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         BUG_ON(!current->journal_info);
  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  * @flags: The inode dirty flags
  69  *
  70  */
  71
  72 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
  73                        struct gfs2_ail *ai)
  74 {
  75         struct gfs2_bufdata *bd = bh->b_private;
  76
  77         BUG_ON(!buffer_uptodate(bh));
  78         BUG_ON(!buffer_pinned(bh));
  79
  80         lock_buffer(bh);
  81         mark_buffer_dirty(bh);
  82         clear_buffer_pinned(bh);
  83
  84         spin_lock(&sdp->sd_ail_lock);
  85         if (bd->bd_ail) {
  86                 list_del(&bd->bd_ail_st_list);
  87                 brelse(bh);
  88         } else {
  89                 struct gfs2_glock *gl = bd->bd_gl;
  90                 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
  91                 atomic_inc(&gl->gl_ail_count);
  92         }
  93         bd->bd_ail = ai;
  94         list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
  95         spin_unlock(&sdp->sd_ail_lock);
  96
  97         clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
  98         trace_gfs2_pin(bd, 0);
  99         unlock_buffer(bh);
 100         atomic_dec(&sdp->sd_log_pinned);
 101 }
 102
 103
 104 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
 105 {
 106         return (struct gfs2_log_descriptor *)bh->b_data;
 107 }
 108
 109 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
 110 {
 111         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 112         return (__force __be64 *)(ld + 1);
 113 }
 114
 115 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
 116 {
 117         return (__force __be64 *)(bh->b_data + bh->b_size);
 118 }
 119
 120
 121 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
 122 {
 123         struct buffer_head *bh = gfs2_log_get_buf(sdp);
 124         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 125         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 126         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 127         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 128         ld->ld_type = cpu_to_be32(ld_type);
 129         ld->ld_length = 0;
 130         ld->ld_data1 = 0;
 131         ld->ld_data2 = 0;
 132         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 133         return bh;
 134 }
 135
 136 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 137 {
 138         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 139         struct gfs2_meta_header *mh;
 140         struct gfs2_trans *tr;
 141
 142         if (!list_empty(&bd->bd_list_tr))
 143                 return;
 144         tr = current->journal_info;
 145         tr->tr_touched = 1;
 146         tr->tr_num_buf++;
 147         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 148         if (!list_empty(&le->le_list))
 149                 return;
 150         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 151         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 152         gfs2_meta_check(sdp, bd->bd_bh);
 153         gfs2_pin(sdp, bd->bd_bh);
 154         mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
 155         mh->__pad0 = cpu_to_be64(0);
 156         mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
 157         sdp->sd_log_num_buf++;
 158         list_add(&le->le_list, &sdp->sd_log_le_buf);
 159         tr->tr_num_buf_new++;
 160 }
 161
 162 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 163 {
 164         struct buffer_head *bh;
 165         struct gfs2_log_descriptor *ld;
 166         struct gfs2_bufdata *bd1 = NULL, *bd2;
 167         unsigned int total;
 168         unsigned int limit;
 169         unsigned int num;
 170         unsigned n;
 171         __be64 *ptr;
 172
 173         limit = buf_limit(sdp);
 174         /* for 4k blocks, limit = 503 */
 175
 176         gfs2_log_lock(sdp);
 177         total = sdp->sd_log_num_buf;
 178         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 179         while(total) {
 180                 num = total;
 181                 if (total > limit)
 182                         num = limit;
 183                 gfs2_log_unlock(sdp);
 184                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
 185                 gfs2_log_lock(sdp);
 186                 ld = bh_log_desc(bh);
 187                 ptr = bh_log_ptr(bh);
 188                 ld->ld_length = cpu_to_be32(num + 1);
 189                 ld->ld_data1 = cpu_to_be32(num);
 190
 191                 n = 0;
 192                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 193                                              bd_le.le_list) {
 194                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 195                         if (++n >= num)
 196                                 break;
 197                 }
 198
 199                 gfs2_log_unlock(sdp);
 200                 submit_bh(WRITE_SYNC, bh);
 201                 gfs2_log_lock(sdp);
 202
 203                 n = 0;
 204                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 205                                              bd_le.le_list) {
 206                         get_bh(bd2->bd_bh);
 207                         gfs2_log_unlock(sdp);
 208                         lock_buffer(bd2->bd_bh);
 209                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 210                         submit_bh(WRITE_SYNC, bh);
