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 = get_v2ip(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 = get_v2ip(jd->jd_inode)->i_sbd; 194 struct gfs2_glock *gl = get_v2ip(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 = get_v2ip(jd->jd_inode)->i_sbd; 240 241 if (error) { 242 gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl, DIO_START | DIO_WAIT); 243 return; 244 } 245 if (pass != 1) 246 return; 247 248 gfs2_meta_sync(get_v2ip(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 = get_v2ip(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 = get_v2ip(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 = get_v2ip(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 } 462 gfs2_log_lock(sdp); 463 if (ip->i_di.di_flags & GFS2_DIF_JDATA) 464 sdp->sd_log_num_jdata++; 465 sdp->sd_log_num_databuf++; 466 list_add(&le->le_list, &sdp->sd_log_le_databuf); 467 gfs2_log_unlock(sdp); 468 } 469 470 static int gfs2_check_magic(struct buffer_head *bh) 471 { 472 struct page *page = bh->b_page; 473 void *kaddr; 474 __be32 *ptr; 475 int rv = 0; 476 477 kaddr = kmap_atomic(page, KM_USER0); 478 ptr = kaddr + bh_offset(bh); 479 if (*ptr == cpu_to_be32(GFS2_MAGIC)) 480 rv = 1; 481 kunmap_atomic(page, KM_USER0); 482 483 return rv; 484 } 485 486 /** 487 * databuf_lo_before_commit - Scan the data buffers, writing as we go 488 * 489 * Here we scan through the lists of buffers and make the assumption 490 * that any buffer thats been pinned is being journaled, and that 491 * any unpinned buffer is an ordered write data buffer and therefore 492 * will be written back rather than journaled. 493 */ 494 static void databuf_lo_before_commit(struct gfs2_sbd *sdp) 495 { 496 LIST_HEAD(started); 497 struct gfs2_bufdata *bd1 = NULL, *bd2, *bdt; 498 struct buffer_head *bh = NULL; 499 unsigned int offset = sizeof(struct gfs2_log_descriptor); 500 struct gfs2_log_descriptor *ld; 501 unsigned int limit; 502 unsigned int total_dbuf = sdp->sd_log_num_databuf; 503 unsigned int total_jdata = sdp->sd_log_num_jdata; 504 unsigned int num, n; 505 __be64 *ptr = NULL; 506 unsigned i; 507 508 offset += (2*sizeof(__be64) - 1); 509 offset &= ~(2*sizeof(__be64) - 1); 510 limit = (sdp->sd_sb.sb_bsize - offset)/sizeof(__be64); 511 512 /* printk(KERN_INFO "totals: jdata=%u dbuf=%u\n", total_jdata, total_dbuf); */ 513 /* 514 * Start writing ordered buffers, write journaled buffers 515 * into the log along with a header 516 */ 517 gfs2_log_lock(sdp); 518 /* printk(KERN_INFO "locked in lops databuf_before_commit\n"); */ 519 bd2 = bd1 = list_prepare_entry(bd1, &sdp->sd_log_le_databuf, bd_le.le_list); 520 while(total_dbuf) { 521 num = total_jdata; 522 if (num > limit) 523 num = limit; 524 /* printk(KERN_INFO "total_dbuf=%u num=%u\n", total_dbuf, num); */ 525 n = 0; 526 i = 0; 527 list_for_each_entry_safe_continue(bd1, bdt, &sdp->sd_log_le_databuf, bd_le.le_list) { 528 /* An ordered write buffer */ 529 if (bd1->bd_bh && !buffer_pinned(bd1->bd_bh)) { 530 list_move(&bd1->bd_le.le_list, &started); 531 if (bd1 == bd2) { 532 bd2 = NULL; 533 bd2 = list_prepare_entry(bd2, &sdp->sd_log_le_databuf, bd_le.le_list); 534 } 535 total_dbuf--; 536 if (bd1->bd_bh) { 537 get_bh(bd1->bd_bh); 538 if (buffer_dirty(bd1->bd_bh)) { 539 gfs2_log_unlock(sdp); 540 wait_on_buffer(bd1->bd_bh); 541 ll_rw_block(WRITE, 1, &bd1->bd_bh); 542 gfs2_log_lock(sdp); 543 } 544 