fs/gfs2/lops.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/mempool.h>
  16 #include <linux/gfs2_ondisk.h>
  17 #include <linux/bio.h>
  18 #include <linux/fs.h>
  19
  20 #include "gfs2.h"
  21 #include "incore.h"
  22 #include "inode.h"
  23 #include "glock.h"
  24 #include "log.h"
  25 #include "lops.h"
  26 #include "meta_io.h"
  27 #include "recovery.h"
  28 #include "rgrp.h"
  29 #include "trans.h"
  30 #include "util.h"
  31 #include "trace_gfs2.h"
  32
  33 /**
  34  * gfs2_pin - Pin a buffer in memory
  35  * @sdp: The superblock
  36  * @bh: The buffer to be pinned
  37  *
  38  * The log lock must be held when calling this function
  39  */
  40 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  41 {
  42         struct gfs2_bufdata *bd;
  43
  44         BUG_ON(!current->journal_info);
  45
  46         clear_buffer_dirty(bh);
  47         if (test_set_buffer_pinned(bh))
  48                 gfs2_assert_withdraw(sdp, 0);
  49         if (!buffer_uptodate(bh))
  50                 gfs2_io_error_bh(sdp, bh);
  51         bd = bh->b_private;
  52         /* If this buffer is in the AIL and it has already been written
  53          * to in-place disk block, remove it from the AIL.
  54          */
  55         spin_lock(&sdp->sd_ail_lock);
  56         if (bd->bd_ail)
  57                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
  58         spin_unlock(&sdp->sd_ail_lock);
  59         get_bh(bh);
  60         atomic_inc(&sdp->sd_log_pinned);
  61         trace_gfs2_pin(bd, 1);
  62 }
  63
  64 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
  65 {
  66         return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
  67 }
  68
  69 static void maybe_release_space(struct gfs2_bufdata *bd)
  70 {
  71         struct gfs2_glock *gl = bd->bd_gl;
  72         struct gfs2_sbd *sdp = gl->gl_sbd;
  73         struct gfs2_rgrpd *rgd = gl->gl_object;
  74         unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
  75         struct gfs2_bitmap *bi = rgd->rd_bits + index;
  76
  77         if (bi->bi_clone == 0)
  78                 return;
  79         if (sdp->sd_args.ar_discard)
  80                 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
  81         memcpy(bi->bi_clone + bi->bi_offset,
  82                bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
  83         clear_bit(GBF_FULL, &bi->bi_flags);
  84         rgd->rd_free_clone = rgd->rd_free;
  85 }
  86
  87 /**
  88  * gfs2_unpin - Unpin a buffer
  89  * @sdp: the filesystem the buffer belongs to
  90  * @bh: The buffer to unpin
  91  * @ai:
  92  * @flags: The inode dirty flags
  93  *
  94  */
  95
  96 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  97                        struct gfs2_ail *ai)
  98 {
  99         struct gfs2_bufdata *bd = bh->b_private;
 100
 101         BUG_ON(!buffer_uptodate(bh));
 102         BUG_ON(!buffer_pinned(bh));
 103
 104         lock_buffer(bh);
 105         mark_buffer_dirty(bh);
 106         clear_buffer_pinned(bh);
 107
 108         if (buffer_is_rgrp(bd))
 109                 maybe_release_space(bd);
 110
 111         spin_lock(&sdp->sd_ail_lock);
 112         if (bd->bd_ail) {
 113                 list_del(&bd->bd_ail_st_list);
 114                 brelse(bh);
 115         } else {
 116                 struct gfs2_glock *gl = bd->bd_gl;
 117                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
 118                 atomic_inc(&gl->gl_ail_count);
 119         }
 120         bd->bd_ail = ai;
 121         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 122         spin_unlock(&sdp->sd_ail_lock);
 123
 124         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 125         trace_gfs2_pin(bd, 0);
 126         unlock_buffer(bh);
 127         atomic_dec(&sdp->sd_log_pinned);
 128 }
 129
 130 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
 131 {
 132         BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
 133                (sdp->sd_log_flush_head != sdp->sd_log_head));
 134
 135         if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks) {
 136                 sdp->sd_log_flush_head = 0;
 137                 sdp->sd_log_flush_wrapped = 1;
 138         }
 139 }
 140
 141 static u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
 142 {
 143         unsigned int lbn = sdp->sd_log_flush_head;
 144         struct gfs2_journal_extent *je;
 145         u64 block;
 146
 147         list_for_each_entry(je, &sdp->sd_jdesc->extent_list, extent_list) {
 148                 if (lbn >= je->lblock && lbn < je->lblock + je->blocks) {
 149                         block = je->dblock + lbn - je->lblock;
 150                         gfs2_log_incr_head(sdp);
 151                         return block;
 152                 }
 153         }
 154
 155         return -1;
 156 }
 157
 158 /**
 159  * gfs2_end_log_write_bh - end log write of pagecache data with buffers
 160  * @sdp: The superblock
 161  * @bvec: The bio_vec
 162  * @error: The i/o status
 163  *
 164  * This finds the relavent buffers and unlocks then and sets the
 165  * error flag according to the status of the i/o request. This is
 166  * used when the log is writing data which has an in-place version
 167  * that is pinned in the pagecache.
