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 <asm/semaphore.h>
  16
  17 #include "gfs2.h"
  18 #include "glock.h"
  19 #include "log.h"
  20 #include "lops.h"
  21 #include "meta_io.h"
  22 #include "recovery.h"
  23 #include "rgrp.h"
  24 #include "trans.h"
  25
  26 static void glock_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  27 {
  28         struct gfs2_glock *gl;
  29
  30         get_transaction->tr_touched = 1;
  31
  32         if (!list_empty(&le->le_list))
  33                 return;
  34
  35         gl = container_of(le, struct gfs2_glock, gl_le);
  36         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
  37                 return;
  38         gfs2_glock_hold(gl);
  39         set_bit(GLF_DIRTY, &gl->gl_flags);
  40
  41         gfs2_log_lock(sdp);
  42         sdp->sd_log_num_gl++;
  43         list_add(&le->le_list, &sdp->sd_log_le_gl);
  44         gfs2_log_unlock(sdp);
  45 }
  46
  47 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  48 {
  49         struct list_head *head = &sdp->sd_log_le_gl;
  50         struct gfs2_glock *gl;
  51
  52         while (!list_empty(head)) {
  53                 gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
  54                 list_del_init(&gl->gl_le.le_list);
  55                 sdp->sd_log_num_gl--;
  56
  57                 gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
  58                 gfs2_glock_put(gl);
  59         }
  60         gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
  61 }
  62
  63 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  64 {
  65         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
  66         struct gfs2_trans *tr;
  67
  68         if (!list_empty(&bd->bd_list_tr))
  69                 return;
  70
  71         tr = get_transaction;
  72         tr->tr_touched = 1;
  73         tr->tr_num_buf++;
  74         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
  75
  76         if (!list_empty(&le->le_list))
  77                 return;
  78
  79         gfs2_trans_add_gl(bd->bd_gl);
  80
  81         gfs2_meta_check(sdp, bd->bd_bh);
  82         gfs2_pin(sdp, bd->bd_bh);
  83
  84         gfs2_log_lock(sdp);
  85         sdp->sd_log_num_buf++;
  86         list_add(&le->le_list, &sdp->sd_log_le_buf);
  87         gfs2_log_unlock(sdp);
  88
  89         tr->tr_num_buf_new++;
  90 }
  91
  92 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
  93 {
  94         struct list_head *head = &tr->tr_list_buf;
  95         struct gfs2_bufdata *bd;
  96
  97         while (!list_empty(head)) {
  98                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
  99                 list_del_init(&bd->bd_list_tr);
 100                 tr->tr_num_buf--;
 101         }
 102         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 103 }
 104
 105 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 106 {
 107         struct buffer_head *bh;
 108         struct gfs2_log_descriptor *ld;
 109         struct gfs2_bufdata *bd1 = NULL, *bd2;
 110         unsigned int total = sdp->sd_log_num_buf;
 111         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 112         unsigned int limit;
 113         unsigned int num;
 114         unsigned n;
 115         __be64 *ptr;
 116
 117         offset += (sizeof(__be64) - 1);
 118         offset &= ~(sizeof(__be64) - 1);
 119         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 120         /* for 4k blocks, limit = 503 */
 121
 122         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 123         while(total) {
 124                 num = total;
 125                 if (total > limit)
 126                         num = limit;
 127                 bh = gfs2_log_get_buf(sdp);
 128                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 129                 ptr = (__be64 *)(bh->b_data + offset);
 130                 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 131                 ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
 132                 ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
 133                 ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
 134                 ld->ld_length = cpu_to_be32(num + 1);
 135                 ld->ld_data1 = cpu_to_be32(num);
 136                 ld->ld_data2 = cpu_to_be32(0);
 137                 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 138
 139                 n = 0;
 140                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 141                                              bd_le.le_list) {
 142                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 143                         if (++n >= num)
 144                                 break;
 145                 }
 146
 147                 set_buffer_dirty(bh);
 148                 ll_rw_block(WRITE, 1, &bh);
 149
 150                 n = 0;
 151                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 152                                              bd_le.le_list) {
 153                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 154                         set_buffer_dirty(bh);
 155                         ll_rw_block(WRITE, 1, &bh);
 156                         if (++n >= num)
 157                                 break;
