&sdp->sd_log_in_flight));
 649                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 650         }
 651
 652         log_write_header(sdp, 0, 0);
 653 }
 654
 655 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 656 {
 657         struct gfs2_bufdata *bd;
 658         struct buffer_head *bh;
 659         LIST_HEAD(written);
 660
 661         gfs2_log_lock(sdp);
 662         while (!list_empty(&sdp->sd_log_le_ordered)) {
 663                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 664                 list_move(&bd->bd_le.le_list, &written);
 665                 bh = bd->bd_bh;
 666                 if (!buffer_dirty(bh))
 667                         continue;
 668                 get_bh(bh);
 669                 gfs2_log_unlock(sdp);
 670                 lock_buffer(bh);
 671                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
 672                         bh->b_end_io = end_buffer_write_sync;
 673                         submit_bh(WRITE_SYNC_PLUG, bh);
 674                 } else {
 675                         unlock_buffer(bh);
 676                         brelse(bh);
 677                 }
 678                 gfs2_log_lock(sdp);
 679         }
 680         list_splice(&written, &sdp->sd_log_le_ordered);
 681         gfs2_log_unlock(sdp);
 682 }
 683
 684 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 685 {
 686         struct gfs2_bufdata *bd;
 687         struct buffer_head *bh;
 688
 689         gfs2_log_lock(sdp);
 690         while (!list_empty(&sdp->sd_log_le_ordered)) {
 691                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 692                 bh = bd->bd_bh;
 693                 if (buffer_locked(bh)) {
 694                         get_bh(bh);
 695                         gfs2_log_unlock(sdp);
 696                         wait_on_buffer(bh);
 697                         brelse(bh);
 698                         gfs2_log_lock(sdp);
 699                         continue;
 700                 }
 701                 list_del_init(&bd->bd_le.le_list);
 702         }
 703         gfs2_log_unlock(sdp);
 704 }
 705
 706 /**
 707  * gfs2_log_flush - flush incore transaction(s)
 708  * @sdp: the filesystem
 709  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 710  *
 711  */
 712
 713 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 714 {
 715         struct gfs2_ail *ai;
 716
 717         down_write(&sdp->sd_log_flush_lock);
 718
 719         /* Log might have been flushed while we waited for the flush lock */
 720         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 721                 up_write(&sdp->sd_log_flush_lock);
 722                 return;
 723         }
 724         trace_gfs2_log_flush(sdp, 1);
 725
 726         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 727         INIT_LIST_HEAD(&ai->ai_ail1_list);
 728         INIT_LIST_HEAD(&ai->ai_ail2_list);
 729
 730         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
 731                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
 732                        sdp->sd_log_commited_buf);
 733                 gfs2_assert_withdraw(sdp, 0);
 734         }
 735         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
 736                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
 737                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
 738                 gfs2_assert_withdraw(sdp, 0);
 739         }
 740         gfs2_assert_withdraw(sdp,
 741                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 742
 743         sdp->sd_log_flush_head = sdp->sd_log_head;
 744         sdp->sd_log_flush_wrapped = 0;
 745         ai->ai_first = sdp->sd_log_flush_head;
 746
 747         gfs2_ordered_write(sdp);
 748         lops_before_commit(sdp);
 749         gfs2_ordered_wait(sdp);
 750
 751         if (sdp->sd_log_head != sdp->sd_log_flush_head)
 752                 log_flush_commit(sdp);
 753         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 754                 gfs2_log_lock(sdp);
 755                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 756                 trace_gfs2_log_blocks(sdp, -1);
 757                 gfs2_log_unlock(sdp);
 758                 log_write_header(sdp, 0, PULL);
 759         }
 760         lops_after_commit(sdp, ai);
 761
 762         gfs2_log_lock(sdp);
 763         sdp->sd_log_head = sdp->sd_log_flush_head;
 764         sdp->sd_log_blks_reserved = 0;
 765         sdp->sd_log_commited_buf = 0;
 766         sdp->sd_log_commited_databuf = 0;
 767         sdp->sd_log_commited_revoke = 0;
 768
 769         if (!list_empty(&ai->ai_ail1_list)) {
 770                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 771                 ai = NULL;
 772         }
 773         gfs2_log_unlock(sdp);
 774         trace_gfs2_log_flush(sdp, 0);
 775         up_write(&sdp->sd_log_flush_lock);
 776
 777         kfree(ai);
 778 }
 779
 780 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 781 {
 782         unsigned int reserved;
 783         unsigned int unused;
 784
 785         gfs2_log_lock(sdp);
