fs/gfs2/lops.c

   1 /*
   2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
   3  * Copyright (C) 2004-2006 Red Hat, Inc.  All rights reserved.
   4  *
   5  * This copyrighted material is made available to anyone wishing to use,
   6  * modify, copy, or redistribute it subject to the terms and conditions
   7  * of the GNU General Public License version 2.
   8  */
   9
  10 #include <linux/sched.h>
  11 #include <linux/slab.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/completion.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/gfs2_ondisk.h>
  16 #include <linux/lm_interface.h>
  17
  18 #include "gfs2.h"
  19 #include "incore.h"
  20 #include "inode.h"
  21 #include "glock.h"
  22 #include "log.h"
  23 #include "lops.h"
  24 #include "meta_io.h"
  25 #include "recovery.h"
  26 #include "rgrp.h"
  27 #include "trans.h"
  28 #include "util.h"
  29
  30 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         gl = container_of(le, struct gfs2_glock, gl_le);
  38         if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl)))
  39                 return;
  40
  41         gfs2_log_lock(sdp);
  42         if (!list_empty(&le->le_list)){
  43                 gfs2_log_unlock(sdp);
  44                 return;
  45         }
  46         gfs2_glock_hold(gl);
  47         set_bit(GLF_DIRTY, &gl->gl_flags);
  48         sdp->sd_log_num_gl++;
  49         list_add(&le->le_list, &sdp->sd_log_le_gl);
  50         gfs2_log_unlock(sdp);
  51 }
  52
  53 static void glock_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
  54 {
  55         struct list_head *head = &sdp->sd_log_le_gl;
  56         struct gfs2_glock *gl;
  57
  58         while (!list_empty(head)) {
  59                 gl = list_entry(head->next, struct gfs2_glock, gl_le.le_list);
  60                 list_del_init(&gl->gl_le.le_list);
  61                 sdp->sd_log_num_gl--;
  62
  63                 gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(gl));
  64                 gfs2_glock_put(gl);
  65         }
  66         gfs2_assert_warn(sdp, !sdp->sd_log_num_gl);
  67 }
  68
  69 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
  70 {
  71         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
  72         struct gfs2_trans *tr;
  73
  74         gfs2_log_lock(sdp);
  75         if (!list_empty(&bd->bd_list_tr)) {
  76                 gfs2_log_unlock(sdp);
  77                 return;
  78         }
  79         tr = current->journal_info;
  80         tr->tr_touched = 1;
  81         tr->tr_num_buf++;
  82         list_add(&bd->bd_list_tr, &tr->tr_list_buf);
  83         gfs2_log_unlock(sdp);
  84
  85         if (!list_empty(&le->le_list))
  86                 return;
  87
  88         gfs2_trans_add_gl(bd->bd_gl);
  89
  90         gfs2_meta_check(sdp, bd->bd_bh);
  91         gfs2_pin(sdp, bd->bd_bh);
  92         gfs2_log_lock(sdp);
  93         sdp->sd_log_num_buf++;
  94         list_add(&le->le_list, &sdp->sd_log_le_buf);
  95         gfs2_log_unlock(sdp);
  96
  97         tr->tr_num_buf_new++;
  98 }
  99
 100 static void buf_lo_incore_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
 101 {
 102         struct list_head *head = &tr->tr_list_buf;
 103         struct gfs2_bufdata *bd;
 104
 105         gfs2_log_lock(sdp);
 106         while (!list_empty(head)) {
 107                 bd = list_entry(head->next, struct gfs2_bufdata, bd_list_tr);
 108                 list_del_init(&bd->bd_list_tr);
 109                 tr->tr_num_buf--;
 110         }
 111         gfs2_log_unlock(sdp);
 112         gfs2_assert_warn(sdp, !tr->tr_num_buf);
 113 }
 114
 115 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
 116 {
 117         struct buffer_head *bh;
 118         struct gfs2_log_descriptor *ld;
 119         struct gfs2_bufdata *bd1 = NULL, *bd2;
 120         unsigned int total;
 121         unsigned int offset = BUF_OFFSET;
 122         unsigned int limit;
 123         unsigned int num;
 124         unsigned n;
 125         __be64 *ptr;
 126
 127         limit = buf_limit(sdp);
 128         /* for 4k blocks, limit = 503 */
 129
 130         gfs2_log_lock(sdp);
 131         total = sdp->sd_log_num_buf;
 132         bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
 133         while(total) {
 134                 num = total;
 135                 if (total > limit)
 136                         num = limit;
 137                 gfs2_log_unlock(sdp);
 138                 bh = gfs2_log_get_buf(sdp);
 139                 gfs2_log_lock(sdp);
 140                 ld = (struct gfs2_log_descriptor *)bh->b_data;
 141                 ptr = (__be64 *)(bh->b_data + offset);
