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/bio.h>
  17 #include <linux/fs.h>
  18
  19 #include "gfs2.h"
  20 #include "incore.h"
  21 #include "inode.h"
  22 #include "glock.h"
  23 #include "log.h"
  24 #include "lops.h"
  25 #include "meta_io.h"
  26 #include "recovery.h"
  27 #include "rgrp.h"
  28 #include "trans.h"
  29 #include "util.h"
  30 #include "trace_gfs2.h"
  31
  32 /**
  33  * gfs2_pin - Pin a buffer in memory
  34  * @sdp: The superblock
  35  * @bh: The buffer to be pinned
  36  *
  37  * The log lock must be held when calling this function
  38  */
  39 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
  40 {
  41         struct gfs2_bufdata *bd;
  42
  43         gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
  44
  45         clear_buffer_dirty(bh);
  46         if (test_set_buffer_pinned(bh))
  47                 gfs2_assert_withdraw(sdp, 0);
  48         if (!buffer_uptodate(bh))
  49                 gfs2_io_error_bh(sdp, bh);
  50         bd = bh->b_private;
  51         /* If this buffer is in the AIL and it has already been written
  52          * to in-place disk block, remove it from the AIL.
  53          */
  54         spin_lock(&sdp->sd_ail_lock);
  55         if (bd->bd_ail)
  56                 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
  57         spin_unlock(&sdp->sd_ail_lock);
  58         get_bh(bh);
  59         atomic_inc(&sdp->sd_log_pinned);
  60         trace_gfs2_pin(bd, 1);
  61 }
  62
  63 /**
  64  * gfs2_unpin - Unpin a buffer
  65  * @sdp: the filesystem the buffer belongs to
  66  * @bh: The buffer to unpin
  67  * @ai:
  68  *
  69  */
  70
  71 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  72                        struct gfs2_ail *ai)
  73 {
  74         struct gfs2_bufdata *bd = bh->b_private;
  75
  76         gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
  77
  78         if (!buffer_pinned(bh))
  79                 gfs2_assert_withdraw(sdp, 0);
  80
  81         lock_buffer(bh);
  82         mark_buffer_dirty(bh);
  83         clear_buffer_pinned(bh);
  84
  85         spin_lock(&sdp->sd_ail_lock);
  86         if (bd->bd_ail) {
  87                 list_del(&bd->bd_ail_st_list);
  88                 brelse(bh);
  89         } else {
  90                 struct gfs2_glock *gl = bd->bd_gl;
  91                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
  92                 atomic_inc(&gl->gl_ail_count);
  93         }
  94         bd->bd_ail = ai;
  95         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
  96         spin_unlock(&sdp->sd_ail_lock);
  97
  98         if (test_and_clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags))
  99                 gfs2_glock_schedule_for_reclaim(bd->bd_gl);
 100         trace_gfs2_pin(bd, 0);
 101         unlock_buffer(bh);
 102         atomic_dec(&sdp->sd_log_pinned);
 103 }
 104
 105
 106 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
 107 {
 108         return (struct gfs2_log_descriptor *)bh->b_data;
 109 }
 110
 111 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
 112 {
 113         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 114         return (__force __be64 *)(ld + 1);
 115 }
 116
 117 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
 118 {
 119         return (__force __be64 *)(bh->b_data + bh->b_size);
 120 }
 121
 122
 123 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
 124 {
 125         struct buffer_head *bh = gfs2_log_get_buf(sdp);
 126         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 127         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 128         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 129         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 130         ld->ld_type = cpu_to_be32(ld_type);
 131         ld->ld_length = 0;
 132         ld->ld_data1 = 0;
 133         ld->ld_data2 = 0;
 134         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 135         return bh;
 136 }
 137
 138 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 139 {
 140         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 141         struct gfs2_meta_header *mh;
 142         struct gfs2_trans *tr;
 143
 144         lock_buffer(bd->bd_bh);
 145         gfs2_log_lock(sdp);
 146         if (!list_empty(&bd->bd_list_tr))
 147                 goto out;
 148         tr = current->journal_info;
 149         tr->tr_touched = 1;
 150         tr->tr_num_buf++;
 151         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 152         if (!list_empty(&le->le_list))
 153                 goto out;
 154         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 155         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 156         gfs2_meta_check(sdp, bd->bd_bh);
 157         gfs2_pin(sdp, bd->bd_bh);
 158         mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
 159         mh->__pad0 = cpu_to_be64(0);
 160         mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 161         sdp->sd_log_num_buf++;
 162         list_add(&le->le_list, &sdp->sd_log_le_buf);
 163         tr->tr_num_buf_new++;
 164 out:
 165         gfs2_log_unlock(sdp);
 166         unlock_buffer(bd->bd_bh);
 167 }
 168
 169 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 170 {
 171         struct buffer_head *bh;
 172         struct gfs2_log_descriptor *ld;
 173         struct gfs2_bufdata *bd1 = NULL, *bd2;
 174         unsigned int total;
 175         unsigned int limit;
 176         unsigned int num;
 177         unsigned n;
 178         __be64 *ptr;
 179
 180         limit = buf_limit(sdp);
 181         /* for 4k blocks, limit = 503 */