 211                         gfs2_log_lock(sdp);
 212                         if (++n >= num)
 213                                 break;
 214                 }
 215
 216                 BUG_ON(total < num);
 217                 total -= num;
 218         }
 219         gfs2_log_unlock(sdp);
 220 }
 221
 222 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 223 {
 224         struct list_head *head = &sdp->sd_log_le_buf;
 225         struct gfs2_bufdata *bd;
 226
 227         while (!list_empty(head)) {
 228                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 229                 list_del_init(&bd->bd_le.le_list);
 230                 sdp->sd_log_num_buf--;
 231
 232                 gfs2_unpin(sdp, bd->bd_bh, ai);
 233         }
 234         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 235 }
 236
 237 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 238                                struct gfs2_log_header_host *head, int pass)
 239 {
 240         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 241
 242         if (pass != 0)
 243                 return;
 244
 245         sdp->sd_found_blocks = 0;
 246         sdp->sd_replayed_blocks = 0;
 247 }
 248
 249 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 250                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 251                                 int pass)
 252 {
 253         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 254         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 255         struct gfs2_glock *gl = ip->i_gl;
 256         unsigned int blks = be32_to_cpu(ld->ld_data1);
 257         struct buffer_head *bh_log, *bh_ip;
 258         u64 blkno;
 259         int error = 0;
 260
 261         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 262                 return 0;
 263
 264         gfs2_replay_incr_blk(sdp, &start);
 265
 266         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 267                 blkno = be64_to_cpu(*ptr++);
 268
 269                 sdp->sd_found_blocks++;
 270
 271                 if (gfs2_revoke_check(sdp, blkno, start))
 272                         continue;
 273
 274                 error = gfs2_replay_read_block(jd, start, &bh_log);
 275                 if (error)
 276                         return error;
 277
 278                 bh_ip = gfs2_meta_new(gl, blkno);
 279                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 280
 281                 if (gfs2_meta_check(sdp, bh_ip))
 282                         error = -EIO;
 283                 else
 284                         mark_buffer_dirty(bh_ip);
 285
 286                 brelse(bh_log);
 287                 brelse(bh_ip);
 288
 289                 if (error)
 290                         break;
 291
 292                 sdp->sd_replayed_blocks++;
 293         }
 294
 295         return error;
 296 }
 297
 298 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 299 {
 300         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 301         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 302
 303         if (error) {
 304                 gfs2_meta_sync(ip->i_gl);
 305                 return;
 306         }
 307         if (pass != 1)
 308                 return;
 309
 310         gfs2_meta_sync(ip->i_gl);
 311
 312         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 313                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 314 }
 315
 316 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 317 {
 318         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 319         struct gfs2_glock *gl = bd->bd_gl;
 320         struct gfs2_trans *tr;
 321
 322         tr = current->journal_info;
 323         tr->tr_touched = 1;
 324         tr->tr_num_revoke++;
 325         sdp->sd_log_num_revoke++;
 326         atomic_inc(&gl->gl_revokes);
 327         set_bit(GLF_LFLUSH, &gl->gl_flags);
 328         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 329 }
 330
 331 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 332 {
 333         struct gfs2_log_descriptor *ld;
 334         struct gfs2_meta_header *mh;
 335         struct buffer_head *bh;
 336         unsigned int offset;
 337         struct list_head *head = &sdp->sd_log_le_revoke;
 338         struct gfs2_bufdata *bd;
 339
 340         if (!sdp->sd_log_num_revoke)
 341                 return;
 342
 343         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
 344         ld = bh_log_desc(bh);
 345         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 346                                                     sizeof(u64)));
 347         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 348         offset = sizeof(struct gfs2_log_descriptor);
 349
 350         list_for_each_entry(bd, head, bd_le.le_list) {
 351                 sdp->sd_log_num_revoke--;
 352
 353                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 354                         submit_bh(WRITE_SYNC, bh);
 355
 356                         bh = gfs2_log_get_buf(sdp);
 357                         mh = (struct gfs2_meta_header *)bh->b_data;
 358                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 359                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 360                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 361                         offset = sizeof(struct gfs2_meta_header);
 362                 }
 363
 364                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
