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         gfs2_log_unlock(sdp);
  91         unlock_buffer(bh);
  92 }
  93
  94
  95 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
  96 {
  97         return (struct gfs2_log_descriptor *)bh->b_data;
  98 }
  99
 100 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
 101 {
 102         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 103         return (__force __be64 *)(ld + 1);
 104 }
 105
 106 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
 107 {
 108         return (__force __be64 *)(bh->b_data + bh->b_size);
 109 }
 110
 111
 112 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
 113 {
 114         struct buffer_head *bh = gfs2_log_get_buf(sdp);
 115         struct gfs2_log_descriptor *ld = bh_log_desc(bh);
 116         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 117         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 118         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 119         ld->ld_type = cpu_to_be32(ld_type);
 120         ld->ld_length = 0;
 121         ld->ld_data1 = 0;
 122         ld->ld_data2 = 0;
 123         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 124         return bh;
 125 }
 126
 127 static void __glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 128 {
 129         struct gfs2_glock *gl;
 130         struct gfs2_trans *tr = current->journal_info;
 131
 132         tr->tr_touched = 1;
 133
 134         gl = container_of(le, struct gfs2_glock, gl_le);
 135         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
 136                 return;
 137
 138         if (!list_empty(&le->le_list))
 139                 return;
 140
 141         gfs2_glock_hold(gl);
 142         set_bit(GLF_DIRTY, &gl->gl_flags);
 143         sdp->sd_log_num_gl++;
 144         list_add(&le->le_list, &sdp->sd_log_le_gl);
 145 }
 146
 147 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 148 {
 149         gfs2_log_lock(sdp);
 150         __glock_lo_add(sdp, le);
 151         gfs2_log_unlock(sdp);
 152 }
 153
 154 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 155 {
 156         struct list_head *head = &sdp->sd_log_le_gl;
 157         struct gfs2_glock *gl;
 158
 159         while (!list_empty(head)) {
 160                 gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
 161                 list_del_init(&gl->gl_le.le_list);
 162                 sdp->sd_log_num_gl--;
 163
 164                 gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
 165                 gfs2_glock_put(gl);
 166         }
 167         gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
 168 }
 169
 170 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 171 {
 172         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 173         struct gfs2_trans *tr;
 174
 175         lock_buffer(bd->bd_bh);
 176         gfs2_log_lock(sdp);
 177         if (!list_empty(&bd->bd_list_tr))
 178                 goto out;
 179         tr = current->journal_info;
 180         tr->tr_touched = 1;
 181         tr->tr_num_buf++;
 182         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 183         if (!list_empty(&le->le_list))
 184                 goto out;
 185         __glock_lo_add(sdp, &bd->bd_gl->gl_le);
 186         gfs2_meta_check(sdp, bd->bd_bh);
 187         gfs2_pin(sdp, bd->bd_bh);
 188         sdp->sd_log_num_buf++;
 189         list_add(&le->le_list, &sdp->sd_log_le_buf);
 190         tr->tr_num_buf_new++;
 191 out:
 192         gfs2_log_unlock(sdp);
 193         unlock_buffer(bd->bd_bh);
 194 }
 195
 196 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 197 {
 198         struct list_head *head = &tr->tr_list_buf;
 199         struct gfs2_bufdata *bd;
 200
 201         gfs2_log_lock(sdp);
 202         while (!list_empty(head)) {
 203                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
 204                 list_del_init(&bd->bd_list_tr);
 205                 tr->tr_num_buf--;
 206         }
 207         gfs2_log_unlock(sdp);
 208         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 209 }
 210
 211 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 212 {
 213         struct buffer_head *bh;
 214         struct gfs2_log_descriptor *ld;
 215         struct gfs2_bufdata *bd1 = NULL, *bd2;
 216         unsigned int total;
 217         unsigned int limit;
 218         unsigned int num;
 219         unsigned n;
 220         __be64 *ptr;
 221
 222         limit = buf_limit(sdp);
 223         /* for 4k blocks, limit = 503 */
 224
 225         gfs2_log_lock(sdp);
 226         total = sdp->sd_log_num_buf;
 227         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 228         while(total) {
 229                 num = total;
 230                 if (total > limit)
 231                         num = limit;
 232                 gfs2_log_unlock(sdp);
 233                 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
 234                 gfs2_log_lock(sdp);
 235                 ld = bh_log_desc(bh);
 236                 ptr = bh_log_ptr(bh);
 237                 ld->ld_length = cpu_to_be32(num + 1);
 238                 ld->ld_data1 = cpu_to_be32(num);
 239
 240                 n = 0;
 241                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 242                                              bd_le.le_list) {
 243                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 244                         if (++n >= num)