 158                 }
 159
 160                 total -= num;
 161         }
 162 }
 163
 164 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 165 {
 166         struct list_head *head = &sdp->sd_log_le_buf;
 167         struct gfs2_bufdata *bd;
 168
 169         while (!list_empty(head)) {
 170                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 171                 list_del_init(&bd->bd_le.le_list);
 172                 sdp->sd_log_num_buf--;
 173
 174                 gfs2_unpin(sdp, bd->bd_bh, ai);
 175         }
 176         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 177 }
 178
 179 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 180                                struct gfs2_log_header *head, int pass)
 181 {
 182         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 183
 184         if (pass != 0)
 185                 return;
 186
 187         sdp->sd_found_blocks = 0;
 188         sdp->sd_replayed_blocks = 0;
 189 }
 190
 191 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 192                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 193                                 int pass)
 194 {
 195         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 196         struct gfs2_glock *gl = get_v2ip(jd->jd_inode)->i_gl;
 197         unsigned int blks = be32_to_cpu(ld->ld_data1);
 198         struct buffer_head *bh_log, *bh_ip;
 199         uint64_t blkno;
 200         int error = 0;
 201
 202         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 203                 return 0;
 204
 205         gfs2_replay_incr_blk(sdp, &start);
 206
 207         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 208                 blkno = be64_to_cpu(*ptr++);
 209
 210                 sdp->sd_found_blocks++;
 211
 212                 if (gfs2_revoke_check(sdp, blkno, start))
 213                         continue;
 214
 215                 error = gfs2_replay_read_block(jd, start, &bh_log);
 216                 if (error)
 217                         return error;
 218
 219                 bh_ip = gfs2_meta_new(gl, blkno);
 220                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 221
 222                 if (gfs2_meta_check(sdp, bh_ip))
 223                         error = -EIO;
 224                 else
 225                         mark_buffer_dirty(bh_ip);
 226
 227                 brelse(bh_log);
 228                 brelse(bh_ip);
 229
 230                 if (error)
 231                         break;
 232
 233                 sdp->sd_replayed_blocks++;
 234         }
 235
 236         return error;
 237 }
 238
 239 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 240 {
 241         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 242
 243         if (error) {
 244                 gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl,
 245                                DIO_START | DIO_WAIT);
 246                 return;
 247         }
 248         if (pass != 1)
 249                 return;
 250
 251         gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl, DIO_START | DIO_WAIT);
 252
 253         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 254                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 255 }
 256
 257 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 258 {
 259         struct gfs2_trans *tr;
 260
 261         tr = get_transaction;
 262         tr->tr_touched = 1;
 263         tr->tr_num_revoke++;
 264
 265         gfs2_log_lock(sdp);
 266         sdp->sd_log_num_revoke++;
 267         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 268         gfs2_log_unlock(sdp);
 269 }
 270
 271 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 272 {
 273         struct gfs2_log_descriptor *ld;
 274         struct gfs2_meta_header *mh;
 275         struct buffer_head *bh;
 276         unsigned int offset;
 277         struct list_head *head = &sdp->sd_log_le_revoke;
 278         struct gfs2_revoke *rv;
 279
 280         if (!sdp->sd_log_num_revoke)
 281                 return;
 282
 283         bh = gfs2_log_get_buf(sdp);
 284         ld = (struct gfs2_log_descriptor *)bh->b_data;
 285         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 286         ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
 287         ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
 288         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
 289         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 290                                                     sizeof(uint64_t)));
 291         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 292         ld->ld_data2 = cpu_to_be32(0);
 293         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 294         offset = sizeof(struct gfs2_log_descriptor);
 295
 296         while (!list_empty(head)) {
 297                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
 298                 list_del_init(&rv->rv_le.le_list);
 299                 sdp->sd_log_num_revoke--;
 300
 301                 if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) {
 302                         set_buffer_dirty(bh);
 303                         ll_rw_block(WRITE, 1, &bh);
 304
 305                         bh = gfs2_log_get_buf(sdp);
