fs/gfs2/lops.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2005 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 v.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 <asm/semaphore.h>
  17
  18 #include "gfs2.h"
  19 #include "lm_interface.h"
  20 #include "incore.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 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  31 {
  32         struct gfs2_glock *gl;
  33         struct gfs2_trans *tr = current->journal_info;
  34
  35         tr->tr_touched = 1;
  36
  37         if (!list_empty(&le->le_list))
  38                 return;
  39
  40         gl = container_of(le, struct gfs2_glock, gl_le);
  41         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
  42                 return;
  43         gfs2_glock_hold(gl);
  44         set_bit(GLF_DIRTY, &gl->gl_flags);
  45
  46         gfs2_log_lock(sdp);
  47         sdp->sd_log_num_gl++;
  48         list_add(&le->le_list, &sdp->sd_log_le_gl);
  49         gfs2_log_unlock(sdp);
  50 }
  51
  52 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  53 {
  54         struct list_head *head = &sdp->sd_log_le_gl;
  55         struct gfs2_glock *gl;
  56
  57         while (!list_empty(head)) {
  58                 gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
  59                 list_del_init(&gl->gl_le.le_list);
  60                 sdp->sd_log_num_gl--;
  61
  62                 gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
  63                 gfs2_glock_put(gl);
  64         }
  65         gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
  66 }
  67
  68 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  69 {
  70         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
  71         struct gfs2_trans *tr;
  72
  73         if (!list_empty(&bd->bd_list_tr))
  74                 return;
  75
  76         tr = current->journal_info;
  77         tr->tr_touched = 1;
  78         tr->tr_num_buf++;
  79         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
  80
  81         if (!list_empty(&le->le_list))
  82                 return;
  83
  84         gfs2_trans_add_gl(bd->bd_gl);
  85
  86         gfs2_meta_check(sdp, bd->bd_bh);
  87         gfs2_pin(sdp, bd->bd_bh);
  88
  89         gfs2_log_lock(sdp);
  90         sdp->sd_log_num_buf++;
  91         list_add(&le->le_list, &sdp->sd_log_le_buf);
  92         gfs2_log_unlock(sdp);
  93
  94         tr->tr_num_buf_new++;
  95 }
  96
  97 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
  98 {
  99         struct list_head *head = &tr->tr_list_buf;
 100         struct gfs2_bufdata *bd;
 101
 102         while (!list_empty(head)) {
 103                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
 104                 list_del_init(&bd->bd_list_tr);
 105                 tr->tr_num_buf--;
 106         }
 107         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 108 }
 109
 110 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 111 {
 112         struct buffer_head *bh;
 113         struct gfs2_log_descriptor *ld;
 114         struct gfs2_bufdata *bd1 = NULL, *bd2;
 115         unsigned int total = sdp->sd_log_num_buf;
 116         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 117         unsigned int limit;
 118         unsigned int num;
 119         unsigned n;
 120         __be64 *ptr;
 121
 122         offset += (sizeof(__be64) - 1);
 123         offset &= ~(sizeof(__be64) - 1);
 124         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 125         /* for 4k blocks, limit = 503 */
 126
 127         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 128         while(total) {
 129                 num = total;
 130                 if (total > limit)
 131                         num = limit;
 132                 bh = gfs2_log_get_buf(sdp);
 133                 sdp->sd_log_num_hdrs++;
 134                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 135                 ptr = (__be64 *)(bh->b_data + offset);
 136                 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 137                 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 138                 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 139                 ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
 140                 ld->ld_length = cpu_to_be32(num + 1);
 141                 ld->ld_data1 = cpu_to_be32(num);
 142                 ld->ld_data2 = cpu_to_be32(0);
 143                 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 144
 145                 n = 0;
 146                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 147                                              bd_le.le_list) {
 148                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 149                         if (++n >= num)
 150                                 break;
 151                 }
 152
 153                 set_buffer_dirty(bh);
 154                 ll_rw_block(WRITE, 1, &bh);
 155
 156                 n = 0;
 157                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 158                                              bd_le.le_list) {
 159                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 160                         set_buffer_dirty(bh);
 161                         ll_rw_block(WRITE, 1, &bh);
 162                         if (++n >= num)
 163                                 break;
