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/gfs2_ondisk.h>
  16 #include <linux/lm_interface.h>
  17
  18 #include "gfs2.h"
  19 #include "incore.h"
  20 #include "inode.h"
  21 #include "glock.h"
  22 #include "log.h"
  23 #include "lops.h"
  24 #include "meta_io.h"
  25 #include "recovery.h"
  26 #include "rgrp.h"
  27 #include "trans.h"
  28 #include "util.h"
  29
  30 /**
  31  * gfs2_pin - Pin a buffer in memory
  32  * @sdp: The superblock
  33  * @bh: The buffer to be pinned
  34  *
  35  * The log lock must be held when calling this function
  36  */
  37 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  38 {
  39         struct gfs2_bufdata *bd;
  40
  41         gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
  42
  43         clear_buffer_dirty(bh);
  44         if (test_set_buffer_pinned(bh))
  45                 gfs2_assert_withdraw(sdp, 0);
  46         if (!buffer_uptodate(bh))
  47                 gfs2_io_error_bh(sdp, bh);
  48         bd = bh->b_private;
  49         /* If this buffer is in the AIL and it has already been written
  50          * to in-place disk block, remove it from the AIL.
  51          */
  52         if (bd->bd_ail)
  53                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
  54         get_bh(bh);
  55 }
  56
  57 /**
  58  * gfs2_unpin - Unpin a buffer
  59  * @sdp: the filesystem the buffer belongs to
  60  * @bh: The buffer to unpin
  61  * @ai:
  62  *
  63  */
  64
  65 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  66                        struct gfs2_ail *ai)
  67 {
  68         struct gfs2_bufdata *bd = bh->b_private;
  69
  70         gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
  71
  72         if (!buffer_pinned(bh))
  73                 gfs2_assert_withdraw(sdp, 0);
  74
  75         lock_buffer(bh);
  76         mark_buffer_dirty(bh);
  77         clear_buffer_pinned(bh);
  78
  79         gfs2_log_lock(sdp);
  80         if (bd->bd_ail) {
  81                 list_del(&bd->bd_ail_st_list);
  82                 brelse(bh);
  83         } else {
  84                 struct gfs2_glock *gl = bd->bd_gl;
  85                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
  86                 atomic_inc(&gl->gl_ail_count);
  87         }
  88         bd->bd_ail = ai;
  89         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
  90         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
  91         gfs2_log_unlock(sdp);
  92         unlock_buffer(bh);
  93 }
  94
  95
  96 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
  97 {
  98         return (struct gfs2_log_descriptor *)bh->b_data;
  99 }
 100
 101 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
 102 {
 103         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 104         return (__force __be64 *)(ld + 1);
 105 }
 106
 107 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
 108 {
 109         return (__force __be64 *)(bh->b_data + bh->b_size);
 110 }
 111
 112
 113 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
 114 {
 115         struct buffer_head *bh = gfs2_log_get_buf(sdp);
 116         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 117         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 118         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 119         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 120         ld->ld_type = cpu_to_be32(ld_type);
 121         ld->ld_length = 0;
 122         ld->ld_data1 = 0;
 123         ld->ld_data2 = 0;
 124         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 125         return bh;
 126 }
 127
 128 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 129 {
 130         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 131         struct gfs2_trans *tr;
 132
 133         lock_buffer(bd->bd_bh);
 134         gfs2_log_lock(sdp);
 135         if (!list_empty(&bd->bd_list_tr))
 136                 goto out;
 137         tr = current->journal_info;
 138         tr->tr_touched = 1;
 139         tr->tr_num_buf++;
 140         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 141         if (!list_empty(&le->le_list))
 142                 goto out;
 143         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 144         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 145         gfs2_meta_check(sdp, bd->bd_bh);
 146         gfs2_pin(sdp, bd->bd_bh);
 147         sdp->sd_log_num_buf++;
 148         list_add(&le->le_list, &sdp->sd_log_le_buf);
 149         tr->tr_num_buf_new++;
 150 out:
 151         gfs2_log_unlock(sdp);
 152         unlock_buffer(bd->bd_bh);
 153 }
 154
 155 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 156 {
 157         struct buffer_head *bh;
 158         struct gfs2_log_descriptor *ld;
 159         struct gfs2_bufdata *bd1 = NULL, *bd2;
 160         unsigned int total;
 161         unsigned int limit;
 162         unsigned int num;
 163         unsigned n;
 164         __be64 *ptr;
 165
 166         limit = buf_limit(sdp);
 167         /* for 4k blocks, limit = 503 */
 168
 169         gfs2_log_lock(sdp);
 170         total = sdp->sd_log_num_buf;
 171         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 172         while(total) {
 173                 num = total;
 174                 if (total > limit)
 175                         num = limit;
 176                 gfs2_log_unlock(sdp);
 177                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
 178                 gfs2_log_lock(sdp);
 179                 ld = bh_log_desc(bh);
 180                 ptr = bh_log_ptr(bh);
 181                 ld->ld_length = cpu_to_be32(num + 1);