 306                         mh = (struct gfs2_meta_header *)bh->b_data;
 307                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 308                         mh->mh_type = cpu_to_be16(GFS2_METATYPE_LB);
 309                         mh->mh_format = cpu_to_be16(GFS2_FORMAT_LB);
 310                         offset = sizeof(struct gfs2_meta_header);
 311                 }
 312
 313                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
 314                 kfree(rv);
 315
 316                 offset += sizeof(uint64_t);
 317         }
 318         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 319
 320         set_buffer_dirty(bh);
 321         ll_rw_block(WRITE, 1, &bh);
 322 }
 323
 324 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 325                                   struct gfs2_log_header *head, int pass)
 326 {
 327         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 328
 329         if (pass != 0)
 330                 return;
 331
 332         sdp->sd_found_revokes = 0;
 333         sdp->sd_replay_tail = head->lh_tail;
 334 }
 335
 336 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 337                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 338                                    int pass)
 339 {
 340         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 341         unsigned int blks = be32_to_cpu(ld->ld_length);
 342         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 343         struct buffer_head *bh;
 344         unsigned int offset;
 345         uint64_t blkno;
 346         int first = 1;
 347         int error;
 348
 349         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 350                 return 0;
 351
 352         offset = sizeof(struct gfs2_log_descriptor);
 353
 354         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 355                 error = gfs2_replay_read_block(jd, start, &bh);
 356                 if (error)
 357                         return error;
 358
 359                 if (!first)
 360                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 361
 362                 while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) {
 363                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 364
 365                         error = gfs2_revoke_add(sdp, blkno, start);
 366                         if (error < 0)
 367                                 return error;
 368                         else if (error)
 369                                 sdp->sd_found_revokes++;
 370
 371                         if (!--revokes)
 372                                 break;
 373                         offset += sizeof(uint64_t);
 374                 }
 375
 376                 brelse(bh);
 377                 offset = sizeof(struct gfs2_meta_header);
 378                 first = 0;
 379         }
 380
 381         return 0;
 382 }
 383
 384 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 385 {
 386         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 387
 388         if (error) {
 389                 gfs2_revoke_clean(sdp);
 390                 return;
 391         }
 392         if (pass != 1)
 393                 return;
 394
 395         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 396                 jd->jd_jid, sdp->sd_found_revokes);
 397
 398         gfs2_revoke_clean(sdp);
 399 }
 400
 401 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 402 {
 403         struct gfs2_rgrpd *rgd;
 404
 405         get_transaction->tr_touched = 1;
 406
 407         if (!list_empty(&le->le_list))
 408                 return;
 409
 410         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 411         gfs2_rgrp_bh_hold(rgd);
 412
 413         gfs2_log_lock(sdp);
 414         sdp->sd_log_num_rg++;
 415         list_add(&le->le_list, &sdp->sd_log_le_rg);
 416         gfs2_log_unlock(sdp);
 417 }
 418
 419 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 420 {
 421         struct list_head *head = &sdp->sd_log_le_rg;
 422         struct gfs2_rgrpd *rgd;
 423
 424         while (!list_empty(head)) {
 425                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 426                 list_del_init(&rgd->rd_le.le_list);
 427                 sdp->sd_log_num_rg--;
 428
 429                 gfs2_rgrp_repolish_clones(rgd);
 430                 gfs2_rgrp_bh_put(rgd);
 431         }
 432         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 433 }
 434
 435 /**
 436  * databuf_lo_add - Add a databuf to the transaction.
 437  *
 438  * This is used in two distinct cases:
 439  * i) In ordered write mode
 440  *    We put the data buffer on a list so that we can ensure that its
 441  *    synced to disk at the right time
 442  * ii) In journaled data mode
 443  *    We need to journal the data block in the same way as metadata in
 444  *    the functions above. The difference is that here we have a tag
 445  *    which is two __be64's being the block number (as per meta data)
 446  *    and a flag which says whether the data block needs escaping or
 447  *    not. This means we need a new log entry for each 251 or so data
 448  *    blocks, which isn't an enormous overhead but twice as much as
 449  *    for normal metadata blocks.