 142                 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 143                 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 144                 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 145                 ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_METADATA);
 146                 ld->ld_length = cpu_to_be32(num + 1);
 147                 ld->ld_data1 = cpu_to_be32(num);
 148                 ld->ld_data2 = cpu_to_be32(0);
 149                 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 150
 151                 n = 0;
 152                 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
 153                                              bd_le.le_list) {
 154                         *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
 155                         if (++n >= num)
 156                                 break;
 157                 }
 158
 159                 gfs2_log_unlock(sdp);
 160                 set_buffer_dirty(bh);
 161                 ll_rw_block(WRITE, 1, &bh);
 162                 gfs2_log_lock(sdp);
 163
 164                 n = 0;
 165                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
 166                                              bd_le.le_list) {
 167                         gfs2_log_unlock(sdp);
 168                         bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 169                         set_buffer_dirty(bh);
 170                         ll_rw_block(WRITE, 1, &bh);
 171                         gfs2_log_lock(sdp);
 172                         if (++n >= num)
 173                                 break;
 174                 }
 175
 176                 BUG_ON(total < num);
 177                 total -= num;
 178         }
 179         gfs2_log_unlock(sdp);
 180 }
 181
 182 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 183 {
 184         struct list_head *head = &sdp->sd_log_le_buf;
 185         struct gfs2_bufdata *bd;
 186
 187         while (!list_empty(head)) {
 188                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 189                 list_del_init(&bd->bd_le.le_list);
 190                 sdp->sd_log_num_buf--;
 191
 192                 gfs2_unpin(sdp, bd->bd_bh, ai);
 193         }
 194         gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
 195 }
 196
 197 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
 198                                struct gfs2_log_header_host *head, int pass)
 199 {
 200         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 201
 202         if (pass != 0)
 203                 return;
 204
 205         sdp->sd_found_blocks = 0;
 206         sdp->sd_replayed_blocks = 0;
 207 }
 208
 209 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 210                                 struct gfs2_log_descriptor *ld, __be64 *ptr,
 211                                 int pass)
 212 {
 213         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 214         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 215         struct gfs2_glock *gl = ip->i_gl;
 216         unsigned int blks = be32_to_cpu(ld->ld_data1);
 217         struct buffer_head *bh_log, *bh_ip;
 218         u64 blkno;
 219         int error = 0;
 220
 221         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
 222                 return 0;
 223
 224         gfs2_replay_incr_blk(sdp, &start);
 225
 226         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 227                 blkno = be64_to_cpu(*ptr++);
 228
 229                 sdp->sd_found_blocks++;
 230
 231                 if (gfs2_revoke_check(sdp, blkno, start))
 232                         continue;
 233
 234                 error = gfs2_replay_read_block(jd, start, &bh_log);
 235                 if (error)
 236                         return error;
 237
 238                 bh_ip = gfs2_meta_new(gl, blkno);
 239                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 240
 241                 if (gfs2_meta_check(sdp, bh_ip))
 242                         error = -EIO;
 243                 else
 244                         mark_buffer_dirty(bh_ip);
 245
 246                 brelse(bh_log);
 247                 brelse(bh_ip);
 248
 249                 if (error)
 250                         break;
 251
 252                 sdp->sd_replayed_blocks++;
 253         }
 254
 255         return error;
 256 }
 257
 258 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 259 {
 260         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 261         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 262
 263         if (error) {
 264                 gfs2_meta_sync(ip->i_gl);
 265                 return;
 266         }
 267         if (pass != 1)
 268                 return;
 269
 270         gfs2_meta_sync(ip->i_gl);
 271
 272         fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
