fs/gfs2/log.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/crc32.h>
  17 #include <linux/lm_interface.h>
  18 #include <linux/delay.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "bmap.h"
  23 #include "glock.h"
  24 #include "log.h"
  25 #include "lops.h"
  26 #include "meta_io.h"
  27 #include "util.h"
  28 #include "dir.h"
  29
  30 #define PULL 1
  31
  32 /**
  33  * gfs2_struct2blk - compute stuff
  34  * @sdp: the filesystem
  35  * @nstruct: the number of structures
  36  * @ssize: the size of the structures
  37  *
  38  * Compute the number of log descriptor blocks needed to hold a certain number
  39  * of structures of a certain size.
  40  *
  41  * Returns: the number of blocks needed (minimum is always 1)
  42  */
  43
  44 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  45                              unsigned int ssize)
  46 {
  47         unsigned int blks;
  48         unsigned int first, second;
  49
  50         blks = 1;
  51         first = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / ssize;
  52
  53         if (nstruct > first) {
  54                 second = (sdp->sd_sb.sb_bsize -
  55                           sizeof(struct gfs2_meta_header)) / ssize;
  56                 blks += DIV_ROUND_UP(nstruct - first, second);
  57         }
  58
  59         return blks;
  60 }
  61
  62 /**
  63  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
  64  * @mapping: The associated mapping (maybe NULL)
  65  * @bd: The gfs2_bufdata to remove
  66  *
  67  * The log lock _must_ be held when calling this function
  68  *
  69  */
  70
  71 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  72 {
  73         bd->bd_ail = NULL;
  74         list_del_init(&bd->bd_ail_st_list);
  75         list_del_init(&bd->bd_ail_gl_list);
  76         atomic_dec(&bd->bd_gl->gl_ail_count);
  77         brelse(bd->bd_bh);
  78 }
  79
  80 /**
  81  * gfs2_ail1_start_one - Start I/O on a part of the AIL
  82  * @sdp: the filesystem
  83  * @tr: the part of the AIL
  84  *
  85  */
  86
  87 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  88 {
  89         struct gfs2_bufdata *bd, *s;
  90         struct buffer_head *bh;
  91         int retry;
  92
  93         BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
  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 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 inode *inode = sdp->sd_jdesc->jd_inode;
 343         int error;
 344         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 345
 346         bh_map.b_size = 1 << inode->i_blkbits;
 347         error = gfs2_block_map(inode, lbn, 0, &bh_map);
 348         if (error || !bh_map.b_blocknr)
 349                 printk(KERN_INFO "error=%d, dbn=%llu lbn=%u", error,
 350                        (unsigned long long)bh_map.b_blocknr, lbn);
 351         gfs2_assert_withdraw(sdp, !error && bh_map.b_blocknr);
 352
 353         return bh_map.b_blocknr;
 354 }
 355
 356 /**
 357  * log_distance - Compute distance between two journal blocks
 358  * @sdp: The GFS2 superblock
 359  * @newer: The most recent journal block of the pair
 360  * @older: The older journal block of the pair
 361  *
 362  *   Compute the distance (in the journal direction) between two
 363  *   blocks in the journal
 364  *
 365  * Returns: the distance in blocks
 366  */
 367
 368 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
 369                                         unsigned int older)
 370 {
 371         int dist;
 372
 373         dist = newer - older;
 374         if (dist < 0)
 375                 dist += sdp->sd_jdesc->jd_blocks;
 376
 377         return dist;
 378 }
 379
 380 /**
 381  * calc_reserved - Calculate the number of blocks to reserve when
 382  *                 refunding a transaction's unused buffers.
 383  * @sdp: The GFS2 superblock
 384  *
 385  * This is complex.  We need to reserve room for all our currently used
 386  * metadata buffers (e.g. normal file I/O rewriting file time stamps) and
 387  * all our journaled data buffers for journaled files (e.g. files in the
 388  * meta_fs like rindex, or files for which chattr +j was done.)
 389  * If we don't reserve enough space, gfs2_log_refund and gfs2_log_flush
 390  * will count it as free space (sd_log_blks_free) and corruption will follow.
 391  *
 392  * We can have metadata bufs and jdata bufs in the same journal.  So each
 393  * type gets its own log header, for which we need to reserve a block.
 394  * In fact, each type has the potential for needing more than one header
 395  * in cases where we have more buffers than will fit on a journal page.
 396  * Metadata journal entries take up half the space of journaled buffer entries.