 450  */
 451 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 452 {
 453         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 454         struct gfs2_trans *tr = get_transaction;
 455         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 456         struct gfs2_inode *ip = get_v2ip(mapping->host);
 457
 458         tr->tr_touched = 1;
 459         if (!list_empty(&bd->bd_list_tr) &&
 460             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
 461                 tr->tr_num_buf++;
 462                 gfs2_trans_add_gl(bd->bd_gl);
 463                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 464                 gfs2_pin(sdp, bd->bd_bh);
 465         }
 466         gfs2_log_lock(sdp);
 467         if (!list_empty(&le->le_list)) {
 468                 if (ip->i_di.di_flags & GFS2_DIF_JDATA)
 469                         sdp->sd_log_num_jdata++;
 470                 sdp->sd_log_num_databuf++;
 471                 list_add(&le->le_list, &sdp->sd_log_le_databuf);
 472         }
 473         gfs2_log_unlock(sdp);
 474 }
 475
 476 static int gfs2_check_magic(struct buffer_head *bh)
 477 {
 478         struct page *page = bh->b_page;
 479         void *kaddr;
 480         __be32 *ptr;
 481         int rv = 0;
 482
 483         kaddr = kmap_atomic(page, KM_USER0);
 484         ptr = kaddr + bh_offset(bh);
 485         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 486                 rv = 1;
 487         kunmap_atomic(page, KM_USER0);
 488
 489         return rv;
 490 }
 491
 492 /**
 493  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 494  *
 495  * Here we scan through the lists of buffers and make the assumption
 496  * that any buffer thats been pinned is being journaled, and that
 497  * any unpinned buffer is an ordered write data buffer and therefore
 498  * will be written back rather than journaled.
 499  */
 500 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 501 {
 502         LIST_HEAD(started);
 503         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
 504         struct buffer_head *bh = NULL;
 505         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 506         struct gfs2_log_descriptor *ld;
 507         unsigned int limit;
 508         unsigned int total_dbuf = sdp->sd_log_num_databuf;
 509         unsigned int total_jdata = sdp->sd_log_num_jdata;
 510         unsigned int num, n;
 511         __be64 *ptr = NULL;
 512
 513         offset += (2*sizeof(__be64) - 1);
 514         offset &= ~(2*sizeof(__be64) - 1);
 515         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 516
 517         /*
 518          * Start writing ordered buffers, write journaled buffers
 519          * into the log along with a header
 520          */
 521         gfs2_log_lock(sdp);
 522         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
 523                                        bd_le.le_list);
 524         while(total_dbuf) {
 525                 num = total_jdata;
 526                 if (num > limit)
 527                         num = limit;
 528                 n = 0;
 529                 list_for_each_entry_safe_continue(bd1, bdt,
 530                                                   &sdp->sd_log_le_databuf,
 531                                                   bd_le.le_list) {
 532                         /* An ordered write buffer */
 533                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
 534                                 list_move(&bd1->bd_le.le_list, &started);
 535                                 if (bd1 == bd2) {
 536                                         bd2 = NULL;
 537                                         bd2 = list_prepare_entry(bd2,
 538                                                         &sdp->sd_log_le_databuf,
 539                                                         bd_le.le_list);
 540                                 }
 541                                 total_dbuf--;
 542                                 if (bd1->bd_bh) {
 543                                         get_bh(bd1->bd_bh);
 544                                         if (buffer_dirty(bd1->bd_bh)) {
 545                                                 gfs2_log_unlock(sdp);
 546                                                 wait_on_buffer(bd1->bd_bh);
 547                                                 ll_rw_block(WRITE, 1,
 548                                                             &bd1->bd_bh);
 549                                                 gfs2_log_lock(sdp);
 550                                         }
 551                                         brelse(bd1->bd_bh);
 552                                         continue;
 553                                 }
 554                                 continue;
 555                         } else if (bd1->bd_bh) { /* A journaled buffer */
 556                                 int magic;