 273                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 274 }
 275
 276 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 277 {
 278         struct gfs2_trans *tr;
 279
 280         tr = current->journal_info;
 281         tr->tr_touched = 1;
 282         tr->tr_num_revoke++;
 283
 284         gfs2_log_lock(sdp);
 285         sdp->sd_log_num_revoke++;
 286         list_add(&le->le_list, &sdp->sd_log_le_revoke);
 287         gfs2_log_unlock(sdp);
 288 }
 289
 290 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
 291 {
 292         struct gfs2_log_descriptor *ld;
 293         struct gfs2_meta_header *mh;
 294         struct buffer_head *bh;
 295         unsigned int offset;
 296         struct list_head *head = &sdp->sd_log_le_revoke;
 297         struct gfs2_revoke *rv;
 298
 299         if (!sdp->sd_log_num_revoke)
 300                 return;
 301
 302         bh = gfs2_log_get_buf(sdp);
 303         ld = (struct gfs2_log_descriptor *)bh->b_data;
 304         ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
 305         ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
 306         ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
 307         ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_REVOKE);
 308         ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
 309                                                     sizeof(u64)));
 310         ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
 311         ld->ld_data2 = cpu_to_be32(0);
 312         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 313         offset = sizeof(struct gfs2_log_descriptor);
 314
 315         while (!list_empty(head)) {
 316                 rv = list_entry(head->next, struct gfs2_revoke, rv_le.le_list);
 317                 list_del_init(&rv->rv_le.le_list);
 318                 sdp->sd_log_num_revoke--;
 319
 320                 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
 321                         set_buffer_dirty(bh);
 322                         ll_rw_block(WRITE, 1, &bh);
 323
 324                         bh = gfs2_log_get_buf(sdp);
 325                         mh = (struct gfs2_meta_header *)bh->b_data;
 326                         mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
 327                         mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
 328                         mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
 329                         offset = sizeof(struct gfs2_meta_header);
 330                 }
 331
 332                 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(rv->rv_blkno);
 333                 kfree(rv);
 334
 335                 offset += sizeof(u64);
 336         }
 337         gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
 338
 339         set_buffer_dirty(bh);
 340         ll_rw_block(WRITE, 1, &bh);
 341 }
 342
 343 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
 344                                   struct gfs2_log_header_host *head, int pass)
 345 {
 346         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 347
 348         if (pass != 0)
 349                 return;
 350
 351         sdp->sd_found_revokes = 0;
 352         sdp->sd_replay_tail = head->lh_tail;
 353 }
 354
 355 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 356                                    struct gfs2_log_descriptor *ld, __be64 *ptr,
 357                                    int pass)
 358 {
 359         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 360         unsigned int blks = be32_to_cpu(ld->ld_length);
 361         unsigned int revokes = be32_to_cpu(ld->ld_data1);
 362         struct buffer_head *bh;
 363         unsigned int offset;
 364         u64 blkno;
 365         int first = 1;
 366         int error;
 367
 368         if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
 369                 return 0;
 370
 371         offset = sizeof(struct gfs2_log_descriptor);
 372
 373         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 374                 error = gfs2_replay_read_block(jd, start, &bh);
 375                 if (error)
 376                         return error;
 377
 378                 if (!first)
 379                         gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
 380
 381                 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
 382                         blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
 383
 384                         error = gfs2_revoke_add(sdp, blkno, start);
 385                         if (error < 0)
 386                                 return error;
 387                         else if (error)
 388                                 sdp->sd_found_revokes++;
 389
 390                         if (!--revokes)