 397  * Thus, metadata entries have buf_limit (502) and journaled buffers have
 398  * databuf_limit (251) before they cause a wrap around.
 399  *
 400  * Also, we need to reserve blocks for revoke journal entries and one for an
 401  * overall header for the lot.
 402  *
 403  * Returns: the number of blocks reserved
 404  */
 405 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
 406 {
 407         unsigned int reserved = 0;
 408         unsigned int mbuf_limit, metabufhdrs_needed;
 409         unsigned int dbuf_limit, databufhdrs_needed;
 410         unsigned int revokes = 0;
 411
 412         mbuf_limit = buf_limit(sdp);
 413         metabufhdrs_needed = (sdp->sd_log_commited_buf +
 414                               (mbuf_limit - 1)) / mbuf_limit;
 415         dbuf_limit = databuf_limit(sdp);
 416         databufhdrs_needed = (sdp->sd_log_commited_databuf +
 417                               (dbuf_limit - 1)) / dbuf_limit;
 418
 419         if (sdp->sd_log_commited_revoke)
 420                 revokes = gfs2_struct2blk(sdp, sdp->sd_log_commited_revoke,
 421                                           sizeof(u64));
 422
 423         reserved = sdp->sd_log_commited_buf + metabufhdrs_needed +
 424                 sdp->sd_log_commited_databuf + databufhdrs_needed +
 425                 revokes;
 426         /* One for the overall header */
 427         if (reserved)
 428                 reserved++;
 429         return reserved;
 430 }
 431
 432 static unsigned int current_tail(struct gfs2_sbd *sdp)
 433 {
 434         struct gfs2_ail *ai;
 435         unsigned int tail;
 436
 437         gfs2_log_lock(sdp);
 438
 439         if (list_empty(&sdp->sd_ail1_list)) {
 440                 tail = sdp->sd_log_head;
 441         } else {
 442                 ai = list_entry(sdp->sd_ail1_list.prev, struct gfs2_ail, ai_list);
 443                 tail = ai->ai_first;
 444         }
 445
 446         gfs2_log_unlock(sdp);
 447
 448         return tail;
 449 }
 450
 451 void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 452 {
 453         if (sdp->sd_log_flush_head == sdp->sd_log_tail)
 454                 BUG_ON(sdp->sd_log_flush_head != sdp->sd_log_head);
 455
 456         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 457                 sdp->sd_log_flush_head = 0;
 458                 sdp->sd_log_flush_wrapped = 1;
 459         }
 460 }
 461
 462 /**
 463  * gfs2_log_write_endio - End of I/O for a log buffer
 464  * @bh: The buffer head
 465  * @uptodate: I/O Status
 466  *
 467  */
 468
 469 static void gfs2_log_write_endio(struct buffer_head *bh, int uptodate)
 470 {
 471         struct gfs2_sbd *sdp = bh->b_private;
 472         bh->b_private = NULL;
 473
 474         end_buffer_write_sync(bh, uptodate);
 475         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 476                 wake_up(&sdp->sd_log_flush_wait);
 477 }
 478
 479 /**
 480  * gfs2_log_get_buf - Get and initialize a buffer to use for log control data
 481  * @sdp: The GFS2 superblock
 482  *
 483  * Returns: the buffer_head
 484  */
 485
 486 struct buffer_head *gfs2_log_get_buf(struct gfs2_sbd *sdp)
 487 {
 488         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 489         struct buffer_head *bh;
 490
 491         bh = sb_getblk(sdp->sd_vfs, blkno);
 492         lock_buffer(bh);
 493         memset(bh->b_data, 0, bh->b_size);
 494         set_buffer_uptodate(bh);
 495         clear_buffer_dirty(bh);
 496         gfs2_log_incr_head(sdp);
 497         atomic_inc(&sdp->sd_log_in_flight);
 498         bh->b_private = sdp;
 499         bh->b_end_io = gfs2_log_write_endio;
 500
 501         return bh;
 502 }
 503
 504 /**
 505  * gfs2_fake_write_endio -
 506  * @bh: The buffer head
 507  * @uptodate: The I/O Status
 508  *
 509  */
 510
 511 static void gfs2_fake_write_endio(struct buffer_head *bh, int uptodate)
 512 {
 513         struct buffer_head *real_bh = bh->b_private;
 514         struct gfs2_bufdata *bd = real_bh->b_private;
 515         struct gfs2_sbd *sdp = bd->bd_gl->gl_sbd;
 516
 517         end_buffer_write_sync(bh, uptodate);
 518         free_buffer_head(bh);
 519         unlock_buffer(real_bh);
 520         brelse(real_bh);