 557                                 gfs2_log_unlock(sdp);
 558                                 if (!bh) {
 559                                         bh = gfs2_log_get_buf(sdp);
 560                                         ld = (struct gfs2_log_descriptor *)
 561                                              bh->b_data;
 562                                         ptr = (__be64 *)(bh->b_data + offset);
 563                                         ld->ld_header.mh_magic =
 564                                                 cpu_to_be32(GFS2_MAGIC);
 565                                         ld->ld_header.mh_type =
 566                                                 cpu_to_be16(GFS2_METATYPE_LD);
 567                                         ld->ld_header.mh_format =
 568                                                 cpu_to_be16(GFS2_FORMAT_LD);
 569                                         ld->ld_type =
 570                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
 571                                         ld->ld_length = cpu_to_be32(num + 1);
 572                                         ld->ld_data1 = cpu_to_be32(num);
 573                                         ld->ld_data2 = cpu_to_be32(0);
 574                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 575                                 }
 576                                 magic = gfs2_check_magic(bd1->bd_bh);
 577                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 578                                 *ptr++ = cpu_to_be64((__u64)magic);
 579                                 clear_buffer_escaped(bd1->bd_bh);
 580                                 if (unlikely(magic != 0))
 581                                         set_buffer_escaped(bd1->bd_bh);
 582                                 gfs2_log_lock(sdp);
 583                                 if (n++ > num)
 584                                         break;
 585                         }
 586                 }
 587                 gfs2_log_unlock(sdp);
 588                 if (bh) {
 589                         set_buffer_dirty(bh);
 590                         ll_rw_block(WRITE, 1, &bh);
 591                         bh = NULL;
 592                 }
 593                 n = 0;
 594                 gfs2_log_lock(sdp);
 595                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
 596                                              bd_le.le_list) {
 597                         if (!bd2->bd_bh)
 598                                 continue;
 599                         /* copy buffer if it needs escaping */
 600                         gfs2_log_unlock(sdp);
 601                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
 602                                 void *kaddr;
 603                                 struct page *page = bd2->bd_bh->b_page;
 604                                 bh = gfs2_log_get_buf(sdp);
 605                                 kaddr = kmap_atomic(page, KM_USER0);
 606                                 memcpy(bh->b_data,
 607                                        kaddr + bh_offset(bd2->bd_bh),
 608                                        sdp->sd_sb.sb_bsize);
 609                                 kunmap_atomic(page, KM_USER0);
 610                                 *(__be32 *)bh->b_data = 0;
 611                         } else {
 612                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 613                         }
 614                         set_buffer_dirty(bh);
 615                         ll_rw_block(WRITE, 1, &bh);
 616                         gfs2_log_lock(sdp);
 617                         if (++n >= num)
 618                                 break;
 619                 }
 620                 bh = NULL;
 621                 total_dbuf -= num;
 622                 total_jdata -= num;
 623         }
 624         gfs2_log_unlock(sdp);
 625
 626         /* Wait on all ordered buffers */
 627         while (!list_empty(&started)) {
 628                 gfs2_log_lock(sdp);
 629                 bd1 = list_entry(started.next, struct gfs2_bufdata,
 630                                  bd_le.le_list);
 631                 list_del(&bd1->bd_le.le_list);
 632                 sdp->sd_log_num_databuf--;
 633
 634                 bh = bd1->bd_bh;
 635                 if (bh) {
 636                         set_v2bd(bh, NULL);
 637                         gfs2_log_unlock(sdp);
 638                         wait_on_buffer(bh);
 639                         brelse(bh);
 640                 } else
 641                         gfs2_log_unlock(sdp);
 642
 643                 kfree(bd1);
 644         }
 645
 646         /* We've removed all the ordered write bufs here, so only jdata left */
 647         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
 648 }
 649
 650 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 651                                     struct gfs2_log_descriptor *ld,