 182                 ld->ld_data1 = cpu_to_be32(num);
 183
 184                 n = 0;
 185                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 186                                              bd_le.le_list) {
 187                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 188                         if (++n >= num)
 189                                 break;
 190                 }
 191
 192                 gfs2_log_unlock(sdp);
 193                 submit_bh(WRITE, bh);
 194                 gfs2_log_lock(sdp);
 195
 196                 n = 0;
 197                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 198                                              bd_le.le_list) {
 199                         get_bh(bd2->bd_bh);
 200                         gfs2_log_unlock(sdp);
 201                         lock_buffer(bd2->bd_bh);
 202                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 203                         submit_bh(WRITE, bh);
 204                         gfs2_log_lock(sdp);
 205                         if (++n >= num)
 206                                 break;
 207                 }
 208
 209                 BUG_ON(total < num);
 210                 total -= num;
 211         }
 212         gfs2_log_unlock(sdp);
 213 }
 214
 215 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 216 {
 217         struct list_head *head = &sdp->sd_log_le_buf;
 218         struct gfs2_bufdata *bd;
 219
 220         while (!list_empty(head)) {
 221                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 222                 list_del_init(&bd->bd_le.le_list);
 223                 sdp->sd_log_num_buf--;
 224
 225                 gfs2_unpin(sdp, bd->bd_bh, ai);
 226         }
 227         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 228 }
 229
 230 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 231                                struct gfs2_log_header_host *head, int pass)
 232 {
 233         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 234
 235         if (pass != 0)
 236                 return;
 237
 238         sdp->sd_found_blocks = 0;
 239         sdp->sd_replayed_blocks = 0;
 240 }
 241
 242 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 243                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 244                                 int pass)
 245 {
 246         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 247         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 248         struct gfs2_glock *gl = ip->i_gl;
 249         unsigned int blks = be32_to_cpu(ld->ld_data1);
 250         struct buffer_head *bh_log, *bh_ip;
 251         u64 blkno;
 252         int error = 0;
 253
 254         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 255                 return 0;
 256
 257         gfs2_replay_incr_blk(sdp, &start);
 258
 259         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 260                 blkno = be64_to_cpu(*ptr++);
 261
 262                 sdp->sd_found_blocks++;
 263
 264                 if (gfs2_revoke_check(sdp, blkno, start))
 265                         continue;
 266
 267                 error = gfs2_replay_read_block(jd, start, &bh_log);
 268                 if (error)
 269                         return error;
 270
 271                 bh_ip = gfs2_meta_new(gl, blkno);
 272                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 273
 274                 if (gfs2_meta_check(sdp, bh_ip))
 275                         error = -EIO;
 276                 else
 277                         mark_buffer_dirty(bh_ip);
 278
 279                 brelse(bh_log);
 280                 brelse(bh_ip);
 281
 282                 if (error)
 283                         break;
 284
 285                 sdp->sd_replayed_blocks++;
 286         }
 287
 288         return error;
 289 }
 290
 291 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 292 {
 293         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 294         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 295
 296         if (error) {
 297                 gfs2_meta_sync(ip->i_gl);
 298                 return;
 299         }
 300         if (pass != 1)
 301                 return;
 302
 303         gfs2_meta_sync(ip->i_gl);
 304
 305         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 306                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 307 }
 308
 309 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 310 {
 311         struct gfs2_trans *tr;
 312
 313         tr = current->journal_info;
 314         tr->tr_touched = 1;
 315         tr->tr_num_revoke++;
 316         sdp->sd_log_num_revoke++;
 317         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 318 }
 319
 320 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 321 {
 322         struct gfs2_log_descriptor *ld;
 323         struct gfs2_meta_header *mh;
 324         struct buffer_head *bh;
 325         unsigned int offset;
 326         struct list_head *head = &sdp->sd_log_le_revoke;
 327         struct gfs2_bufdata *bd;
 328
 329         if (!sdp->sd_log_num_revoke)
 330                 return;
 331
 332         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
 333         ld = bh_log_desc(bh);
 334         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 335                                                     sizeof(u64)));
 336         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 337         offset = sizeof(struct gfs2_log_descriptor);
 338
 339         while (!list_empty(head)) {
 340                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 341                 list_del_init(&bd->bd_le.le_list);
