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 <linux/gfs2_ondisk.h>
  21 #include <asm/semaphore.h>
  22
  23 #include "gfs2.h"
  24 #include "lm_interface.h"
  25 #include "incore.h"
  26 #include "glock.h"
  27 #include "glops.h"
  28 #include "inode.h"
  29 #include "log.h"
  30 #include "lops.h"
  31 #include "meta_io.h"
  32 #include "rgrp.h"
  33 #include "trans.h"
  34 #include "util.h"
  35
  36 #define buffer_busy(bh) \
  37 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
  38 #define buffer_in_io(bh) \
  39 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
  40
  41 static int aspace_get_block(struct inode *inode, sector_t lblock,
  42                             struct buffer_head *bh_result, int create)
  43 {
  44         gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
  45         return -EOPNOTSUPP;
  46 }
  47
  48 static int gfs2_aspace_writepage(struct page *page,
  49                                  struct writeback_control *wbc)
  50 {
  51         return block_write_full_page(page, aspace_get_block, wbc);
  52 }
  53
  54 /**
  55  * stuck_releasepage - We're stuck in gfs2_releasepage().  Print stuff out.
  56  * @bh: the buffer we're stuck on
  57  *
  58  */
  59
  60 static void stuck_releasepage(struct buffer_head *bh)
  61 {
  62         struct gfs2_sbd *sdp = bh->b_page->mapping->host->i_sb->s_fs_info;
  63         struct gfs2_bufdata *bd = bh->b_private;
  64         struct gfs2_glock *gl;
  65
  66         fs_warn(sdp, "stuck in gfs2_releasepage()\n");
  67         fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
  68                 (uint64_t)bh->b_blocknr, atomic_read(&bh->b_count));
  69         fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
  70         fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
  71
  72         if (!bd)
  73                 return;
  74
  75         gl = bd->bd_gl;
  76
  77         fs_warn(sdp, "gl = (%u, %llu)\n",
  78                 gl->gl_name.ln_type, gl->gl_name.ln_number);
  79
  80         fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
  81                 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
  82                 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
  83
  84         if (gl->gl_ops == &gfs2_inode_glops) {
  85                 struct gfs2_inode *ip = gl->gl_object;
  86                 unsigned int x;
  87
  88                 if (!ip)
  89                         return;
  90
  91                 fs_warn(sdp, "ip = %llu %llu\n",
  92                         ip->i_num.no_formal_ino, ip->i_num.no_addr);
  93                 fs_warn(sdp, "ip->i_count = %d, ip->i_vnode = %s\n",
  94                         atomic_read(&ip->i_count),
  95                         (ip->i_vnode) ? "!NULL" : "NULL");
  96
  97                 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
  98                         fs_warn(sdp, "ip->i_cache[%u] = %s\n",
  99                                 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
 100         }
 101 }
 102
 103 /**
 104  * gfs2_aspace_releasepage - free the metadata associated with a page
 105  * @page: the page that's being released
 106  * @gfp_mask: passed from Linux VFS, ignored by us
 107  *
 108  * Call try_to_free_buffers() if the buffers in this page can be
 109  * released.
 110  *
 111  * Returns: 0
 112  */
 113
 114 static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask)
 115 {
 116         struct inode *aspace = page->mapping->host;
 117         struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
 118         struct buffer_head *bh, *head;
 119         struct gfs2_bufdata *bd;
 120         unsigned long t;
 121
 122         if (!page_has_buffers(page))
 123                 goto out;
 124
 125         head = bh = page_buffers(page);
 126         do {
 127                 t = jiffies;
 128
 129                 while (atomic_read(&bh->b_count)) {
 130                         if (atomic_read(&aspace->i_writecount)) {
 131                                 if (time_after_eq(jiffies, t +
 132                                     gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
 133                                         stuck_releasepage(bh);
 134                                         t = jiffies;
 135                                 }
 136
 137                                 yield();
 138                                 continue;
 139                         }
 140
 141                         return 0;
 142                 }
 143
 144                 gfs2_assert_warn(sdp, !buffer_pinned(bh));
 145
 146                 bd = bh->b_private;
 147                 if (bd) {
 148                         gfs2_assert_warn(sdp, bd->bd_bh == bh);
 149                         gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
 150                         gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
 151                         gfs2_assert_warn(sdp, !bd->bd_ail);
 152                         kmem_cache_free(gfs2_bufdata_cachep, bd);
 153                         bh->b_private = NULL;
 154                 }
 155
 156                 bh = bh->b_this_page;
 157         }
 158         while (bh != head);
 159
 160  out:
 161         return try_to_free_buffers(page);
 162 }
 163
 164 static struct address_space_operations aspace_aops = {
 165         .writepage = gfs2_aspace_writepage,
 166         .releasepage = gfs2_aspace_releasepage,
 167 };
 168
 169 /**
 170  * gfs2_aspace_get - Create and initialize a struct inode structure
 171  * @sdp: the filesystem the aspace is in
 172  *
 173  * Right now a struct inode is just a struct inode.  Maybe Linux
 174  * will supply a more lightweight address space construct (that works)
 175  * in the future.