 245                                 break;
 246                 }
 247
 248                 gfs2_log_unlock(sdp);
 249                 submit_bh(WRITE, bh);
 250                 gfs2_log_lock(sdp);
 251
 252                 n = 0;
 253                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 254                                              bd_le.le_list) {
 255                         get_bh(bd2->bd_bh);
 256                         gfs2_log_unlock(sdp);
 257                         lock_buffer(bd2->bd_bh);
 258                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 259                         submit_bh(WRITE, bh);
 260                         gfs2_log_lock(sdp);
 261                         if (++n >= num)
 262                                 break;
 263                 }
 264
 265                 BUG_ON(total < num);
 266                 total -= num;
 267         }
 268         gfs2_log_unlock(sdp);
 269 }
 270
 271 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 272 {
 273         struct list_head *head = &sdp->sd_log_le_buf;
 274         struct gfs2_bufdata *bd;
 275
 276         while (!list_empty(head)) {
 277                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 278                 list_del_init(&bd->bd_le.le_list);
 279                 sdp->sd_log_num_buf--;
 280
 281                 gfs2_unpin(sdp, bd->bd_bh, ai);
 282         }
 283         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 284 }
 285
 286 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 287                                struct gfs2_log_header_host *head, int pass)
 288 {
 289         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 290
 291         if (pass != 0)
 292                 return;
 293
 294         sdp->sd_found_blocks = 0;
 295         sdp->sd_replayed_blocks = 0;
 296 }
 297
 298 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 299                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 300                                 int pass)
 301 {
 302         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 303         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 304         struct gfs2_glock *gl = ip->i_gl;
 305         unsigned int blks = be32_to_cpu(ld->ld_data1);
 306         struct buffer_head *bh_log, *bh_ip;
 307         u64 blkno;
 308         int error = 0;
 309
 310         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 311                 return 0;
 312
 313         gfs2_replay_incr_blk(sdp, &start);
 314
 315         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 316                 blkno = be64_to_cpu(*ptr++);
 317
 318                 sdp->sd_found_blocks++;
 319
 320                 if (gfs2_revoke_check(sdp, blkno, start))
 321                         continue;
 322
 323                 error = gfs2_replay_read_block(jd, start, &bh_log);
 324                 if (error)
 325                         return error;
 326
 327                 bh_ip = gfs2_meta_new(gl, blkno);
 328                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 329
 330                 if (gfs2_meta_check(sdp, bh_ip))
 331                         error = -EIO;
 332                 else
 333                         mark_buffer_dirty(bh_ip);
 334
 335                 brelse(bh_log);
 336                 brelse(bh_ip);
 337
 338                 if (error)
 339                         break;
 340
 341                 sdp->sd_replayed_blocks++;
 342         }
 343
 344         return error;
 345 }
 346
 347 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 348 {
 349         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 350         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 351
 352         if (error) {
 353                 gfs2_meta_sync(ip->i_gl);
 354                 return;
 355         }
 356         if (pass != 1)
 357                 return;
 358
 359         gfs2_meta_sync(ip->i_gl);
 360
 361         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 362                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 363 }
 364
 365 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 366 {
 367         struct gfs2_trans *tr;
 368
 369         tr = current->journal_info;
 370         tr->tr_touched = 1;
 371         tr->tr_num_revoke++;
 372         sdp->sd_log_num_revoke++;
 373         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 374 }
 375
 376 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 377 {
 378         struct gfs2_log_descriptor *ld;
 379         struct gfs2_meta_header *mh;
 380         struct buffer_head *bh;
 381         unsigned int offset;
 382         struct list_head *head = &sdp->sd_log_le_revoke;
 383         struct gfs2_bufdata *bd;
 384
 385         if (!sdp->sd_log_num_revoke)
 386                 return;
 387
 388         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
 389         ld = bh_log_desc(bh);
 390         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 391                                                     sizeof(u64)));
 392         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 393         offset = sizeof(struct gfs2_log_descriptor);
 394
 395         while (!list_empty(head)) {
 396                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 397                 list_del_init(&bd->bd_le.le_list);
 398                 sdp->sd_log_num_revoke--;
 399