 521         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 522                 wake_up(&sdp->sd_log_flush_wait);
 523 }
 524
 525 /**
 526  * gfs2_log_fake_buf - Build a fake buffer head to write metadata buffer to log
 527  * @sdp: the filesystem
 528  * @data: the data the buffer_head should point to
 529  *
 530  * Returns: the log buffer descriptor
 531  */
 532
 533 struct buffer_head *gfs2_log_fake_buf(struct gfs2_sbd *sdp,
 534                                       struct buffer_head *real)
 535 {
 536         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 537         struct buffer_head *bh;
 538
 539         bh = alloc_buffer_head(GFP_NOFS | __GFP_NOFAIL);
 540         atomic_set(&bh->b_count, 1);
 541         bh->b_state = (1 << BH_Mapped) | (1 << BH_Uptodate) | (1 << BH_Lock);
 542         set_bh_page(bh, real->b_page, bh_offset(real));
 543         bh->b_blocknr = blkno;
 544         bh->b_size = sdp->sd_sb.sb_bsize;
 545         bh->b_bdev = sdp->sd_vfs->s_bdev;
 546         bh->b_private = real;
 547         bh->b_end_io = gfs2_fake_write_endio;
 548
 549         gfs2_log_incr_head(sdp);
 550         atomic_inc(&sdp->sd_log_in_flight);
 551
 552         return bh;
 553 }
 554
 555 static void log_pull_tail(struct gfs2_sbd *sdp, unsigned int new_tail)
 556 {
 557         unsigned int dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
 558
 559         ail2_empty(sdp, new_tail);
 560
 561         gfs2_log_lock(sdp);
 562         atomic_add(dist, &sdp->sd_log_blks_free);
 563         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <= sdp->sd_jdesc->jd_blocks);
 564         gfs2_log_unlock(sdp);
 565
 566         sdp->sd_log_tail = new_tail;
 567 }
 568
 569 /**
 570  * log_write_header - Get and initialize a journal header buffer
 571  * @sdp: The GFS2 superblock
 572  *
 573  * Returns: the initialized log buffer descriptor
 574  */
 575
 576 static void log_write_header(struct gfs2_sbd *sdp, u32 flags, int pull)
 577 {
 578         u64 blkno = log_bmap(sdp, sdp->sd_log_flush_head);
 579         struct buffer_head *bh;
 580         struct gfs2_log_header *lh;
 581         unsigned int tail;
 582         u32 hash;
 583
 584         bh = sb_getblk(sdp->sd_vfs, blkno);
 585         lock_buffer(bh);
 586         memset(bh->b_data, 0, bh->b_size);
 587         set_buffer_uptodate(bh);
 588         clear_buffer_dirty(bh);
 589         unlock_buffer(bh);
 590
 591         gfs2_ail1_empty(sdp, 0);
 592         tail = current_tail(sdp);
 593
 594         lh = (struct gfs2_log_header *)bh->b_data;
 595         memset(lh, 0, sizeof(struct gfs2_log_header));
 596         lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 597         lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
 598         lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
 599         lh->lh_sequence = cpu_to_be64(sdp->sd_log_sequence++);
 600         lh->lh_flags = cpu_to_be32(flags);
 601         lh->lh_tail = cpu_to_be32(tail);
 602         lh->lh_blkno = cpu_to_be32(sdp->sd_log_flush_head);
 603         hash = gfs2_disk_hash(bh->b_data, sizeof(struct gfs2_log_header));
 604         lh->lh_hash = cpu_to_be32(hash);
 605
 606         set_buffer_dirty(bh);
 607         if (sync_dirty_buffer(bh))
 608                 gfs2_io_error_bh(sdp, bh);
 609         brelse(bh);
 610
 611         if (sdp->sd_log_tail != tail)
 612                 log_pull_tail(sdp, tail);
 613         else
 614                 gfs2_assert_withdraw(sdp, !pull);
 615
 616         sdp->sd_log_idle = (tail == sdp->sd_log_flush_head);
 617         gfs2_log_incr_head(sdp);
 618 }
 619
 620 static void log_flush_commit(struct gfs2_sbd *sdp)
 621 {
 622         DEFINE_WAIT(wait);
 623
 624         if (atomic_read(&sdp->sd_log_in_flight)) {
 625                 do {
 626                         prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
 627                                         TASK_UNINTERRUPTIBLE);
 628                         if (atomic_read(&sdp->sd_log_in_flight))
 629                                 io_schedule();
 630                 } while(atomic_read(&sdp->sd_log_in_flight));
 631                 finish_wait(&sdp->sd_log_flush_wait, &wait);
 632         }
 633
 634         log_write_header(sdp, 0, 0);
 635 }
 636