 342                 sdp->sd_log_num_revoke--;
 343
 344                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 345                         submit_bh(WRITE, bh);
 346
 347                         bh = gfs2_log_get_buf(sdp);
 348                         mh = (struct gfs2_meta_header *)bh->b_data;
 349                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 350                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 351                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 352                         offset = sizeof(struct gfs2_meta_header);
 353                 }
 354
 355                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
 356                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 357
 358                 offset += sizeof(u64);
 359         }
 360         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 361
 362         submit_bh(WRITE, bh);
 363 }
 364
 365 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 366                                   struct gfs2_log_header_host *head, int pass)
 367 {
 368         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 369
 370         if (pass != 0)
 371                 return;
 372
 373         sdp->sd_found_revokes = 0;
 374         sdp->sd_replay_tail = head->lh_tail;
 375 }
 376
 377 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 378                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 379                                    int pass)
 380 {
 381         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 382         unsigned int blks = be32_to_cpu(ld->ld_length);
 383         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 384         struct buffer_head *bh;
 385         unsigned int offset;
 386         u64 blkno;
 387         int first = 1;
 388         int error;
 389
 390         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 391                 return 0;
 392
 393         offset = sizeof(struct gfs2_log_descriptor);
 394
 395         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 396                 error = gfs2_replay_read_block(jd, start, &bh);
 397                 if (error)
 398                         return error;
 399
 400                 if (!first)
 401                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 402
 403                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 404                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 405
 406                         error = gfs2_revoke_add(sdp, blkno, start);
 407                         if (error < 0) {
 408                                 brelse(bh);
 409                                 return error;
 410                         }
 411                         else if (error)
 412                                 sdp->sd_found_revokes++;
 413
 414                         if (!--revokes)
 415                                 break;
 416                         offset += sizeof(u64);
 417                 }
 418
 419                 brelse(bh);
 420                 offset = sizeof(struct gfs2_meta_header);
 421                 first = 0;
 422         }
 423
 424         return 0;
 425 }
 426
 427 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 428 {
 429         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 430
 431         if (error) {
 432                 gfs2_revoke_clean(sdp);
 433                 return;
 434         }
 435         if (pass != 1)
 436                 return;
 437
 438         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 439                 jd->jd_jid, sdp->sd_found_revokes);
 440
 441         gfs2_revoke_clean(sdp);
 442 }
 443
 444 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 445 {
 446         struct gfs2_rgrpd *rgd;
 447         struct gfs2_trans *tr = current->journal_info;
 448
 449         tr->tr_touched = 1;
 450
 451         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 452
 453         gfs2_log_lock(sdp);
 454         if (!list_empty(&le->le_list)){
 455                 gfs2_log_unlock(sdp);
 456                 return;
 457         }
 458         gfs2_rgrp_bh_hold(rgd);
 459         sdp->sd_log_num_rg++;
 460         list_add(&le->le_list, &sdp->sd_log_le_rg);
 461         gfs2_log_unlock(sdp);
 462 }
 463
 464 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 465 {
 466         struct list_head *head = &sdp->sd_log_le_rg;
 467         struct gfs2_rgrpd *rgd;
 468
 469         while (!list_empty(head)) {
 470                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 471                 list_del_init(&rgd->rd_le.le_list);
 472                 sdp->sd_log_num_rg--;
 473
 474                 gfs2_rgrp_repolish_clones(rgd);
 475                 gfs2_rgrp_bh_put(rgd);
 476         }
 477         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 478 }
 479
 480 /**
 481  * databuf_lo_add - Add a databuf to the transaction.
 482  *
 483  * This is used in two distinct cases:
 484  * i) In ordered write mode
 485  *    We put the data buffer on a list so that we can ensure that its
 486  *    synced to disk at the right time
 487  * ii) In journaled data mode
 488  *    We need to journal the data block in the same way as metadata in
 489  *    the functions above. The difference is that here we have a tag
 490  *    which is two __be64's being the block number (as per meta data)
 491  *    and a flag which says whether the data block needs escaping or
 492  *    not. This means we need a new log entry for each 251 or so data
 493  *    blocks, which isn't an enormous overhead but twice as much as
 494  *    for normal metadata blocks.