 391                                 break;
 392                         offset += sizeof(u64);
 393                 }
 394
 395                 brelse(bh);
 396                 offset = sizeof(struct gfs2_meta_header);
 397                 first = 0;
 398         }
 399
 400         return 0;
 401 }
 402
 403 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 404 {
 405         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 406
 407         if (error) {
 408                 gfs2_revoke_clean(sdp);
 409                 return;
 410         }
 411         if (pass != 1)
 412                 return;
 413
 414         fs_info(sdp, "jid=%u: Found %u revoke tags\n",
 415                 jd->jd_jid, sdp->sd_found_revokes);
 416
 417         gfs2_revoke_clean(sdp);
 418 }
 419
 420 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 421 {
 422         struct gfs2_rgrpd *rgd;
 423         struct gfs2_trans *tr = current->journal_info;
 424
 425         tr->tr_touched = 1;
 426
 427         rgd = container_of(le, struct gfs2_rgrpd, rd_le);
 428
 429         gfs2_log_lock(sdp);
 430         if (!list_empty(&le->le_list)){
 431                 gfs2_log_unlock(sdp);
 432                 return;
 433         }
 434         gfs2_rgrp_bh_hold(rgd);
 435         sdp->sd_log_num_rg++;
 436         list_add(&le->le_list, &sdp->sd_log_le_rg);
 437         gfs2_log_unlock(sdp);
 438 }
 439
 440 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 441 {
 442         struct list_head *head = &sdp->sd_log_le_rg;
 443         struct gfs2_rgrpd *rgd;
 444
 445         while (!list_empty(head)) {
 446                 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
 447                 list_del_init(&rgd->rd_le.le_list);
 448                 sdp->sd_log_num_rg--;
 449
 450                 gfs2_rgrp_repolish_clones(rgd);
 451                 gfs2_rgrp_bh_put(rgd);
 452         }
 453         gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
 454 }
 455
 456 /**
 457  * databuf_lo_add - Add a databuf to the transaction.
 458  *
 459  * This is used in two distinct cases:
 460  * i) In ordered write mode
 461  *    We put the data buffer on a list so that we can ensure that its
 462  *    synced to disk at the right time
 463  * ii) In journaled data mode
 464  *    We need to journal the data block in the same way as metadata in
 465  *    the functions above. The difference is that here we have a tag
 466  *    which is two __be64's being the block number (as per meta data)
 467  *    and a flag which says whether the data block needs escaping or
 468  *    not. This means we need a new log entry for each 251 or so data
 469  *    blocks, which isn't an enormous overhead but twice as much as
 470  *    for normal metadata blocks.
 471  */
 472 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
 473 {
 474         struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
 475         struct gfs2_trans *tr = current->journal_info;
 476         struct address_space *mapping = bd->bd_bh->b_page->mapping;
 477         struct gfs2_inode *ip = GFS2_I(mapping->host);
 478
 479         gfs2_log_lock(sdp);
 480         if (!list_empty(&bd->bd_list_tr)) {
 481                 gfs2_log_unlock(sdp);
 482                 return;
 483         }
 484         tr->tr_touched = 1;
 485         if (gfs2_is_jdata(ip)) {
 486                 tr->tr_num_buf++;
 487                 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
 488         }
 489         gfs2_log_unlock(sdp);
 490         if (!list_empty(&le->le_list))
 491                 return;
 492
 493         gfs2_trans_add_gl(bd->bd_gl);
 494         if (gfs2_is_jdata(ip)) {
 495                 sdp->sd_log_num_jdata++;
 496                 gfs2_pin(sdp, bd->bd_bh);
 497                 tr->tr_num_databuf_new++;
 498         }
 499         gfs2_log_lock(sdp);
 500         sdp->sd_log_num_databuf++;
 501         list_add(&le->le_list, &sdp->sd_log_le_databuf);
 502         gfs2_log_unlock(sdp);
 503 }
 504
 505 static int gfs2_check_magic(struct buffer_head *bh)
 506 {
 507         struct page *page = bh->b_page;
 508         void *kaddr;
 509         __be32 *ptr;
 510         int rv = 0;
 511
 512         kaddr = kmap_atomic(page, KM_USER0);
 513         ptr = kaddr + bh_offset(bh);
 514         if (*ptr == cpu_to_be32(GFS2_MAGIC))
 515                 rv = 1;
 516         kunmap_atomic(kaddr, KM_USER0);
 517
 518         return rv;
 519 }
 520
 521 /**
 522  * databuf_lo_before_commit - Scan the data buffers, writing as we go
 523  *
 524  * Here we scan through the lists of buffers and make the assumption
 525  * that any buffer thats been pinned is being journaled, and that
 526  * any unpinned buffer is an ordered write data buffer and therefore
 527  * will be written back rather than journaled.
