fs/gfs2/meta_io.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 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 v.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/mm.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/writeback.h>
  18 #include <linux/swap.h>
  19 #include <linux/delay.h>
  20 #include <asm/semaphore.h>
  21
  22 #include "gfs2.h"
  23 #include "glock.h"
  24 #include "glops.h"
  25 #include "inode.h"
  26 #include "log.h"
  27 #include "lops.h"
  28 #include "meta_io.h"
  29 #include "rgrp.h"
  30 #include "trans.h"
  31
  32 #define buffer_busy(bh) \
  33 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
  34 #define buffer_in_io(bh) \
  35 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
  36
  37 static int aspace_get_block(struct inode *inode, sector_t lblock,
  38                             struct buffer_head *bh_result, int create)
  39 {
  40         gfs2_assert_warn(get_v2sdp(inode->i_sb), 0);
  41         return -EOPNOTSUPP;
  42 }
  43
  44 static int gfs2_aspace_writepage(struct page *page,
  45                                  struct writeback_control *wbc)
  46 {
  47         return block_write_full_page(page, aspace_get_block, wbc);
  48 }
  49
  50 /**
  51  * stuck_releasepage - We're stuck in gfs2_releasepage().  Print stuff out.
  52  * @bh: the buffer we're stuck on
  53  *
  54  */
  55
  56 static void stuck_releasepage(struct buffer_head *bh)
  57 {
  58         struct gfs2_sbd *sdp = get_v2sdp(bh->b_page->mapping->host->i_sb);
  59         struct gfs2_bufdata *bd = get_v2bd(bh);
  60         struct gfs2_glock *gl;
  61
  62         fs_warn(sdp, "stuck in gfs2_releasepage()\n");
  63         fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
  64                 (uint64_t)bh->b_blocknr, atomic_read(&bh->b_count));
  65         fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
  66         fs_warn(sdp, "get_v2bd(bh) = %s\n", (bd) ? "!NULL" : "NULL");
  67
  68         if (!bd)
  69                 return;
  70
  71         gl = bd->bd_gl;
  72
  73         fs_warn(sdp, "gl = (%u, %llu)\n",
  74                 gl->gl_name.ln_type, gl->gl_name.ln_number);
  75
  76         fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
  77                 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
  78                 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
  79
  80         if (gl->gl_ops == &gfs2_inode_glops) {
  81                 struct gfs2_inode *ip = get_gl2ip(gl);
  82                 unsigned int x;
  83
  84                 if (!ip)
  85                         return;
  86
  87                 fs_warn(sdp, "ip = %llu %llu\n",
  88                         ip->i_num.no_formal_ino, ip->i_num.no_addr);
  89                 fs_warn(sdp, "ip->i_count = %d, ip->i_vnode = %s\n",
  90                         atomic_read(&ip->i_count),
  91                         (ip->i_vnode) ? "!NULL" : "NULL");
  92
  93                 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
  94                         fs_warn(sdp, "ip->i_cache[%u] = %s\n",
  95                                 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
  96         }
  97 }
  98
  99 /**
 100  * gfs2_aspace_releasepage - free the metadata associated with a page
 101  * @page: the page that's being released
 102  * @gfp_mask: passed from Linux VFS, ignored by us
 103  *
 104  * Call try_to_free_buffers() if the buffers in this page can be
 105  * released.
 106  *
 107  * Returns: 0
 108  */
 109
 110 static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask)
 111 {
 112         struct inode *aspace = page->mapping->host;
 113         struct gfs2_sbd *sdp = get_v2sdp(aspace->i_sb);
 114         struct buffer_head *bh, *head;
 115         struct gfs2_bufdata *bd;
 116         unsigned long t;
 117
 118         if (!page_has_buffers(page))
 119                 goto out;
 120
 121         head = bh = page_buffers(page);
 122         do {
 123                 t = jiffies;
 124
 125                 while (atomic_read(&bh->b_count)) {
 126                         if (atomic_read(&aspace->i_writecount)) {
 127                                 if (time_after_eq(jiffies, t +
 128                                     gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
 129                                         stuck_releasepage(bh);
 130                                         t = jiffies;
 131                                 }
 132
 133                                 yield();
 134                                 continue;
 135                         }
 136
 137                         return 0;
 138                 }
 139
 140                 gfs2_assert_warn(sdp, !buffer_pinned(bh));
 141
 142                 bd = get_v2bd(bh);
 143                 if (bd) {
 144                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
 145                         gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
 146                         gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
 147                         gfs2_assert_warn(sdp, !bd->bd_ail);
 148                         kmem_cache_free(gfs2_bufdata_cachep, bd);
 149                         atomic_dec(&sdp->sd_bufdata_count);
 150                         set_v2bd(bh, NULL);
 151                 }
 152
 153                 bh = bh->b_this_page;
 154         }
 155         while (bh != head);
 156
 157  out:
 158         return try_to_free_buffers(page);
 159 }
 160
 161 static struct address_space_operations aspace_aops = {
 162         .writepage = gfs2_aspace_writepage,
 163         .releasepage = gfs2_aspace_releasepage,
 164 };
 165
 166 /**
 167  * gfs2_aspace_get - Create and initialize a struct inode structure
 168  * @sdp: the filesystem the aspace is in
 169  *
 170  * Right now a struct inode is just a struct inode.  Maybe Linux
 171  * will supply a more lightweight address space construct (that works)
 172  * in the future.
