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, bd_le.le_list) {
 141                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 142                         if (++n >= num)
 143                                 break;
 144                 }
 145
 146                 set_buffer_dirty(bh);
 147                 ll_rw_block(WRITE, 1, &bh);
 148
 149                 n = 0;
 150                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf, bd_le.le_list) {
 151                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 152                         set_buffer_dirty(bh);
 153                         ll_rw_block(WRITE, 1, &bh);
 154                         if (++n >= num)
 155                                 break;
 156                 }
 157
 158                 total -= num;
 159         }
 160 }
 161
 162 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 163 {
 164         struct list_head *head = &sdp->sd_log_le_buf;
 165         struct gfs2_bufdata *bd;
 166
 167         while (!list_empty(head)) {
 168                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 169                 list_del_init(&bd->bd_le.le_list);
 170                 sdp->sd_log_num_buf--;
 171
 172                 gfs2_unpin(sdp, bd->bd_bh, ai);
 173         }
 174         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 175 }
 176
 177 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 178                                struct gfs2_log_header *head, int pass)
 179 {
 180         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 181
 182         if (pass != 0)
 183                 return;
 184
 185         sdp->sd_found_blocks = 0;
 186         sdp->sd_replayed_blocks = 0;
 187 }
 188
 189 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 190                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 191                                 int pass)
 192 {
 193         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 194         struct gfs2_glock *gl = jd->jd_inode->i_gl;
 195         unsigned int blks = be32_to_cpu(ld->ld_data1);
 196         struct buffer_head *bh_log, *bh_ip;
 197         uint64_t blkno;
 198         int error = 0;
 199
 200         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 201                 return 0;
 202
 203         gfs2_replay_incr_blk(sdp, &start);
 204
 205         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 206                 blkno = be64_to_cpu(*ptr++);
 207
 208                 sdp->sd_found_blocks++;
 209
 210                 if (gfs2_revoke_check(sdp, blkno, start))
 211                         continue;
 212
 213                 error = gfs2_replay_read_block(jd, start, &bh_log);
 214                 if (error)
 215                         return error;
 216
 217                 bh_ip = gfs2_meta_new(gl, blkno);
 218                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 219
 220                 if (gfs2_meta_check(sdp, bh_ip))
 221                         error = -EIO;
 222                 else
 223                         mark_buffer_dirty(bh_ip);
 224
 225                 brelse(bh_log);
 226                 brelse(bh_ip);
 227
 228                 if (error)
 229                         break;
 230
 231                 sdp->sd_replayed_blocks++;
 232         }
 233
 234         return error;
 235 }
 236
 237 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 238 {
 239         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 240
 241         if (error) {
 242                 gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
 243                 return;
 244         }
 245         if (pass != 1)
 246                 return;
 247
 248         gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
 249
 250         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 251                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 252 }
 253
 254 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 255 {
 256         struct gfs2_trans *tr;
 257
 258         tr = get_transaction;
 259         tr->tr_touched = 1;
 260         tr->tr_num_revoke++;
 261
 262         gfs2_log_lock(sdp);
 263         sdp->sd_log_num_revoke++;
 264         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 265         gfs2_log_unlock(sdp);
 266 }
 267
 268 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 269 {
 270         struct gfs2_log_descriptor *ld;
 271         struct gfs2_meta_header *mh;
 272         struct buffer_head *bh;
 273         unsigned int offset;
 274         struct list_head *head = &sdp->sd_log_le_revoke;
 275         struct gfs2_revoke *rv;
 276
 277         if (!sdp->sd_log_num_revoke)
 278                 return;
 279
 280         bh = gfs2_log_get_buf(sdp);
 281         ld = (struct gfs2_log_descriptor *)bh->b_data;
 282         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 283         ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
 284         ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
 285         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
 286         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(uint64_t)));
 287         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 288         ld->ld_data2 = cpu_to_be32(0);
 289         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 290         offset = sizeof(struct gfs2_log_descriptor);
 291
 292         while (!list_empty(head)) {
 293                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
 294                 list_del(&rv->rv_le.le_list);