 400                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 401                         submit_bh(WRITE, bh);
 402
 403                         bh = gfs2_log_get_buf(sdp);
 404                         mh = (struct gfs2_meta_header *)bh->b_data;
 405                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 406                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 407                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 408                         offset = sizeof(struct gfs2_meta_header);
 409                 }
 410
 411                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
 412                 kmem_cache_free(gfs2_bufdata_cachep, bd);
 413
 414                 offset += sizeof(u64);
 415         }
 416         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 417
 418         submit_bh(WRITE, bh);
 419 }
 420
 421 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 422                                   struct gfs2_log_header_host *head, int pass)
 423 {
 424         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 425
 426         if (pass != 0)
 427                 return;
 428
 429         sdp->sd_found_revokes = 0;
 430         sdp->sd_replay_tail = head->lh_tail;
 431 }
 432
 433 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 434                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 435                                    int pass)
 436 {
 437         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 438         unsigned int blks = be32_to_cpu(ld->ld_length);
 439         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 440         struct buffer_head *bh;
 441         unsigned int offset;
 442         u64 blkno;
 443         int first = 1;
 444         int error;
 445
 446         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 447                 return 0;
 448
 449         offset = sizeof(struct gfs2_log_descriptor);
 450
 451         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 452                 error = gfs2_replay_read_block(jd, start, &bh);
 453                 if (error)
 454                         return error;
 455
 456                 if (!first)
 457                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 458
 459                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 460                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 461
 462                         error = gfs2_revoke_add(sdp, blkno, start);
 463                         if (error < 0)
 464                                 return error;
 465                         else if (error)
 466                                 sdp->sd_found_revokes++;
 467
 468                         if (!--revokes)
 469                                 break;
 470                         offset += sizeof(u64);
 471                 }
 472
 473                 brelse(bh);
 474                 offset = sizeof(struct gfs2_meta_header);
 475                 first = 0;
 476         }
 477
 478         return 0;
 479 }
 480
 481 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 482 {
 483         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 484
 485         if (error) {
 486                 gfs2_revoke_clean(sdp);
 487                 return;
 488         }
 489         if (pass != 1)
 490                 return;
 491
 492         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 493                 jd->jd_jid, sdp->sd_found_revokes);
 494
 495         gfs2_revoke_clean(sdp);
 496 }
 497
 498 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 499 {
 500         struct gfs2_rgrpd *rgd;
 501         struct gfs2_trans *tr = current->journal_info;
 502
 503         tr->tr_touched = 1;
 504
 505         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 506
 507         gfs2_log_lock(sdp);
 508         if (!list_empty(&le->le_list)){
 509                 gfs2_log_unlock(sdp);
 510                 return;
 511         }
 512         gfs2_rgrp_bh_hold(rgd);
 513         sdp->sd_log_num_rg++;
 514         list_add(&le->le_list, &sdp->sd_log_le_rg);
 515         gfs2_log_unlock(sdp);
 516 }
 517
 518 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 519 {
 520         struct list_head *head = &sdp->sd_log_le_rg;
 521         struct gfs2_rgrpd *rgd;
 522
 523         while (!list_empty(head)) {
 524                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 525                 list_del_init(&rgd->rd_le.le_list);
 526                 sdp->sd_log_num_rg--;
 527
 528                 gfs2_rgrp_repolish_clones(rgd);
 529                 gfs2_rgrp_bh_put(rgd);
 530         }
 531         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 532 }
 533
 534 /**
 535  * databuf_lo_add - Add a databuf to the transaction.
 536  *
 537  * This is used in two distinct cases:
 538  * i) In ordered write mode
 539  *    We put the data buffer on a list so that we can ensure that its
 540  *    synced to disk at the right time
 541  * ii) In journaled data mode
 542  *    We need to journal the data block in the same way as metadata in
 543  *    the functions above. The difference is that here we have a tag
 544  *    which is two __be64's being the block number (as per meta data)
 545  *    and a flag which says whether the data block needs escaping or
 546  *    not. This means we need a new log entry for each 251 or so data
 547  *    blocks, which isn't an enormous overhead but twice as much as
 548  *    for normal metadata blocks.