 173  *
 174  * Make sure pages/buffers in this aspace aren't in high memory.
 175  *
 176  * Returns: the aspace
 177  */
 178
 179 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
 180 {
 181         struct inode *aspace;
 182
 183         aspace = new_inode(sdp->sd_vfs);
 184         if (aspace) {
 185                 mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL);
 186                 aspace->i_mapping->a_ops = &aspace_aops;
 187                 aspace->i_size = ~0ULL;
 188                 set_v2ip(aspace, NULL);
 189                 insert_inode_hash(aspace);
 190         }
 191
 192         return aspace;
 193 }
 194
 195 void gfs2_aspace_put(struct inode *aspace)
 196 {
 197         remove_inode_hash(aspace);
 198         iput(aspace);
 199 }
 200
 201 /**
 202  * gfs2_ail1_start_one - Start I/O on a part of the AIL
 203  * @sdp: the filesystem
 204  * @tr: the part of the AIL
 205  *
 206  */
 207
 208 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 209 {
 210         struct gfs2_bufdata *bd, *s;
 211         struct buffer_head *bh;
 212         int retry;
 213
 214         do {
 215                 retry = 0;
 216
 217                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 218                                                  bd_ail_st_list) {
 219                         bh = bd->bd_bh;
 220
 221                         gfs2_assert(sdp, bd->bd_ail == ai);
 222
 223                         if (!buffer_busy(bh)) {
 224                                 if (!buffer_uptodate(bh))
 225                                         gfs2_io_error_bh(sdp, bh);
 226                                 list_move(&bd->bd_ail_st_list,
 227                                           &ai->ai_ail2_list);
 228                                 continue;
 229                         }
 230
 231                         if (!buffer_dirty(bh))
 232                                 continue;
 233
 234                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 235
 236                         gfs2_log_unlock(sdp);
 237                         wait_on_buffer(bh);
 238                         ll_rw_block(WRITE, 1, &bh);
 239                         gfs2_log_lock(sdp);
 240
 241                         retry = 1;
 242                         break;
 243                 }
 244         } while (retry);
 245 }
 246
 247 /**
 248  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 249  * @sdp: the filesystem
 250  * @ai: the AIL entry
 251  *
 252  */
 253
 254 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 255 {
 256         struct gfs2_bufdata *bd, *s;
 257         struct buffer_head *bh;
 258
 259         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 260                                          bd_ail_st_list) {
 261                 bh = bd->bd_bh;
 262
 263                 gfs2_assert(sdp, bd->bd_ail == ai);
 264
 265                 if (buffer_busy(bh)) {
 266                         if (flags & DIO_ALL)
 267                                 continue;
 268                         else
 269                                 break;
 270                 }
 271
 272                 if (!buffer_uptodate(bh))
 273                         gfs2_io_error_bh(sdp, bh);
 274
 275                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 276         }
 277
 278         return list_empty(&ai->ai_ail1_list);
 279 }
 280
 281 /**
 282  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 283  * @sdp: the filesystem
 284  * @ai: the AIL entry
 285  *
 286  */
 287
 288 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 289 {
 290         struct list_head *head = &ai->ai_ail2_list;
 291         struct gfs2_bufdata *bd;
 292
 293         while (!list_empty(head)) {
 294                 bd = list_entry(head->prev, struct gfs2_bufdata,
 295                                 bd_ail_st_list);
 296                 gfs2_assert(sdp, bd->bd_ail == ai);
 297                 bd->bd_ail = NULL;
 298                 list_del(&bd->bd_ail_st_list);
 299                 list_del(&bd->bd_ail_gl_list);
 300                 atomic_dec(&bd->bd_gl->gl_ail_count);
 301                 brelse(bd->bd_bh);
 302         }
 303 }
 304
 305 /**
 306  * ail_empty_gl - remove all buffers for a given lock from the AIL
 307  * @gl: the glock
 308  *
 309  * None of the buffers should be dirty, locked, or pinned.