 295                 sdp->sd_log_num_revoke--;
 296
 297                 if (offset + sizeof(uint64_t) > sdp->sd_sb.sb_bsize) {
 298                         set_buffer_dirty(bh);
 299                         ll_rw_block(WRITE, 1, &bh);
 300
 301                         bh = gfs2_log_get_buf(sdp);
 302                         mh = (struct gfs2_meta_header *)bh->b_data;
 303                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 304                         mh->mh_type = cpu_to_be16(GFS2_METATYPE_LB);
 305                         mh->mh_format = cpu_to_be16(GFS2_FORMAT_LB);
 306                         offset = sizeof(struct gfs2_meta_header);
 307                 }
 308
 309                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
 310                 kfree(rv);
 311
 312                 offset += sizeof(uint64_t);
 313         }
 314         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 315
 316         set_buffer_dirty(bh);
 317         ll_rw_block(WRITE, 1, &bh);
 318 }
 319
 320 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 321                                   struct gfs2_log_header *head, int pass)
 322 {
 323         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 324
 325         if (pass != 0)
 326                 return;
 327
 328         sdp->sd_found_revokes = 0;
 329         sdp->sd_replay_tail = head->lh_tail;
 330 }
 331
 332 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 333                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 334                                    int pass)
 335 {
 336         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 337         unsigned int blks = be32_to_cpu(ld->ld_length);
 338         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 339         struct buffer_head *bh;
 340         unsigned int offset;
 341         uint64_t blkno;
 342         int first = 1;
 343         int error;
 344
 345         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 346                 return 0;
 347
 348         offset = sizeof(struct gfs2_log_descriptor);
 349
 350         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 351                 error = gfs2_replay_read_block(jd, start, &bh);
 352                 if (error)
 353                         return error;
 354
 355                 if (!first)
 356                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 357
 358                 while (offset + sizeof(uint64_t) <= sdp->sd_sb.sb_bsize) {
 359                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 360
 361                         error = gfs2_revoke_add(sdp, blkno, start);
 362                         if (error < 0)
 363                                 return error;
 364                         else if (error)
 365                                 sdp->sd_found_revokes++;
 366
 367                         if (!--revokes)
 368                                 break;
 369                         offset += sizeof(uint64_t);
 370                 }
 371
 372                 brelse(bh);
 373                 offset = sizeof(struct gfs2_meta_header);
 374                 first = 0;
 375         }
 376
 377         return 0;
 378 }
 379
 380 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 381 {
 382         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 383
 384         if (error) {
 385                 gfs2_revoke_clean(sdp);
 386                 return;
 387         }
 388         if (pass != 1)
 389                 return;
 390
 391         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 392                 jd->jd_jid, sdp->sd_found_revokes);
 393
 394         gfs2_revoke_clean(sdp);
 395 }
 396
 397 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 398 {
 399         struct gfs2_rgrpd *rgd;
 400
 401         get_transaction->tr_touched = 1;
 402
 403         if (!list_empty(&le->le_list))
 404                 return;
 405
 406         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 407         gfs2_rgrp_bh_hold(rgd);
 408
 409         gfs2_log_lock(sdp);
 410         sdp->sd_log_num_rg++;
 411         list_add(&le->le_list, &sdp->sd_log_le_rg);
 412         gfs2_log_unlock(sdp);
 413 }
 414
 415 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 416 {
 417         struct list_head *head = &sdp->sd_log_le_rg;
 418         struct gfs2_rgrpd *rgd;
 419
 420         while (!list_empty(head)) {
 421                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 422                 list_del_init(&rgd->rd_le.le_list);
 423                 sdp->sd_log_num_rg--;
 424
 425                 gfs2_rgrp_repolish_clones(rgd);
 426                 gfs2_rgrp_bh_put(rgd);
 427         }
 428         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 429 }
 430
 431 /**
 432  * databuf_lo_add - Add a databuf to the transaction.
 433  *
 434  * This is used in two distinct cases:
 435  * i) In ordered write mode
 436  *    We put the data buffer on a list so that we can ensure that its
 437  *    synced to disk at the right time
 438  * ii) In journaled data mode
 439  *    We need to journal the data block in the same way as metadata in
 440  *    the functions above. The difference is that here we have a tag
 441  *    which is two __be64's being the block number (as per meta data)
 442  *    and a flag which says whether the data block needs escaping or
 443  *    not. This means we need a new log entry for each 251 or so data
 444  *    blocks, which isn't an enormous overhead but twice as much as
 445  *    for normal metadata blocks.