 365                 offset += sizeof(u64);
 366         }
 367         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 368
 369         submit_bh(WRITE_SYNC, bh);
 370 }
 371
 372 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 373 {
 374         struct list_head *head = &sdp->sd_log_le_revoke;
 375         struct gfs2_bufdata *bd;
 376         struct gfs2_glock *gl;
 377
 378         while (!list_empty(head)) {
 379                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 380                 list_del_init(&bd->bd_le.le_list);
 381                 gl = bd->bd_gl;
 382                 atomic_dec(&gl->gl_revokes);
 383                 clear_bit(GLF_LFLUSH, &gl->gl_flags);
 384                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 385         }
 386 }
 387
 388 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 389                                   struct gfs2_log_header_host *head, int pass)
 390 {
 391         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 392
 393         if (pass != 0)
 394                 return;
 395
 396         sdp->sd_found_revokes = 0;
 397         sdp->sd_replay_tail = head->lh_tail;
 398 }
 399
 400 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 401                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 402                                    int pass)
 403 {
 404         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 405         unsigned int blks = be32_to_cpu(ld->ld_length);
 406         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 407         struct buffer_head *bh;
 408         unsigned int offset;
 409         u64 blkno;
 410         int first = 1;
 411         int error;
 412
 413         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 414                 return 0;
 415
 416         offset = sizeof(struct gfs2_log_descriptor);
 417
 418         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 419                 error = gfs2_replay_read_block(jd, start, &bh);
 420                 if (error)
 421                         return error;
 422
 423                 if (!first)
 424                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 425
 426                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 427                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 428
 429                         error = gfs2_revoke_add(sdp, blkno, start);
 430                         if (error < 0) {
 431                                 brelse(bh);
 432                                 return error;
 433                         }
 434                         else if (error)
 435                                 sdp->sd_found_revokes++;
 436
 437                         if (!--revokes)
 438                                 break;
 439                         offset += sizeof(u64);
 440                 }
 441
 442                 brelse(bh);
 443                 offset = sizeof(struct gfs2_meta_header);
 444                 first = 0;
 445         }
 446
 447         return 0;
 448 }
 449
 450 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 451 {
 452         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 453
 454         if (error) {
 455                 gfs2_revoke_clean(sdp);
 456                 return;
 457         }
 458         if (pass != 1)
 459                 return;
 460
 461         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 462                 jd->jd_jid, sdp->sd_found_revokes);
 463
 464         gfs2_revoke_clean(sdp);
 465 }
 466
 467 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 468 {
 469         struct gfs2_rgrpd *rgd;
 470         struct gfs2_trans *tr = current->journal_info;
 471
 472         tr->tr_touched = 1;
 473
 474         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 475
 476         gfs2_log_lock(sdp);
 477         if (!list_empty(&le->le_list)){
 478                 gfs2_log_unlock(sdp);
 479                 return;
 480         }
 481         gfs2_rgrp_bh_hold(rgd);
 482         sdp->sd_log_num_rg++;
 483         list_add(&le->le_list, &sdp->sd_log_le_rg);
 484         gfs2_log_unlock(sdp);
 485 }
 486
 487 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 488 {
 489         struct list_head *head = &sdp->sd_log_le_rg;
 490         struct gfs2_rgrpd *rgd;
 491
 492         while (!list_empty(head)) {
 493                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 494                 list_del_init(&rgd->rd_le.le_list);
 495                 sdp->sd_log_num_rg--;
 496
 497                 gfs2_rgrp_repolish_clones(rgd);
 498                 gfs2_rgrp_bh_put(rgd);
 499         }
 500         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 501 }
 502
 503 /**
 504  * databuf_lo_add - Add a databuf to the transaction.
 505  *
 506  * This is used in two distinct cases:
 507  * i) In ordered write mode
 508  *    We put the data buffer on a list so that we can ensure that its
 509  *    synced to disk at the right time
 510  * ii) In journaled data mode
 511  *    We need to journal the data block in the same way as metadata in
 512  *    the functions above. The difference is that here we have a tag
 513  *    which is two __be64's being the block number (as per meta data)
 514  *    and a flag which says whether the data block needs escaping or
 515  *    not. This means we need a new log entry for each 251 or so data
 516  *    blocks, which isn't an enormous overhead but twice as much as
 517  *    for normal metadata blocks.
