1 /* $NetBSD: udf_strat_rmw.c,v 1.16 2008/12/16 16:18:25 pooka Exp $ */ 2 3 /* 4 * Copyright (c) 2006, 2008 Reinoud Zandijk 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 * 27 */ 28 29 #include <sys/cdefs.h> 30 #ifndef lint 31 __KERNEL_RCSID(0, "$NetBSD: udf_strat_rmw.c,v 1.16 2008/12/16 16:18:25 pooka Exp $"); 32 #endif /* not lint */ 33 34 35 #if defined(_KERNEL_OPT) 36 #include "opt_compat_netbsd.h" 37 #endif 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/sysctl.h> 42 #include <sys/namei.h> 43 #include <sys/proc.h> 44 #include <sys/kernel.h> 45 #include <sys/vnode.h> 46 #include <miscfs/genfs/genfs_node.h> 47 #include <sys/mount.h> 48 #include <sys/buf.h> 49 #include <sys/file.h> 50 #include <sys/device.h> 51 #include <sys/disklabel.h> 52 #include <sys/ioctl.h> 53 #include <sys/malloc.h> 54 #include <sys/dirent.h> 55 #include <sys/stat.h> 56 #include <sys/conf.h> 57 #include <sys/kauth.h> 58 #include <sys/kthread.h> 59 #include <dev/clock_subr.h> 60 61 #include <fs/udf/ecma167-udf.h> 62 #include <fs/udf/udf_mount.h> 63 64 #include "udf.h" 65 #include "udf_subr.h" 66 #include "udf_bswap.h" 67 68 69 #define VTOI(vnode) ((struct udf_node *) (vnode)->v_data) 70 #define PRIV(ump) ((struct strat_private *) (ump)->strategy_private) 71 #define BTOE(buf) ((struct udf_eccline *) ((buf)->b_private)) 72 73 /* --------------------------------------------------------------------- */ 74 75 #define UDF_MAX_PACKET_SIZE 64 /* DONT change this */ 76 77 /* sheduler states */ 78 #define UDF_SHED_WAITING 1 /* waiting on timeout */ 79 #define UDF_SHED_READING 2 80 #define UDF_SHED_WRITING 3 81 #define UDF_SHED_SEQWRITING 4 82 #define UDF_SHED_IDLE 5 /* resting */ 83 #define UDF_SHED_FREE 6 /* recycleable */ 84 #define UDF_SHED_MAX 6+1 85 86 /* flags */ 87 #define ECC_LOCKED 0x01 /* prevent access */ 88 #define ECC_WANTED 0x02 /* trying access */ 89 #define ECC_SEQWRITING 0x04 /* sequential queue */ 90 #define ECC_FLOATING 0x08 /* not queued yet */ 91 92 #define ECC_WAITTIME 10 93 94 95 TAILQ_HEAD(ecclineq, udf_eccline); 96 struct udf_eccline { 97 struct udf_mount *ump; 98 uint64_t present; /* preserve these */ 99 uint64_t readin; /* bitmap */ 100 uint64_t dirty; /* bitmap */ 101 uint64_t error; /* bitmap */ 102 uint32_t refcnt; 103 104 struct timespec wait_time; 105 uint32_t flags; 106 uint32_t start_sector; /* physical */ 107 108 struct buf *buf; 109 void *blob; 110 111 struct buf *bufs[UDF_MAX_PACKET_SIZE]; 112 uint32_t bufs_bpos[UDF_MAX_PACKET_SIZE]; 113 int bufs_len[UDF_MAX_PACKET_SIZE]; 114 115 int queued_on; /* on which BUFQ list */ 116 LIST_ENTRY(udf_eccline) hashchain; /* on sector lookup */ 117 }; 118 119 120 struct strat_private { 121 lwp_t *queue_lwp; 122 kcondvar_t discstrat_cv; /* to wait on */ 123 kmutex_t discstrat_mutex; /* disc strategy */ 124 kmutex_t seqwrite_mutex; /* protect mappings */ 125 126 int thread_running; /* thread control */ 127 int run_thread; /* thread control */ 128 int thread_finished; /* thread control */ 129 int cur_queue; 130 131 int num_floating; 132 int num_queued[UDF_SHED_MAX]; 133 struct bufq_state *queues[UDF_SHED_MAX]; 134 struct timespec last_queued[UDF_SHED_MAX]; 135 struct disk_strategy old_strategy_setting; 136 137 struct pool eccline_pool; 138 struct pool ecclineblob_pool; 139 LIST_HEAD(, udf_eccline) eccline_hash[UDF_ECCBUF_HASHSIZE]; 140 }; 141 142 /* --------------------------------------------------------------------- */ 143 144 #define UDF_LOCK_ECCLINE(eccline) udf_lock_eccline(eccline) 145 #define UDF_UNLOCK_ECCLINE(eccline) udf_unlock_eccline(eccline) 146 147 /* can be called with or without discstrat lock */ 148 static void 149 udf_lock_eccline(struct udf_eccline *eccline) 150 { 151 struct strat_private *priv = PRIV(eccline->ump); 152 int waslocked, ret; 153 154 waslocked = mutex_owned(&priv->discstrat_mutex); 155 if (!waslocked) 156 mutex_enter(&priv->discstrat_mutex); 157 158 /* wait until its unlocked first */ 159 while (eccline->flags & ECC_LOCKED) { 160 eccline->flags |= ECC_WANTED; 161 ret = cv_timedwait(&priv->discstrat_cv, &priv->discstrat_mutex, 162 hz/8); 163 if (ret == EWOULDBLOCK) 164 DPRINTF(LOCKING, ("eccline lock helt, waiting for " 165 "release")); 166 } 167 eccline->flags |= ECC_LOCKED; 168 eccline->flags &= ~ECC_WANTED; 169 170 if (!waslocked) 171 mutex_exit(&priv->discstrat_mutex); 172 } 173 174 175 /* can be called with or without discstrat lock */ 176 static void 177 udf_unlock_eccline(struct udf_eccline *eccline) 178 { 179 struct strat_private *priv = PRIV(eccline->ump); 180 int waslocked; 181 182 waslocked = mutex_owned(&priv->discstrat_mutex); 183 if (!waslocked) 184 mutex_enter(&priv->discstrat_mutex); 185 186 eccline->flags &= ~ECC_LOCKED; 187 cv_broadcast(&priv->discstrat_cv); 188 189 if (!waslocked) 190 mutex_exit(&priv->discstrat_mutex); 191 } 192 193 194 /* NOTE discstrat_mutex should be held! */ 195 static void 196 udf_dispose_eccline(struct udf_eccline *eccline) 197 { 198 struct strat_private *priv = PRIV(eccline->ump); 199 struct buf *ret; 200 201 KASSERT(mutex_owned(&priv->discstrat_mutex)); 202 203 KASSERT(eccline->refcnt == 0); 204 KASSERT(eccline->dirty == 0); 205 206 DPRINTF(ECCLINE, ("dispose eccline with start sector %d, " 207 "present %0"PRIx64"\n", eccline->start_sector, 208 eccline->present)); 209 210 if (eccline->queued_on) { 211 ret = BUFQ_CANCEL(priv->queues[eccline->queued_on], eccline->buf); 212 KASSERT(ret == eccline->buf); 213 priv->num_queued[eccline->queued_on]--; 214 } 215 LIST_REMOVE(eccline, hashchain); 216 217 if (eccline->flags & ECC_FLOATING) { 218 eccline->flags &= ~ECC_FLOATING; 219 priv->num_floating--; 220 } 221 222 putiobuf(eccline->buf); 223 pool_put(&priv->ecclineblob_pool, eccline->blob); 224 pool_put(&priv->eccline_pool, eccline); 225 } 226 227 228 /* NOTE discstrat_mutex should be held! */ 229 static void 230 udf_push_eccline(struct udf_eccline *eccline, int newqueue) 231 { 232 struct strat_private *priv = PRIV(eccline->ump); 233 struct buf *ret; 234 int curqueue; 235 236 KASSERT(mutex_owned(&priv->discstrat_mutex)); 237 238 DPRINTF(PARANOIA, ("DEBUG: buf %p pushed on queue %d\n", eccline->buf, newqueue)); 239 240 /* requeue */ 241 curqueue = eccline->queued_on; 242 if (curqueue) { 243 ret = BUFQ_CANCEL(priv->queues[curqueue], eccline->buf); 244 245 DPRINTF(PARANOIA, ("push_eccline BUFQ_CANCEL returned %p when " 246 "requested to remove %p from queue %d\n", ret, 247 eccline->buf, curqueue)); 248 #ifdef DIAGNOSTIC 249 if (ret == NULL) { 250 int i; 251 252 printf("udf_push_eccline: bufq_cancel can't find " 253 "buffer; dumping queues\n"); 254 for (i = 1; i < UDF_SHED_MAX; i++) { 255 printf("queue %d\n\t", i); 256 ret = BUFQ_GET(priv->queues[i]); 257 while (ret) { 258 printf("%p ", ret); 259 if (ret == eccline->buf) 260 printf("[<-] "); 261 ret = BUFQ_GET(priv->queues[i]); 262 } 263 printf("\n"); 264 } 265 panic("fatal queue bug; exit"); 266 } 267 #endif 268 269 KASSERT(ret == eccline->buf); 270 priv->num_queued[curqueue]--; 271 } 272 273 /* set buffer block numbers to make sure its queued correctly */ 274 eccline->buf->b_lblkno = eccline->start_sector; 275 eccline->buf->b_blkno = eccline->start_sector; 276 eccline->buf->b_rawblkno = eccline->start_sector; 277 278 BUFQ_PUT(priv->queues[newqueue], eccline->buf); 279 eccline->queued_on = newqueue; 280 priv->num_queued[newqueue]++; 281 vfs_timestamp(&priv->last_queued[newqueue]); 282 283 if (eccline->flags & ECC_FLOATING) { 284 eccline->flags &= ~ECC_FLOATING; 285 priv->num_floating--; 286 } 287 288 /* tickle disc strategy statemachine */ 289 if (newqueue != UDF_SHED_IDLE) 290 cv_signal(&priv->discstrat_cv); 291 } 292 293 294 static struct udf_eccline * 295 udf_pop_eccline(struct strat_private *priv, int queued_on) 296 { 297 struct udf_eccline *eccline; 298 struct buf *buf; 299 300 KASSERT(mutex_owned(&priv->discstrat_mutex)); 301 302 buf = BUFQ_GET(priv->queues[queued_on]); 303 if (!buf) { 304 KASSERT(priv->num_queued[queued_on] == 0); 305 return NULL; 306 } 307 308 eccline = BTOE(buf); 309 KASSERT(eccline->queued_on == queued_on); 310 eccline->queued_on = 0; 311 priv->num_queued[queued_on]--; 312 313 if (eccline->flags & ECC_FLOATING) 314 panic("popping already marked floating eccline"); 315 eccline->flags |= ECC_FLOATING; 316 priv->num_floating++; 317 318 DPRINTF(PARANOIA, ("DEBUG: buf %p popped from queue %d\n", 319 eccline->buf, queued_on)); 320 321 return eccline; 322 } 323 324 325 static struct udf_eccline * 326 udf_geteccline(struct udf_mount *ump, uint32_t sector, int flags) 327 { 328 struct strat_private *priv = PRIV(ump); 329 struct udf_eccline *eccline; 330 uint32_t start_sector, lb_size, blobsize; 331 uint8_t *eccline_blob; 332 int line, line_offset; 333 int num_busy, ret; 334 335 line_offset = sector % ump->packet_size; 336 start_sector = sector - line_offset; 337 line = (start_sector/ump->packet_size) & UDF_ECCBUF_HASHMASK; 338 339 mutex_enter(&priv->discstrat_mutex); 340 KASSERT(priv->thread_running); 341 342 retry: 343 DPRINTF(ECCLINE, ("get line sector %d, line %d\n", sector, line)); 344 LIST_FOREACH(eccline, &priv->eccline_hash[line], hashchain) { 345 if (eccline->start_sector == start_sector) { 346 DPRINTF(ECCLINE, ("\tfound eccline, start_sector %d\n", 347 eccline->start_sector)); 348 349 UDF_LOCK_ECCLINE(eccline); 350 /* move from freelist (!) */ 351 if (eccline->queued_on == UDF_SHED_FREE) { 352 DPRINTF(ECCLINE, ("was on freelist\n")); 353 KASSERT(eccline->refcnt == 0); 354 udf_push_eccline(eccline, UDF_SHED_IDLE); 355 } 356 eccline->refcnt++; 357 mutex_exit(&priv->discstrat_mutex); 358 return eccline; 359 } 360 } 361 362 DPRINTF(ECCLINE, ("\tnot found in eccline cache\n")); 363 /* not found in eccline cache */ 364 365 lb_size = udf_rw32(ump->logical_vol->lb_size); 366 blobsize = ump->packet_size * lb_size; 367 368 /* dont allow too many pending requests */ 369 DPRINTF(ECCLINE, ("\tallocating new eccline\n")); 370 num_busy = (priv->num_queued[UDF_SHED_SEQWRITING] + priv->num_floating); 371 if ((flags & ECC_SEQWRITING) && (num_busy > UDF_ECCLINE_MAXBUSY)) { 372 ret = cv_timedwait(&priv->discstrat_cv, 373 &priv->discstrat_mutex, hz/8); 374 goto retry; 375 } 376 377 eccline_blob = pool_get(&priv->ecclineblob_pool, PR_NOWAIT); 378 eccline = pool_get(&priv->eccline_pool, PR_NOWAIT); 379 if ((eccline_blob == NULL) || (eccline == NULL)) { 380 if (eccline_blob) 381 pool_put(&priv->ecclineblob_pool, eccline_blob); 382 if (eccline) 383 pool_put(&priv->eccline_pool, eccline); 384 385 /* out of memory for now; canibalise freelist */ 386 eccline = udf_pop_eccline(priv, UDF_SHED_FREE); 387 if (eccline == NULL) { 388 /* serious trouble; wait and retry */ 389 cv_timedwait(&priv->discstrat_cv, 