brelse(bd1->bd_bh); 545 /* printk(KERN_INFO "db write %p\n", bd1); */ 546 if (++i > 100000) { 547 printk(KERN_INFO "looping bd1=%p bdt=%p eol=%p started=%p\n", bd1, bdt, &sdp->sd_log_le_databuf, &started); 548 dump_stack(); 549 BUG(); 550 } 551 continue; 552 } 553 /* printk(KERN_INFO "db skip\n"); */ 554 continue; 555 } else if (bd1->bd_bh) { /* A journaled buffer */ 556 int magic; 557 gfs2_log_unlock(sdp); 558 printk(KERN_INFO "journaled buffer %p\n", bd1->bd_bh); 559 printk(KERN_INFO "%lu %u %p %p\n", bd1->bd_bh->b_blocknr, bd1->bd_bh->b_size, bd1->bd_bh->b_data, bd1->bd_bh->b_page); 560 if (!bh) { 561 bh = gfs2_log_get_buf(sdp); 562 ld = (struct gfs2_log_descriptor *)bh->b_data; 563 ptr = (__be64 *)(bh->b_data + offset); 564 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC); 565 ld->ld_header.mh_type = cpu_to_be16(GFS2_METATYPE_LD); 566 ld->ld_header.mh_format = cpu_to_be16(GFS2_FORMAT_LD); 567 ld->ld_type = cpu_to_be32(GFS2_LOG_DESC_JDATA); 568 ld->ld_length = cpu_to_be32(num + 1); 569 ld->ld_data1 = cpu_to_be32(num); 570 ld->ld_data2 = cpu_to_be32(0); 571 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved)); 572 } 573 /* printk(KERN_INFO "check_magic\n"); */ 574 magic = gfs2_check_magic(bd1->bd_bh); 575 /* printk(KERN_INFO "write data\n"); */ 576 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr); 577 *ptr++ = cpu_to_be64((__u64)magic); 578 /* printk(KERN_INFO "mark escaped or not\n"); */ 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 /* printk(KERN_INFO "totals2: jdata=%u dbuf=%u\n", total_jdata, total_dbuf); */ 595 gfs2_log_lock(sdp); 596 list_for_each_entry_continue(bd2, &sdp->sd_log_le_databuf, 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, kaddr + bh_offset(bd2->bd_bh), sdp->sd_sb.sb_bsize); 607 kunmap_atomic(page, KM_USER0); 608 *(__be32 *)bh->b_data = 0; 609 } else { 610 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh); 611 } 612 set_buffer_dirty(bh); 613 ll_rw_block(WRITE, 1, &bh); 614 gfs2_log_lock(sdp); 615 if (++n >= num) 616 break; 617 } 618 bh = NULL; 619 total_dbuf -= num; 620 total_jdata -= num; 621 } 622 gfs2_log_unlock(sdp); 623 624 /* printk(KERN_INFO "wait on ordered data buffers\n"); */ 625 /* Wait on all ordered buffers */ 626 while (!list_empty(&started)) { 627 bd1 = list_entry(started.next, struct gfs2_bufdata, bd_le.le_list); 628 list_del(&bd1->bd_le.le_list); 629 sdp->sd_log_num_databuf--; 630 631 gfs2_log_lock(sdp); 632 bh = bd1->bd_bh; 633 if (bh) { 634 set_v2bd(bh, NULL); 635 gfs2_log_unlock(sdp); 636 wait_on_buffer(bh); 637 brelse(bh); 638 } else 639 gfs2_log_unlock(sdp); 640 641 kfree(bd1); 642 } 643 644 /* printk(KERN_INFO "sd_log_num_databuf %u sd_log_num_jdata %u\n", sdp->sd_log_num_databuf, sdp->sd_log_num_jdata); */ 645 /* We've removed all the ordered write bufs here, so only jdata left */ 646 gfs2_assert_warn(sdp, sdp->sd_log_num_databuf == sdp->sd_log_num_jdata); 647 } 648 649 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start, 650 struct gfs2_log_descriptor *ld, 651 __be64 *ptr, int pass) 652 { 653 struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd; 654 struct gfs2_glock *gl = get_v2ip(jd->jd_inode)->i_gl; 655 unsigned int blks = be32_to_cpu(ld->ld_data1); 656 struct buffer_head *bh_log, *bh_ip; 657 uint64_t blkno; 658 uint64_t esc; 659 int error = 0; 660 661 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA) 662 