 637 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
 638 {
 639         struct gfs2_bufdata *bd;
 640         struct buffer_head *bh;
 641         LIST_HEAD(written);
 642
 643         gfs2_log_lock(sdp);
 644         while (!list_empty(&sdp->sd_log_le_ordered)) {
 645                 bd = list_entry(sdp->sd_log_le_ordered.next, struct gfs2_bufdata, bd_le.le_list);
 646                 list_move(&bd->bd_le.le_list, &written);
 647                 bh = bd->bd_bh;
 648                 if (!buffer_dirty(bh))
 649                         continue;
 650                 get_bh(bh);
 651                 gfs2_log_unlock(sdp);
 652                 lock_buffer(bh);
 653                 if (buffer_mapped(bh) && test_clear_buffer_dirty(bh)) {
 654                         bh->b_end_io = end_buffer_write_sync;
 655                         submit_bh(WRITE, bh);
 656                 } else {
 657                         unlock_buffer(bh);
 658                         brelse(bh);
 659                 }
 660                 gfs2_log_lock(sdp);
 661         }
 662         list_splice(&written, &sdp->sd_log_le_ordered);
 663         gfs2_log_unlock(sdp);
 664 }
 665
 666 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
 667 {
 668         struct gfs2_bufdata *bd;
 669         struct buffer_head *bh;
 670
 671         gfs2_log_lock(sdp);
 672         while (!list_empty(&sdp->sd_log_le_ordered)) {
 673                 bd = list_entry(sdp->sd_log_le_ordered.prev, struct gfs2_bufdata, bd_le.le_list);
 674                 bh = bd->bd_bh;
 675                 if (buffer_locked(bh)) {
 676                         get_bh(bh);
 677                         gfs2_log_unlock(sdp);
 678                         wait_on_buffer(bh);
 679                         brelse(bh);
 680                         gfs2_log_lock(sdp);
 681                         continue;
 682                 }
 683                 list_del_init(&bd->bd_le.le_list);
 684         }
 685         gfs2_log_unlock(sdp);
 686 }
 687
 688 /**
 689  * gfs2_log_flush - flush incore transaction(s)
 690  * @sdp: the filesystem
 691  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
 692  *
 693  */
 694
 695 void __gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 696 {
 697         struct gfs2_ail *ai;
 698
 699         down_write(&sdp->sd_log_flush_lock);
 700
 701         /* Log might have been flushed while we waited for the flush lock */
 702         if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags)) {
 703                 up_write(&sdp->sd_log_flush_lock);
 704                 return;
 705         }
 706
 707         ai = kzalloc(sizeof(struct gfs2_ail), GFP_NOFS | __GFP_NOFAIL);
 708         INIT_LIST_HEAD(&ai->ai_ail1_list);
 709         INIT_LIST_HEAD(&ai->ai_ail2_list);
 710
 711         if (sdp->sd_log_num_buf != sdp->sd_log_commited_buf) {
 712                 printk(KERN_INFO "GFS2: log buf %u %u\n", sdp->sd_log_num_buf,
 713                        sdp->sd_log_commited_buf);
 714                 gfs2_assert_withdraw(sdp, 0);
 715         }
 716         if (sdp->sd_log_num_databuf != sdp->sd_log_commited_databuf) {
 717                 printk(KERN_INFO "GFS2: log databuf %u %u\n",
 718                        sdp->sd_log_num_databuf, sdp->sd_log_commited_databuf);
 719                 gfs2_assert_withdraw(sdp, 0);
 720         }
 721         gfs2_assert_withdraw(sdp,
 722                         sdp->sd_log_num_revoke == sdp->sd_log_commited_revoke);
 723
 724         sdp->sd_log_flush_head = sdp->sd_log_head;
 725         sdp->sd_log_flush_wrapped = 0;
 726         ai->ai_first = sdp->sd_log_flush_head;
 727
 728         gfs2_ordered_write(sdp);
 729         lops_before_commit(sdp);
 730         gfs2_ordered_wait(sdp);
 731
 732         if (sdp->sd_log_head != sdp->sd_log_flush_head)
 733                 log_flush_commit(sdp);
 734         else if (sdp->sd_log_tail != current_tail(sdp) && !sdp->sd_log_idle){
 735                 gfs2_log_lock(sdp);
 736                 atomic_dec(&sdp->sd_log_blks_free); /* Adjust for unreserved buffer */
 737                 gfs2_log_unlock(sdp);
 738                 log_write_header(sdp, 0, PULL);
 739         }
 740         lops_after_commit(sdp, ai);
 741