 528  */
 529 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
 530 {
 531         LIST_HEAD(started);
 532         struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt;
 533         struct buffer_head *bh = NULL,*bh1 = NULL;
 534         struct gfs2_log_descriptor *ld;
 535         unsigned int limit;
 536         unsigned int total_dbuf;
 537         unsigned int total_jdata;
 538         unsigned int num, n;
 539         __be64 *ptr = NULL;
 540
 541         limit = databuf_limit(sdp);
 542
 543         /*
 544          * Start writing ordered buffers, write journaled buffers
 545          * into the log along with a header
 546          */
 547         gfs2_log_lock(sdp);
 548         total_dbuf = sdp->sd_log_num_databuf;
 549         total_jdata = sdp->sd_log_num_jdata;
 550         bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf,
 551                                        bd_le.le_list);
 552         while(total_dbuf) {
 553                 num = total_jdata;
 554                 if (num > limit)
 555                         num = limit;
 556                 n = 0;
 557                 list_for_each_entry_safe_continue(bd1, bdt,
 558                                                   &sdp->sd_log_le_databuf,
 559                                                   bd_le.le_list) {
 560                         /* store off the buffer head in a local ptr since
 561                          * gfs2_bufdata might change when we drop the log lock
 562                          */
 563                         bh1 = bd1->bd_bh;
 564
 565                         /* An ordered write buffer */
 566                         if (bh1 && !buffer_pinned(bh1)) {
 567                                 list_move(&bd1->bd_le.le_list, &started);
 568                                 if (bd1 == bd2) {
 569                                         bd2 = NULL;
 570                                         bd2 = list_prepare_entry(bd2,
 571                                                         &sdp->sd_log_le_databuf,
 572                                                         bd_le.le_list);
 573                                 }
 574                                 total_dbuf--;
 575                                 if (bh1) {
 576                                         if (buffer_dirty(bh1)) {
 577                                                 get_bh(bh1);
 578
 579                                                 gfs2_log_unlock(sdp);
 580
 581                                                 ll_rw_block(SWRITE, 1, &bh1);
 582                                                 brelse(bh1);
 583
 584                                                 gfs2_log_lock(sdp);
 585                                         }
 586                                         continue;
 587                                 }
 588                                 continue;
 589                         } else if (bh1) { /* A journaled buffer */
 590                                 int magic;
 591                                 gfs2_log_unlock(sdp);
 592                                 if (!bh) {
 593                                         bh = gfs2_log_get_buf(sdp);
 594                                         ld = (struct gfs2_log_descriptor *)
 595                                              bh->b_data;
 596                                         ptr = (__be64 *)(bh->b_data +
 597                                                          DATABUF_OFFSET);
 598                                         ld->ld_header.mh_magic =
 599                                                 cpu_to_be32(GFS2_MAGIC);
 600                                         ld->ld_header.mh_type =
 601                                                 cpu_to_be32(GFS2_METATYPE_LD);
 602                                         ld->ld_header.mh_format =
 603                                                 cpu_to_be32(GFS2_FORMAT_LD);
 604                                         ld->ld_type =
 605                                                 cpu_to_be32(GFS2_LOG_DESC_JDATA);
 606                                         ld->ld_length = cpu_to_be32(num + 1);
 607                                         ld->ld_data1 = cpu_to_be32(num);
 608                                         ld->ld_data2 = cpu_to_be32(0);
 609                                         memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
 610                                 }
 611                                 magic = gfs2_check_magic(bh1);
 612                                 *ptr++ = cpu_to_be64(bh1->b_blocknr);