 182
 183         gfs2_log_lock(sdp);
 184         total = sdp->sd_log_num_buf;
 185         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 186         while(total) {
 187                 num = total;
 188                 if (total > limit)
 189                         num = limit;
 190                 gfs2_log_unlock(sdp);
 191                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
 192                 gfs2_log_lock(sdp);
 193                 ld = bh_log_desc(bh);
 194                 ptr = bh_log_ptr(bh);
 195                 ld->ld_length = cpu_to_be32(num + 1);
 196                 ld->ld_data1 = cpu_to_be32(num);
 197
 198                 n = 0;
 199                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 200                                              bd_le.le_list) {
 201                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 202                         if (++n >= num)
 203                                 break;
 204                 }
 205
 206                 gfs2_log_unlock(sdp);
 207                 submit_bh(WRITE_SYNC, bh);
 208                 gfs2_log_lock(sdp);
 209
 210                 n = 0;
 211                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 212                                              bd_le.le_list) {
 213                         get_bh(bd2->bd_bh);
 214                         gfs2_log_unlock(sdp);
 215                         lock_buffer(bd2->bd_bh);
 216                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 217                         submit_bh(WRITE_SYNC, bh);
 218                         gfs2_log_lock(sdp);
 219                         if (++n >= num)
 220                                 break;
 221                 }
 222
 223                 BUG_ON(total < num);
 224                 total -= num;
 225         }
 226         gfs2_log_unlock(sdp);
 227 }
 228
 229 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 230 {
 231         struct list_head *head = &sdp->sd_log_le_buf;
 232         struct gfs2_bufdata *bd;
 233
 234         while (!list_empty(head)) {
 235                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 236                 list_del_init(&bd->bd_le.le_list);
 237                 sdp->sd_log_num_buf--;
 238
 239                 gfs2_unpin(sdp, bd->bd_bh, ai);
 240         }
 241         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 242 }
 243
 244 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 245                                struct gfs2_log_header_host *head, int pass)
 246 {
 247         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 248
 249         if (pass != 0)
 250                 return;
 251
 252         sdp->sd_found_blocks = 0;
 253         sdp->sd_replayed_blocks = 0;
 254 }
 255
 256 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 257                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 258                                 int pass)
 259 {
 260         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 261         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 262         struct gfs2_glock *gl = ip->i_gl;
 263         unsigned int blks = be32_to_cpu(ld->ld_data1);
 264         struct buffer_head *bh_log, *bh_ip;
 265         u64 blkno;
 266         int error = 0;
 267
 268         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 269                 return 0;
 270
 271         gfs2_replay_incr_blk(sdp, &start);
 272
 273         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 274                 blkno = be64_to_cpu(*ptr++);
 275
 276                 sdp->sd_found_blocks++;
 277
 278                 if (gfs2_revoke_check(sdp, blkno, start))
 279                         continue;
 280
 281                 error = gfs2_replay_read_block(jd, start, &bh_log);
 282                 if (error)
 283                         return error;
 284
 285                 bh_ip = gfs2_meta_new(gl, blkno);
 286                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 287
 288                 if (gfs2_meta_check(sdp, bh_ip))
 289                         error = -EIO;
 290                 else
 291                         mark_buffer_dirty(bh_ip);
 292
 293                 brelse(bh_log);
 294                 brelse(bh_ip);
 295
 296                 if (error)
 297                         break;
 298
 299                 sdp->sd_replayed_blocks++;
 300         }
 301
 302         return error;
 303 }
 304
 305 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 306 {
 307         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 308         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 309
 310         if (error) {
 311                 gfs2_meta_sync(ip->i_gl);
 312                 return;
 313         }
 314         if (pass != 1)
 315                 return;
 316
 317         gfs2_meta_sync(ip->i_gl);
 318
 319         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 320                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 321 }
 322
 323 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 324 {
 325         struct gfs2_trans *tr;
 326
 327         tr = current->journal_info;
 328         tr->tr_touched = 1;
 329         tr->tr_num_revoke++;
 330         sdp->sd_log_num_revoke++;
 331         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 332 }
 333
 334 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 335 {
 336         struct gfs2_log_descriptor *ld;
 337         struct gfs2_meta_header *mh;
 338         struct buffer_head *bh;
 339         unsigned int offset;
 340         struct list_head *head = &sdp->sd_log_le_revoke;