390 &priv->discstrat_mutex, hz/8); 391 goto retry; 392 } 393 /* push back line if we're waiting for it */ 394 if (eccline->flags & ECC_WANTED) { 395 udf_push_eccline(eccline, UDF_SHED_IDLE); 396 goto retry; 397 } 398 399 /* unlink this entry */ 400 LIST_REMOVE(eccline, hashchain); 401 402 KASSERT(eccline->flags & ECC_FLOATING); 403 404 eccline_blob = eccline->blob; 405 memset(eccline, 0, sizeof(struct udf_eccline)); 406 eccline->flags = ECC_FLOATING; 407 } else { 408 memset(eccline, 0, sizeof(struct udf_eccline)); 409 eccline->flags = ECC_FLOATING; 410 priv->num_floating++; 411 } 412 413 eccline->queued_on = 0; 414 eccline->blob = eccline_blob; 415 eccline->buf = getiobuf(NULL, true); 416 eccline->buf->b_private = eccline; /* IMPORTANT */ 417 418 /* initialise eccline blob */ 419 memset(eccline->blob, 0, blobsize); 420 421 eccline->ump = ump; 422 eccline->present = eccline->readin = eccline->dirty = 0; 423 eccline->error = 0; 424 eccline->refcnt = 0; 425 426 eccline->start_sector = start_sector; 427 eccline->buf->b_lblkno = start_sector; 428 eccline->buf->b_blkno = start_sector; 429 eccline->buf->b_rawblkno = start_sector; 430 431 LIST_INSERT_HEAD(&priv->eccline_hash[line], eccline, hashchain); 432 433 /* 434 * TODO possible optimalisation for checking overlap with partitions 435 * to get a clue on future eccline usage 436 */ 437 eccline->refcnt++; 438 UDF_LOCK_ECCLINE(eccline); 439 440 mutex_exit(&priv->discstrat_mutex); 441 442 return eccline; 443 } 444 445 446 static void 447 udf_puteccline(struct udf_eccline *eccline) 448 { 449 struct strat_private *priv = PRIV(eccline->ump); 450 struct udf_mount *ump = eccline->ump; 451 uint64_t allbits = ((uint64_t) 1 << ump->packet_size)-1; 452 453 mutex_enter(&priv->discstrat_mutex); 454 455 /* clear directly all readin requests from present ones */ 456 if (eccline->readin & eccline->present) { 457 /* clear all read bits that are already read in */ 458 eccline->readin &= (~eccline->present) & allbits; 459 wakeup(eccline); 460 } 461 462 DPRINTF(ECCLINE, ("put eccline start sector %d, refcnt %d\n", 463 eccline->start_sector, eccline->refcnt)); 464 465 /* if we have active nodes we dont set it on seqwriting */ 466 if (eccline->refcnt > 1) 467 eccline->flags &= ~ECC_SEQWRITING; 468 469 vfs_timestamp(&eccline->wait_time); 470 eccline->wait_time.tv_sec += ECC_WAITTIME; 471 udf_push_eccline(eccline, UDF_SHED_WAITING); 472 473 KASSERT(eccline->refcnt >= 1); 474 eccline->refcnt--; 475 UDF_UNLOCK_ECCLINE(eccline); 476 477 wakeup(eccline); 478 mutex_exit(&priv->discstrat_mutex); 479 } 480 481 /* --------------------------------------------------------------------- */ 482 483 static int 484 udf_create_nodedscr_rmw(struct udf_strat_args *args) 485 { 486 union dscrptr **dscrptr = &args->dscr; 487 struct udf_mount *ump = args->ump; 488 struct long_ad *icb = args->icb; 489 struct udf_eccline *eccline; 490 uint64_t bit; 491 uint32_t sectornr, lb_size, dummy; 492 uint8_t *mem; 493 int error, eccsect; 494 495 error = udf_translate_vtop(ump, icb, §ornr, &dummy); 496 if (error) 497 return error; 498 499 lb_size = udf_rw32(ump->logical_vol->lb_size); 500 501 /* get our eccline */ 502 eccline = udf_geteccline(ump, sectornr, 0); 503 eccsect = sectornr - eccline->start_sector; 504 505 bit = (uint64_t) 1 << eccsect; 506 eccline->readin &= ~bit; /* just in case */ 507 eccline->present |= bit; 508 eccline->dirty &= ~bit; /* Err... euhm... clean? */ 509 510 eccline->refcnt++; 511 512 /* clear space */ 513 mem = ((uint8_t *) eccline->blob) + eccsect * lb_size; 514 memset(mem, 0, lb_size); 515 516 udf_puteccline(eccline); 517 518 *dscrptr = (union dscrptr *) mem; 519 return 0; 520 } 521 522 523 static void 524 udf_free_nodedscr_rmw(struct udf_strat_args *args) 525 { 526 struct udf_mount *ump = args->ump; 527 struct long_ad *icb = args->icb; 528 struct udf_eccline *eccline; 529 uint64_t bit; 530 uint32_t sectornr, dummy; 531 int error, eccsect; 532 533 error = udf_translate_vtop(ump, icb, §ornr, &dummy); 534 if (error) 535 return; 536 537 /* get our eccline */ 538 eccline = udf_geteccline(ump, sectornr, 0); 539 eccsect = sectornr - eccline->start_sector; 540 541 bit = (uint64_t) 1 << eccsect; 542 eccline->readin &= ~bit; /* just in case */ 543 544 KASSERT(eccline->refcnt >= 1); 545 eccline->refcnt--; 546 547 udf_puteccline(eccline); 548 } 549 550 551 static int 552 udf_read_nodedscr_rmw(struct udf_strat_args *args) 553 { 554 union dscrptr **dscrptr = &args->dscr; 555 struct udf_mount *ump = args->ump; 556 struct long_ad *icb = args->icb; 557 struct udf_eccline *eccline; 558 uint64_t bit; 559 uint32_t sectornr, dummy; 560 uint8_t *pos; 561 int sector_size = ump->discinfo.sector_size; 562 int lb_size = udf_rw32(ump->logical_vol->lb_size); 563 int i, error, dscrlen, eccsect; 564 565 lb_size = lb_size; 566 KASSERT(sector_size == lb_size); 567 error = udf_translate_vtop(ump, icb, §ornr, &dummy); 568 if (error) 569 return error; 570 571 /* get our eccline */ 572 eccline = udf_geteccline(ump, sectornr, 0); 573 eccsect = sectornr - eccline->start_sector; 574 575 bit = (uint64_t) 1 << eccsect; 576 if ((eccline->present & bit) == 0) { 577 /* mark bit for readin */ 578 eccline->readin |= bit; 579 eccline->refcnt++; /* prevent recycling */ 580 KASSERT(eccline->bufs[eccsect] == NULL); 581 udf_puteccline(eccline); 582 583 /* wait for