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_tr)
  57                 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_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_trans *tr)
  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_tr) {
 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_tr = tr;
 121         list_add(&bd->bd_ail_st_list, &tr->tr_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_trans *tr)
 484 {
 485         struct list_head *head = &sdp->sd_log_le_buf;
 486         struct gfs2_bufdata *bd;
 487
 488         if (tr == NULL) {
 489                 gfs2_assert(sdp, list_empty(head));
 490                 return;
 491         }
 492
 493         while (!list_empty(head)) {
 494                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 495                 list_del_init(&bd->bd_list);
 496                 sdp->sd_log_num_buf--;
 497
 498                 gfs2_unpin(sdp, bd->bd_bh, tr);
 499         }
 500         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 501 }
 502
 503 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 504                                struct gfs2_log_header_host *head, int pass)
 505 {
 506         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 507
 508         if (pass != 0)
 509                 return;
 510
 511         sdp->sd_found_blocks = 0;
 512         sdp->sd_replayed_blocks = 0;
 513 }
 514
 515 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 516                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 517                                 int pass)
 518 {
 519         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 520         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 521         struct gfs2_glock *gl = ip->i_gl;
 522         unsigned int blks = be32_to_cpu(ld->ld_data1);
 523         struct buffer_head *bh_log, *bh_ip;
 524         u64 blkno;
 525         int error = 0;
 526
 527         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 528                 return 0;
 529
 530         gfs2_replay_incr_blk(sdp, &start);
 531
 532         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 533                 blkno = be64_to_cpu(*ptr++);
 534
 535                 sdp->sd_found_blocks++;
 536
 537                 if (gfs2_revoke_check(sdp, blkno, start))
 538                         continue;
 539
 540                 error = gfs2_replay_read_block(jd, start, &bh_log);
 541                 if (error)
 542                         return error;
 543
 544                 bh_ip = gfs2_meta_new(gl, blkno);
 545                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 546
 547                 if (gfs2_meta_check(sdp, bh_ip))
 548                         error = -EIO;
 549                 else
 550                         mark_buffer_dirty(bh_ip);
 551
 552                 brelse(bh_log);
 553                 brelse(bh_ip);
 554
 555                 if (error)
 556                         break;
 557
 558                 sdp->sd_replayed_blocks++;
 559         }
 560
 561         return error;
 562 }
 563
 564 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 565 {
 566         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 567         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 568
 569         if (error) {
 570                 gfs2_meta_sync(ip->i_gl);
 571                 return;
 572         }
 573         if (pass != 1)
 574                 return;
 575
 576         gfs2_meta_sync(ip->i_gl);
 577
 578         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 579                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 580 }
 581
 582 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 583 {
 584         struct gfs2_meta_header *mh;
 585         unsigned int offset;
 586         struct list_head *head = &sdp->sd_log_le_revoke;
 587         struct gfs2_bufdata *bd;
 588         struct page *page;
 589         unsigned int length;
 590
 591         if (!sdp->sd_log_num_revoke)
 592                 return;
 593
 594         length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
 595         page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
 596         offset = sizeof(struct gfs2_log_descriptor);
 597
 598         list_for_each_entry(bd, head, bd_list) {
 599                 sdp->sd_log_num_revoke--;
 600
 601                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 602
 603                         gfs2_log_write_page(sdp, page);
 604                         page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
 605                         mh = page_address(page);
 606                         clear_page(mh);
 607                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 608                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 609                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 610                         offset = sizeof(struct gfs2_meta_header);
 611                 }
 612
 613                 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
 614                 offset += sizeof(u64);
 615         }
 616         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 617
 618         gfs2_log_write_page(sdp, page);
 619 }
 620
 621 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 622 {
 623         struct list_head *head = &sdp->sd_log_le_revoke;
 624         struct gfs2_bufdata *bd;
 625         struct gfs2_glock *gl;
 626
 627         while (!list_empty(head)) {
 628                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 629                 list_del_init(&bd->bd_list);
 630                 gl = bd->bd_gl;
 631                 atomic_dec(&gl->gl_revokes);
 632                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 633                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 634         }
 635 }
 636
 637 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 638                                   struct gfs2_log_header_host *head, int pass)
 639 {
 640         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 641
 642         if (pass != 0)
 643                 return;
 644
 645         sdp->sd_found_revokes = 0;
 646         sdp->sd_replay_tail = head->lh_tail;
 647 }
 648
 649 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 650                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 651                                    int pass)
 652 {
 653         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 654         unsigned int blks = be32_to_cpu(ld->ld_length);
 655         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 656         struct buffer_head *bh;
 657         unsigned int offset;
 658         u64 blkno;