 341         struct gfs2_bufdata *bd;
 342
 343         if (!sdp->sd_log_num_revoke)
 344                 return;
 345
 346         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
 347         ld = bh_log_desc(bh);
 348         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 349                                                     sizeof(u64)));
 350         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 351         offset = sizeof(struct gfs2_log_descriptor);
 352
 353         while (!list_empty(head)) {
 354                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 355                 list_del_init(&bd->bd_le.le_list);
 356                 sdp->sd_log_num_revoke--;
 357
 358                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 359                         submit_bh(WRITE_SYNC, bh);
 360
 361                         bh = gfs2_log_get_buf(sdp);
 362                         mh = (struct gfs2_meta_header *)bh->b_data;
 363                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 364                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 365                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 366                         offset = sizeof(struct gfs2_meta_header);
 367                 }
 368
 369                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
 370                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 371
 372                 offset += sizeof(u64);
 373         }
 374         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 375
 376         submit_bh(WRITE_SYNC, bh);
 377 }
 378
 379 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 380                                   struct gfs2_log_header_host *head, int pass)
 381 {
 382         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 383
 384         if (pass != 0)
 385                 return;
 386
 387         sdp->sd_found_revokes = 0;
 388         sdp->sd_replay_tail = head->lh_tail;
 389 }
 390
 391 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 392                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 393                                    int pass)
 394 {
 395         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 396         unsigned int blks = be32_to_cpu(ld->ld_length);
 397         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 398         struct buffer_head *bh;
 399         unsigned int offset;
 400         u64 blkno;
 401         int first = 1;
 402         int error;
 403
 404         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 405                 return 0;
 406
 407         offset = sizeof(struct gfs2_log_descriptor);
 408
 409         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 410                 error = gfs2_replay_read_block(jd, start, &bh);
 411                 if (error)
 412                         return error;
 413
 414                 if (!first)
 415                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 416
 417                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 418                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 419
 420                         error = gfs2_revoke_add(sdp, blkno, start);
 421                         if (error < 0) {
 422                                 brelse(bh);
 423                                 return error;
 424                         }
 425                         else if (error)
 426                                 sdp->sd_found_revokes++;
 427
 428                         if (!--revokes)
 429                                 break;
 430                         offset += sizeof(u64);
 431                 }
 432
 433                 brelse(bh);
 434                 offset = sizeof(struct gfs2_meta_header);
 435                 first = 0;
 436         }
 437
 438         return 0;
 439 }
 440
 441 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 442 {
 443         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 444
 445         if (error) {
 446                 gfs2_revoke_clean(sdp);
 447                 return;
 448         }
 449         if (pass != 1)
 450                 return;
 451
 452         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 453                 jd->jd_jid, sdp->sd_found_revokes);
 454
 455         gfs2_revoke_clean(sdp);
 456 }
 457
 458 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 459 {
 460         struct gfs2_rgrpd *rgd;
 461         struct gfs2_trans *tr = current->journal_info;
 462
 463         tr->tr_touched = 1;
 464
 465         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 466
 467         gfs2_log_lock(sdp);
 468         if (!list_empty(&le->le_list)){
 469                 gfs2_log_unlock(sdp);
 470                 return;
 471         }
 472         gfs2_rgrp_bh_hold(rgd);
 473         sdp->sd_log_num_rg++;
 474         list_add(&le->le_list, &sdp->sd_log_le_rg);
 475         gfs2_log_unlock(sdp);
 476 }
 477
 478 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 479 {
 480         struct list_head *head = &sdp->sd_log_le_rg;
 481         struct gfs2_rgrpd *rgd;
 482
 483         while (!list_empty(head)) {
 484                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 485                 list_del_init(&rgd->rd_le.le_list);
 486                 sdp->sd_log_num_rg--;
 487
 488                 gfs2_rgrp_repolish_clones(rgd);
 489                 gfs2_rgrp_bh_put(rgd);
 490         }
 491         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 492 }
 493
 494 /**
 495  * databuf_lo_add - Add a databuf to the transaction.