 176  *
 177  * Make sure pages/buffers in this aspace aren't in high memory.
 178  *
 179  * Returns: the aspace
 180  */
 181
 182 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
 183 {
 184         struct inode *aspace;
 185
 186         aspace = new_inode(sdp->sd_vfs);
 187         if (aspace) {
 188                 mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL);
 189                 aspace->i_mapping->a_ops = &aspace_aops;
 190                 aspace->i_size = ~0ULL;
 191                 aspace->u.generic_ip = NULL;
 192                 insert_inode_hash(aspace);
 193         }
 194
 195         return aspace;
 196 }
 197
 198 void gfs2_aspace_put(struct inode *aspace)
 199 {
 200         remove_inode_hash(aspace);
 201         iput(aspace);
 202 }
 203
 204 /**
 205  * gfs2_ail1_start_one - Start I/O on a part of the AIL
 206  * @sdp: the filesystem
 207  * @tr: the part of the AIL
 208  *
 209  */
 210
 211 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 212 {
 213         struct gfs2_bufdata *bd, *s;
 214         struct buffer_head *bh;
 215         int retry;
 216
 217         do {
 218                 retry = 0;
 219
 220                 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 221                                                  bd_ail_st_list) {
 222                         bh = bd->bd_bh;
 223
 224                         gfs2_assert(sdp, bd->bd_ail == ai);
 225
 226                         if (!buffer_busy(bh)) {
 227                                 if (!buffer_uptodate(bh))
 228                                         gfs2_io_error_bh(sdp, bh);
 229                                 list_move(&bd->bd_ail_st_list,
 230                                           &ai->ai_ail2_list);
 231                                 continue;
 232                         }
 233
 234                         if (!buffer_dirty(bh))
 235                                 continue;
 236
 237                         list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 238
 239                         gfs2_log_unlock(sdp);
 240                         wait_on_buffer(bh);
 241                         ll_rw_block(WRITE, 1, &bh);
 242                         gfs2_log_lock(sdp);
 243
 244                         retry = 1;
 245                         break;
 246                 }
 247         } while (retry);
 248 }
 249
 250 /**
 251  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
 252  * @sdp: the filesystem
 253  * @ai: the AIL entry
 254  *
 255  */
 256
 257 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
 258 {
 259         struct gfs2_bufdata *bd, *s;
 260         struct buffer_head *bh;
 261
 262         list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
 263                                          bd_ail_st_list) {
 264                 bh = bd->bd_bh;
 265
 266                 gfs2_assert(sdp, bd->bd_ail == ai);
 267
 268                 if (buffer_busy(bh)) {
 269                         if (flags & DIO_ALL)
 270                                 continue;
 271                         else
 272                                 break;
 273                 }
 274
 275                 if (!buffer_uptodate(bh))
 276                         gfs2_io_error_bh(sdp, bh);
 277
 278                 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
 279         }
 280
 281         return list_empty(&ai->ai_ail1_list);
 282 }
 283
 284 /**
 285  * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
 286  * @sdp: the filesystem
 287  * @ai: the AIL entry
 288  *
 289  */
 290
 291 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 292 {
 293         struct list_head *head = &ai->ai_ail2_list;
 294         struct gfs2_bufdata *bd;
 295
 296         while (!list_empty(head)) {
 297                 bd = list_entry(head->prev, struct gfs2_bufdata,
 298                                 bd_ail_st_list);
 299                 gfs2_assert(sdp, bd->bd_ail == ai);
 300                 bd->bd_ail = NULL;
 301                 list_del(&bd->bd_ail_st_list);
 302                 list_del(&bd->bd_ail_gl_list);
 303                 atomic_dec(&bd->bd_gl->gl_ail_count);
 304                 brelse(bd->bd_bh);
 305         }
 306 }
 307
 308 /**
 309  * ail_empty_gl - remove all buffers for a given lock from the AIL
 310  * @gl: the glock
 311  *
 312  * None of the buffers should be dirty, locked, or pinned.