 613                                 *ptr++ = cpu_to_be64((__u64)magic);
 614                                 clear_buffer_escaped(bh1);
 615                                 if (unlikely(magic != 0))
 616                                         set_buffer_escaped(bh1);
 617                                 gfs2_log_lock(sdp);
 618                                 if (++n >= num)
 619                                         break;
 620                         } else if (!bh1) {
 621                                 total_dbuf--;
 622                                 sdp->sd_log_num_databuf--;
 623                                 list_del_init(&bd1->bd_le.le_list);
 624                                 if (bd1 == bd2) {
 625                                         bd2 = NULL;
 626                                         bd2 = list_prepare_entry(bd2,
 627                                                 &sdp->sd_log_le_databuf,
 628                                                 bd_le.le_list);
 629                                 }
 630                                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
 631                         }
 632                 }
 633                 gfs2_log_unlock(sdp);
 634                 if (bh) {
 635                         set_buffer_dirty(bh);
 636                         ll_rw_block(WRITE, 1, &bh);
 637                         bh = NULL;
 638                         ptr = NULL;
 639                 }
 640                 n = 0;
 641                 gfs2_log_lock(sdp);
 642                 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf,
 643                                              bd_le.le_list) {
 644                         if (!bd2->bd_bh)
 645                                 continue;
 646                         /* copy buffer if it needs escaping */
 647                         gfs2_log_unlock(sdp);
 648                         if (unlikely(buffer_escaped(bd2->bd_bh))) {
 649                                 void *kaddr;
 650                                 struct page *page = bd2->bd_bh->b_page;
 651                                 bh = gfs2_log_get_buf(sdp);
 652                                 kaddr = kmap_atomic(page, KM_USER0);
 653                                 memcpy(bh->b_data,
 654                                        kaddr + bh_offset(bd2->bd_bh),
 655                                        sdp->sd_sb.sb_bsize);
 656                                 kunmap_atomic(kaddr, KM_USER0);
 657                                 *(__be32 *)bh->b_data = 0;
 658                         } else {
 659                                 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
 660                         }
 661                         set_buffer_dirty(bh);
 662                         ll_rw_block(WRITE, 1, &bh);
 663                         gfs2_log_lock(sdp);
 664                         if (++n >= num)
 665                                 break;
 666                 }
 667                 bh = NULL;
 668                 BUG_ON(total_dbuf < num);
 669                 total_dbuf -= num;
 670                 total_jdata -= num;
 671         }
 672         gfs2_log_unlock(sdp);
 673
 674         /* Wait on all ordered buffers */
 675         while (!list_empty(&started)) {
 676                 gfs2_log_lock(sdp);
 677                 bd1 = list_entry(started.next, struct gfs2_bufdata,
 678                                  bd_le.le_list);
 679                 list_del_init(&bd1->bd_le.le_list);
 680                 sdp->sd_log_num_databuf--;
 681                 bh = bd1->bd_bh;
 682                 if (bh) {
 683                         bh->b_private = NULL;
 684                         get_bh(bh);
 685                         gfs2_log_unlock(sdp);
 686                         wait_on_buffer(bh);
 687                         brelse(bh);
 688                 } else
 689                         gfs2_log_unlock(sdp);
 690
 691                 kmem_cache_free(gfs2_bufdata_cachep, bd1);
 692         }
 693
 694         /* We've removed all the ordered write bufs here, so only jdata left */
 695         gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata);
 696 }
 697
 698 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
 699                                     struct gfs2_log_descriptor *ld,
 700                                     __be64 *ptr, int pass)
 701 {
 702         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 703         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 704         struct gfs2_glock *gl = ip->i_gl;
 705         unsigned int blks = be32_to_cpu(ld->ld_data1);