 495  */
 496 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 497 {
 498         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 499         struct gfs2_trans *tr = current->journal_info;
 500         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 501         struct gfs2_inode *ip = GFS2_I(mapping->host);
 502
 503         lock_buffer(bd->bd_bh);
 504         gfs2_log_lock(sdp);
 505         if (tr) {
 506                 if (!list_empty(&bd->bd_list_tr))
 507                         goto out;
 508                 tr->tr_touched = 1;
 509                 if (gfs2_is_jdata(ip)) {
 510                         tr->tr_num_buf++;
 511                         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 512                 }
 513         }
 514         if (!list_empty(&le->le_list))
 515                 goto out;
 516
 517         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 518         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 519         if (gfs2_is_jdata(ip)) {
 520                 gfs2_pin(sdp, bd->bd_bh);
 521                 tr->tr_num_databuf_new++;
 522                 sdp->sd_log_num_databuf++;
 523                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
 524         } else {
 525                 list_add(&le->le_list, &sdp->sd_log_le_ordered);
 526         }
 527 out:
 528         gfs2_log_unlock(sdp);
 529         unlock_buffer(bd->bd_bh);
 530 }
 531
 532 static void gfs2_check_magic(struct buffer_head *bh)
 533 {
 534         void *kaddr;
 535         __be32 *ptr;
 536
 537         clear_buffer_escaped(bh);
 538         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 539         ptr = kaddr + bh_offset(bh);
 540         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 541                 set_buffer_escaped(bh);
 542         kunmap_atomic(kaddr, KM_USER0);
 543 }
 544
 545 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
 546                               struct list_head *list, struct list_head *done,
 547                               unsigned int n)
 548 {
 549         struct buffer_head *bh1;
 550         struct gfs2_log_descriptor *ld;
 551         struct gfs2_bufdata *bd;
 552         __be64 *ptr;
 553
 554         if (!bh)
 555                 return;
 556
 557         ld = bh_log_desc(bh);
 558         ld->ld_length = cpu_to_be32(n + 1);
 559         ld->ld_data1 = cpu_to_be32(n);
 560
 561         ptr = bh_log_ptr(bh);
 562
 563         get_bh(bh);
 564         submit_bh(WRITE, bh);
 565         gfs2_log_lock(sdp);
 566         while(!list_empty(list)) {
 567                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
 568                 list_move_tail(&bd->bd_le.le_list, done);
 569                 get_bh(bd->bd_bh);
 570                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
 571                         gfs2_log_incr_head(sdp);
 572                         ptr += 2;
 573                 }
 574                 gfs2_log_unlock(sdp);
 575                 lock_buffer(bd->bd_bh);
 576                 if (buffer_escaped(bd->bd_bh)) {
 577                         void *kaddr;
 578                         bh1 = gfs2_log_get_buf(sdp);
 579                         kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
 580                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
 581                                bh1->b_size);
 582                         kunmap_atomic(kaddr, KM_USER0);
 583                         *(__be32 *)bh1->b_data = 0;
 584                         clear_buffer_escaped(bd->bd_bh);
 585                         unlock_buffer(bd->bd_bh);
 586                         brelse(bd->bd_bh);
 587                 } else {
 588                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
 589                 }
 590                 submit_bh(WRITE, bh1);
 591                 gfs2_log_lock(sdp);
 592                 ptr += 2;
 593         }
 594         gfs2_log_unlock(sdp);
 595         brelse(bh);
 596 }
 597
 598 /**
 599  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 600  *
 601  */
 602
 603 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 604 {
 605         struct gfs2_bufdata *bd = NULL;
 606         struct buffer_head *bh = NULL;
 607         unsigned int n = 0;
 608         __be64 *ptr = NULL, *end = NULL;
 609         LIST_HEAD(processed);
 610         LIST_HEAD(in_progress);
 611
 612         gfs2_log_lock(sdp);
 613         while (!list_empty(&sdp->sd_log_le_databuf)) {
 614                 if (ptr == end) {
 615                         gfs2_log_unlock(sdp);
 616                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 617                         n = 0;
 618                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
 619                         ptr = bh_log_ptr(bh);
 620                         end = bh_ptr_end(bh) - 1;
 621                         gfs2_log_lock(sdp);
 622                         continue;
 623                 }
 624                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
 625                 list_move_tail(&bd->bd_le.le_list, &in_progress);