 313  */
 314
 315 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
 316 {
 317         struct gfs2_sbd *sdp = gl->gl_sbd;
 318         unsigned int blocks;
 319         struct list_head *head = &gl->gl_ail_list;
 320         struct gfs2_bufdata *bd;
 321         struct buffer_head *bh;
 322         uint64_t blkno;
 323         int error;
 324
 325         blocks = atomic_read(&gl->gl_ail_count);
 326         if (!blocks)
 327                 return;
 328
 329         error = gfs2_trans_begin(sdp, 0, blocks);
 330         if (gfs2_assert_withdraw(sdp, !error))
 331                 return;
 332
 333         gfs2_log_lock(sdp);
 334         while (!list_empty(head)) {
 335                 bd = list_entry(head->next, struct gfs2_bufdata,
 336                                 bd_ail_gl_list);
 337                 bh = bd->bd_bh;
 338                 blkno = bh->b_blocknr;
 339                 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
 340
 341                 bd->bd_ail = NULL;
 342                 list_del(&bd->bd_ail_st_list);
 343                 list_del(&bd->bd_ail_gl_list);
 344                 atomic_dec(&gl->gl_ail_count);
 345                 brelse(bh);
 346                 gfs2_log_unlock(sdp);
 347
 348                 gfs2_trans_add_revoke(sdp, blkno);
 349
 350                 gfs2_log_lock(sdp);
 351         }
 352         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 353         gfs2_log_unlock(sdp);
 354
 355         gfs2_trans_end(sdp);
 356         gfs2_log_flush(sdp);
 357 }
 358
 359 /**
 360  * gfs2_meta_inval - Invalidate all buffers associated with a glock
 361  * @gl: the glock
 362  *
 363  */
 364
 365 void gfs2_meta_inval(struct gfs2_glock *gl)
 366 {
 367         struct gfs2_sbd *sdp = gl->gl_sbd;
 368         struct inode *aspace = gl->gl_aspace;
 369         struct address_space *mapping = gl->gl_aspace->i_mapping;
 370
 371         gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
 372
 373         atomic_inc(&aspace->i_writecount);
 374         truncate_inode_pages(mapping, 0);
 375         atomic_dec(&aspace->i_writecount);
 376
 377         gfs2_assert_withdraw(sdp, !mapping->nrpages);
 378 }
 379
 380 /**
 381  * gfs2_meta_sync - Sync all buffers associated with a glock
 382  * @gl: The glock
 383  * @flags: DIO_START | DIO_WAIT
 384  *
 385  */
 386
 387 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
 388 {
 389         struct address_space *mapping = gl->gl_aspace->i_mapping;
 390         int error = 0;
 391
 392         if (flags & DIO_START)
 393                 filemap_fdatawrite(mapping);
 394         if (!error && (flags & DIO_WAIT))
 395                 error = filemap_fdatawait(mapping);
 396
 397         if (error)
 398                 gfs2_io_error(gl->gl_sbd);
 399 }
 400
 401 /**
 402  * getbuf - Get a buffer with a given address space
 403  * @sdp: the filesystem
 404  * @aspace: the address space
 405  * @blkno: the block number (filesystem scope)
 406  * @create: 1 if the buffer should be created
 407  *
 408  * Returns: the buffer
 409  */
 410
 411 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
 412                                   uint64_t blkno, int create)
 413 {
 414         struct page *page;
 415         struct buffer_head *bh;
 416         unsigned int shift;
 417         unsigned long index;
 418         unsigned int bufnum;
 419
 420         shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
 421         index = blkno >> shift;             /* convert block to page */
 422         bufnum = blkno - (index << shift);  /* block buf index within page */
 423
 424         if (create) {
 425                 for (;;) {
 426                         page = grab_cache_page(aspace->i_mapping, index);
 427                         if (page)
 428                                 break;
 429                         yield();
 430                 }
 431         } else {
 432                 page = find_lock_page(aspace->i_mapping, index);
 433                 if (!page)
 434                         return NULL;
 435         }
 436
 437         if (!page_has_buffers(page))
 438                 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
 439
 440         /* Locate header for our buffer within our page */
 441         for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
 442                 /* Do nothing */;
 443         get_bh(bh);
 444
 445         if (!buffer_mapped(bh))
 446                 map_bh(bh, sdp->sd_vfs, blkno);
 447
 448         unlock_page(page);
 449         mark_page_accessed(page);
 450         page_cache_release(page);
 451
 452         return bh;
 453 }
 454
 455 static void meta_prep_new(struct buffer_head *bh)
 456 {
 457         struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