 496  *
 497  * This is used in two distinct cases:
 498  * i) In ordered write mode
 499  *    We put the data buffer on a list so that we can ensure that its
 500  *    synced to disk at the right time
 501  * ii) In journaled data mode
 502  *    We need to journal the data block in the same way as metadata in
 503  *    the functions above. The difference is that here we have a tag
 504  *    which is two __be64's being the block number (as per meta data)
 505  *    and a flag which says whether the data block needs escaping or
 506  *    not. This means we need a new log entry for each 251 or so data
 507  *    blocks, which isn't an enormous overhead but twice as much as
 508  *    for normal metadata blocks.
 509  */
 510 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 511 {
 512         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 513         struct gfs2_trans *tr = current->journal_info;
 514         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 515         struct gfs2_inode *ip = GFS2_I(mapping->host);
 516
 517         lock_buffer(bd->bd_bh);
 518         gfs2_log_lock(sdp);
 519         if (tr) {
 520                 if (!list_empty(&bd->bd_list_tr))
 521                         goto out;
 522                 tr->tr_touched = 1;
 523                 if (gfs2_is_jdata(ip)) {
 524                         tr->tr_num_buf++;
 525                         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 526                 }
 527         }
 528         if (!list_empty(&le->le_list))
 529                 goto out;
 530
 531         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 532         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 533         if (gfs2_is_jdata(ip)) {
 534                 gfs2_pin(sdp, bd->bd_bh);
 535                 tr->tr_num_databuf_new++;
 536                 sdp->sd_log_num_databuf++;
 537                 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
 538         } else {
 539                 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
 540         }
 541 out:
 542         gfs2_log_unlock(sdp);
 543         unlock_buffer(bd->bd_bh);
 544 }
 545
 546 static void gfs2_check_magic(struct buffer_head *bh)
 547 {
 548         void *kaddr;
 549         __be32 *ptr;
 550
 551         clear_buffer_escaped(bh);
 552         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 553         ptr = kaddr + bh_offset(bh);
 554         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 555                 set_buffer_escaped(bh);
 556         kunmap_atomic(kaddr, KM_USER0);
 557 }
 558
 559 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
 560                               struct list_head *list, struct list_head *done,
 561                               unsigned int n)
 562 {
 563         struct buffer_head *bh1;
 564         struct gfs2_log_descriptor *ld;
 565         struct gfs2_bufdata *bd;
 566         __be64 *ptr;
 567
 568         if (!bh)
 569                 return;
 570
 571         ld = bh_log_desc(bh);
 572         ld->ld_length = cpu_to_be32(n + 1);
 573         ld->ld_data1 = cpu_to_be32(n);
 574
 575         ptr = bh_log_ptr(bh);
 576
 577         get_bh(bh);
 578         submit_bh(WRITE_SYNC, bh);
 579         gfs2_log_lock(sdp);
 580         while(!list_empty(list)) {
 581                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
 582                 list_move_tail(&bd->bd_le.le_list, done);
 583                 get_bh(bd->bd_bh);
 584                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
 585                         gfs2_log_incr_head(sdp);
 586                         ptr += 2;
 587                 }
 588                 gfs2_log_unlock(sdp);
 589                 lock_buffer(bd->bd_bh);
 590                 if (buffer_escaped(bd->bd_bh)) {
 591                         void *kaddr;
 592                         bh1 = gfs2_log_get_buf(sdp);
 593                         kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
 594                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
 595                                bh1->b_size);
 596                         kunmap_atomic(kaddr, KM_USER0);
 597                         *(__be32 *)bh1->b_data = 0;
 598                         clear_buffer_escaped(bd->bd_bh);
 599                         unlock_buffer(bd->bd_bh);
 600                         brelse(bd->bd_bh);
 601                 } else {
 602                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
 603                 }
 604                 submit_bh(WRITE_SYNC, bh1);
 605                 gfs2_log_lock(sdp);
 606                 ptr += 2;
 607         }
 608         gfs2_log_unlock(sdp);
 609         brelse(bh);
 610 }
 611
 612 /**
 613  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 614  *
 615  */
 616
 617 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 618 {
 619         struct gfs2_bufdata *bd = NULL;
 620         struct buffer_head *bh = NULL;
 621         unsigned int n = 0;
 622         __be64 *ptr = NULL, *end = NULL;
 623         LIST_HEAD(processed);
 624         LIST_HEAD(in_progress);
 625
 626         gfs2_log_lock(sdp);
 627         while (!list_empty(&sdp->sd_log_le_databuf)) {
 628                 if (ptr == end) {
 629                         gfs2_log_unlock(sdp);
 630                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 631                         n = 0;