 626                 gfs2_check_magic(bd->bd_bh);
 627                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
 628                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
 629                 n++;
 630         }
 631         gfs2_log_unlock(sdp);
 632         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 633         gfs2_log_lock(sdp);
 634         list_splice(&processed, &sdp->sd_log_le_databuf);
 635         gfs2_log_unlock(sdp);
 636 }
 637
 638 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 639                                     struct gfs2_log_descriptor *ld,
 640                                     __be64 *ptr, int pass)
 641 {
 642         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 643         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 644         struct gfs2_glock *gl = ip->i_gl;
 645         unsigned int blks = be32_to_cpu(ld->ld_data1);
 646         struct buffer_head *bh_log, *bh_ip;
 647         u64 blkno;
 648         u64 esc;
 649         int error = 0;
 650
 651         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 652                 return 0;
 653
 654         gfs2_replay_incr_blk(sdp, &start);
 655         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 656                 blkno = be64_to_cpu(*ptr++);
 657                 esc = be64_to_cpu(*ptr++);
 658
 659                 sdp->sd_found_blocks++;
 660
 661                 if (gfs2_revoke_check(sdp, blkno, start))
 662                         continue;
 663
 664                 error = gfs2_replay_read_block(jd, start, &bh_log);
 665                 if (error)
 666                         return error;
 667
 668                 bh_ip = gfs2_meta_new(gl, blkno);
 669                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 670
 671                 /* Unescape */
 672                 if (esc) {
 673                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 674                         *eptr = cpu_to_be32(GFS2_MAGIC);
 675                 }
 676                 mark_buffer_dirty(bh_ip);
 677
 678                 brelse(bh_log);
 679                 brelse(bh_ip);
 680                 if (error)
 681                         break;
 682
 683                 sdp->sd_replayed_blocks++;
 684         }
 685
 686         return error;
 687 }
 688
 689 /* FIXME: sort out accounting for log blocks etc. */
 690
 691 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 692 {
 693         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 694         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 695
 696         if (error) {
 697                 gfs2_meta_sync(ip->i_gl);
 698                 return;
 699         }
 700         if (pass != 1)
 701                 return;
 702
 703         /* data sync? */
 704         gfs2_meta_sync(ip->i_gl);
 705
 706         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 707                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 708 }
 709
 710 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 711 {
 712         struct list_head *head = &sdp->sd_log_le_databuf;
 713         struct gfs2_bufdata *bd;
 714
 715         while (!list_empty(head)) {
 716                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 717                 list_del_init(&bd->bd_le.le_list);
 718                 sdp->sd_log_num_databuf--;
 719                 gfs2_unpin(sdp, bd->bd_bh, ai);
 720         }
 721         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 722 }
 723
 724
 725 const struct gfs2_log_operations gfs2_buf_lops = {
 726         .lo_add = buf_lo_add,
 727         .lo_before_commit = buf_lo_before_commit,
 728         .lo_after_commit = buf_lo_after_commit,
 729         .lo_before_scan = buf_lo_before_scan,
 730         .lo_scan_elements = buf_lo_scan_elements,
 731         .lo_after_scan = buf_lo_after_scan,
 732         .lo_name = "buf",
 733 };
 734
 735 const struct gfs2_log_operations gfs2_revoke_lops = {
 736         .lo_add = revoke_lo_add,
 737         .lo_before_commit = revoke_lo_before_commit,
 738         .lo_before_scan = revoke_lo_before_scan,
 739         .lo_scan_elements = revoke_lo_scan_elements,
 740         .lo_after_scan = revoke_lo_after_scan,
 741         .lo_name = "revoke",
 742 };
 743
 744 const struct gfs2_log_operations gfs2_rg_lops = {
 745         .lo_add = rg_lo_add,
 746         .lo_after_commit = rg_lo_after_commit,
 747         .lo_name = "rg",
 748 };
 749
 750 const struct gfs2_log_operations gfs2_databuf_lops = {
 751         .lo_add = databuf_lo_add,
 752         .lo_before_commit = databuf_lo_before_commit,
 753         .lo_after_commit = databuf_lo_after_commit,
 754         .lo_scan_elements = databuf_lo_scan_elements,
 755         .lo_after_scan = databuf_lo_after_scan,
 756         .lo_name = "databuf",
 757 };
 758
 759 const struct gfs2_log_operations *gfs2_log_ops[] = {
 760         &gfs2_databuf_lops,
 761         &gfs2_buf_lops,
 762         &gfs2_rg_lops,
 763         &gfs2_revoke_lops,
 764         NULL,
 765 };
 766