return 0; 663 664 gfs2_replay_incr_blk(sdp, &start); 665 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) { 666 blkno = be64_to_cpu(*ptr++); 667 esc = be64_to_cpu(*ptr++); 668 669 sdp->sd_found_blocks++; 670 671 if (gfs2_revoke_check(sdp, blkno, start)) 672 continue; 673 674 error = gfs2_replay_read_block(jd, start, &bh_log); 675 if (error) 676 return error; 677 678 bh_ip = gfs2_meta_new(gl, blkno); 679 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size); 680 681 /* Unescape */ 682 if (esc) { 683 __be32 *eptr = (__be32 *)bh_ip->b_data; 684 *eptr = cpu_to_be32(GFS2_MAGIC); 685 } 686 mark_buffer_dirty(bh_ip); 687 688 brelse(bh_log); 689 brelse(bh_ip); 690 if (error) 691 break; 692 693 sdp->sd_replayed_blocks++; 694 } 695 696 return error; 697 } 698 699 /* FIXME: sort out accounting for log blocks etc. */ 700 701 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass) 702 { 703 struct gfs2_sbd *sdp = get_v2ip(jd->jd_inode)->i_sbd; 704 705 if (error) { 706 gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl, DIO_START | DIO_WAIT); 707 return; 708 } 709 if (pass != 1) 710 return; 711 712 /* data sync? */ 713 gfs2_meta_sync(get_v2ip(jd->jd_inode)->i_gl, DIO_START | DIO_WAIT); 714 715 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n", 716 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks); 717 } 718 719 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai) 720 { 721 struct list_head *head = &sdp->sd_log_le_databuf; 722 struct gfs2_bufdata *bd; 723 724 while (!list_empty(head)) { 725 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list); 726 list_del(&bd->bd_le.le_list); 727 sdp->sd_log_num_databuf--; 728 sdp->sd_log_num_jdata--; 729 gfs2_unpin(sdp, bd->bd_bh, ai); 730 } 731 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf); 732 gfs2_assert_warn(sdp, !sdp->sd_log_num_jdata); 733 } 734 735 736 struct gfs2_log_operations gfs2_glock_lops = { 737 .lo_add = glock_lo_add, 738 .lo_after_commit = glock_lo_after_commit, 739 .lo_name = "glock" 740 }; 741 742 struct gfs2_log_operations gfs2_buf_lops = { 743 .lo_add = buf_lo_add, 744 .lo_incore_commit = buf_lo_incore_commit, 745 .lo_before_commit = buf_lo_before_commit, 746 .lo_after_commit = buf_lo_after_commit, 747 .lo_before_scan = buf_lo_before_scan, 748 .lo_scan_elements = buf_lo_scan_elements, 749 .lo_after_scan = buf_lo_after_scan, 750 .lo_name = "buf" 751 }; 752 753 struct gfs2_log_operations gfs2_revoke_lops = { 754 .lo_add = revoke_lo_add, 755 .lo_before_commit = revoke_lo_before_commit, 756 .lo_before_scan = revoke_lo_before_scan, 757 .lo_scan_elements = revoke_lo_scan_elements, 758 .lo_after_scan = revoke_lo_after_scan, 759 .lo_name = "revoke" 760 }; 761 762 struct gfs2_log_operations gfs2_rg_lops = { 763 .lo_add = rg_lo_add, 764 .lo_after_commit = rg_lo_after_commit, 765 .lo_name = "rg" 766 }; 767 768 struct gfs2_log_operations gfs2_databuf_lops = { 769 .lo_add = databuf_lo_add, 770 .lo_incore_commit = buf_lo_incore_commit, 771 .lo_before_commit = databuf_lo_before_commit, 772 .lo_after_commit = databuf_lo_after_commit, 773 .lo_scan_elements = databuf_lo_scan_elements, 774 .lo_after_scan = databuf_lo_after_scan, 775 .lo_name = "databuf" 776 }; 777 778 struct gfs2_log_operations *gfs2_log_ops[] = { 779 &gfs2_glock_lops, 780 &gfs2_buf_lops, 781 &gfs2_revoke_lops, 782 &gfs2_rg_lops, 783 &gfs2_databuf_lops, 784 NULL 785 }; 786 787