 632                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
 633                         ptr = bh_log_ptr(bh);
 634                         end = bh_ptr_end(bh) - 1;
 635                         gfs2_log_lock(sdp);
 636                         continue;
 637                 }
 638                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
 639                 list_move_tail(&bd->bd_le.le_list, &in_progress);
 640                 gfs2_check_magic(bd->bd_bh);
 641                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
 642                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
 643                 n++;
 644         }
 645         gfs2_log_unlock(sdp);
 646         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 647         gfs2_log_lock(sdp);
 648         list_splice(&processed, &sdp->sd_log_le_databuf);
 649         gfs2_log_unlock(sdp);
 650 }
 651
 652 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 653                                     struct gfs2_log_descriptor *ld,
 654                                     __be64 *ptr, int pass)
 655 {
 656         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 657         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 658         struct gfs2_glock *gl = ip->i_gl;
 659         unsigned int blks = be32_to_cpu(ld->ld_data1);
 660         struct buffer_head *bh_log, *bh_ip;
 661         u64 blkno;
 662         u64 esc;
 663         int error = 0;
 664
 665         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 666                 return 0;
 667
 668         gfs2_replay_incr_blk(sdp, &start);
 669         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 670                 blkno = be64_to_cpu(*ptr++);
 671                 esc = be64_to_cpu(*ptr++);
 672
 673                 sdp->sd_found_blocks++;
 674
 675                 if (gfs2_revoke_check(sdp, blkno, start))
 676                         continue;
 677
 678                 error = gfs2_replay_read_block(jd, start, &bh_log);
 679                 if (error)
 680                         return error;
 681
 682                 bh_ip = gfs2_meta_new(gl, blkno);
 683                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 684
 685                 /* Unescape */
 686                 if (esc) {
 687                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 688                         *eptr = cpu_to_be32(GFS2_MAGIC);
 689                 }
 690                 mark_buffer_dirty(bh_ip);
 691
 692                 brelse(bh_log);
 693                 brelse(bh_ip);
 694                 if (error)
 695                         break;
 696
 697                 sdp->sd_replayed_blocks++;
 698         }
 699
 700         return error;
 701 }
 702
 703 /* FIXME: sort out accounting for log blocks etc. */
 704
 705 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 706 {
 707         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 708         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 709
 710         if (error) {
 711                 gfs2_meta_sync(ip->i_gl);
 712                 return;
 713         }
 714         if (pass != 1)
 715                 return;
 716
 717         /* data sync? */
 718         gfs2_meta_sync(ip->i_gl);
 719
 720         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 721                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 722 }
 723
 724 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 725 {
 726         struct list_head *head = &sdp->sd_log_le_databuf;
 727         struct gfs2_bufdata *bd;
 728
 729         while (!list_empty(head)) {
 730                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 731                 list_del_init(&bd->bd_le.le_list);
 732                 sdp->sd_log_num_databuf--;
 733                 gfs2_unpin(sdp, bd->bd_bh, ai);
 734         }
 735         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 736 }
 737
 738
 739 const struct gfs2_log_operations gfs2_buf_lops = {
 740         .lo_add = buf_lo_add,
 741         .lo_before_commit = buf_lo_before_commit,
 742         .lo_after_commit = buf_lo_after_commit,
 743         .lo_before_scan = buf_lo_before_scan,
 744         .lo_scan_elements = buf_lo_scan_elements,
 745         .lo_after_scan = buf_lo_after_scan,
 746         .lo_name = "buf",
 747 };
 748
 749 const struct gfs2_log_operations gfs2_revoke_lops = {
 750         .lo_add = revoke_lo_add,
 751         .lo_before_commit = revoke_lo_before_commit,
 752         .lo_before_scan = revoke_lo_before_scan,
 753         .lo_scan_elements = revoke_lo_scan_elements,
 754         .lo_after_scan = revoke_lo_after_scan,
 755         .lo_name = "revoke",
 756 };
 757
 758 const struct gfs2_log_operations gfs2_rg_lops = {
 759         .lo_add = rg_lo_add,
 760         .lo_after_commit = rg_lo_after_commit,
 761         .lo_name = "rg",
 762 };
 763
 764 const struct gfs2_log_operations gfs2_databuf_lops = {
 765         .lo_add = databuf_lo_add,
 766         .lo_before_commit = databuf_lo_before_commit,
 767         .lo_after_commit = databuf_lo_after_commit,
 768         .lo_scan_elements = databuf_lo_scan_elements,
 769         .lo_after_scan = databuf_lo_after_scan,
 770         .lo_name = "databuf",
 771 };
 772
 773 const struct gfs2_log_operations *gfs2_log_ops[] = {
 774         &gfs2_databuf_lops,
 775         &gfs2_buf_lops,
 776         &gfs2_rg_lops,
 777         &gfs2_revoke_lops,
 778         NULL,
 779 };
 780