 549  */
 550 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 551 {
 552         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 553         struct gfs2_trans *tr = current->journal_info;
 554         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 555         struct gfs2_inode *ip = GFS2_I(mapping->host);
 556
 557         lock_buffer(bd->bd_bh);
 558         gfs2_log_lock(sdp);
 559         if (!list_empty(&bd->bd_list_tr))
 560                 goto out;
 561         tr->tr_touched = 1;
 562         if (gfs2_is_jdata(ip)) {
 563                 tr->tr_num_buf++;
 564                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 565         }
 566         if (!list_empty(&le->le_list))
 567                 goto out;
 568
 569         __glock_lo_add(sdp, &bd->bd_gl->gl_le);
 570         if (gfs2_is_jdata(ip)) {
 571                 gfs2_pin(sdp, bd->bd_bh);
 572                 tr->tr_num_databuf_new++;
 573                 sdp->sd_log_num_databuf++;
 574                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
 575         } else {
 576                 list_add(&le->le_list, &sdp->sd_log_le_ordered);
 577         }
 578 out:
 579         gfs2_log_unlock(sdp);
 580         unlock_buffer(bd->bd_bh);
 581 }
 582
 583 static void gfs2_check_magic(struct buffer_head *bh)
 584 {
 585         void *kaddr;
 586         __be32 *ptr;
 587
 588         clear_buffer_escaped(bh);
 589         kaddr = kmap_atomic(bh->b_page, KM_USER0);
 590         ptr = kaddr + bh_offset(bh);
 591         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 592                 set_buffer_escaped(bh);
 593         kunmap_atomic(kaddr, KM_USER0);
 594 }
 595
 596 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
 597                               struct list_head *list, struct list_head *done,
 598                               unsigned int n)
 599 {
 600         struct buffer_head *bh1;
 601         struct gfs2_log_descriptor *ld;
 602         struct gfs2_bufdata *bd;
 603         __be64 *ptr;
 604
 605         if (!bh)
 606                 return;
 607
 608         ld = bh_log_desc(bh);
 609         ld->ld_length = cpu_to_be32(n + 1);
 610         ld->ld_data1 = cpu_to_be32(n);
 611
 612         ptr = bh_log_ptr(bh);
 613
 614         get_bh(bh);
 615         submit_bh(WRITE, bh);
 616         gfs2_log_lock(sdp);
 617         while(!list_empty(list)) {
 618                 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
 619                 list_move_tail(&bd->bd_le.le_list, done);
 620                 get_bh(bd->bd_bh);
 621                 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
 622                         gfs2_log_incr_head(sdp);
 623                         ptr += 2;
 624                 }
 625                 gfs2_log_unlock(sdp);
 626                 lock_buffer(bd->bd_bh);
 627                 if (buffer_escaped(bd->bd_bh)) {
 628                         void *kaddr;
 629                         bh1 = gfs2_log_get_buf(sdp);
 630                         kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
 631                         memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
 632                                bh1->b_size);
 633                         kunmap_atomic(kaddr, KM_USER0);
 634                         *(__be32 *)bh1->b_data = 0;
 635                         clear_buffer_escaped(bd->bd_bh);
 636                         unlock_buffer(bd->bd_bh);
 637                         brelse(bd->bd_bh);
 638                 } else {
 639                         bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
 640                 }
 641                 submit_bh(WRITE, bh1);
 642                 gfs2_log_lock(sdp);
 643                 ptr += 2;
 644         }
 645         gfs2_log_unlock(sdp);
 646         brelse(bh);
 647 }
 648
 649 /**
 650  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 651  *
 652  */
 653
 654 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 655 {
 656         struct gfs2_bufdata *bd = NULL;
 657         struct buffer_head *bh = NULL;
 658         unsigned int n = 0;
 659         __be64 *ptr = NULL, *end = NULL;
 660         LIST_HEAD(processed);
 661         LIST_HEAD(in_progress);
 662
 663         gfs2_log_lock(sdp);
 664         while (!list_empty(&sdp->sd_log_le_databuf)) {
 665                 if (ptr == end) {
 666                         gfs2_log_unlock(sdp);
 667                         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 668                         n = 0;
 669                         bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
 670                         ptr = bh_log_ptr(bh);
 671                         end = bh_ptr_end(bh) - 1;
 672                         gfs2_log_lock(sdp);
 673                         continue;
 674                 }
 675                 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
 676                 list_move_tail(&bd->bd_le.le_list, &in_progress);
 677                 gfs2_check_magic(bd->bd_bh);
 678                 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
 679                 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