 446  */
 447 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 448 {
 449         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 450         struct gfs2_trans *tr = get_transaction;
 451         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 452         struct gfs2_inode *ip = get_v2ip(mapping->host);
 453
 454         tr->tr_touched = 1;
 455         if (!list_empty(&bd->bd_list_tr) &&
 456             (ip->i_di.di_flags & GFS2_DIF_JDATA)) {
 457                 tr->tr_num_buf++;
 458                 gfs2_trans_add_gl(bd->bd_gl);
 459                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 460                 gfs2_pin(sdp, bd->bd_bh);
 461         } else {
 462                 clear_buffer_pinned(bd->bd_bh);
 463         }
 464         gfs2_log_lock(sdp);
 465         if (ip->i_di.di_flags & GFS2_DIF_JDATA)
 466                 sdp->sd_log_num_jdata++;
 467         sdp->sd_log_num_databuf++;
 468         list_add(&le->le_list, &sdp->sd_log_le_databuf);
 469         gfs2_log_unlock(sdp);
 470 }
 471
 472 static int gfs2_check_magic(struct buffer_head *bh)
 473 {
 474         struct page *page = bh->b_page;
 475         void *kaddr;
 476         __be32 *ptr;
 477         int rv = 0;
 478
 479         kaddr = kmap_atomic(page, KM_USER0);
 480         ptr = kaddr + bh_offset(bh);
 481         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 482                 rv = 1;
 483         kunmap_atomic(page, KM_USER0);
 484
 485         return rv;
 486 }
 487
 488 /**
 489  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 490  *
 491  * Here we scan through the lists of buffers and make the assumption
 492  * that any buffer thats been pinned is being journaled, and that
 493  * any unpinned buffer is an ordered write data buffer and therefore
 494  * will be written back rather than journaled.
 495  */
 496 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 497 {
 498         LIST_HEAD(started);
 499         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
 500         struct buffer_head *bh = NULL;
 501         unsigned int offset = sizeof(struct gfs2_log_descriptor);
 502         struct gfs2_log_descriptor *ld;
 503         unsigned int limit;
 504         unsigned int total_dbuf = sdp->sd_log_num_databuf;
 505         unsigned int total_jdata = sdp->sd_log_num_jdata;
 506         unsigned int num, n;
 507         __be64 *ptr;
 508
 509         offset += (2*sizeof(__be64) - 1);
 510         offset &= ~(2*sizeof(__be64) - 1);
 511         limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64);
 512
 513         /* printk(KERN_INFO "totals: jdata=%u dbuf=%u\n", total_jdata, total_dbuf); */
 514         /*
 515          * Start writing ordered buffers, write journaled buffers
 516          * into the log along with a header
 517          */
 518         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf, bd_le.le_list);
 519         while(total_dbuf) {
 520                 num = total_jdata;
 521                 if (num > limit)
 522                         num = limit;
 523                 n = 0;
 524                 list_for_each_entry_safe_continue(bd1, bdt, &sdp->sd_log_le_databuf, bd_le.le_list) {
 525                         gfs2_log_lock(sdp);
 526                         /* An ordered write buffer */
 527                         if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) {
 528                                 list_move(&bd1->bd_le.le_list, &started);
 529                                 if (bd1 == bd2) {
 530                                         bd2 = NULL;
 531                                         bd2 = list_prepare_entry(bd2, &sdp->sd_log_le_databuf, bd_le.le_list);
 532                                 }
 533                                 total_dbuf--;
 534                                 if (bd1->bd_bh) {
 535                                         get_bh(bd1->bd_bh);
 536                                         gfs2_log_unlock(sdp);
 537                                         if (buffer_dirty(bd1->bd_bh)) {
 538                                                 wait_on_buffer(bd1->bd_bh);
 539                                                 ll_rw_block(WRITE, 1, &bd1->bd_bh);
 540                                         }
 541                                         brelse(bd1->bd_bh);
 542                                         continue;
 543                                 }
 544                                 gfs2_log_unlock(sdp);
 545                                 continue;
 546                         } else if (bd1->bd_bh) { /* A journaled buffer */
 547                                 int magic;
 548                                 gfs2_log_unlock(sdp);