 652                                     __be64 *ptr, int pass)
 653 {
 654         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 655         struct gfs2_glock *gl = get_v2ip(jd->jd_inode)->i_gl;
 656         unsigned int blks = be32_to_cpu(ld->ld_data1);
 657         struct buffer_head *bh_log, *bh_ip;
 658         uint64_t blkno;
 659         uint64_t esc;
 660         int error = 0;
 661
 662         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 663                 return 0;
 664
 665         gfs2_replay_incr_blk(sdp, &start);
 666         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 667                 blkno = be64_to_cpu(*ptr++);
 668                 esc = be64_to_cpu(*ptr++);
 669
 670                 sdp->sd_found_blocks++;
 671
 672                 if (gfs2_revoke_check(sdp, blkno, start))
 673                         continue;
 674
 675                 error = gfs2_replay_read_block(jd, start, &bh_log);
 676                 if (error)
 677                         return error;
 678
 679                 bh_ip = gfs2_meta_new(gl, blkno);
 680                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 681
 682                 /* Unescape */
 683                 if (esc) {
 684                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 685                         *eptr = cpu_to_be32(GFS2_MAGIC);
 686                 }
 687                 mark_buffer_dirty(bh_ip);
 688
 689                 brelse(bh_log);
 690                 brelse(bh_ip);
 691                 if (error)
 692                         break;
 693
 694                 sdp->sd_replayed_blocks++;
 695         }
 696
 697         return error;
 698 }
 699
 700 /* FIXME: sort out accounting for log blocks etc. */
 701
 702 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 703 {
 704         struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd;
 705
 706         if (error) {
 707                 gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl,
 708                                DIO_START | DIO_WAIT);
 709                 return;
 710         }
 711         if (pass != 1)
 712                 return;
 713
 714         /* data sync? */
 715         gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl, DIO_START | DIO_WAIT);
 716
 717         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 718                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 719 }
 720
 721 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 722 {
 723         struct list_head *head = &sdp->sd_log_le_databuf;
 724         struct gfs2_bufdata *bd;
 725
 726         while (!list_empty(head)) {
 727                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 728                 list_del(&bd->bd_le.le_list);
 729                 sdp->sd_log_num_databuf--;
 730                 sdp->sd_log_num_jdata--;
 731                 gfs2_unpin(sdp, bd->bd_bh, ai);
 732         }
 733         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 734         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
 735 }
 736
 737
 738 struct gfs2_log_operations gfs2_glock_lops = {
 739         .lo_add = glock_lo_add,
 740         .lo_after_commit = glock_lo_after_commit,
 741         .lo_name = "glock"
 742 };
 743
 744 struct gfs2_log_operations gfs2_buf_lops = {
 745         .lo_add = buf_lo_add,
 746         .lo_incore_commit = buf_lo_incore_commit,
 747         .lo_before_commit = buf_lo_before_commit,
 748         .lo_after_commit = buf_lo_after_commit,
 749         .lo_before_scan = buf_lo_before_scan,
 750         .lo_scan_elements = buf_lo_scan_elements,
 751         .lo_after_scan = buf_lo_after_scan,
 752         .lo_name = "buf"
 753 };
 754
 755 struct gfs2_log_operations gfs2_revoke_lops = {
 756         .lo_add = revoke_lo_add,
 757         .lo_before_commit = revoke_lo_before_commit,
 758         .lo_before_scan = revoke_lo_before_scan,
 759         .lo_scan_elements = revoke_lo_scan_elements,
 760         .lo_after_scan = revoke_lo_after_scan,
 761         .lo_name = "revoke"
 762 };
 763
 764 struct gfs2_log_operations gfs2_rg_lops = {
 765         .lo_add = rg_lo_add,
 766         .lo_after_commit = rg_lo_after_commit,
 767         .lo_name = "rg"
 768 };
 769
 770 struct gfs2_log_operations gfs2_databuf_lops = {
 771         .lo_add = databuf_lo_add,
 772         .lo_incore_commit = buf_lo_incore_commit,
 773         .lo_before_commit = databuf_lo_before_commit,
 774         .lo_after_commit = databuf_lo_after_commit,
 775         .lo_scan_elements = databuf_lo_scan_elements,
 776         .lo_after_scan = databuf_lo_after_scan,
 777         .lo_name = "databuf"
 778 };
 779
 780 struct gfs2_log_operations *gfs2_log_ops[] = {
 781         &gfs2_glock_lops,
 782         &gfs2_buf_lops,
 783         &gfs2_revoke_lops,
 784         &gfs2_rg_lops,
 785         &gfs2_databuf_lops,
 786         NULL
 787 };
 788