 164                 }
 165
 166                 total -= num;
 167         }
 168 }
 169
 170 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 171 {
 172         struct list_head *head = &sdp->sd_log_le_buf;
 173         struct gfs2_bufdata *bd;
 174
 175         while (!list_empty(head)) {
 176                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 177                 list_del_init(&bd->bd_le.le_list);
 178                 sdp->sd_log_num_buf--;
 179
 180                 gfs2_unpin(sdp, bd->bd_bh, ai);
 181         }
 182         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 183 }
 184
 185 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 186                                struct gfs2_log_header *head, int pass)
 187 {
 188         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 189         struct gfs2_sbd *sdp = ip->i_sbd;
 190
 191         if (pass != 0)
 192                 return;
 193
 194         sdp->sd_found_blocks = 0;
 195         sdp->sd_replayed_blocks = 0;
 196 }
 197
 198 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 199                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 200                                 int pass)
 201 {
 202         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 203         struct gfs2_sbd *sdp = ip->i_sbd;
 204         struct gfs2_glock *gl = ip->i_gl;
 205         unsigned int blks = be32_to_cpu(ld->ld_data1);
 206         struct buffer_head *bh_log, *bh_ip;
 207         uint64_t blkno;
 208         int error = 0;
 209
 210         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 211                 return 0;
 212
 213         gfs2_replay_incr_blk(sdp, &start);
 214
 215         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 216                 blkno = be64_to_cpu(*ptr++);
 217
 218                 sdp->sd_found_blocks++;
 219
 220                 if (gfs2_revoke_check(sdp, blkno, start))
 221                         continue;
 222
 223                 error = gfs2_replay_read_block(jd, start, &bh_log);
 224                 if (error)
 225                         return error;
 226
 227                 bh_ip = gfs2_meta_new(gl, blkno);
 228                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 229
 230                 if (gfs2_meta_check(sdp, bh_ip))
 231                         error = -EIO;
 232                 else
 233                         mark_buffer_dirty(bh_ip);
 234
 235                 brelse(bh_log);
 236                 brelse(bh_ip);
 237
 238                 if (error)
 239                         break;
 240
 241                 sdp->sd_replayed_blocks++;
 242         }
 243
 244         return error;
 245 }
 246
 247 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 248 {
 249         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 250         struct gfs2_sbd *sdp = ip->i_sbd;
 251
 252         if (error) {
 253                 gfs2_meta_sync(ip->i_gl,
 254                                DIO_START | DIO_WAIT);
 255                 return;
 256         }
 257         if (pass != 1)
 258                 return;
 259
 260         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
 261
 262         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 263                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 264 }
 265
 266 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 267 {
 268         struct gfs2_trans *tr;
 269
 270         tr = current->journal_info;
 271         tr->tr_touched = 1;
 272         tr->tr_num_revoke++;
 273
 274         gfs2_log_lock(sdp);
 275         sdp->sd_log_num_revoke++;
 276         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 277         gfs2_log_unlock(sdp);
 278 }
 279
 280 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 281 {
 282         struct gfs2_log_descriptor *ld;
 283         struct gfs2_meta_header *mh;
 284         struct buffer_head *bh;
 285         unsigned int offset;
 286         struct list_head *head = &sdp->sd_log_le_revoke;
 287         struct gfs2_revoke *rv;
 288
 289         if (!sdp->sd_log_num_revoke)
 290                 return;
 291
 292         bh = gfs2_log_get_buf(sdp);
 293         ld = (struct gfs2_log_descriptor *)bh->b_data;
 294         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 295         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 296         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 297         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
 298         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 299                                                     sizeof(uint64_t)));
 300         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 301         ld->ld_data2 = cpu_to_be32(0);
 302         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 303         offset = sizeof(struct gfs2_log_descriptor);
 304
 305         while (!list_empty(head)) {
 306                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
 307                 list_del_init(&rv->rv_le.le_list);
 308                 sdp->sd_log_num_revoke--;
 309
 310                 if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) {
 311                         set_buffer_dirty(bh);
 312                         ll_rw_block(WRITE, 1, &bh);
 313
 314                         bh = gfs2_log_get_buf(sdp);
 315                         mh = (struct gfs2_meta_header *)bh->b_data;