 680                 n++;
 681         }
 682         gfs2_log_unlock(sdp);
 683         gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
 684         gfs2_log_lock(sdp);
 685         list_splice(&processed, &sdp->sd_log_le_databuf);
 686         gfs2_log_unlock(sdp);
 687 }
 688
 689 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 690                                     struct gfs2_log_descriptor *ld,
 691                                     __be64 *ptr, int pass)
 692 {
 693         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 694         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 695         struct gfs2_glock *gl = ip->i_gl;
 696         unsigned int blks = be32_to_cpu(ld->ld_data1);
 697         struct buffer_head *bh_log, *bh_ip;
 698         u64 blkno;
 699         u64 esc;
 700         int error = 0;
 701
 702         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 703                 return 0;
 704
 705         gfs2_replay_incr_blk(sdp, &start);
 706         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 707                 blkno = be64_to_cpu(*ptr++);
 708                 esc = be64_to_cpu(*ptr++);
 709
 710                 sdp->sd_found_blocks++;
 711
 712                 if (gfs2_revoke_check(sdp, blkno, start))
 713                         continue;
 714
 715                 error = gfs2_replay_read_block(jd, start, &bh_log);
 716                 if (error)
 717                         return error;
 718
 719                 bh_ip = gfs2_meta_new(gl, blkno);
 720                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 721
 722                 /* Unescape */
 723                 if (esc) {
 724                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 725                         *eptr = cpu_to_be32(GFS2_MAGIC);
 726                 }
 727                 mark_buffer_dirty(bh_ip);
 728
 729                 brelse(bh_log);
 730                 brelse(bh_ip);
 731                 if (error)
 732                         break;
 733
 734                 sdp->sd_replayed_blocks++;
 735         }
 736
 737         return error;
 738 }
 739
 740 /* FIXME: sort out accounting for log blocks etc. */
 741
 742 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 743 {
 744         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 745         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 746
 747         if (error) {
 748                 gfs2_meta_sync(ip->i_gl);
 749                 return;
 750         }
 751         if (pass != 1)
 752                 return;
 753
 754         /* data sync? */
 755         gfs2_meta_sync(ip->i_gl);
 756
 757         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 758                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 759 }
 760
 761 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 762 {
 763         struct list_head *head = &sdp->sd_log_le_databuf;
 764         struct gfs2_bufdata *bd;
 765
 766         while (!list_empty(head)) {
 767                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 768                 list_del_init(&bd->bd_le.le_list);
 769                 sdp->sd_log_num_databuf--;
 770                 gfs2_unpin(sdp, bd->bd_bh, ai);
 771         }
 772         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 773 }
 774
 775
 776 const struct gfs2_log_operations gfs2_glock_lops = {
 777         .lo_add = glock_lo_add,
 778         .lo_after_commit = glock_lo_after_commit,
 779         .lo_name = "glock",
 780 };
 781
 782 const struct gfs2_log_operations gfs2_buf_lops = {
 783         .lo_add = buf_lo_add,
 784         .lo_incore_commit = buf_lo_incore_commit,
 785         .lo_before_commit = buf_lo_before_commit,
 786         .lo_after_commit = buf_lo_after_commit,
 787         .lo_before_scan = buf_lo_before_scan,
 788         .lo_scan_elements = buf_lo_scan_elements,
 789         .lo_after_scan = buf_lo_after_scan,
 790         .lo_name = "buf",
 791 };
 792
 793 const struct gfs2_log_operations gfs2_revoke_lops = {
 794         .lo_add = revoke_lo_add,
 795         .lo_before_commit = revoke_lo_before_commit,
 796         .lo_before_scan = revoke_lo_before_scan,
 797         .lo_scan_elements = revoke_lo_scan_elements,
 798         .lo_after_scan = revoke_lo_after_scan,
 799         .lo_name = "revoke",
 800 };
 801
 802 const struct gfs2_log_operations gfs2_rg_lops = {
 803         .lo_add = rg_lo_add,
 804         .lo_after_commit = rg_lo_after_commit,
 805         .lo_name = "rg",
 806 };
 807
 808 const struct gfs2_log_operations gfs2_databuf_lops = {
 809         .lo_add = databuf_lo_add,
 810         .lo_incore_commit = buf_lo_incore_commit,
 811         .lo_before_commit = databuf_lo_before_commit,
 812         .lo_after_commit = databuf_lo_after_commit,
 813         .lo_scan_elements = databuf_lo_scan_elements,
 814         .lo_after_scan = databuf_lo_after_scan,
 815         .lo_name = "databuf",
 816 };
 817
 818 const struct gfs2_log_operations *gfs2_log_ops[] = {
 819         &gfs2_glock_lops,
 820         &gfs2_databuf_lops,
 821         &gfs2_buf_lops,
 822         &gfs2_rg_lops,
 823         &gfs2_revoke_lops,
 824         NULL,
 825 };
 826