 706         struct buffer_head *bh_log, *bh_ip;
 707         u64 blkno;
 708         u64 esc;
 709         int error = 0;
 710
 711         if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
 712                 return 0;
 713
 714         gfs2_replay_incr_blk(sdp, &start);
 715         for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
 716                 blkno = be64_to_cpu(*ptr++);
 717                 esc = be64_to_cpu(*ptr++);
 718
 719                 sdp->sd_found_blocks++;
 720
 721                 if (gfs2_revoke_check(sdp, blkno, start))
 722                         continue;
 723
 724                 error = gfs2_replay_read_block(jd, start, &bh_log);
 725                 if (error)
 726                         return error;
 727
 728                 bh_ip = gfs2_meta_new(gl, blkno);
 729                 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
 730
 731                 /* Unescape */
 732                 if (esc) {
 733                         __be32 *eptr = (__be32 *)bh_ip->b_data;
 734                         *eptr = cpu_to_be32(GFS2_MAGIC);
 735                 }
 736                 mark_buffer_dirty(bh_ip);
 737
 738                 brelse(bh_log);
 739                 brelse(bh_ip);
 740                 if (error)
 741                         break;
 742
 743                 sdp->sd_replayed_blocks++;
 744         }
 745
 746         return error;
 747 }
 748
 749 /* FIXME: sort out accounting for log blocks etc. */
 750
 751 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
 752 {
 753         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
 754         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
 755
 756         if (error) {
 757                 gfs2_meta_sync(ip->i_gl);
 758                 return;
 759         }
 760         if (pass != 1)
 761                 return;
 762
 763         /* data sync? */
 764         gfs2_meta_sync(ip->i_gl);
 765
 766         fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
 767                 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
 768 }
 769
 770 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
 771 {
 772         struct list_head *head = &sdp->sd_log_le_databuf;
 773         struct gfs2_bufdata *bd;
 774
 775         while (!list_empty(head)) {
 776                 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
 777                 list_del_init(&bd->bd_le.le_list);
 778                 sdp->sd_log_num_databuf--;
 779                 sdp->sd_log_num_jdata--;
 780                 gfs2_unpin(sdp, bd->bd_bh, ai);
 781         }
 782         gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
 783         gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata);
 784 }
 785
 786
 787 const struct gfs2_log_operations gfs2_glock_lops = {
 788         .lo_add = glock_lo_add,
 789         .lo_after_commit = glock_lo_after_commit,
 790         .lo_name = "glock",
 791 };
 792
 793 const struct gfs2_log_operations gfs2_buf_lops = {
 794         .lo_add = buf_lo_add,
 795         .lo_incore_commit = buf_lo_incore_commit,
 796         .lo_before_commit = buf_lo_before_commit,
 797         .lo_after_commit = buf_lo_after_commit,
 798         .lo_before_scan = buf_lo_before_scan,
 799         .lo_scan_elements = buf_lo_scan_elements,
 800         .lo_after_scan = buf_lo_after_scan,
 801         .lo_name = "buf",
 802 };
 803
 804 const struct gfs2_log_operations gfs2_revoke_lops = {
 805         .lo_add = revoke_lo_add,
 806         .lo_before_commit = revoke_lo_before_commit,
 807         .lo_before_scan = revoke_lo_before_scan,
 808         .lo_scan_elements = revoke_lo_scan_elements,
 809         .lo_after_scan = revoke_lo_after_scan,
 810         .lo_name = "revoke",
 811 };
 812
 813 const struct gfs2_log_operations gfs2_rg_lops = {
 814         .lo_add = rg_lo_add,
 815         .lo_after_commit = rg_lo_after_commit,
 816         .lo_name = "rg",
 817 };
 818
 819 const struct gfs2_log_operations gfs2_databuf_lops = {
 820         .lo_add = databuf_lo_add,
 821         .lo_incore_commit = buf_lo_incore_commit,
 822         .lo_before_commit = databuf_lo_before_commit,
 823         .lo_after_commit = databuf_lo_after_commit,
 824         .lo_scan_elements = databuf_lo_scan_elements,
 825         .lo_after_scan = databuf_lo_after_scan,
 826         .lo_name = "databuf",
 827 };
 828
 829 const struct gfs2_log_operations *gfs2_log_ops[] = {
 830         &gfs2_glock_lops,
 831         &gfs2_buf_lops,
 832         &gfs2_revoke_lops,
 833         &gfs2_rg_lops,
 834         &gfs2_databuf_lops,
 835         NULL,
 836 };
 837