 549                                 /* printk(KERN_INFO "journaled buffer\n"); */
 550                                 if (!bh) {
 551                                         bh = gfs2_log_get_buf(sdp);
 552                                         ld = (struct gfs2_log_descriptor *)bh->b_data;
 553                                         ptr = (__be64 *)(bh->b_data + offset);
 554                                         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 555                                         ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD);
 556                                         ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD);
 557                                         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_JDATA);
 558                                         ld->ld_length = cpu_to_be32(num + 1);
 559                                         ld->ld_data1 = cpu_to_be32(num);
 560                                         ld->ld_data2 = cpu_to_be32(0);
 561                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 562                                 }
 563                                 magic = gfs2_check_magic(bd1->bd_bh);
 564                                 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 565                                 *ptr++ = cpu_to_be64((__u64)magic);
 566                                 clear_buffer_escaped(bd1->bd_bh);
 567                                 if (unlikely(magic != 0))
 568                                         set_buffer_escaped(bd1->bd_bh);
 569                                 if (n++ > num)
 570                                         break;
 571                         }
 572                 }
 573                 if (bh) {
 574                         set_buffer_dirty(bh);
 575                         ll_rw_block(WRITE, 1, &bh);
 576                         bh = NULL;
 577                 }
 578                 n = 0;
 579                 /* printk(KERN_INFO "totals2: jdata=%u dbuf=%u\n", total_jdata, total_dbuf); */
 580                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf, bd_le.le_list) {
 581                         if (!bd2->bd_bh)
 582                                 continue;
 583                         /* copy buffer if it needs escaping */
 584                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
 585                                 void *kaddr;
 586                                 struct page *page = bd2->bd_bh->b_page;
 587                                 bh = gfs2_log_get_buf(sdp);
 588                                 kaddr = kmap_atomic(page, KM_USER0);
 589                                 memcpy(bh->b_data, kaddr + bh_offset(bd2->bd_bh), sdp->sd_sb.sb_bsize);
 590                                 kunmap_atomic(page, KM_USER0);
 591                                 *(__be32 *)bh->b_data = 0;
 592                         } else {
 593                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 594                         }
 595                         set_buffer_dirty(bh);
 596                         ll_rw_block(WRITE, 1, &bh);
 597                         if (++n >= num)
 598                                 break;
 599                 }
 600                 bh = NULL;
 601                 total_dbuf -= num;
 602                 total_jdata -= num;
 603         }
 604         /* printk(KERN_INFO "wait on ordered data buffers\n"); */
 605         /* Wait on all ordered buffers */
 606         while (!list_empty(&started)) {
 607                 bd1 = list_entry(started.next, struct gfs2_bufdata, bd_le.le_list);
 608                 list_del(&bd1->bd_le.le_list);
 609                 sdp->sd_log_num_databuf--;
 610
 611                 gfs2_log_lock(sdp);
 612                 bh = bd1->bd_bh;
 613                 if (bh) {
 614                         set_v2bd(bh, NULL);
 615                         gfs2_log_unlock(sdp);
 616                         wait_on_buffer(bh);
 617                         brelse(bh);
 618                 } else
 619                         gfs2_log_unlock(sdp);
 620
 621                 kfree(bd1);
 622         }
 623
 624         /* printk(KERN_INFO "sd_log_num_databuf %u sd_log_num_jdata %u\n", sdp->sd_log_num_databuf, sdp->sd_log_num_jdata); */
 625         /* We've removed all the ordered write bufs here, so only jdata left */
 626         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
 627 }
 628
 629 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 630                                     struct gfs2_log_descriptor *ld,
 631                                     __be64 *ptr, int pass)
 632 {
 633         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 634         struct gfs2_glock *gl = jd->jd_inode->i_gl;