 310  */
 311
 312 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
 313 {
 314         struct gfs2_sbd *sdp = gl->gl_sbd;
 315         unsigned int blocks;
 316         struct list_head *head = &gl->gl_ail_list;
 317         struct gfs2_bufdata *bd;
 318         struct buffer_head *bh;
 319         uint64_t blkno;
 320         int error;
 321
 322         blocks = atomic_read(&gl->gl_ail_count);
 323         if (!blocks)
 324                 return;
 325
 326         error = gfs2_trans_begin(sdp, 0, blocks);
 327         if (gfs2_assert_withdraw(sdp, !error))
 328                 return;
 329
 330         gfs2_log_lock(sdp);
 331         while (!list_empty(head)) {
 332                 bd = list_entry(head->next, struct gfs2_bufdata,
 333                                 bd_ail_gl_list);
 334                 bh = bd->bd_bh;
 335                 blkno = bh->b_blocknr;
 336                 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
 337
 338                 bd->bd_ail = NULL;
 339                 list_del(&bd->bd_ail_st_list);
 340                 list_del(&bd->bd_ail_gl_list);
 341                 atomic_dec(&gl->gl_ail_count);
 342                 brelse(bh);
 343                 gfs2_log_unlock(sdp);
 344
 345                 gfs2_trans_add_revoke(sdp, blkno);
 346
 347                 gfs2_log_lock(sdp);
 348         }
 349         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 350         gfs2_log_unlock(sdp);
 351
 352         gfs2_trans_end(sdp);
 353         gfs2_log_flush(sdp);
 354 }
 355
 356 /**
 357  * gfs2_meta_inval - Invalidate all buffers associated with a glock
 358  * @gl: the glock
 359  *
 360  */
 361
 362 void gfs2_meta_inval(struct gfs2_glock *gl)
 363 {
 364         struct gfs2_sbd *sdp = gl->gl_sbd;
 365         struct inode *aspace = gl->gl_aspace;
 366         struct address_space *mapping = gl->gl_aspace->i_mapping;
 367
 368         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 369
 370         atomic_inc(&aspace->i_writecount);
 371         truncate_inode_pages(mapping, 0);
 372         atomic_dec(&aspace->i_writecount);
 373
 374         gfs2_assert_withdraw(sdp, !mapping->nrpages);
 375 }
 376
 377 /**
 378  * gfs2_meta_sync - Sync all buffers associated with a glock
 379  * @gl: The glock
 380  * @flags: DIO_START | DIO_WAIT
 381  *
 382  */
 383
 384 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
 385 {
 386         struct address_space *mapping = gl->gl_aspace->i_mapping;
 387         int error = 0;
 388
 389         if (flags & DIO_START)
 390                 filemap_fdatawrite(mapping);
 391         if (!error && (flags & DIO_WAIT))
 392                 error = filemap_fdatawait(mapping);
 393
 394         if (error)
 395                 gfs2_io_error(gl->gl_sbd);
 396 }
 397
 398 /**
 399  * getbuf - Get a buffer with a given address space
 400  * @sdp: the filesystem
 401  * @aspace: the address space
 402  * @blkno: the block number (filesystem scope)
 403  * @create: 1 if the buffer should be created
 404  *
 405  * Returns: the buffer
 406  */
 407
 408 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
 409                                   uint64_t blkno, int create)
 410 {
 411         struct page *page;
 412         struct buffer_head *bh;
 413         unsigned int shift;
 414         unsigned long index;
 415         unsigned int bufnum;
 416
 417         shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
 418         index = blkno >> shift;             /* convert block to page */
 419         bufnum = blkno - (index << shift);  /* block buf index within page */
 420
 421         if (create) {
 422                 for (;;) {
 423                         page = grab_cache_page(aspace->i_mapping, index);
 424                         if (page)
 425                                 break;
 426                         yield();
 427                 }
 428         } else {
 429                 page = find_lock_page(aspace->i_mapping, index);
 430                 if (!page)
 431                         return NULL;
 432         }
 433
 434         if (!page_has_buffers(page))
 435                 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
 436
 437         /* Locate header for our buffer within our page */
 438         for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