 786
 787         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 788         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 789                 tr->tr_num_databuf_rm;
 790         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 791                              (((int)sdp->sd_log_commited_databuf) >= 0));
 792         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 793         reserved = calc_reserved(sdp);
 794         gfs2_assert_withdraw(sdp, sdp->sd_log_blks_reserved + tr->tr_reserved >= reserved);
 795         unused = sdp->sd_log_blks_reserved - reserved + tr->tr_reserved;
 796         atomic_add(unused, &sdp->sd_log_blks_free);
 797         trace_gfs2_log_blocks(sdp, unused);
 798         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 799                              sdp->sd_jdesc->jd_blocks);
 800         sdp->sd_log_blks_reserved = reserved;
 801
 802         gfs2_log_unlock(sdp);
 803 }
 804
 805 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 806 {
 807         struct list_head *head = &tr->tr_list_buf;
 808         struct gfs2_bufdata *bd;
 809
 810         gfs2_log_lock(sdp);
 811         while (!list_empty(head)) {
 812                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
 813                 list_del_init(&bd->bd_list_tr);
 814                 tr->tr_num_buf--;
 815         }
 816         gfs2_log_unlock(sdp);
 817         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 818 }
 819
 820 /**
 821  * gfs2_log_commit - Commit a transaction to the log
 822  * @sdp: the filesystem
 823  * @tr: the transaction
 824  *
 825  * Returns: errno
 826  */
 827
 828 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 829 {
 830         log_refund(sdp, tr);
 831         buf_lo_incore_commit(sdp, tr);
 832
 833         up_read(&sdp->sd_log_flush_lock);
 834
 835         gfs2_log_lock(sdp);
 836         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 837                 wake_up_process(sdp->sd_logd_process);
 838         gfs2_log_unlock(sdp);
 839 }
 840
 841 /**
 842  * gfs2_log_shutdown - write a shutdown header into a journal
 843  * @sdp: the filesystem
 844  *
 845  */
 846
 847 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 848 {
 849         down_write(&sdp->sd_log_flush_lock);
 850
 851         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 852         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 853         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 854         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 855         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 856         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 857
 858         sdp->sd_log_flush_head = sdp->sd_log_head;
 859         sdp->sd_log_flush_wrapped = 0;
 860
 861         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 862                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 863
 864         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 865         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 866         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 867
 868         sdp->sd_log_head = sdp->sd_log_flush_head;
 869         sdp->sd_log_tail = sdp->sd_log_head;
 870
 871         up_write(&sdp->sd_log_flush_lock);
 872 }
 873
 874
 875 /**
 876  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 877  * @sdp: the filesystem
 878  *
 879  */
 880
 881 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 882 {
 883         gfs2_log_flush(sdp, NULL);
 884         for (;;) {
 885                 gfs2_ail1_start(sdp, DIO_ALL);
 886                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 887                         break;
 888                 msleep(10);
 889         }
 890 }
 891
 892
 893 /**
 894  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
 895  * @sdp: Pointer to GFS2 superblock
 896  *
 897  * Also, periodically check to make sure that we're using the most recent
 898  * journal index.
 899  */
 900
 901 int gfs2_logd(void *data)
 902 {
 903         struct gfs2_sbd *sdp = data;
 904         unsigned long t;
 905         int need_flush;
 906
 907         while (!kthread_should_stop()) {
 908                 /* Advance the log tail */
 909
 910                 t = sdp->sd_log_flush_time +
 911                     gfs2_tune_get(sdp, gt_log_flush_secs) * HZ;
 912
 913                 gfs2_ail1_empty(sdp, DIO_ALL);
 914                 gfs2_log_lock(sdp);
 915                 need_flush = sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks);
 916                 gfs2_log_unlock(sdp);
 917                 if (need_flush || time_after_eq(jiffies, t)) {
 918                         gfs2_log_flush(sdp, NULL);
 919                         sdp->sd_log_flush_time = jiffies;
 920                 }
 921
 922                 t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
 923                 if (freezing(current))
 924                         refrigerator();
 925                 schedule_timeout_interruptible(t);
 926         }
 927
 928         return 0;
 929 }
 930