 458
 459         lock_buffer(bh);
 460         clear_buffer_dirty(bh);
 461         set_buffer_uptodate(bh);
 462         unlock_buffer(bh);
 463
 464         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 465 }
 466
 467 /**
 468  * gfs2_meta_new - Get a block
 469  * @gl: The glock associated with this block
 470  * @blkno: The block number
 471  *
 472  * Returns: The buffer
 473  */
 474
 475 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
 476 {
 477         struct buffer_head *bh;
 478         bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
 479         meta_prep_new(bh);
 480         return bh;
 481 }
 482
 483 /**
 484  * gfs2_meta_read - Read a block from disk
 485  * @gl: The glock covering the block
 486  * @blkno: The block number
 487  * @flags: flags to gfs2_dreread()
 488  * @bhp: the place where the buffer is returned (NULL on failure)
 489  *
 490  * Returns: errno
 491  */
 492
 493 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
 494                    struct buffer_head **bhp)
 495 {
 496         int error;
 497
 498         *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
 499         error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
 500         if (error)
 501                 brelse(*bhp);
 502
 503         return error;
 504 }
 505
 506 /**
 507  * gfs2_meta_reread - Reread a block from disk
 508  * @sdp: the filesystem
 509  * @bh: The block to read
 510  * @flags: Flags that control the read
 511  *
 512  * Returns: errno
 513  */
 514
 515 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
 516 {
 517         if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 518                 return -EIO;
 519
 520         if (flags & DIO_FORCE)
 521                 clear_buffer_uptodate(bh);
 522
 523         if ((flags & DIO_START) && !buffer_uptodate(bh))
 524                 ll_rw_block(READ, 1, &bh);
 525
 526         if (flags & DIO_WAIT) {
 527                 wait_on_buffer(bh);
 528
 529                 if (!buffer_uptodate(bh)) {
 530                         struct gfs2_trans *tr = current->journal_info;
 531                         if (tr && tr->tr_touched)
 532                                 gfs2_io_error_bh(sdp, bh);
 533                         return -EIO;
 534                 }
 535                 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
 536                         return -EIO;
 537         }
 538
 539         return 0;
 540 }
 541
 542 /**
 543  * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
 544  * @gl: the glock the buffer belongs to
 545  * @bh: The buffer to be attached to
 546  * @meta: Flag to indicate whether its metadata or not
 547  */
 548
 549 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
 550                          int meta)
 551 {
 552         struct gfs2_bufdata *bd;
 553
 554         if (meta)
 555                 lock_page(bh->b_page);
 556
 557         if (bh->b_private) {
 558                 if (meta)
 559                         unlock_page(bh->b_page);
 560                 return;
 561         }
 562
 563         bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
 564         memset(bd, 0, sizeof(struct gfs2_bufdata));
 565
 566         bd->bd_bh = bh;
 567         bd->bd_gl = gl;
 568
 569         INIT_LIST_HEAD(&bd->bd_list_tr);
 570         if (meta) {
 571                 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
 572         } else {
 573                 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
 574                 get_bh(bh);
 575         }
 576         bh->b_private = bd;
 577
 578         if (meta)
 579                 unlock_page(bh->b_page);
 580 }
 581
 582 /**
 583  * gfs2_pin - Pin a buffer in memory
 584  * @sdp: the filesystem the buffer belongs to
 585  * @bh: The buffer to be pinned
 586  *
 587  */
 588
 589 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
 590 {
 591         struct gfs2_bufdata *bd = bh->b_private;
 592
 593         gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
 594
 595         if (test_set_buffer_pinned(bh))
 596                 gfs2_assert_withdraw(sdp, 0);
 597
 598         wait_on_buffer(bh);
 599
 600         /* If this buffer is in the AIL and it has already been written
 601            to in-place disk block, remove it from the AIL. */
 602
 603         gfs2_log_lock(sdp);
 604         if (bd->bd_ail && !buffer_in_io(bh))
 605                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
 606         gfs2_log_unlock(sdp);
 607
 608         clear_buffer_dirty(bh);
 609         wait_on_buffer(bh);
 610
 611         if (!buffer_uptodate(bh))
 612                 gfs2_io_error_bh(sdp, bh);
 613
 614         get_bh(bh);
 615 }
 616
 617 /**