completion; XXX remodel to lock bit code */ 584 error = 0; 585 while ((eccline->present & bit) == 0) { 586 tsleep(eccline, PRIBIO+1, "udflvdrd", hz/8); 587 if (eccline->error & bit) { 588 KASSERT(eccline->refcnt >= 1); 589 eccline->refcnt--; /* undo temp refcnt */ 590 *dscrptr = NULL; 591 return EIO; /* XXX error code */ 592 } 593 } 594 595 /* reget our line */ 596 eccline = udf_geteccline(ump, sectornr, 0); 597 KASSERT(eccline->refcnt >= 1); 598 eccline->refcnt--; /* undo refcnt */ 599 } 600 601 *dscrptr = (union dscrptr *) 602 (((uint8_t *) eccline->blob) + eccsect * sector_size); 603 604 /* code from read_phys_descr */ 605 /* check if its a valid tag */ 606 error = udf_check_tag(*dscrptr); 607 if (error) { 608 /* check if its an empty block */ 609 pos = (uint8_t *) *dscrptr; 610 for (i = 0; i < sector_size; i++, pos++) { 611 if (*pos) break; 612 } 613 if (i == sector_size) { 614 /* return no error but with no dscrptr */ 615 error = 0; 616 } 617 *dscrptr = NULL; 618 udf_puteccline(eccline); 619 return error; 620 } 621 622 /* calculate descriptor size */ 623 dscrlen = udf_tagsize(*dscrptr, sector_size); 624 error = udf_check_tag_payload(*dscrptr, dscrlen); 625 if (error) { 626 *dscrptr = NULL; 627 udf_puteccline(eccline); 628 return error; 629 } 630 631 eccline->refcnt++; 632 udf_puteccline(eccline); 633 634 return 0; 635 } 636 637 638 static int 639 udf_write_nodedscr_rmw(struct udf_strat_args *args) 640 { 641 union dscrptr *dscrptr = args->dscr; 642 struct udf_mount *ump = args->ump; 643 struct long_ad *icb = args->icb; 644 struct udf_node *udf_node = args->udf_node; 645 struct udf_eccline *eccline; 646 uint64_t bit; 647 uint32_t sectornr, logsectornr, dummy; 648 // int waitfor = args->waitfor; 649 int sector_size = ump->discinfo.sector_size; 650 int lb_size = udf_rw32(ump->logical_vol->lb_size); 651 int error, eccsect; 652 653 lb_size = lb_size; 654 KASSERT(sector_size == lb_size); 655 sectornr = 0; 656 error = udf_translate_vtop(ump, icb, §ornr, &dummy); 657 if (error) 658 return error; 659 660 /* add reference to the vnode to prevent recycling */ 661 vhold(udf_node->vnode); 662 663 /* get our eccline */ 664 eccline = udf_geteccline(ump, sectornr, 0); 665 eccsect = sectornr - eccline->start_sector; 666 667 bit = (uint64_t) 1 << eccsect; 668 669 /* old callback still pending? */ 670 if (eccline->bufs[eccsect]) { 671 DPRINTF(WRITE, ("udf_write_nodedscr_rmw: writing descriptor" 672 " over buffer?\n")); 673 nestiobuf_done(eccline->bufs[eccsect], 674 eccline->bufs_len[eccsect], 675 0); 676 eccline->bufs[eccsect] = NULL; 677 } 678 679 /* set sector number in the descriptor and validate */ 680 dscrptr = (union dscrptr *) 681 (((uint8_t *) eccline->blob) + eccsect * sector_size); 682 KASSERT(dscrptr == args->dscr); 683 684 logsectornr = udf_rw32(icb->loc.lb_num); 685 dscrptr->tag.tag_loc = udf_rw32(logsectornr); 686 udf_validate_tag_and_crc_sums(dscrptr); 687 688 udf_fixup_node_internals(ump, (uint8_t *) dscrptr, UDF_C_NODE); 689 690 /* set our flags */ 691 KASSERT(eccline->present & bit); 692 eccline->dirty |= bit; 693 694 KASSERT(udf_tagsize(dscrptr, sector_size) <= sector_size); 695 696 udf_puteccline(eccline); 697 698 holdrele(udf_node->vnode); 699 udf_node->outstanding_nodedscr--; 700 if (udf_node->outstanding_nodedscr == 0) { 701 UDF_UNLOCK_NODE(udf_node, udf_node->i_flags & IN_CALLBACK_ULK); 702 wakeup(&udf_node->outstanding_nodedscr); 703 } 704 705 /* XXX waitfor not used */ 706 return 0; 707 } 708 709 710 static void 711 udf_queuebuf_rmw(struct udf_strat_args *args) 712 { 713 struct udf_mount *ump = args->ump; 714 struct buf *buf = args->nestbuf; 715 struct desc_tag *tag; 716 struct strat_private *priv = PRIV(ump); 717 struct udf_eccline *eccline; 718 struct long_ad *node_ad_cpy; 719 uint64_t bit, *lmapping, *pmapping, *lmappos, *pmappos, blknr; 720 uint32_t buf_len, len, sectors, sectornr, our_sectornr; 721 uint32_t bpos; 722 uint16_t vpart_num; 723 uint8_t *fidblk, *src, *dst; 724 int sector_size = ump->discinfo.sector_size; 725 int blks = sector_size / DEV_BSIZE; 726 int eccsect, what, queue, error; 727 728 KASSERT(ump); 729 KASSERT(buf); 730 KASSERT(buf->b_iodone == nestiobuf_iodone); 731 732 blknr = buf->b_blkno; 733 our_sectornr = blknr / blks; 734 735 what = buf->b_udf_c_type; 736 queue = UDF_SHED_READING; 737 if ((buf->b_flags & B_READ) == 0) { 738 /* writing */ 739 queue = UDF_SHED_SEQWRITING; 740 if (what == UDF_C_DSCR) 741 queue = UDF_SHED_WRITING; 742 if (what == UDF_C_NODE) 743 queue = UDF_SHED_WRITING; 744 } 745 746 if (queue == UDF_SHED_READING) { 747 DPRINTF(SHEDULE, ("\nudf_queuebuf_rmw READ %p : sector %d type %d," 748 "b_resid %d, b_bcount %d, b_bufsize %d\n", 749 buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type, 750 buf->b_resid, buf->b_bcount, buf->b_bufsize)); 751 752 /* mark bits for reading */ 753 buf_len = buf->b_bcount; 754 sectornr = our_sectornr; 755 eccline = udf_geteccline(ump, sectornr, 0); 756 eccsect = sectornr - eccline->start_sector; 757 bpos = 0; 758 while (buf_len) { 759 len = MIN(buf_len, sector_size); 760 if (eccsect == ump->packet_size) { 761 udf_puteccline(eccline); 762 eccline = udf_geteccline(ump, sectornr, 0); 763 eccsect = sectornr - eccline->start_sector; 764 } 765 bit = (uint64_t) 1 << eccsect; 766 error = eccline->error & bit ? EIO : 0; 767 if (eccline->present & bit) { 