 316                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 317                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 318                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 319                         offset = sizeof(struct gfs2_meta_header);
 320                 }
 321
 322                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
 323                 kfree(rv);
 324
 325                 offset += sizeof(uint64_t);
 326         }
 327         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 328
 329         set_buffer_dirty(bh);
 330         ll_rw_block(WRITE, 1, &bh);
 331 }
 332
 333 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 334                                   struct gfs2_log_header *head, int pass)
 335 {
 336         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 337         struct gfs2_sbd *sdp = ip->i_sbd;
 338
 339         if (pass != 0)
 340                 return;
 341
 342         sdp->sd_found_revokes = 0;
 343         sdp->sd_replay_tail = head->lh_tail;
 344 }
 345
 346 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 347                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 348                                    int pass)
 349 {
 350         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 351         struct gfs2_sbd *sdp = ip->i_sbd;
 352         unsigned int blks = be32_to_cpu(ld->ld_length);
 353         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 354         struct buffer_head *bh;
 355         unsigned int offset;
 356         uint64_t blkno;
 357         int first = 1;
 358         int error;
 359
 360         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 361                 return 0;
 362
 363         offset = sizeof(struct gfs2_log_descriptor);
 364
 365         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 366                 error = gfs2_replay_read_block(jd, start, &bh);
 367                 if (error)
 368                         return error;
 369
 370                 if (!first)
 371                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 372
 373                 while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) {
 374                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 375
 376                         error = gfs2_revoke_add(sdp, blkno, start);
 377                         if (error < 0)
 378                                 return error;
 379                         else if (error)
 380                                 sdp->sd_found_revokes++;
 381
 382                         if (!--revokes)
 383                                 break;
 384                         offset += sizeof(uint64_t);
 385                 }
 386
 387                 brelse(bh);
 388                 offset = sizeof(struct gfs2_meta_header);
 389                 first = 0;
 390         }
 391
 392         return 0;
 393 }
 394
 395 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 396 {
 397         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 398         struct gfs2_sbd *sdp = ip->i_sbd;
 399
 400         if (error) {
 401                 gfs2_revoke_clean(sdp);
 402                 return;
 403         }
 404         if (pass != 1)
 405                 return;
 406
 407         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 408                 jd->jd_jid, sdp->sd_found_revokes);
 409
 410         gfs2_revoke_clean(sdp);
 411 }
 412
 413 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 414 {
 415         struct gfs2_rgrpd *rgd;
 416         struct gfs2_trans *tr = current->journal_info;
 417
 418         tr->tr_touched = 1;
 419
 420         if (!list_empty(&le->le_list))
 421                 return;
 422
 423         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 424         gfs2_rgrp_bh_hold(rgd);
 425
 426         gfs2_log_lock(sdp);
 427         sdp->sd_log_num_rg++;
 428         list_add(&le->le_list, &sdp->sd_log_le_rg);
 429         gfs2_log_unlock(sdp);
 430 }
 431
 432 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 433 {
 434         struct list_head *head = &sdp->sd_log_le_rg;
 435         struct gfs2_rgrpd *rgd;
 436
 437         while (!list_empty(head)) {
 438                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 439                 list_del_init(&rgd->rd_le.le_list);
 440                 sdp->sd_log_num_rg--;
 441
 442                 gfs2_rgrp_repolish_clones(rgd);
 443                 gfs2_rgrp_bh_put(rgd);
 444         }
 445         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 446 }
 447
 448 /**
 449  * databuf_lo_add - Add a databuf to the transaction.
 450  *
 451  * This is used in two distinct cases:
 452  * i) In ordered write mode
 453  *    We put the data buffer on a list so that we can ensure that its
 454  *    synced to disk at the right time
 455  * ii) In journaled data mode
 456  *    We need to journal the data block in the same way as metadata in
 457  *    the functions above. The difference is that here we have a tag
 458  *    which is two __be64's being the block number (as per meta data)
 459  *    and a flag which says whether the data block needs escaping or
 460  *    not. This means we need a new log entry for each 251 or so data
 461  *    blocks, which isn't an enormous overhead but twice as much as
 462  *    for normal metadata blocks.