 742         gfs2_log_lock(sdp);
 743         sdp->sd_log_head = sdp->sd_log_flush_head;
 744         sdp->sd_log_blks_reserved = 0;
 745         sdp->sd_log_commited_buf = 0;
 746         sdp->sd_log_commited_databuf = 0;
 747         sdp->sd_log_commited_revoke = 0;
 748
 749         if (!list_empty(&ai->ai_ail1_list)) {
 750                 list_add(&ai->ai_list, &sdp->sd_ail1_list);
 751                 ai = NULL;
 752         }
 753         gfs2_log_unlock(sdp);
 754
 755         sdp->sd_vfs->s_dirt = 0;
 756         up_write(&sdp->sd_log_flush_lock);
 757
 758         kfree(ai);
 759 }
 760
 761 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 762 {
 763         unsigned int reserved;
 764         unsigned int old;
 765
 766         gfs2_log_lock(sdp);
 767
 768         sdp->sd_log_commited_buf += tr->tr_num_buf_new - tr->tr_num_buf_rm;
 769         sdp->sd_log_commited_databuf += tr->tr_num_databuf_new -
 770                 tr->tr_num_databuf_rm;
 771         gfs2_assert_withdraw(sdp, (((int)sdp->sd_log_commited_buf) >= 0) ||
 772                              (((int)sdp->sd_log_commited_databuf) >= 0));
 773         sdp->sd_log_commited_revoke += tr->tr_num_revoke - tr->tr_num_revoke_rm;
 774         gfs2_assert_withdraw(sdp, ((int)sdp->sd_log_commited_revoke) >= 0);
 775         reserved = calc_reserved(sdp);
 776         old = atomic_read(&sdp->sd_log_blks_free);
 777         atomic_add(tr->tr_reserved - (reserved - sdp->sd_log_blks_reserved),
 778                    &sdp->sd_log_blks_free);
 779
 780         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) >= old);
 781         gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
 782                              sdp->sd_jdesc->jd_blocks);
 783
 784         sdp->sd_log_blks_reserved = reserved;
 785
 786         gfs2_log_unlock(sdp);
 787 }
 788
 789 /**
 790  * gfs2_log_commit - Commit a transaction to the log
 791  * @sdp: the filesystem
 792  * @tr: the transaction
 793  *
 794  * Returns: errno
 795  */
 796
 797 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 798 {
 799         log_refund(sdp, tr);
 800         lops_incore_commit(sdp, tr);
 801
 802         sdp->sd_vfs->s_dirt = 1;
 803         up_read(&sdp->sd_log_flush_lock);
 804
 805         gfs2_log_lock(sdp);
 806         if (sdp->sd_log_num_buf > gfs2_tune_get(sdp, gt_incore_log_blocks))
 807                 wake_up_process(sdp->sd_logd_process);
 808         gfs2_log_unlock(sdp);
 809 }
 810
 811 /**
 812  * gfs2_log_shutdown - write a shutdown header into a journal
 813  * @sdp: the filesystem
 814  *
 815  */
 816
 817 void gfs2_log_shutdown(struct gfs2_sbd *sdp)
 818 {
 819         down_write(&sdp->sd_log_flush_lock);
 820
 821         gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
 822         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_buf);
 823         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 824         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_rg);
 825         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_databuf);
 826         gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
 827
 828         sdp->sd_log_flush_head = sdp->sd_log_head;
 829         sdp->sd_log_flush_wrapped = 0;
 830
 831         log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT,
 832                          (sdp->sd_log_tail == current_tail(sdp)) ? 0 : PULL);
 833
 834         gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
 835         gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
 836         gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
 837
 838         sdp->sd_log_head = sdp->sd_log_flush_head;
 839         sdp->sd_log_tail = sdp->sd_log_head;
 840
 841         up_write(&sdp->sd_log_flush_lock);
 842 }
 843
 844
 845 /**
 846  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 847  * @sdp: the filesystem
 848  *
 849  */
 850
 851 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 852 {
 853         gfs2_log_flush(sdp, NULL);
 854         for (;;) {
 855                 gfs2_ail1_start(sdp, DIO_ALL);
 856                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 857                         break;
 858                 msleep(10);
 859         }
 860 }
 861