768 src = (uint8_t *) eccline->blob + 769 eccsect * sector_size; 770 dst = (uint8_t *) buf->b_data + bpos; 771 if (!error) 772 memcpy(dst, src, len); 773 nestiobuf_done(buf, len, error); 774 } else { 775 eccline->readin |= bit; 776 KASSERT(eccline->bufs[eccsect] == NULL); 777 eccline->bufs[eccsect] = buf; 778 eccline->bufs_bpos[eccsect] = bpos; 779 eccline->bufs_len[eccsect] = len; 780 } 781 bpos += sector_size; 782 eccsect++; 783 sectornr++; 784 buf_len -= len; 785 } 786 udf_puteccline(eccline); 787 return; 788 } 789 790 if (queue == UDF_SHED_WRITING) { 791 DPRINTF(SHEDULE, ("\nudf_queuebuf_rmw WRITE %p : sector %d " 792 "type %d, b_resid %d, b_bcount %d, b_bufsize %d\n", 793 buf, (uint32_t) buf->b_blkno / blks, buf->b_udf_c_type, 794 buf->b_resid, buf->b_bcount, buf->b_bufsize)); 795 /* if we have FIDs fixup using buffer's sector number(s) */ 796 if (buf->b_udf_c_type == UDF_C_FIDS) { 797 panic("UDF_C_FIDS in SHED_WRITING!\n"); 798 #if 0 799 buf_len = buf->b_bcount; 800 sectornr = our_sectornr; 801 bpos = 0; 802 while (buf_len) { 803 len = MIN(buf_len, sector_size); 804 fidblk = (uint8_t *) buf->b_data + bpos; 805 udf_fixup_fid_block(fidblk, sector_size, 806 0, len, sectornr); 807 sectornr++; 808 bpos += len; 809 buf_len -= len; 810 } 811 #endif 812 } 813 udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type); 814 815 /* copy parts into the bufs and set for writing */ 816 buf_len = buf->b_bcount; 817 sectornr = our_sectornr; 818 eccline = udf_geteccline(ump, sectornr, 0); 819 eccsect = sectornr - eccline->start_sector; 820 bpos = 0; 821 while (buf_len) { 822 len = MIN(buf_len, sector_size); 823 if (eccsect == ump->packet_size) { 824 udf_puteccline(eccline); 825 eccline = udf_geteccline(ump, sectornr, 0); 826 eccsect = sectornr - eccline->start_sector; 827 } 828 bit = (uint64_t) 1 << eccsect; 829 KASSERT((eccline->readin & bit) == 0); 830 eccline->present |= bit; 831 eccline->dirty |= bit; 832 if (eccline->bufs[eccsect]) { 833 /* old callback still pending */ 834 nestiobuf_done(eccline->bufs[eccsect], 835 eccline->bufs_len[eccsect], 836 0); 837 eccline->bufs[eccsect] = NULL; 838 } 839 840 src = (uint8_t *) buf->b_data + bpos; 841 dst = (uint8_t *) eccline->blob + eccsect * sector_size; 842 if (len != sector_size) 843 memset(dst, 0, sector_size); 844 memcpy(dst, src, len); 845 846 /* note that its finished for this extent */ 847 eccline->bufs[eccsect] = NULL; 848 nestiobuf_done(buf, len, 0); 849 850 bpos += sector_size; 851 eccsect++; 852 sectornr++; 853 buf_len -= len; 854 } 855 udf_puteccline(eccline); 856 return; 857 858 } 859 860 /* sequential writing */ 861 KASSERT(queue == UDF_SHED_SEQWRITING); 862 DPRINTF(SHEDULE, ("\nudf_queuebuf_rmw SEQWRITE %p : sector XXXX " 863 "type %d, b_resid %d, b_bcount %d, b_bufsize %d\n", 864 buf, buf->b_udf_c_type, buf->b_resid, buf->b_bcount, 865 buf->b_bufsize)); 866 /* 867 * Buffers should not have been allocated to disc addresses yet on 868 * this queue. Note that a buffer can get multiple extents allocated. 869 * Note that it *looks* like the normal writing but its different in 870 * the details. 871 * 872 * lmapping contains lb_num relative to base partition. 873 * 874 * XXX should we try to claim/organize the allocated memory to 875 * block-aligned pieces? 876 */ 877 mutex_enter(&priv->seqwrite_mutex); 878 879 lmapping = ump->la_lmapping; 880 node_ad_cpy = ump->la_node_ad_cpy; 881 882 /* logically allocate buf and map it in the file */ 883 udf_late_allocate_buf(ump, buf, lmapping, node_ad_cpy, &vpart_num); 884 885 /* if we have FIDs, fixup using the new allocation table */ 886 if (buf->b_udf_c_type == UDF_C_FIDS) { 887 buf_len = buf->b_bcount; 888 bpos = 0; 889 lmappos = lmapping; 890 while (buf_len) { 891 sectornr = *lmappos++; 892 len = MIN(buf_len, sector_size); 893 fidblk = (uint8_t *) buf->b_data + bpos; 894 udf_fixup_fid_block(fidblk, sector_size, 895 0, len, sectornr); 896 bpos += len; 897 buf_len -= len; 898 } 899 } 900 if (buf->b_udf_c_type == UDF_C_METADATA_SBM) { 901 if (buf->b_lblkno == 0) { 902 /* update the tag location inside */ 903 tag = (struct desc_tag *) buf->b_data; 904 tag->tag_loc = udf_rw32(*lmapping); 905 udf_validate_tag_and_crc_sums(buf->b_data); 906 } 907 } 908 udf_fixup_node_internals(ump, buf->b_data, buf->b_udf_c_type); 909 910 /* 911 * Translate new mappings in lmapping to pmappings. 912 * pmapping to contain lb_nums as used for disc adressing. 913 */ 914 pmapping = ump->la_pmapping; 915 sectors = (buf->b_bcount + sector_size -1) / sector_size; 916 udf_translate_vtop_list(ump, sectors, vpart_num, lmapping, pmapping); 917 918 /* copy parts into the bufs and set for writing */ 919 pmappos = pmapping; 920 buf_len = buf->b_bcount; 921 sectornr = *pmappos++; 922 eccline = udf_geteccline(ump, sectornr, ECC_SEQWRITING); 923 eccsect = sectornr - eccline->start_sector; 924 bpos = 0; 925 while (buf_len) { 926 len = MIN(buf_len, sector_size); 927 eccsect = sectornr - eccline->start_sector; 928 if ((eccsect < 0) || (eccsect >= ump->packet_size)) { 929 eccline->flags |= ECC_SEQWRITING; 930 udf_puteccline(eccline); 931 eccline = udf_geteccline(ump, sectornr, ECC_SEQWRITING); 932 eccsect = sectornr - eccline->start_sector; 933 } 934 bit = (uint64_t) 1 << eccsect; 935 KASSERT((eccline->readin & bit) == 0); 936 eccline->present |= bit; 937 eccline->dirty |= bit; 938 