 618  * gfs2_unpin - Unpin a buffer
 619  * @sdp: the filesystem the buffer belongs to
 620  * @bh: The buffer to unpin
 621  * @ai:
 622  *
 623  */
 624
 625 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 626                 struct gfs2_ail *ai)
 627 {
 628         struct gfs2_bufdata *bd = bh->b_private;
 629
 630         gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
 631
 632         if (!buffer_pinned(bh))
 633                 gfs2_assert_withdraw(sdp, 0);
 634
 635         mark_buffer_dirty(bh);
 636         clear_buffer_pinned(bh);
 637
 638         gfs2_log_lock(sdp);
 639         if (bd->bd_ail) {
 640                 list_del(&bd->bd_ail_st_list);
 641                 brelse(bh);
 642         } else {
 643                 struct gfs2_glock *gl = bd->bd_gl;
 644                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 645                 atomic_inc(&gl->gl_ail_count);
 646         }
 647         bd->bd_ail = ai;
 648         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 649         gfs2_log_unlock(sdp);
 650 }
 651
 652 /**
 653  * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
 654  * @ip: the inode who owns the buffers
 655  * @bstart: the first buffer in the run
 656  * @blen: the number of buffers in the run
 657  *
 658  */
 659
 660 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
 661 {
 662         struct gfs2_sbd *sdp = ip->i_sbd;
 663         struct inode *aspace = ip->i_gl->gl_aspace;
 664         struct buffer_head *bh;
 665
 666         while (blen) {
 667                 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
 668                 if (bh) {
 669                         struct gfs2_bufdata *bd = bh->b_private;
 670
 671                         if (test_clear_buffer_pinned(bh)) {
 672                                 struct gfs2_trans *tr = current->journal_info;
 673                                 gfs2_log_lock(sdp);
 674                                 list_del_init(&bd->bd_le.le_list);
 675                                 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
 676                                 sdp->sd_log_num_buf--;
 677                                 gfs2_log_unlock(sdp);
 678                                 tr->tr_num_buf_rm++;
 679                                 brelse(bh);
 680                         }
 681                         if (bd) {
 682                                 gfs2_log_lock(sdp);
 683                                 if (bd->bd_ail) {
 684                                         uint64_t blkno = bh->b_blocknr;
 685                                         bd->bd_ail = NULL;
 686                                         list_del(&bd->bd_ail_st_list);
 687                                         list_del(&bd->bd_ail_gl_list);
 688                                         atomic_dec(&bd->bd_gl->gl_ail_count);
 689                                         brelse(bh);
 690                                         gfs2_log_unlock(sdp);
 691                                         gfs2_trans_add_revoke(sdp, blkno);
 692                                 } else
 693                                         gfs2_log_unlock(sdp);
 694                         }
 695
 696                         lock_buffer(bh);
 697                         clear_buffer_dirty(bh);
 698                         clear_buffer_uptodate(bh);
 699                         unlock_buffer(bh);
 700
 701                         brelse(bh);
 702                 }
 703
 704                 bstart++;
 705                 blen--;
 706         }
 707 }
 708
 709 /**
 710  * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
 711  * @ip: The GFS2 inode
 712  *
 713  * This releases buffers that are in the most-recently-used array of
 714  * blocks used for indirect block addressing for this inode.
 715  */
 716
 717 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
 718 {
 719         struct buffer_head **bh_slot;
 720         unsigned int x;
 721
 722         spin_lock(&ip->i_spin);
 723
 724         for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
 725                 bh_slot = &ip->i_cache[x];
 726                 if (!*bh_slot)
 727                         break;
 728                 brelse(*bh_slot);
 729                 *bh_slot = NULL;
 730         }
 731
 732         spin_unlock(&ip->i_spin);
 733 }
 734
 735 /**
 736  * gfs2_meta_indirect_buffer - Get a metadata buffer
 737  * @ip: The GFS2 inode
 738  * @height: The level of this buf in the metadata (indir addr) tree (if any)
 739  * @num: The block number (device relative) of the buffer
 740  * @new: Non-zero if we may create a new buffer
 741  * @bhp: the buffer is returned here
 742  *
 743  * Try to use the gfs2_inode's MRU metadata tree cache.