 439                 /* Do nothing */;
 440         get_bh(bh);
 441
 442         if (!buffer_mapped(bh))
 443                 map_bh(bh, sdp->sd_vfs, blkno);
 444
 445         unlock_page(page);
 446         mark_page_accessed(page);
 447         page_cache_release(page);
 448
 449         return bh;
 450 }
 451
 452 static void meta_prep_new(struct buffer_head *bh)
 453 {
 454         struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
 455
 456         lock_buffer(bh);
 457         clear_buffer_dirty(bh);
 458         set_buffer_uptodate(bh);
 459         unlock_buffer(bh);
 460
 461         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 462 }
 463
 464 /**
 465  * gfs2_meta_new - Get a block
 466  * @gl: The glock associated with this block
 467  * @blkno: The block number
 468  *
 469  * Returns: The buffer
 470  */
 471
 472 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
 473 {
 474         struct buffer_head *bh;
 475         bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
 476         meta_prep_new(bh);
 477         return bh;
 478 }
 479
 480 /**
 481  * gfs2_meta_read - Read a block from disk
 482  * @gl: The glock covering the block
 483  * @blkno: The block number
 484  * @flags: flags to gfs2_dreread()
 485  * @bhp: the place where the buffer is returned (NULL on failure)
 486  *
 487  * Returns: errno
 488  */
 489
 490 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
 491                    struct buffer_head **bhp)
 492 {
 493         int error;
 494
 495         *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
 496         error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
 497         if (error)
 498                 brelse(*bhp);
 499
 500         return error;
 501 }
 502
 503 /**
 504  * gfs2_meta_reread - Reread a block from disk
 505  * @sdp: the filesystem
 506  * @bh: The block to read
 507  * @flags: Flags that control the read
 508  *
 509  * Returns: errno
 510  */
 511
 512 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
 513 {
 514         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 515                 return -EIO;
 516
 517         if (flags & DIO_FORCE)
 518                 clear_buffer_uptodate(bh);
 519
 520         if ((flags & DIO_START) && !buffer_uptodate(bh))
 521                 ll_rw_block(READ, 1, &bh);
 522
 523         if (flags & DIO_WAIT) {
 524                 wait_on_buffer(bh);
 525
 526                 if (!buffer_uptodate(bh)) {
 527                         struct gfs2_trans *tr = get_transaction;
 528                         if (tr && tr->tr_touched)
 529                                 gfs2_io_error_bh(sdp, bh);
 530                         return -EIO;
 531                 }
 532                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 533                         return -EIO;
 534         }
 535
 536         return 0;
 537 }
 538
 539 /**
 540  * gfs2_meta_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
 541  * @gl: the glock the buffer belongs to
 542  * @bh: The buffer to be attached to
 543  *
 544  */
 545
 546 void gfs2_meta_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh)
 547 {
 548         struct gfs2_bufdata *bd;
 549
 550         lock_page(bh->b_page);
 551
 552         if (get_v2bd(bh)) {
 553                 unlock_page(bh->b_page);
 554                 return;
 555         }
 556
 557         bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_KERNEL | __GFP_NOFAIL),
 558         atomic_inc(&gl->gl_sbd->sd_bufdata_count);
 559
 560         memset(bd, 0, sizeof(struct gfs2_bufdata));
 561
 562         bd->bd_bh = bh;
 563         bd->bd_gl = gl;
 564
 565         INIT_LIST_HEAD(&bd->bd_list_tr);
 566         lops_init_le(&bd->bd_le, &gfs2_buf_lops);
 567
 568         set_v2bd(bh, bd);
 569
 570         unlock_page(bh->b_page);
 571 }
 572
 573 /**
 574  * gfs2_meta_pin - Pin a metadata buffer in memory
 575  * @sdp: the filesystem the buffer belongs to
 576  * @bh: The buffer to be pinned
 577  *
 578  */
 579
 580 void gfs2_meta_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
 581 {
 582         struct gfs2_bufdata *bd = get_v2bd(bh);
 583
 584         gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
 585
 586         if (test_set_buffer_pinned(bh))
 587                 gfs2_assert_withdraw(sdp, 0);
 588
 589         wait_on_buffer(bh);