 518  */
 519 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 520 {
 521         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 522         struct gfs2_trans *tr = current->journal_info;
 523         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 524         struct gfs2_inode *ip = GFS2_I(mapping->host);
 525
 526         if (tr) {
 527                 if (!list_empty(&bd->bd_list_tr))
 528                         return;
 529                 tr->tr_touched = 1;
 530                 if (gfs2_is_jdata(ip)) {
 531                         tr->tr_num_buf++;
 532                         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 533                 }
 534         }
 535         if (!list_empty(&le->le_list))
 536                 return;
 537
 538         set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
 539         set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
 540         if (gfs2_is_jdata(ip)) {
 541                 gfs2_pin(sdp, bd->bd_bh);
 542                 tr->tr_num_databuf_new++;
 543                 sdp->sd_log_num_databuf++;
 544                 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
 545         } else {
 546                 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
 547         }
 548 }
 549
 550 static void gfs2_check_magic(struct buffer_head *bh)
 551 {
 552         void *kaddr;
 553         __be32 *ptr;
 554
 555         clear_buffer_escaped(bh);
 556         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 557         ptr = kaddr + bh_offset(bh);
 558         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 559                 set_buffer_escaped(bh);
 560         kunmap_atomic(kaddr, KM_USER0);
 561 }
 562
 563 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
 564                               struct list_head *list, struct list_head *done,
 565                               unsigned int n)
 566 {
 567         struct buffer_head *bh1;
 568         struct gfs2_log_descriptor *ld;
 569         struct gfs2_bufdata *bd;
 570         __be64 *ptr;
 571
 572         if (!bh)
 573                 return;
 574
 575         ld = bh_log_desc(bh);
 576         ld->ld_length = cpu_to_be32(n + 1);
 577         ld->ld_data1 = cpu_to_be32(n);
 578
 579         ptr = bh_log_ptr(bh);
 580
 581         get_bh(bh);
 582         submit_bh(WRITE_SYNC, bh);
 583         gfs2_log_lock(sdp);
 584         while(!list_empty(list)) {
 585                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
 586                 list_move_tail(&bd->bd_le.le_list, done);
 587                 get_bh(bd->bd_bh);
 588                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
 589                         gfs2_log_incr_head(sdp);
 590                         ptr += 2;
 591                 }
 592                 gfs2_log_unlock(sdp);
 593                 lock_buffer(bd->bd_bh);
 594                 if (buffer_escaped(bd->bd_bh)) {
 595                         void *kaddr;
 596                         bh1 = gfs2_log_get_buf(sdp);
 597                         kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
 598                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
 599                                bh1->b_size);
 600                         kunmap_atomic(kaddr, KM_USER0);
 601                         *(__be32 *)bh1->b_data = 0;
 602                         clear_buffer_escaped(bd->bd_bh);
 603                         unlock_buffer(bd->bd_bh);
 604                         brelse(bd->bd_bh);
 605                 } else {
 606                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
 607                 }
 608                 submit_bh(WRITE_SYNC, bh1);
 609                 gfs2_log_lock(sdp);
 610                 ptr += 2;
 611         }
 612         gfs2_log_unlock(sdp);
 613         brelse(bh);
 614 }
 615
 616 /**
 617  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 618  *
 619  */
 620
 621 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 622 {
 623         struct gfs2_bufdata *bd = NULL;
 624         struct buffer_head *bh = NULL;
 625         unsigned int n = 0;
 626         __be64 *ptr = NULL, *end = NULL;
 627         LIST_HEAD(processed);
 628         LIST_HEAD(in_progress);
 629
 630         gfs2_log_lock(sdp);
 631         while (!list_empty(&sdp->sd_log_le_databuf)) {
 632                 if (ptr == end) {
 633                         gfs2_log_unlock(sdp);
 634                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 635                         n = 0;
 636                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
 637                         ptr = bh_log_ptr(bh);
 638                         end = bh_ptr_end(bh) - 1;
 639                         gfs2_log_lock(sdp);
 640                         continue;
 641                 }
 642                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
 643                 list_move_tail(&bd->bd_le.le_list, &in_progress);
 644                 gfs2_check_magic(bd->bd_bh);