 463  */
 464 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 465 {
 466         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 467         struct gfs2_trans *tr = current->journal_info;
 468         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 469         struct gfs2_inode *ip = mapping->host->u.generic_ip;
 470
 471         tr->tr_touched = 1;
 472         if (!list_empty(&bd->bd_list_tr) &&
 473             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
 474                 tr->tr_num_buf++;
 475                 gfs2_trans_add_gl(bd->bd_gl);
 476                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 477                 gfs2_pin(sdp, bd->bd_bh);
 478                 tr->tr_num_buf_new++;
 479         }
 480         gfs2_log_lock(sdp);
 481         if (!list_empty(&le->le_list)) {
 482                 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
 483                         sdp->sd_log_num_jdata++;
 484                 sdp->sd_log_num_databuf++;
 485                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
 486         }
 487         gfs2_log_unlock(sdp);
 488 }
 489
 490 static int gfs2_check_magic(struct buffer_head *bh)
 491 {
 492         struct page *page = bh->b_page;
 493         void *kaddr;
 494         __be32 *ptr;
 495         int rv = 0;
 496
 497         kaddr = kmap_atomic(page, KM_USER0);
 498         ptr = kaddr + bh_offset(bh);
 499         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 500                 rv = 1;
 501         kunmap_atomic(page, KM_USER0);
 502
 503         return rv;
 504 }
 505
 506 /**
 507  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 508  *
 509  * Here we scan through the lists of buffers and make the assumption
 510  * that any buffer thats been pinned is being journaled, and that
 511  * any unpinned buffer is an ordered write data buffer and therefore
 512  * will be written back rather than journaled.
 513  */
 514 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 515 {
 516         LIST_HEAD(started);
 517         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
 518         struct buffer_head *bh = NULL;
 519         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 520         struct gfs2_log_descriptor *ld;
 521         unsigned int limit;
 522         unsigned int total_dbuf = sdp->sd_log_num_databuf;
 523         unsigned int total_jdata = sdp->sd_log_num_jdata;
 524         unsigned int num, n;
 525         __be64 *ptr = NULL;
 526
 527         offset += (2*sizeof(__be64) - 1);
 528         offset &= ~(2*sizeof(__be64) - 1);
 529         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 530
 531         /*
 532          * Start writing ordered buffers, write journaled buffers
 533          * into the log along with a header
 534          */
 535         gfs2_log_lock(sdp);
 536         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
 537                                        bd_le.le_list);
 538         while(total_dbuf) {
 539                 num = total_jdata;
 540                 if (num > limit)
 541                         num = limit;
 542                 n = 0;
 543                 list_for_each_entry_safe_continue(bd1, bdt,
 544                                                   &sdp->sd_log_le_databuf,
 545                                                   bd_le.le_list) {
 546                         /* An ordered write buffer */
 547                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
 548                                 list_move(&bd1->bd_le.le_list, &started);
 549                                 if (bd1 == bd2) {
 550                                         bd2 = NULL;
 551                                         bd2 = list_prepare_entry(bd2,
 552                                                         &sdp->sd_log_le_databuf,
 553                                                         bd_le.le_list);
 554                                 }
 555                                 total_dbuf--;
 556                                 if (bd1->bd_bh) {
 557                                         get_bh(bd1->bd_bh);
 558                                         if (buffer_dirty(bd1->bd_bh)) {
 559                                                 gfs2_log_unlock(sdp);
 560                                                 wait_on_buffer(bd1->bd_bh);
 561                                                 ll_rw_block(WRITE, 1,
 562                                                             &bd1->bd_bh);
 563                                                 gfs2_log_lock(sdp);
 564                                         }
 565                                         brelse(bd1->bd_bh);
 566                                         continue;
 567                                 }
 568                                 continue;
 569                         } else if (bd1->bd_bh) { /* A journaled buffer */
 570                                 int magic;