 168  */
 169
 170 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp, struct bio_vec *bvec,
 171                                   int error)
 172 {
 173         struct buffer_head *bh, *next;
 174         struct page *page = bvec->bv_page;
 175         unsigned size;
 176
 177         bh = page_buffers(page);
 178         size = bvec->bv_len;
 179         while (bh_offset(bh) < bvec->bv_offset)
 180                 bh = bh->b_this_page;
 181         do {
 182                 if (error)
 183                         set_buffer_write_io_error(bh);
 184                 unlock_buffer(bh);
 185                 next = bh->b_this_page;
 186                 size -= bh->b_size;
 187                 brelse(bh);
 188                 bh = next;
 189         } while(bh && size);
 190 }
 191
 192 /**
 193  * gfs2_end_log_write - end of i/o to the log
 194  * @bio: The bio
 195  * @error: Status of i/o request
 196  *
 197  * Each bio_vec contains either data from the pagecache or data
 198  * relating to the log itself. Here we iterate over the bio_vec
 199  * array, processing both kinds of data.
 200  *
 201  */
 202
 203 static void gfs2_end_log_write(struct bio *bio, int error)
 204 {
 205         struct gfs2_sbd *sdp = bio->bi_private;
 206         struct bio_vec *bvec;
 207         struct page *page;
 208         int i;
 209
 210         if (error) {
 211                 sdp->sd_log_error = error;
 212                 fs_err(sdp, "Error %d writing to log\n", error);
 213         }
 214
 215         bio_for_each_segment(bvec, bio, i) {
 216                 page = bvec->bv_page;
 217                 if (page_has_buffers(page))
 218                         gfs2_end_log_write_bh(sdp, bvec, error);
 219                 else
 220                         mempool_free(page, gfs2_page_pool);
 221         }
 222
 223         bio_put(bio);
 224         if (atomic_dec_and_test(&sdp->sd_log_in_flight))
 225                 wake_up(&sdp->sd_log_flush_wait);
 226 }
 227
 228 /**
 229  * gfs2_log_flush_bio - Submit any pending log bio
 230  * @sdp: The superblock
 231  * @rw: The rw flags
 232  *
 233  * Submit any pending part-built or full bio to the block device. If
 234  * there is no pending bio, then this is a no-op.
 235  */
 236
 237 void gfs2_log_flush_bio(struct gfs2_sbd *sdp, int rw)
 238 {
 239         if (sdp->sd_log_bio) {
 240                 atomic_inc(&sdp->sd_log_in_flight);
 241                 submit_bio(rw, sdp->sd_log_bio);
 242                 sdp->sd_log_bio = NULL;
 243         }
 244 }
 245
 246 /**
 247  * gfs2_log_alloc_bio - Allocate a new bio for log writing
 248  * @sdp: The superblock
 249  * @blkno: The next device block number we want to write to
 250  *
 251  * This should never be called when there is a cached bio in the
 252  * super block. When it returns, there will be a cached bio in the
 253  * super block which will have as many bio_vecs as the device is
 254  * happy to handle.