 645                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
 646                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
 647                 n++;
 648         }
 649         gfs2_log_unlock(sdp);
 650         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 651         gfs2_log_lock(sdp);
 652         list_splice(&processed, &sdp->sd_log_le_databuf);
 653         gfs2_log_unlock(sdp);
 654 }
 655
 656 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 657                                     struct gfs2_log_descriptor *ld,
 658                                     __be64 *ptr, int pass)
 659 {
 660         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 661         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 662         struct gfs2_glock *gl = ip->i_gl;
 663         unsigned int blks = be32_to_cpu(ld->ld_data1);
 664         struct buffer_head *bh_log, *bh_ip;
 665         u64 blkno;
 666         u64 esc;
 667         int error = 0;
 668
 669         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 670                 return 0;
 671
 672         gfs2_replay_incr_blk(sdp, &start);
 673         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 674                 blkno = be64_to_cpu(*ptr++);
 675                 esc = be64_to_cpu(*ptr++);
 676
 677                 sdp->sd_found_blocks++;
 678
 679                 if (gfs2_revoke_check(sdp, blkno, start))
 680                         continue;
 681
 682                 error = gfs2_replay_read_block(jd, start, &bh_log);
 683                 if (error)
 684                         return error;
 685
 686                 bh_ip = gfs2_meta_new(gl, blkno);
 687                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 688
 689                 /* Unescape */
 690                 if (esc) {
 691                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 692                         *eptr = cpu_to_be32(GFS2_MAGIC);
 693                 }
 694                 mark_buffer_dirty(bh_ip);
 695
 696                 brelse(bh_log);
 697                 brelse(bh_ip);
 698                 if (error)
 699                         break;
 700
 701                 sdp->sd_replayed_blocks++;
 702         }
 703
 704         return error;
 705 }
 706
 707 /* FIXME: sort out accounting for log blocks etc. */
 708
 709 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 710 {
 711         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 712         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 713
 714         if (error) {
 715                 gfs2_meta_sync(ip->i_gl);
 716                 return;
 717         }
 718         if (pass != 1)
 719                 return;
 720
 721         /* data sync? */
 722         gfs2_meta_sync(ip->i_gl);
 723
 724         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 725                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 726 }
 727
 728 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 729 {
 730         struct list_head *head = &sdp->sd_log_le_databuf;
 731         struct gfs2_bufdata *bd;
 732
 733         while (!list_empty(head)) {
 734                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 735                 list_del_init(&bd->bd_le.le_list);
 736                 sdp->sd_log_num_databuf--;
 737                 gfs2_unpin(sdp, bd->bd_bh, ai);
 738         }
 739         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 740 }
 741
 742
 743 const struct gfs2_log_operations gfs2_buf_lops = {
 744         .lo_add = buf_lo_add,
 745         .lo_before_commit = buf_lo_before_commit,
 746         .lo_after_commit = buf_lo_after_commit,
 747         .lo_before_scan = buf_lo_before_scan,
 748         .lo_scan_elements = buf_lo_scan_elements,
 749         .lo_after_scan = buf_lo_after_scan,
 750         .lo_name = "buf",
 751 };
 752
 753 const struct gfs2_log_operations gfs2_revoke_lops = {
 754         .lo_add = revoke_lo_add,
 755         .lo_before_commit = revoke_lo_before_commit,
 756         .lo_after_commit = revoke_lo_after_commit,
 757         .lo_before_scan = revoke_lo_before_scan,
 758         .lo_scan_elements = revoke_lo_scan_elements,
 759         .lo_after_scan = revoke_lo_after_scan,
 760         .lo_name = "revoke",
 761 };
 762
 763 const struct gfs2_log_operations gfs2_rg_lops = {
 764         .lo_add = rg_lo_add,
 765         .lo_after_commit = rg_lo_after_commit,
 766         .lo_name = "rg",
 767 };
 768
 769 const struct gfs2_log_operations gfs2_databuf_lops = {
 770         .lo_add = databuf_lo_add,
 771         .lo_before_commit = databuf_lo_before_commit,
 772         .lo_after_commit = databuf_lo_after_commit,
 773         .lo_scan_elements = databuf_lo_scan_elements,
 774         .lo_after_scan = databuf_lo_after_scan,
 775         .lo_name = "databuf",
 776 };
 777
 778 const struct gfs2_log_operations *gfs2_log_ops[] = {
 779         &gfs2_databuf_lops,
 780         &gfs2_buf_lops,
 781         &gfs2_rg_lops,
 782         &gfs2_revoke_lops,
 783         NULL,
 784 };
 785