 659         int first = 1;
 660         int error;
 661
 662         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 663                 return 0;
 664
 665         offset = sizeof(struct gfs2_log_descriptor);
 666
 667         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 668                 error = gfs2_replay_read_block(jd, start, &bh);
 669                 if (error)
 670                         return error;
 671
 672                 if (!first)
 673                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 674
 675                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 676                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 677
 678                         error = gfs2_revoke_add(sdp, blkno, start);
 679                         if (error < 0) {
 680                                 brelse(bh);
 681                                 return error;
 682                         }
 683                         else if (error)
 684                                 sdp->sd_found_revokes++;
 685
 686                         if (!--revokes)
 687                                 break;
 688                         offset += sizeof(u64);
 689                 }
 690
 691                 brelse(bh);
 692                 offset = sizeof(struct gfs2_meta_header);
 693                 first = 0;
 694         }
 695
 696         return 0;
 697 }
 698
 699 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 700 {
 701         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 702
 703         if (error) {
 704                 gfs2_revoke_clean(sdp);
 705                 return;
 706         }
 707         if (pass != 1)
 708                 return;
 709
 710         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 711                 jd->jd_jid, sdp->sd_found_revokes);
 712
 713         gfs2_revoke_clean(sdp);
 714 }
 715
 716 /**
 717  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 718  *
 719  */
 720
 721 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 722 {
 723         unsigned int limit = buf_limit(sdp) / 2;
 724
 725         gfs2_before_commit(sdp, limit, sdp->sd_log_num_databuf,
 726                            &sdp->sd_log_le_databuf, 1);
 727 }
 728
 729 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 730                                     struct gfs2_log_descriptor *ld,
 731                                     __be64 *ptr, int pass)
 732 {
 733         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 734         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 735         struct gfs2_glock *gl = ip->i_gl;
 736         unsigned int blks = be32_to_cpu(ld->ld_data1);
 737         struct buffer_head *bh_log, *bh_ip;
 738         u64 blkno;
 739         u64 esc;
 740         int error = 0;
 741
 742         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 743                 return 0;
 744
 745         gfs2_replay_incr_blk(sdp, &start);
 746         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 747                 blkno = be64_to_cpu(*ptr++);
 748                 esc = be64_to_cpu(*ptr++);
 749
 750                 sdp->sd_found_blocks++;
 751
 752                 if (gfs2_revoke_check(sdp, blkno, start))
 753                         continue;
 754
 755                 error = gfs2_replay_read_block(jd, start, &bh_log);
 756                 if (error)
 757                         return error;
 758
 759                 bh_ip = gfs2_meta_new(gl, blkno);
 760                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 761
 762                 /* Unescape */
 763                 if (esc) {
 764                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 765                         *eptr = cpu_to_be32(GFS2_MAGIC);
 766                 }
 767                 mark_buffer_dirty(bh_ip);
 768
 769                 brelse(bh_log);
 770                 brelse(bh_ip);
 771
 772                 sdp->sd_replayed_blocks++;
 773         }
 774
 775         return error;
 776 }
 777
 778 /* FIXME: sort out accounting for log blocks etc. */
 779
 780 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 781 {
 782         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 783         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 784
 785         if (error) {
 786                 gfs2_meta_sync(ip->i_gl);
 787                 return;
 788         }
 789         if (pass != 1)
 790                 return;
 791
 792         /* data sync? */
 793         gfs2_meta_sync(ip->i_gl);
 794
 795         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 796                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 797 }
 798
 799 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 800 {
 801         struct list_head *head = &sdp->sd_log_le_databuf;
 802         struct gfs2_bufdata *bd;
 803
 804         if (tr == NULL) {
 805                 gfs2_assert(sdp, list_empty(head));
 806                 return;
 807         }
 808
 809         while (!list_empty(head)) {
 810                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
 811                 list_del_init(&bd->bd_list);
 812                 sdp->sd_log_num_databuf--;
 813                 gfs2_unpin(sdp, bd->bd_bh, tr);
 814         }
 815         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 816 }
 817
 818
 819 const struct gfs2_log_operations gfs2_buf_lops = {
 820         .lo_before_commit = buf_lo_before_commit,
 821         .lo_after_commit = buf_lo_after_commit,
 822         .lo_before_scan = buf_lo_before_scan,
 823         .lo_scan_elements = buf_lo_scan_elements,
 824         .lo_after_scan = buf_lo_after_scan,
 825         .lo_name = "buf",
 826 };
 827
 828 const struct gfs2_log_operations gfs2_revoke_lops = {
 829         .lo_before_commit = revoke_lo_before_commit,
 830         .lo_after_commit = revoke_lo_after_commit,
 831         .lo_before_scan = revoke_lo_before_scan,
 832         .lo_scan_elements = revoke_lo_scan_elements,
 833         .lo_after_scan = revoke_lo_after_scan,
 834         .lo_name = "revoke",
 835 };
 836
 837 const struct gfs2_log_operations gfs2_rg_lops = {
 838         .lo_name = "rg",
 839 };
 840
 841 const struct gfs2_log_operations gfs2_databuf_lops = {
 842         .lo_before_commit = databuf_lo_before_commit,
 843         .lo_after_commit = databuf_lo_after_commit,
 844         .lo_scan_elements = databuf_lo_scan_elements,
 845         .lo_after_scan = databuf_lo_after_scan,
 846         .lo_name = "databuf",
 847 };
 848
 849 const struct gfs2_log_operations *gfs2_log_ops[] = {
 850         &gfs2_databuf_lops,
 851         &gfs2_buf_lops,
 852         &gfs2_rg_lops,
 853         &gfs2_revoke_lops,
 854         NULL,
 855 };
 856