 255  *
 256  * Returns: Newly allocated bio
 257  */
 258
 259 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno)
 260 {
 261         struct super_block *sb = sdp->sd_vfs;
 262         unsigned nrvecs = bio_get_nr_vecs(sb->s_bdev);
 263         struct bio *bio;
 264
 265         BUG_ON(sdp->sd_log_bio);
 266
 267         while (1) {
 268                 bio = bio_alloc(GFP_NOIO, nrvecs);
 269                 if (likely(bio))
 270                         break;
 271                 nrvecs = max(nrvecs/2, 1U);
 272         }
 273
 274         bio->bi_sector = blkno * (sb->s_blocksize >> 9);
 275         bio->bi_bdev = sb->s_bdev;
 276         bio->bi_end_io = gfs2_end_log_write;
 277         bio->bi_private = sdp;
 278
 279         sdp->sd_log_bio = bio;
 280
 281         return bio;
 282 }
 283
 284 /**
 285  * gfs2_log_get_bio - Get cached log bio, or allocate a new one
 286  * @sdp: The superblock
 287  * @blkno: The device block number we want to write to
 288  *
 289  * If there is a cached bio, then if the next block number is sequential
 290  * with the previous one, return it, otherwise flush the bio to the
 291  * device. If there is not a cached bio, or we just flushed it, then
 292  * allocate a new one.
 293  *
 294  * Returns: The bio to use for log writes
 295  */
 296
 297 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno)
 298 {
 299         struct bio *bio = sdp->sd_log_bio;
 300         u64 nblk;
 301
 302         if (bio) {
 303                 nblk = bio->bi_sector + bio_sectors(bio);
 304                 nblk >>= sdp->sd_fsb2bb_shift;
 305                 if (blkno == nblk)
 306                         return bio;
 307                 gfs2_log_flush_bio(sdp, WRITE);
 308         }
 309
 310         return gfs2_log_alloc_bio(sdp, blkno);
 311 }
 312
 313
 314 /**
 315  * gfs2_log_write - write to log
 316  * @sdp: the filesystem
 317  * @page: the page to write
 318  * @size: the size of the data to write
 319  * @offset: the offset within the page
 320  *
 321  * Try and add the page segment to the current bio. If that fails,
 322  * submit the current bio to the device and create a new one, and
 323  * then add the page segment to that.
 324  */
 325
 326 static void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
 327                            unsigned size, unsigned offset)
 328 {
 329         u64 blkno = gfs2_log_bmap(sdp);
 330         struct bio *bio;
 331         int ret;
 332
 333         bio = gfs2_log_get_bio(sdp, blkno);
 334         ret = bio_add_page(bio, page, size, offset);
 335         if (ret == 0) {
 336                 gfs2_log_flush_bio(sdp, WRITE);
 337                 bio = gfs2_log_alloc_bio(sdp, blkno);
 338                 ret = bio_add_page(bio, page, size, offset);
 339                 WARN_ON(ret == 0);
 340         }
 341 }
 342
 343 /**
 344  * gfs2_log_write_bh - write a buffer's content to the log
 345  * @sdp: The super block
 346  * @bh: The buffer pointing to the in-place location
 347  *
 348  * This writes the content of the buffer to the next available location
 349  * in the log. The buffer will be unlocked once the i/o to the log has
 350  * completed.
 351  */
 352
 353 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
 354 {
 355         gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh));
 356 }
 357
 358 /**
 359  * gfs2_log_write_page - write one block stored in a page, into the log
 360  * @sdp: The superblock
 361  * @page: The struct page
 362  *
 363  * This writes the first block-sized part of the page into the log. Note
 364  * that the page must have been allocated from the gfs2_page_pool mempool
 365  * and that after this has been called, ownership has been transferred and
 366  * the page may be freed at any time.