 590
 591         /* If this buffer is in the AIL and it has already been written
 592            to in-place disk block, remove it from the AIL. */
 593
 594         gfs2_log_lock(sdp);
 595         if (bd->bd_ail && !buffer_in_io(bh))
 596                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
 597         gfs2_log_unlock(sdp);
 598
 599         clear_buffer_dirty(bh);
 600         wait_on_buffer(bh);
 601
 602         if (!buffer_uptodate(bh))
 603                 gfs2_io_error_bh(sdp, bh);
 604
 605         get_bh(bh);
 606 }
 607
 608 /**
 609  * gfs2_meta_unpin - Unpin a buffer
 610  * @sdp: the filesystem the buffer belongs to
 611  * @bh: The buffer to unpin
 612  * @ai:
 613  *
 614  */
 615
 616 void gfs2_meta_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 617                      struct gfs2_ail *ai)
 618 {
 619         struct gfs2_bufdata *bd = get_v2bd(bh);
 620
 621         gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
 622
 623         if (!buffer_pinned(bh))
 624                 gfs2_assert_withdraw(sdp, 0);
 625
 626         mark_buffer_dirty(bh);
 627         clear_buffer_pinned(bh);
 628
 629         gfs2_log_lock(sdp);
 630         if (bd->bd_ail) {
 631                 list_del(&bd->bd_ail_st_list);
 632                 brelse(bh);
 633         } else {
 634                 struct gfs2_glock *gl = bd->bd_gl;
 635                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 636                 atomic_inc(&gl->gl_ail_count);
 637         }
 638         bd->bd_ail = ai;
 639         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 640         gfs2_log_unlock(sdp);
 641 }
 642
 643 /**
 644  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
 645  * @ip: the inode who owns the buffers
 646  * @bstart: the first buffer in the run
 647  * @blen: the number of buffers in the run
 648  *
 649  */
 650
 651 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
 652 {
 653         struct gfs2_sbd *sdp = ip->i_sbd;
 654         struct inode *aspace = ip->i_gl->gl_aspace;
 655         struct buffer_head *bh;
 656
 657         while (blen) {
 658                 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
 659                 if (bh) {
 660                         struct gfs2_bufdata *bd = get_v2bd(bh);
 661
 662                         if (test_clear_buffer_pinned(bh)) {
 663                                 gfs2_log_lock(sdp);
 664                                 list_del_init(&bd->bd_le.le_list);
 665                                 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
 666                                 sdp->sd_log_num_buf--;
 667                                 gfs2_log_unlock(sdp);
 668                                 get_transaction->tr_num_buf_rm++;
 669                                 brelse(bh);
 670                         }
 671                         if (bd) {
 672                                 gfs2_log_lock(sdp);
 673                                 if (bd->bd_ail) {
 674                                         uint64_t blkno = bh->b_blocknr;
 675                                         bd->bd_ail = NULL;
 676                                         list_del(&bd->bd_ail_st_list);
 677                                         list_del(&bd->bd_ail_gl_list);
 678                                         atomic_dec(&bd->bd_gl->gl_ail_count);
 679                                         brelse(bh);
 680                                         gfs2_log_unlock(sdp);
 681                                         gfs2_trans_add_revoke(sdp, blkno);
 682                                 } else
 683                                         gfs2_log_unlock(sdp);
 684                         }
 685
 686                         lock_buffer(bh);
 687                         clear_buffer_dirty(bh);
 688                         clear_buffer_uptodate(bh);
 689                         unlock_buffer(bh);
 690
 691                         brelse(bh);
 692                 }
 693
 694                 bstart++;
 695                 blen--;
 696         }
 697 }
 698
 699 /**
 700  * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
 701  * @ip: The GFS2 inode
 702  *
 703  * This releases buffers that are in the most-recently-used array of
 704  * blocks used for indirect block addressing for this inode.