eccline->bufs[eccsect] = NULL; 939 940 src = (uint8_t *) buf->b_data + bpos; 941 dst = (uint8_t *) 942 eccline->blob + eccsect * sector_size; 943 if (len != sector_size) 944 memset(dst, 0, sector_size); 945 memcpy(dst, src, len); 946 947 /* note that its finished for this extent */ 948 nestiobuf_done(buf, len, 0); 949 950 bpos += sector_size; 951 sectornr = *pmappos++; 952 buf_len -= len; 953 } 954 eccline->flags |= ECC_SEQWRITING; 955 udf_puteccline(eccline); 956 mutex_exit(&priv->seqwrite_mutex); 957 } 958 959 /* --------------------------------------------------------------------- */ 960 961 static void 962 udf_shedule_read_callback(struct buf *buf) 963 { 964 struct udf_eccline *eccline = BTOE(buf); 965 struct udf_mount *ump = eccline->ump; 966 uint64_t bit; 967 uint8_t *src, *dst; 968 int sector_size = ump->discinfo.sector_size; 969 int error, i, len; 970 971 DPRINTF(ECCLINE, ("read callback called\n")); 972 /* post process read action */ 973 error = buf->b_error; 974 for (i = 0; i < ump->packet_size; i++) { 975 bit = (uint64_t) 1 << i; 976 src = (uint8_t *) buf->b_data + i * sector_size; 977 dst = (uint8_t *) eccline->blob + i * sector_size; 978 if (eccline->present & bit) 979 continue; 980 eccline->present |= bit; 981 if (error) 982 eccline->error |= bit; 983 if (eccline->bufs[i]) { 984 dst = (uint8_t *) eccline->bufs[i]->b_data + 985 eccline->bufs_bpos[i]; 986 len = eccline->bufs_len[i]; 987 if (!error) 988 memcpy(dst, src, len); 989 nestiobuf_done(eccline->bufs[i], len, error); 990 eccline->bufs[i] = NULL; 991 } 992 993 } 994 KASSERT(buf->b_data == eccline->blob); 995 KASSERT(eccline->present == ((uint64_t) 1 << ump->packet_size)-1); 996 997 /* 998 * XXX TODO what to do on read errors? read in all sectors 999 * synchronously and allocate a sparable entry? 1000 */ 1001 1002 udf_puteccline(eccline); 1003 DPRINTF(ECCLINE, ("read callback finished\n")); 1004 } 1005 1006 1007 static void 1008 udf_shedule_write_callback(struct buf *buf) 1009 { 1010 struct udf_eccline *eccline = BTOE(buf); 1011 struct udf_mount *ump = eccline->ump; 1012 uint64_t bit; 1013 int error, i, len; 1014 1015 DPRINTF(ECCLINE, ("write callback called\n")); 1016 /* post process write action */ 1017 error = buf->b_error; 1018 for (i = 0; i < ump->packet_size; i++) { 1019 bit = (uint64_t) 1 << i; 1020 if ((eccline->dirty & bit) == 0) 1021 continue; 1022 if (error) { 1023 eccline->error |= bit; 1024 } else { 1025 eccline->dirty &= ~bit; 1026 } 1027 if (eccline->bufs[i]) { 1028 len = eccline->bufs_len[i]; 1029 nestiobuf_done(eccline->bufs[i], len, error); 1030 eccline->bufs[i] = NULL; 1031 } 1032 } 1033 KASSERT(eccline->dirty == 0); 1034 1035 KASSERT(error == 0); 1036 /* 1037 * XXX TODO on write errors allocate a sparable entry and reissue 1038 */ 1039 1040 udf_puteccline(eccline); 1041 } 1042 1043 1044 static void 1045 udf_issue_eccline(struct udf_eccline *eccline, int queued_on) 1046 { 1047 struct udf_mount *ump = eccline->ump; 1048 struct strat_private *priv = PRIV(ump); 1049 struct buf *buf, *nestbuf; 1050 uint64_t bit, allbits = ((uint64_t) 1 << ump->packet_size)-1; 1051 uint32_t start; 1052 int sector_size = ump->discinfo.sector_size; 1053 int blks = sector_size / DEV_BSIZE; 1054 int i; 1055 1056 if (queued_on == UDF_SHED_READING) { 1057 DPRINTF(SHEDULE, ("udf_issue_eccline reading : ")); 1058 /* read all bits that are not yet present */ 1059 eccline->readin = (~eccline->present) & allbits; 1060 KASSERT(eccline->readin); 1061 start = eccline->start_sector; 1062 buf = eccline->buf; 1063 buf->b_flags = B_READ | B_ASYNC; 1064 SET(buf->b_cflags, BC_BUSY); /* mark buffer busy */ 1065 buf->b_oflags = 0; 1066 buf->b_iodone = udf_shedule_read_callback; 1067 buf->b_data = eccline->blob; 1068 buf->b_bcount = ump->packet_size * sector_size; 1069 buf->b_resid = buf->b_bcount; 1070 buf->b_bufsize = buf->b_bcount; 1071 buf->b_private = eccline; 1072 BIO_SETPRIO(buf, BPRIO_DEFAULT); 1073 buf->b_lblkno = buf->b_blkno = buf->b_rawblkno = start * blks; 1074 buf->b_proc = NULL; 1075 1076 if (eccline->present != 0) { 1077 for (i = 0; i < ump->packet_size; i++) { 1078 bit = (uint64_t) 1 << i; 1079 if (eccline->present & bit) { 1080 nestiobuf_done(buf, sector_size, 0); 1081 continue; 1082 } 1083 nestbuf = getiobuf(NULL, true); 1084 nestiobuf_setup(buf, nestbuf, i * sector_size, 1085 sector_size); 1086 /* adjust blocknumber to read */ 1087 nestbuf->b_blkno = buf->b_blkno + i*blks; 1088 nestbuf->b_rawblkno = buf->b_rawblkno + i*blks; 1089 1090 DPRINTF(SHEDULE, ("sector %d ", 1091 start + i)); 1092 /* call asynchronous */ 1093 VOP_STRATEGY(ump->devvp, nestbuf); 1094 } 1095 DPRINTF(SHEDULE, ("\n")); 1096 return; 1097 } 1098 } else { 1099 /* write or seqwrite */ 1100 DPRINTF(SHEDULE, ("udf_issue_eccline writing or seqwriting : ")); 1101 DPRINTF(SHEDULE, ("\n\tpresent %"PRIx64", readin %"PRIx64", " 1102 "dirty %"PRIx64"\n\t", eccline->present, eccline->readin, 1103 eccline->dirty)); 1104 if (eccline->present != allbits) { 1105 /* requeue to read-only */ 1106 DPRINTF(SHEDULE, ("\n\t-> not complete, requeue to " 1107 "reading\n")); 1108 udf_push_eccline(eccline, UDF_SHED_READING); 1109 return; 1110 } 1111 start = eccline->start_sector; 1112 buf = eccline->buf; 1113 buf->b_flags = B_WRITE | B_ASYNC; 1114 SET(buf->b_cflags, BC_BUSY); /* mark buffer busy */ 1115 buf->b_oflags = 0; 1116 