 744  *
 745  * Returns: errno
 746  */
 747
 748 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
 749                               int new, struct buffer_head **bhp)
 750 {
 751         struct buffer_head *bh, **bh_slot = ip->i_cache + height;
 752         int error;
 753
 754         spin_lock(&ip->i_spin);
 755         bh = *bh_slot;
 756         if (bh) {
 757                 if (bh->b_blocknr == num)
 758                         get_bh(bh);
 759                 else
 760                         bh = NULL;
 761         }
 762         spin_unlock(&ip->i_spin);
 763
 764         if (bh) {
 765                 if (new)
 766                         meta_prep_new(bh);
 767                 else {
 768                         error = gfs2_meta_reread(ip->i_sbd, bh,
 769                                                  DIO_START | DIO_WAIT);
 770                         if (error) {
 771                                 brelse(bh);
 772                                 return error;
 773                         }
 774                 }
 775         } else {
 776                 if (new)
 777                         bh = gfs2_meta_new(ip->i_gl, num);
 778                 else {
 779                         error = gfs2_meta_read(ip->i_gl, num,
 780                                                DIO_START | DIO_WAIT, &bh);
 781                         if (error)
 782                                 return error;
 783                 }
 784
 785                 spin_lock(&ip->i_spin);
 786                 if (*bh_slot != bh) {
 787                         brelse(*bh_slot);
 788                         *bh_slot = bh;
 789                         get_bh(bh);
 790                 }
 791                 spin_unlock(&ip->i_spin);
 792         }
 793
 794         if (new) {
 795                 if (gfs2_assert_warn(ip->i_sbd, height)) {
 796                         brelse(bh);
 797                         return -EIO;
 798                 }
 799                 gfs2_trans_add_bh(ip->i_gl, bh, 1);
 800                 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
 801                 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
 802
 803         } else if (gfs2_metatype_check(ip->i_sbd, bh,
 804                              (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
 805                 brelse(bh);
 806                 return -EIO;
 807         }
 808
 809         *bhp = bh;
 810
 811         return 0;
 812 }
 813
 814 /**
 815  * gfs2_meta_ra - start readahead on an extent of a file
 816  * @gl: the glock the blocks belong to
 817  * @dblock: the starting disk block
 818  * @extlen: the number of blocks in the extent
 819  *
 820  */
 821
 822 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
 823 {
 824         struct gfs2_sbd *sdp = gl->gl_sbd;
 825         struct inode *aspace = gl->gl_aspace;
 826         struct buffer_head *first_bh, *bh;
 827         uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
 828                           sdp->sd_sb.sb_bsize_shift;
 829         int error;
 830
 831         if (!extlen || !max_ra)
 832                 return;
 833         if (extlen > max_ra)
 834                 extlen = max_ra;
 835
 836         first_bh = getbuf(sdp, aspace, dblock, CREATE);
 837
 838         if (buffer_uptodate(first_bh))
 839                 goto out;
 840         if (!buffer_locked(first_bh)) {
 841                 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
 842                 if (error)
 843                         goto out;
 844         }
 845
 846         dblock++;
 847         extlen--;
 848
 849         while (extlen) {
 850                 bh = getbuf(sdp, aspace, dblock, CREATE);
 851
 852                 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
 853                         error = gfs2_meta_reread(sdp, bh, DIO_START);
 854                         brelse(bh);
 855                         if (error)
 856                                 goto out;
 857                 } else
 858                         brelse(bh);
 859
 860                 dblock++;
 861                 extlen--;
 862
 863                 if (buffer_uptodate(first_bh))
 864                         break;
 865         }
 866
 867  out:
 868         brelse(first_bh);
 869 }
 870
 871 /**
 872  * gfs2_meta_syncfs - sync all the buffers in a filesystem
 873  * @sdp: the filesystem
 874  *
 875  */
 876
 877 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
 878 {
 879         gfs2_log_flush(sdp);
 880         for (;;) {
 881                 gfs2_ail1_start(sdp, DIO_ALL);
 882                 if (gfs2_ail1_empty(sdp, DIO_ALL))
 883                         break;
 884                 msleep(100);
 885         }
 886 }
 887