 705  */
 706
 707 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
 708 {
 709         struct buffer_head **bh_slot;
 710         unsigned int x;
 711
 712         spin_lock(&ip->i_spin);
 713
 714         for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
 715                 bh_slot = &ip->i_cache[x];
 716                 if (!*bh_slot)
 717                         break;
 718                 brelse(*bh_slot);
 719                 *bh_slot = NULL;
 720         }
 721
 722         spin_unlock(&ip->i_spin);
 723 }
 724
 725 /**
 726  * gfs2_meta_indirect_buffer - Get a metadata buffer
 727  * @ip: The GFS2 inode
 728  * @height: The level of this buf in the metadata (indir addr) tree (if any)
 729  * @num: The block number (device relative) of the buffer
 730  * @new: Non-zero if we may create a new buffer
 731  * @bhp: the buffer is returned here
 732  *
 733  * Try to use the gfs2_inode's MRU metadata tree cache.
 734  *
 735  * Returns: errno
 736  */
 737
 738 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
 739                               int new, struct buffer_head **bhp)
 740 {
 741         struct buffer_head *bh, **bh_slot = ip->i_cache + height;
 742         int error;
 743
 744         spin_lock(&ip->i_spin);
 745         bh = *bh_slot;
 746         if (bh) {
 747                 if (bh->b_blocknr == num)
 748                         get_bh(bh);
 749                 else
 750                         bh = NULL;
 751         }
 752         spin_unlock(&ip->i_spin);
 753
 754         if (bh) {
 755                 if (new)
 756                         meta_prep_new(bh);
 757                 else {
 758                         error = gfs2_meta_reread(ip->i_sbd, bh,
 759                                                  DIO_START | DIO_WAIT);
 760                         if (error) {
 761                                 brelse(bh);
 762                                 return error;
 763                         }
 764                 }
 765         } else {
 766                 if (new)
 767                         bh = gfs2_meta_new(ip->i_gl, num);
 768                 else {
 769                         error = gfs2_meta_read(ip->i_gl, num,
 770                                                DIO_START | DIO_WAIT, &bh);
 771                         if (error)
 772                                 return error;
 773                 }
 774
 775                 spin_lock(&ip->i_spin);
 776                 if (*bh_slot != bh) {
 777                         brelse(*bh_slot);
 778                         *bh_slot = bh;
 779                         get_bh(bh);
 780                 }
 781                 spin_unlock(&ip->i_spin);
 782         }
 783
 784         if (new) {
 785                 if (gfs2_assert_warn(ip->i_sbd, height)) {
 786                         brelse(bh);
 787                         return -EIO;
 788                 }
 789                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 790                 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 791                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
 792
 793         } else if (gfs2_metatype_check(ip->i_sbd, bh,
 794                              (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
 795                 brelse(bh);
 796                 return -EIO;
 797         }
 798
 799         *bhp = bh;
 800
 801         return 0;
 802 }
 803
 804 /**
 805  * gfs2_meta_ra - start readahead on an extent of a file
 806  * @gl: the glock the blocks belong to
 807  * @dblock: the starting disk block
 808  * @extlen: the number of blocks in the extent
 809  *
 810  */
 811
 812 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
 813 {
 814         struct gfs2_sbd *sdp = gl->gl_sbd;
 815         struct inode *aspace = gl->gl_aspace;
 816         struct buffer_head *first_bh, *bh;
 817         uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >> sdp->sd_sb.sb_bsize_shift;
 818         int error;
 819
 820         if (!extlen || !max_ra)
 821                 return;
 822         if (extlen > max_ra)
 823                 extlen = max_ra;
 824
 825         first_bh = getbuf(sdp, aspace, dblock, CREATE);
 826
 827         if (buffer_uptodate(first_bh))
 828                 goto out;
 829         if (!buffer_locked(first_bh)) {
 830                 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
 831                 if (error)
 832                         goto out;
 833         }
 834
 835         dblock++;
 836         extlen--;
 837
 838         while (extlen) {
 839                 bh = getbuf(sdp, aspace, dblock, CREATE);
 840
 841                 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
 842                         error = gfs2_meta_reread(sdp, bh, DIO_START);
 843                         brelse(bh);
 844                         if (error)
 845                                 goto out;
 846                 } else
 847                         brelse(bh);
 848
 849                 dblock++;
 850                 extlen--;
 851
 852                 if (buffer_uptodate(first_bh))
 853                         break;
 854         }
 855
 856  out:
 857         brelse(first_bh);
 858 }
 859
 860 /**
 861  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 862  * @sdp: the filesystem
 863  *
 864  */
 865
 866 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 867 {
 868         gfs2_log_flush(sdp);
 869         for (;;) {
 870                 gfs2_ail1_start(sdp, DIO_ALL);
 871                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 872                         break;
 873                 msleep(100);
 874         }
 875 }
 876