 635         unsigned int blks = be32_to_cpu(ld->ld_data1);
 636         struct buffer_head *bh_log, *bh_ip;
 637         uint64_t blkno;
 638         uint64_t esc;
 639         int error = 0;
 640
 641         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 642                 return 0;
 643
 644         gfs2_replay_incr_blk(sdp, &start);
 645         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 646                 blkno = be64_to_cpu(*ptr++);
 647                 esc = be64_to_cpu(*ptr++);
 648
 649                 sdp->sd_found_blocks++;
 650
 651                 if (gfs2_revoke_check(sdp, blkno, start))
 652                         continue;
 653
 654                 error = gfs2_replay_read_block(jd, start, &bh_log);
 655                 if (error)
 656                         return error;
 657
 658                 bh_ip = gfs2_meta_new(gl, blkno);
 659                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 660
 661                 /* Unescape */
 662                 if (esc) {
 663                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 664                         *eptr = cpu_to_be32(GFS2_MAGIC);
 665                 }
 666                 mark_buffer_dirty(bh_ip);
 667
 668                 brelse(bh_log);
 669                 brelse(bh_ip);
 670                 if (error)
 671                         break;
 672
 673                 sdp->sd_replayed_blocks++;
 674         }
 675
 676         return error;
 677 }
 678
 679 /* FIXME: sort out accounting for log blocks etc. */
 680
 681 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 682 {
 683         struct gfs2_sbd *sdp = jd->jd_inode->i_sbd;
 684
 685         if (error) {
 686                 gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
 687                 return;
 688         }
 689         if (pass != 1)
 690                 return;
 691
 692         /* data sync? */
 693         gfs2_meta_sync(jd->jd_inode->i_gl, DIO_START | DIO_WAIT);
 694
 695         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 696                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 697 }
 698
 699 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 700 {
 701         struct list_head *head = &sdp->sd_log_le_databuf;
 702         struct gfs2_bufdata *bd;
 703
 704         while (!list_empty(head)) {
 705                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 706                 list_del_init(&bd->bd_le.le_list);
 707                 sdp->sd_log_num_databuf--;
 708                 sdp->sd_log_num_jdata--;
 709                 gfs2_unpin(sdp, bd->bd_bh, ai);
 710                 brelse(bd->bd_bh);
 711                 kfree(bd);
 712         }
 713         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 714         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
 715 }
 716
 717
 718 struct gfs2_log_operations gfs2_glock_lops = {
 719         .lo_add = glock_lo_add,
 720         .lo_after_commit = glock_lo_after_commit,
 721         .lo_name = "glock"
 722 };
 723
 724 struct gfs2_log_operations gfs2_buf_lops = {
 725         .lo_add = buf_lo_add,
 726         .lo_incore_commit = buf_lo_incore_commit,
 727         .lo_before_commit = buf_lo_before_commit,
 728         .lo_after_commit = buf_lo_after_commit,
 729         .lo_before_scan = buf_lo_before_scan,
 730         .lo_scan_elements = buf_lo_scan_elements,
 731         .lo_after_scan = buf_lo_after_scan,
 732         .lo_name = "buf"
 733 };
 734
 735 struct gfs2_log_operations gfs2_revoke_lops = {
 736         .lo_add = revoke_lo_add,
 737         .lo_before_commit = revoke_lo_before_commit,
 738         .lo_before_scan = revoke_lo_before_scan,
 739         .lo_scan_elements = revoke_lo_scan_elements,
 740         .lo_after_scan = revoke_lo_after_scan,
 741         .lo_name = "revoke"
 742 };
 743
 744 struct gfs2_log_operations gfs2_rg_lops = {
 745         .lo_add = rg_lo_add,
 746         .lo_after_commit = rg_lo_after_commit,
 747         .lo_name = "rg"
 748 };
 749
 750 struct gfs2_log_operations gfs2_databuf_lops = {
 751         .lo_add = databuf_lo_add,
 752         .lo_incore_commit = buf_lo_incore_commit,
 753         .lo_before_commit = databuf_lo_before_commit,
 754         .lo_after_commit = databuf_lo_after_commit,
 755         .lo_scan_elements = databuf_lo_scan_elements,
 756         .lo_after_scan = databuf_lo_after_scan,
 757         .lo_name = "databuf"
 758 };
 759
 760 struct gfs2_log_operations *gfs2_log_ops[] = {
 761         &gfs2_glock_lops,
 762         &gfs2_buf_lops,
 763         &gfs2_revoke_lops,
 764         &gfs2_rg_lops,
 765         &gfs2_databuf_lops,
 766         NULL
 767 };
 768