 571                                 gfs2_log_unlock(sdp);
 572                                 if (!bh) {
 573                                         bh = gfs2_log_get_buf(sdp);
 574                                         sdp->sd_log_num_hdrs++;
 575                                         ld = (struct gfs2_log_descriptor *)
 576                                              bh->b_data;
 577                                         ptr = (__be64 *)(bh->b_data + offset);
 578                                         ld->ld_header.mh_magic =
 579                                                 cpu_to_be32(GFS2_MAGIC);
 580                                         ld->ld_header.mh_type =
 581                                                 cpu_to_be32(GFS2_METATYPE_LD);
 582                                         ld->ld_header.mh_format =
 583                                                 cpu_to_be32(GFS2_FORMAT_LD);
 584                                         ld->ld_type =
 585                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
 586                                         ld->ld_length = cpu_to_be32(num + 1);
 587                                         ld->ld_data1 = cpu_to_be32(num);
 588                                         ld->ld_data2 = cpu_to_be32(0);
 589                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 590                                 }
 591                                 magic = gfs2_check_magic(bd1->bd_bh);
 592                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 593                                 *ptr++ = cpu_to_be64((__u64)magic);
 594                                 clear_buffer_escaped(bd1->bd_bh);
 595                                 if (unlikely(magic != 0))
 596                                         set_buffer_escaped(bd1->bd_bh);
 597                                 gfs2_log_lock(sdp);
 598                                 if (n++ > num)
 599                                         break;
 600                         }
 601                 }
 602                 gfs2_log_unlock(sdp);
 603                 if (bh) {
 604                         set_buffer_dirty(bh);
 605                         ll_rw_block(WRITE, 1, &bh);
 606                         bh = NULL;
 607                 }
 608                 n = 0;
 609                 gfs2_log_lock(sdp);
 610                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
 611                                              bd_le.le_list) {
 612                         if (!bd2->bd_bh)
 613                                 continue;
 614                         /* copy buffer if it needs escaping */
 615                         gfs2_log_unlock(sdp);
 616                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
 617                                 void *kaddr;
 618                                 struct page *page = bd2->bd_bh->b_page;
 619                                 bh = gfs2_log_get_buf(sdp);
 620                                 kaddr = kmap_atomic(page, KM_USER0);
 621                                 memcpy(bh->b_data,
 622                                        kaddr + bh_offset(bd2->bd_bh),
 623                                        sdp->sd_sb.sb_bsize);
 624                                 kunmap_atomic(page, KM_USER0);
 625                                 *(__be32 *)bh->b_data = 0;
 626                         } else {
 627                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 628                         }
 629                         set_buffer_dirty(bh);
 630                         ll_rw_block(WRITE, 1, &bh);
 631                         gfs2_log_lock(sdp);
 632                         if (++n >= num)
 633                                 break;
 634                 }
 635                 bh = NULL;
 636                 total_dbuf -= num;
 637                 total_jdata -= num;
 638         }
 639         gfs2_log_unlock(sdp);
 640
 641         /* Wait on all ordered buffers */
 642         while (!list_empty(&started)) {
 643                 gfs2_log_lock(sdp);
 644                 bd1 = list_entry(started.next, struct gfs2_bufdata,
 645                                  bd_le.le_list);
 646                 list_del(&bd1->bd_le.le_list);
 647                 sdp->sd_log_num_databuf--;
 648
 649                 bh = bd1->bd_bh;
 650                 if (bh) {
 651                         bh->b_private = NULL;
 652                         gfs2_log_unlock(sdp);
 653                         wait_on_buffer(bh);
 654                         brelse(bh);
 655                 } else
 656                         gfs2_log_unlock(sdp);
 657
 658                 kfree(bd1);
 659         }
 660
 661         /* We've removed all the ordered write bufs here, so only jdata left */
 662         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
 663 }
 664
 665 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 666                                     struct gfs2_log_descriptor *ld,