 367  */
 368
 369 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
 370 {
 371         struct super_block *sb = sdp->sd_vfs;
 372         gfs2_log_write(sdp, page, sb->s_blocksize, 0);
 373 }
 374
 375 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
 376                                       u32 ld_length, u32 ld_data1)
 377 {
 378         struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 379         struct gfs2_log_descriptor *ld = page_address(page);
 380         clear_page(ld);
 381         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 382         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 383         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 384         ld->ld_type = cpu_to_be32(ld_type);
 385         ld->ld_length = cpu_to_be32(ld_length);
 386         ld->ld_data1 = cpu_to_be32(ld_data1);
 387         ld->ld_data2 = 0;
 388         return page;
 389 }
 390
 391 static void gfs2_check_magic(struct buffer_head *bh)
 392 {
 393         void *kaddr;
 394         __be32 *ptr;
 395
 396         clear_buffer_escaped(bh);
 397         kaddr = kmap_atomic(bh->b_page);
 398         ptr = kaddr + bh_offset(bh);
 399         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 400                 set_buffer_escaped(bh);
 401         kunmap_atomic(kaddr);
 402 }
 403
 404 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
 405                                 unsigned int total, struct list_head *blist,
 406                                 bool is_databuf)
 407 {
 408         struct gfs2_log_descriptor *ld;
 409         struct gfs2_bufdata *bd1 = NULL, *bd2;
 410         struct page *page;
 411         unsigned int num;
 412         unsigned n;
 413         __be64 *ptr;
 414
 415         gfs2_log_lock(sdp);
 416         bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
 417         while(total) {
 418                 num = total;
 419                 if (total > limit)
 420                         num = limit;
 421                 gfs2_log_unlock(sdp);
 422                 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA, num + 1, num);
 423                 ld = page_address(page);
 424                 gfs2_log_lock(sdp);
 425                 ptr = (__be64 *)(ld + 1);
 426
 427                 n = 0;
 428                 list_for_each_entry_continue(bd1, blist, bd_list) {
 429                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 430                         if (is_databuf) {
 431                                 gfs2_check_magic(bd1->bd_bh);
 432                                 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
 433                         }
 434                         if (++n >= num)
 435                                 break;
 436                 }
 437
 438                 gfs2_log_unlock(sdp);
 439                 gfs2_log_write_page(sdp, page);
 440                 gfs2_log_lock(sdp);
 441
 442                 n = 0;
 443                 list_for_each_entry_continue(bd2, blist, bd_list) {
 444                         get_bh(bd2->bd_bh);
 445                         gfs2_log_unlock(sdp);
 446                         lock_buffer(bd2->bd_bh);
 447
 448                         if (buffer_escaped(bd2->bd_bh)) {
 449                                 void *kaddr;
 450                                 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 451                                 ptr = page_address(page);
 452                                 kaddr = kmap_atomic(bd2->bd_bh->b_page);
 453                                 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
 454                                        bd2->bd_bh->b_size);
 455                                 kunmap_atomic(kaddr);
 456                                 *(__be32 *)ptr = 0;
 457                                 clear_buffer_escaped(bd2->bd_bh);
 458                                 unlock_buffer(bd2->bd_bh);
 459                                 brelse(bd2->bd_bh);
 460                                 gfs2_log_write_page(sdp, page);
 461                         } else {
 462                                 gfs2_log_write_bh(sdp, bd2->bd_bh);
 463                         }
 464                         gfs2_log_lock(sdp);
 465                         if (++n >= num)
 466                                 break;
 467                 }
 468
 469                 BUG_ON(total < num);
 470                 total -= num;
 471         }
 472         gfs2_log_unlock(sdp);
 473 }
 474
 475 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 476 {
 477         unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
 478
 479         gfs2_before_commit(sdp, limit, sdp->sd_log_num_buf,
 480                            &sdp->sd_log_le_buf, 0);
 481 }
 482
 483 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 484 {
 485         struct list_head *head = &sdp->sd_log_le_buf;
 486         struct gfs2_bufdata *bd;
 487
 488         while (!list_empty(head)) {
 489                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 490                 list_del_init(&bd->bd_list);
 491                 sdp->sd_log_num_buf--;
 492
 493                 gfs2_unpin(sdp, bd->bd_bh, ai);
 494         }
 495         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 496 }
 497
 498 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 499                                struct gfs2_log_header_host *head, int pass)
 500 {
 501         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 502
 503         if (pass != 0)
 504                 return;
 505
 506         sdp->sd_found_blocks = 0;
 507         sdp->sd_replayed_blocks = 0;
 508 }
 509
 510 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 511                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 512                                 int pass)
 513 {