buf->b_iodone = udf_shedule_write_callback; 1117 buf->b_data = eccline->blob; 1118 buf->b_bcount = ump->packet_size * sector_size; 1119 buf->b_resid = buf->b_bcount; 1120 buf->b_bufsize = buf->b_bcount; 1121 buf->b_private = eccline; 1122 BIO_SETPRIO(buf, BPRIO_DEFAULT); 1123 buf->b_lblkno = buf->b_blkno = buf->b_rawblkno = start * blks; 1124 buf->b_proc = NULL; 1125 } 1126 1127 mutex_exit(&priv->discstrat_mutex); 1128 /* call asynchronous */ 1129 DPRINTF(SHEDULE, ("sector %d for %d\n", 1130 start, ump->packet_size)); 1131 VOP_STRATEGY(ump->devvp, buf); 1132 mutex_enter(&priv->discstrat_mutex); 1133 } 1134 1135 1136 static void 1137 udf_discstrat_thread(void *arg) 1138 { 1139 struct udf_mount *ump = (struct udf_mount *) arg; 1140 struct strat_private *priv = PRIV(ump); 1141 struct udf_eccline *eccline; 1142 struct timespec now, *last; 1143 uint64_t allbits = ((uint64_t) 1 << ump->packet_size)-1; 1144 int new_queue, wait, work, num, cnt; 1145 1146 work = 1; 1147 priv->thread_running = 1; 1148 mutex_enter(&priv->discstrat_mutex); 1149 priv->num_floating = 0; 1150 while (priv->run_thread || work || priv->num_floating) { 1151 /* get our time */ 1152 vfs_timestamp(&now); 1153 1154 /* maintenance: handle eccline state machine */ 1155 num = priv->num_queued[UDF_SHED_WAITING]; 1156 cnt = 0; 1157 while (cnt < num) { 1158 eccline = udf_pop_eccline(priv, UDF_SHED_WAITING); 1159 /* requeue */ 1160 new_queue = UDF_SHED_FREE; 1161 if (eccline->refcnt > 0) 1162 new_queue = UDF_SHED_IDLE; 1163 if (eccline->flags & ECC_WANTED) 1164 new_queue = UDF_SHED_IDLE; 1165 if (eccline->readin) 1166 new_queue = UDF_SHED_READING; 1167 if (eccline->dirty) { 1168 new_queue = UDF_SHED_WAITING; 1169 if ((eccline->wait_time.tv_sec - now.tv_sec <= 0) || 1170 ((eccline->present == allbits) && 1171 (eccline->flags & ECC_SEQWRITING))) 1172 { 1173 new_queue = UDF_SHED_WRITING; 1174 if (eccline->flags & ECC_SEQWRITING) 1175 new_queue = UDF_SHED_SEQWRITING; 1176 if (eccline->present != allbits) 1177 new_queue = UDF_SHED_READING; 1178 } 1179 } 1180 udf_push_eccline(eccline, new_queue); 1181 cnt++; 1182 } 1183 1184 /* maintenance: free exess ecclines */ 1185 while (priv->num_queued[UDF_SHED_FREE] > UDF_ECCLINE_MAXFREE) { 1186 eccline = udf_pop_eccline(priv, UDF_SHED_FREE); 1187 KASSERT(eccline); 1188 KASSERT(eccline->refcnt == 0); 1189 if (eccline->flags & ECC_WANTED) { 1190 udf_push_eccline(eccline, UDF_SHED_IDLE); 1191 DPRINTF(ECCLINE, ("Tried removing, pushed back to free list\n")); 1192 } else { 1193 DPRINTF(ECCLINE, ("Removing entry from free list\n")); 1194 udf_dispose_eccline(eccline); 1195 } 1196 } 1197 1198 /* process the current selected queue */ 1199 /* get our time */ 1200 vfs_timestamp(&now); 1201 last = &priv->last_queued[priv->cur_queue]; 1202 1203 /* get our line */ 1204 eccline = udf_pop_eccline(priv, priv->cur_queue); 1205 if (eccline) { 1206 wait = 0; 1207 new_queue = priv->cur_queue; 1208 DPRINTF(ECCLINE, ("UDF_ISSUE_ECCLINE\n")); 1209 1210 /* complete the `get' by locking and refcounting it */ 1211 UDF_LOCK_ECCLINE(eccline); 1212 eccline->refcnt++; 1213 1214 udf_issue_eccline(eccline, priv->cur_queue); 1215 } else { 1216 /* don't switch too quickly */ 1217 if (now.tv_sec - last->tv_sec < 2) { 1218 /* wait some time */ 1219 cv_timedwait(&priv->discstrat_cv, 1220 &priv->discstrat_mutex, hz); 1221 /* we assume there is work to be done */ 1222 work = 1; 1223 continue; 1224 } 1225 1226 /* XXX select on queue lengths ? */ 1227 wait = 1; 1228 /* check if we can/should switch */ 1229 new_queue = priv->cur_queue; 1230 if (BUFQ_PEEK(priv->queues[UDF_SHED_READING])) 1231 new_queue = UDF_SHED_READING; 1232 if (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING])) 1233 new_queue = UDF_SHED_WRITING; 1234 if (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING])) 1235 new_queue = UDF_SHED_SEQWRITING; 1236 } 1237 1238 /* give room */ 1239 mutex_exit(&priv->discstrat_mutex); 1240 1241 if (new_queue != priv->cur_queue) { 1242 wait = 0; 1243 DPRINTF(SHEDULE, ("switching from %d to %d\n", 1244 priv->cur_queue, new_queue)); 1245 priv->cur_queue = new_queue; 1246 } 1247 mutex_enter(&priv->discstrat_mutex); 1248 1249 /* wait for more if needed */ 1250 if (wait) 1251 cv_timedwait(&priv->discstrat_cv, 1252 &priv->discstrat_mutex, hz/4); /* /8 */ 1253 1254 work = (BUFQ_PEEK(priv->queues[UDF_SHED_WAITING]) != NULL); 1255 work |= (BUFQ_PEEK(priv->queues[UDF_SHED_READING]) != NULL); 1256 work |= (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]) != NULL); 1257 work |= (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]) != NULL); 1258 1259 DPRINTF(PARANOIA, ("work : (%d, %d, %d) -> work %d, float %d\n", 1260 (BUFQ_PEEK(priv->queues[UDF_SHED_READING]) != NULL), 1261 (BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]) != NULL), 1262 (BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]) != NULL), 1263 work, priv->num_floating)); 1264 } 1265 1266 mutex_exit(&priv->discstrat_mutex); 1267 1268 /* tear down remaining ecclines */ 1269 mutex_enter(&priv->discstrat_mutex); 1270 KASSERT(priv->num_queued[UDF_SHED_WAITING] == 0); 1271 KASSERT(priv->num_queued[UDF_SHED_IDLE] == 0); 1272 KASSERT(priv->num_queued[UDF_SHED_READING] == 0); 1273 KASSERT(priv->num_queued[UDF_SHED_WRITING] == 0); 1274 KASSERT(priv->num_queued[UDF_SHED_SEQWRITING] == 0); 1275 1276 