 667                                     __be64 *ptr, int pass)
 668 {
 669         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 670         struct gfs2_sbd *sdp = ip->i_sbd;
 671         struct gfs2_glock *gl = ip->i_gl;
 672         unsigned int blks = be32_to_cpu(ld->ld_data1);
 673         struct buffer_head *bh_log, *bh_ip;
 674         uint64_t blkno;
 675         uint64_t esc;
 676         int error = 0;
 677
 678         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 679                 return 0;
 680
 681         gfs2_replay_incr_blk(sdp, &start);
 682         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 683                 blkno = be64_to_cpu(*ptr++);
 684                 esc = be64_to_cpu(*ptr++);
 685
 686                 sdp->sd_found_blocks++;
 687
 688                 if (gfs2_revoke_check(sdp, blkno, start))
 689                         continue;
 690
 691                 error = gfs2_replay_read_block(jd, start, &bh_log);
 692                 if (error)
 693                         return error;
 694
 695                 bh_ip = gfs2_meta_new(gl, blkno);
 696                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 697
 698                 /* Unescape */
 699                 if (esc) {
 700                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 701                         *eptr = cpu_to_be32(GFS2_MAGIC);
 702                 }
 703                 mark_buffer_dirty(bh_ip);
 704
 705                 brelse(bh_log);
 706                 brelse(bh_ip);
 707                 if (error)
 708                         break;
 709
 710                 sdp->sd_replayed_blocks++;
 711         }
 712
 713         return error;
 714 }
 715
 716 /* FIXME: sort out accounting for log blocks etc. */
 717
 718 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 719 {
 720         struct gfs2_inode *ip = jd->jd_inode->u.generic_ip;
 721         struct gfs2_sbd *sdp = ip->i_sbd;
 722
 723         if (error) {
 724                 gfs2_meta_sync(ip->i_gl,
 725                                DIO_START | DIO_WAIT);
 726                 return;
 727         }
 728         if (pass != 1)
 729                 return;
 730
 731         /* data sync? */
 732         gfs2_meta_sync(ip->i_gl, DIO_START | DIO_WAIT);
 733
 734         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 735                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 736 }
 737
 738 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 739 {
 740         struct list_head *head = &sdp->sd_log_le_databuf;
 741         struct gfs2_bufdata *bd;
 742
 743         while (!list_empty(head)) {
 744                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 745                 list_del(&bd->bd_le.le_list);
 746                 sdp->sd_log_num_databuf--;
 747                 sdp->sd_log_num_jdata--;
 748                 gfs2_unpin(sdp, bd->bd_bh, ai);
 749         }
 750         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 751         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
 752 }
 753
 754
 755 const struct gfs2_log_operations gfs2_glock_lops = {
 756         .lo_add = glock_lo_add,
 757         .lo_after_commit = glock_lo_after_commit,
 758         .lo_name = "glock"
 759 };
 760
 761 const struct gfs2_log_operations gfs2_buf_lops = {
 762         .lo_add = buf_lo_add,
 763         .lo_incore_commit = buf_lo_incore_commit,
 764         .lo_before_commit = buf_lo_before_commit,
 765         .lo_after_commit = buf_lo_after_commit,
 766         .lo_before_scan = buf_lo_before_scan,
 767         .lo_scan_elements = buf_lo_scan_elements,
 768         .lo_after_scan = buf_lo_after_scan,
 769         .lo_name = "buf"
 770 };
 771
 772 const struct gfs2_log_operations gfs2_revoke_lops = {
 773         .lo_add = revoke_lo_add,
 774         .lo_before_commit = revoke_lo_before_commit,
 775         .lo_before_scan = revoke_lo_before_scan,
 776         .lo_scan_elements = revoke_lo_scan_elements,
 777         .lo_after_scan = revoke_lo_after_scan,
 778         .lo_name = "revoke"
 779 };
 780
 781 const struct gfs2_log_operations gfs2_rg_lops = {
 782         .lo_add = rg_lo_add,
 783         .lo_after_commit = rg_lo_after_commit,
 784         .lo_name = "rg"
 785 };
 786
 787 const struct gfs2_log_operations gfs2_databuf_lops = {
 788         .lo_add = databuf_lo_add,
 789         .lo_incore_commit = buf_lo_incore_commit,
 790         .lo_before_commit = databuf_lo_before_commit,
 791         .lo_after_commit = databuf_lo_after_commit,
 792         .lo_scan_elements = databuf_lo_scan_elements,
 793         .lo_after_scan = databuf_lo_after_scan,
 794         .lo_name = "databuf"
 795 };
 796
 797 const struct gfs2_log_operations *gfs2_log_ops[] = {
 798         &gfs2_glock_lops,
 799         &gfs2_buf_lops,
 800         &gfs2_revoke_lops,
 801         &gfs2_rg_lops,
 802         &gfs2_databuf_lops,
 803         NULL
 804 };
 805