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