KASSERT(BUFQ_PEEK(priv->queues[UDF_SHED_WAITING]) == NULL); 1277 KASSERT(BUFQ_PEEK(priv->queues[UDF_SHED_IDLE]) == NULL); 1278 KASSERT(BUFQ_PEEK(priv->queues[UDF_SHED_READING]) == NULL); 1279 KASSERT(BUFQ_PEEK(priv->queues[UDF_SHED_WRITING]) == NULL); 1280 KASSERT(BUFQ_PEEK(priv->queues[UDF_SHED_SEQWRITING]) == NULL); 1281 eccline = udf_pop_eccline(priv, UDF_SHED_FREE); 1282 while (eccline) { 1283 udf_dispose_eccline(eccline); 1284 eccline = udf_pop_eccline(priv, UDF_SHED_FREE); 1285 } 1286 KASSERT(priv->num_queued[UDF_SHED_FREE] == 0); 1287 mutex_exit(&priv->discstrat_mutex); 1288 1289 priv->thread_running = 0; 1290 priv->thread_finished = 1; 1291 wakeup(&priv->run_thread); 1292 kthread_exit(0); 1293 /* not reached */ 1294 } 1295 1296 /* --------------------------------------------------------------------- */ 1297 1298 /* 1299 * Buffer memory pool allocator. 1300 */ 1301 1302 static void * 1303 ecclinepool_page_alloc(struct pool *pp, int flags) 1304 { 1305 return (void *)uvm_km_alloc(kernel_map, 1306 MAXBSIZE, MAXBSIZE, 1307 ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) 1308 | UVM_KMF_WIRED /* UVM_KMF_PAGABLE? */); 1309 } 1310 1311 static void 1312 ecclinepool_page_free(struct pool *pp, void *v) 1313 { 1314 uvm_km_free(kernel_map, (vaddr_t)v, MAXBSIZE, UVM_KMF_WIRED); 1315 } 1316 1317 static struct pool_allocator ecclinepool_allocator = { 1318 .pa_alloc = ecclinepool_page_alloc, 1319 .pa_free = ecclinepool_page_free, 1320 .pa_pagesz = MAXBSIZE, 1321 }; 1322 1323 1324 static void 1325 udf_discstrat_init_rmw(struct udf_strat_args *args) 1326 { 1327 struct udf_mount *ump = args->ump; 1328 struct strat_private *priv = PRIV(ump); 1329 uint32_t lb_size, blobsize, hashline; 1330 int i; 1331 1332 KASSERT(ump); 1333 KASSERT(ump->logical_vol); 1334 KASSERT(priv == NULL); 1335 1336 lb_size = udf_rw32(ump->logical_vol->lb_size); 1337 blobsize = ump->packet_size * lb_size; 1338 KASSERT(lb_size > 0); 1339 KASSERT(ump->packet_size <= 64); 1340 1341 /* initialise our memory space */ 1342 ump->strategy_private = malloc(sizeof(struct strat_private), 1343 M_UDFTEMP, M_WAITOK); 1344 priv = ump->strategy_private; 1345 memset(priv, 0 , sizeof(struct strat_private)); 1346 1347 /* initialise locks */ 1348 cv_init(&priv->discstrat_cv, "udfstrat"); 1349 mutex_init(&priv->discstrat_mutex, MUTEX_DEFAULT, IPL_NONE); 1350 mutex_init(&priv->seqwrite_mutex, MUTEX_DEFAULT, IPL_NONE); 1351 1352 /* initialise struct eccline pool */ 1353 pool_init(&priv->eccline_pool, sizeof(struct udf_eccline), 1354 0, 0, 0, "udf_eccline_pool", NULL, IPL_NONE); 1355 1356 /* initialise eccline blob pool */ 1357 ecclinepool_allocator.pa_pagesz = blobsize; 1358 pool_init(&priv->ecclineblob_pool, blobsize, 1359 0, 0, 0, "udf_eccline_blob", &ecclinepool_allocator, IPL_NONE); 1360 1361 /* initialise main queues */ 1362 for (i = 0; i < UDF_SHED_MAX; i++) { 1363 priv->num_queued[i] = 0; 1364 vfs_timestamp(&priv->last_queued[i]); 1365 } 1366 bufq_alloc(&priv->queues[UDF_SHED_WAITING], "fcfs", 1367 BUFQ_SORT_RAWBLOCK); 1368 bufq_alloc(&priv->queues[UDF_SHED_READING], "disksort", 1369 BUFQ_SORT_RAWBLOCK); 1370 bufq_alloc(&priv->queues[UDF_SHED_WRITING], "disksort", 1371 BUFQ_SORT_RAWBLOCK); 1372 bufq_alloc(&priv->queues[UDF_SHED_SEQWRITING], "disksort", 0); 1373 1374 /* initialise administrative queues */ 1375 bufq_alloc(&priv->queues[UDF_SHED_IDLE], "fcfs", 0); 1376 bufq_alloc(&priv->queues[UDF_SHED_FREE], "fcfs", 0); 1377 1378 for (hashline = 0; hashline < UDF_ECCBUF_HASHSIZE; hashline++) { 1379 LIST_INIT(&priv->eccline_hash[hashline]); 1380 } 1381 1382 /* create our disk strategy thread */ 1383 priv->cur_queue = UDF_SHED_READING; 1384 priv->thread_finished = 0; 1385 priv->thread_running = 0; 1386 priv->run_thread = 1; 1387 if (kthread_create(PRI_NONE, 0 /* KTHREAD_MPSAFE*/, NULL /* cpu_info*/, 1388 udf_discstrat_thread, ump, &priv->queue_lwp, 1389 "%s", "udf_rw")) { 1390 panic("fork udf_rw"); 1391 } 1392 1393 /* wait for thread to spin up */ 1394 while (!priv->thread_running) { 1395 tsleep(&priv->thread_running, PRIBIO+1, "udfshedstart", hz); 1396 } 1397 } 1398 1399 1400 static void 1401 udf_discstrat_finish_rmw(struct udf_strat_args *args) 1402 { 1403 struct udf_mount *ump = args->ump; 1404 struct strat_private *priv = PRIV(ump); 1405 int error; 1406 1407 if (ump == NULL) 1408 return; 1409 1410 /* stop our sheduling thread */ 1411 KASSERT(priv->run_thread == 1); 1412 priv->run_thread = 0; 1413 wakeup(priv->queue_lwp); 1414 while (!priv->thread_finished) { 1415 error = tsleep(&priv->run_thread, PRIBIO+1, 1416 "udfshedfin", hz); 1417 } 1418 /* kthread should be finished now */ 1419 1420 /* cleanup our pools */ 1421 pool_destroy(&priv->eccline_pool); 1422 pool_destroy(&priv->ecclineblob_pool); 1423 1424 cv_destroy(&priv->discstrat_cv); 1425 mutex_destroy(&priv->discstrat_mutex); 1426 mutex_destroy(&priv->seqwrite_mutex); 1427 1428 /* free our private space */ 1429 free(ump->strategy_private, M_UDFTEMP); 1430 ump->strategy_private = NULL; 1431 } 1432 1433 /* --------------------------------------------------------------------- */ 1434 1435 struct udf_strategy udf_strat_rmw = 1436 { 1437 udf_create_nodedscr_rmw, 1438 udf_free_nodedscr_rmw, 1439 udf_read_nodedscr_rmw, 1440 udf_write_nodedscr_rmw, 1441 udf_queuebuf_rmw, 1442 udf_discstrat_init_rmw, 1443 udf_discstrat_finish_rmw 1444 }; 1445 1446