 514         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 515         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 516         struct gfs2_glock *gl = ip->i_gl;
 517         unsigned int blks = be32_to_cpu(ld->ld_data1);
 518         struct buffer_head *bh_log, *bh_ip;
 519         u64 blkno;
 520         int error = 0;
 521
 522         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 523                 return 0;
 524
 525         gfs2_replay_incr_blk(sdp, &start);
 526
 527         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 528                 blkno = be64_to_cpu(*ptr++);
 529
 530                 sdp->sd_found_blocks++;
 531
 532                 if (gfs2_revoke_check(sdp, blkno, start))
 533                         continue;
 534
 535                 error = gfs2_replay_read_block(jd, start, &bh_log);
 536                 if (error)
 537                         return error;
 538
 539                 bh_ip = gfs2_meta_new(gl, blkno);
 540                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 541
 542                 if (gfs2_meta_check(sdp, bh_ip))
 543                         error = -EIO;
 544                 else
 545                         mark_buffer_dirty(bh_ip);
 546
 547                 brelse(bh_log);
 548                 brelse(bh_ip);
 549
 550                 if (error)
 551                         break;
 552
 553                 sdp->sd_replayed_blocks++;
 554         }
 555
 556         return error;
 557 }
 558
 559 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 560 {
 561         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 562         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 563
 564         if (error) {
 565                 gfs2_meta_sync(ip->i_gl);
 566                 return;
 567         }
 568         if (pass != 1)
 569                 return;
 570
 571         gfs2_meta_sync(ip->i_gl);
 572
 573         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 574                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 575 }
 576
 577 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 578 {
 579         struct gfs2_meta_header *mh;
 580         unsigned int offset;
 581         struct list_head *head = &sdp->sd_log_le_revoke;
 582         struct gfs2_bufdata *bd;
 583         struct page *page;
 584         unsigned int length;
 585
 586         if (!sdp->sd_log_num_revoke)
 587                 return;
 588
 589         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 590         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 591         offset = sizeof(struct gfs2_log_descriptor);
 592
 593         list_for_each_entry(bd, head, bd_list) {
 594                 sdp->sd_log_num_revoke--;
 595
 596                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 597
 598                         gfs2_log_write_page(sdp, page);
 599                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 600                         mh = page_address(page);
 601                         clear_page(mh);
 602                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 603                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 604                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 605                         offset = sizeof(struct gfs2_meta_header);
 606                 }
 607
 608                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 609                 offset += sizeof(u64);
 610         }
 611         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 612
 613         gfs2_log_write_page(sdp, page);
 614 }
 615
 616 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 617 {
 618         struct list_head *head = &sdp->sd_log_le_revoke;
 619         struct gfs2_bufdata *bd;
 620         struct gfs2_glock *gl;
 621
 622         while (!list_empty(head)) {
 623                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 624                 list_del_init(&bd->bd_list);
 625                 gl = bd->bd_gl;
 626                 atomic_dec(&gl->gl_revokes);
 627                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 628                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 629         }
 630 }
 631
 632 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 633                                   struct gfs2_log_header_host *head, int pass)
 634 {
 635         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 636
 637         if (pass != 0)
 638                 return;
 639
 640         sdp->sd_found_revokes = 0;
 641         sdp->sd_replay_tail = head->lh_tail;
 642 }
 643
 644 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 645                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 646                                    int pass)
 647 {
 648         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 649         unsigned int blks = be32_to_cpu(ld->ld_length);
 650         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 651         struct buffer_head *bh;
 652         unsigned int offset;
 653         u64 blkno;
 654         int first = 1;
 655         int error;
 656
 657         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 658                 return 0;
 659
 660         offset = sizeof(struct gfs2_log_descriptor);
 661
 662         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 663                 error = gfs2_replay_read_block(jd, start, &bh);
 664                 if (error)
 665                         return error;
 666
 667                 if (!first)
 668                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 669
 670                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 671                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 672
 673                         error = gfs2_revoke_add(sdp, blkno, start);
 674                         if (error < 0) {
 675                                 brelse(bh);
 676                                 return error;
 677                         }
 678                         else if (error)
 679                                 sdp->sd_found_revokes++;
 680
 681                         if (!--revokes)
 682                                 break;
 683                         offset += sizeof(u64);
 684                 }
 685
 686                 brelse(bh);
 687                 offset = sizeof(struct gfs2_meta_header);
 688                 first = 0;
 689         }
 690
 691         return 0;
 692 }
 693
 694 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 695 {
 696         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 697
 698         if (error) {
 699                 gfs2_revoke_clean(sdp);
 700                 return;
 701         }
 702         if (pass != 1)
 703                 return;
 704
 705         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 706                 jd->jd_jid, sdp->sd_found_revokes);
 707
 708         gfs2_revoke_clean(sdp);
 709 }
 710
 711 /**
 712  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 713  *
 714  */
 715
 716 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 717 {
 718         unsigned int limit = buf_limit(sdp) / 2;
 719
 720         gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf,
 721                            &sdp->sd_log_le_databuf, 1);
 722 }
 723
 724 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 725                                     struct gfs2_log_descriptor *ld,
 726                                     __be64 *ptr, int pass)
 727 {
 728         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 729         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 730         struct gfs2_glock *gl = ip->i_gl;
 731         unsigned int blks = be32_to_cpu(ld->ld_data1);
 732         struct buffer_head *bh_log, *bh_ip;
 733         u64 blkno;
 734         u64 esc;
 735         int error = 0;
 736
 737         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 738                 return 0;
 739
 740         gfs2_replay_incr_blk(sdp, &start);
 741         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 742                 blkno = be64_to_cpu(*ptr++);
 743                 esc = be64_to_cpu(*ptr++);
 744
 745                 sdp->sd_found_blocks++;
 746
 747                 if (gfs2_revoke_check(sdp, blkno, start))
 748                         continue;
 749
 750                 error = gfs2_replay_read_block(jd, start, &bh_log);
 751                 if (error)
 752                         return error;
 753
 754                 bh_ip = gfs2_meta_new(gl, blkno);
 755                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 756
 757                 /* Unescape */
 758                 if (esc) {
 759                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 760                         *eptr = cpu_to_be32(GFS2_MAGIC);
 761                 }
 762                 mark_buffer_dirty(bh_ip);
 763
 764                 brelse(bh_log);
 765                 brelse(bh_ip);
 766
 767                 sdp->sd_replayed_blocks++;
 768         }
 769
 770         return error;
 771 }
 772
 773 /* FIXME: sort out accounting for log blocks etc. */
 774
 775 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 776 {
 777         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 778         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 779
 780         if (error) {
 781                 gfs2_meta_sync(ip->i_gl);
 782                 return;
 783         }
 784         if (pass != 1)
 785                 return;
 786
 787         /* data sync? */
 788         gfs2_meta_sync(ip->i_gl);
 789
 790         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 791                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 792 }
 793
 794 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 795 {
 796         struct list_head *head = &sdp->sd_log_le_databuf;
 797         struct gfs2_bufdata *bd;
 798
 799         while (!list_empty(head)) {
 800                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 801                 list_del_init(&bd->bd_list);
 802                 sdp->sd_log_num_databuf--;
 803                 gfs2_unpin(sdp, bd->bd_bh, ai);
 804         }
 805         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 806 }
 807
 808
 809 const struct gfs2_log_operations gfs2_buf_lops = {
 810         .lo_before_commit = buf_lo_before_commit,
 811         .lo_after_commit = buf_lo_after_commit,
 812         .lo_before_scan = buf_lo_before_scan,
 813         .lo_scan_elements = buf_lo_scan_elements,
 814         .lo_after_scan = buf_lo_after_scan,
 815         .lo_name = "buf",
 816 };
 817
 818 const struct gfs2_log_operations gfs2_revoke_lops = {
 819         .lo_before_commit = revoke_lo_before_commit,
 820         .lo_after_commit = revoke_lo_after_commit,
 821         .lo_before_scan = revoke_lo_before_scan,
 822         .lo_scan_elements = revoke_lo_scan_elements,
 823         .lo_after_scan = revoke_lo_after_scan,
 824         .lo_name = "revoke",
 825 };
 826
 827 const struct gfs2_log_operations gfs2_rg_lops = {
 828         .lo_name = "rg",
 829 };
 830
 831 const struct gfs2_log_operations gfs2_databuf_lops = {
 832         .lo_before_commit = databuf_lo_before_commit,
 833         .lo_after_commit = databuf_lo_after_commit,
 834         .lo_scan_elements = databuf_lo_scan_elements,
 835         .lo_after_scan = databuf_lo_after_scan,
 836         .lo_name = "databuf",
 837 };
 838
 839 const struct gfs2_log_operations *gfs2_log_ops[] = {
 840         &gfs2_databuf_lops,
 841         &gfs2_buf_lops,
 842         &gfs2_rg_lops,
 843         &gfs2_revoke_lops,
 844         NULL,
 845 };
 846