1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2013 Peter Grehan <grehan@freebsd.org> 5 * All rights reserved. 6 * Copyright 2020 Joyent, Inc. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $FreeBSD$ 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include <sys/param.h> 36 #ifndef WITHOUT_CAPSICUM 37 #include <sys/capsicum.h> 38 #endif 39 #include <sys/queue.h> 40 #include <sys/errno.h> 41 #include <sys/stat.h> 42 #include <sys/ioctl.h> 43 #include <sys/disk.h> 44 45 #include <assert.h> 46 #ifndef WITHOUT_CAPSICUM 47 #include <capsicum_helpers.h> 48 #endif 49 #include <err.h> 50 #include <fcntl.h> 51 #include <stdio.h> 52 #include <stdlib.h> 53 #include <string.h> 54 #include <pthread.h> 55 #include <pthread_np.h> 56 #include <signal.h> 57 #include <sysexits.h> 58 #include <unistd.h> 59 60 #include <machine/atomic.h> 61 #include <machine/vmm_snapshot.h> 62 63 #include "bhyverun.h" 64 #include "config.h" 65 #include "debug.h" 66 #include "mevent.h" 67 #include "pci_emul.h" 68 #include "block_if.h" 69 70 #define BLOCKIF_SIG 0xb109b109 71 72 #define BLOCKIF_NUMTHR 8 73 #define BLOCKIF_MAXREQ (BLOCKIF_RING_MAX + BLOCKIF_NUMTHR) 74 75 enum blockop { 76 BOP_READ, 77 BOP_WRITE, 78 BOP_FLUSH, 79 BOP_DELETE 80 }; 81 82 enum blockstat { 83 BST_FREE, 84 BST_BLOCK, 85 BST_PEND, 86 BST_BUSY, 87 BST_DONE 88 }; 89 90 struct blockif_elem { 91 TAILQ_ENTRY(blockif_elem) be_link; 92 struct blockif_req *be_req; 93 enum blockop be_op; 94 enum blockstat be_status; 95 pthread_t be_tid; 96 off_t be_block; 97 }; 98 99 struct blockif_ctxt { 100 unsigned int bc_magic; 101 int bc_fd; 102 int bc_ischr; 103 int bc_isgeom; 104 int bc_candelete; 105 int bc_rdonly; 106 off_t bc_size; 107 int bc_sectsz; 108 int bc_psectsz; 109 int bc_psectoff; 110 int bc_closing; 111 int bc_paused; 112 pthread_t bc_btid[BLOCKIF_NUMTHR]; 113 pthread_mutex_t bc_mtx; 114 pthread_cond_t bc_cond; 115 pthread_cond_t bc_work_done_cond; 116 blockif_resize_cb *bc_resize_cb; 117 void *bc_resize_cb_arg; 118 struct mevent *bc_resize_event; 119 120 /* Request elements and free/pending/busy queues */ 121 TAILQ_HEAD(, blockif_elem) bc_freeq; 122 TAILQ_HEAD(, blockif_elem) bc_pendq; 123 TAILQ_HEAD(, blockif_elem) bc_busyq; 124 struct blockif_elem bc_reqs[BLOCKIF_MAXREQ]; 125 }; 126 127 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT; 128 129 struct blockif_sig_elem { 130 pthread_mutex_t bse_mtx; 131 pthread_cond_t bse_cond; 132 int bse_pending; 133 struct blockif_sig_elem *bse_next; 134 }; 135 136 static struct blockif_sig_elem *blockif_bse_head; 137 138 static int 139 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq, 140 enum blockop op) 141 { 142 struct blockif_elem *be, *tbe; 143 off_t off; 144 int i; 145 146 be = TAILQ_FIRST(&bc->bc_freeq); 147 assert(be != NULL); 148 assert(be->be_status == BST_FREE); 149 TAILQ_REMOVE(&bc->bc_freeq, be, be_link); 150 be->be_req = breq; 151 be->be_op = op; 152 switch (op) { 153 case BOP_READ: 154 case BOP_WRITE: 155 case BOP_DELETE: 156 off = breq->br_offset; 157 for (i = 0; i < breq->br_iovcnt; i++) 158 off += breq->br_iov[i].iov_len; 159 break; 160 default: 161 off = OFF_MAX; 162 } 163 be->be_block = off; 164 TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) { 165 if (tbe->be_block == breq->br_offset) 166 break; 167 } 168 if (tbe == NULL) { 169 TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) { 170 if (tbe->be_block == breq->br_offset) 171 break; 172 } 173 } 174 if (tbe == NULL) 175 be->be_status = BST_PEND; 176 else 177 be->be_status = BST_BLOCK; 178 TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link); 179 return (be->be_status == BST_PEND); 180 } 181 182 static int 183 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep) 184 { 185 struct blockif_elem *be; 186 187 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) { 188 if (be->be_status == BST_PEND) 189 break; 190 assert(be->be_status == BST_BLOCK); 191 } 192 if (be == NULL) 193 return (0); 194 TAILQ_REMOVE(&bc->bc_pendq, be, be_link); 195 be->be_status = BST_BUSY; 196 be->be_tid = t; 197 TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link); 198 *bep = be; 199 return (1); 200 } 201 202 static void 203 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be) 204 { 205 struct blockif_elem *tbe; 206 207 if (be->be_status == BST_DONE || be->be_status == BST_BUSY) 208 TAILQ_REMOVE(&bc->bc_busyq, be, be_link); 209 else 210 TAILQ_REMOVE(&bc->bc_pendq, be, be_link); 211 TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) { 212 if (tbe->be_req->br_offset == be->be_block) 213 tbe->be_status = BST_PEND; 214 } 215 be->be_tid = 0; 216 be->be_status = BST_FREE; 217 be->be_req = NULL; 218 TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link); 219 } 220 221 static int 222 blockif_flush_bc(struct blockif_ctxt *bc) 223 { 224 if (bc->bc_ischr) { 225 if (ioctl(bc->bc_fd, DIOCGFLUSH)) 226 return (errno); 227 } else if (fsync(bc->bc_fd)) 228 return (errno); 229 230 return (0); 231 } 232 233 static void 234 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf) 235 { 236 struct blockif_req *br; 237 off_t arg[2]; 238 ssize_t clen, len, off, boff, voff; 239 int i, err; 240 struct spacectl_range range; 241 242 br = be->be_req; 243 if (br->br_iovcnt <= 1) 244 buf = NULL; 245 err = 0; 246 switch (be->be_op) { 247 case BOP_READ: 248 if (buf == NULL) { 249 if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt, 250 br->br_offset)) < 0) 251 err = errno; 252 else 253 br->br_resid -= len; 254 break; 255 } 256 i = 0; 257 off = voff = 0; 258 while (br->br_resid > 0) { 259 len = MIN(br->br_resid, MAXPHYS); 260 if (pread(bc->bc_fd, buf, len, br->br_offset + 261 off) < 0) { 262 err = errno; 263 break; 264 } 265 boff = 0; 266 do { 267 clen = MIN(len - boff, br->br_iov[i].iov_len - 268 voff); 269 memcpy(br->br_iov[i].iov_base + voff, 270 buf + boff, clen); 271 if (clen < br->br_iov[i].iov_len - voff) 272 voff += clen; 273 else { 274 i++; 275 voff = 0; 276 } 277 boff += clen; 278 } while (boff < len); 279 off += len; 280 br->br_resid -= len; 281 } 282 break; 283 case BOP_WRITE: 284 if (bc->bc_rdonly) { 285 err = EROFS; 286 break; 287 } 288 if (buf == NULL) { 289 if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt, 290 br->br_offset)) < 0) 291 err = errno; 292 else 293 br->br_resid -= len; 294 break; 295 } 296 i = 0; 297 off = voff = 0; 298 while (br->br_resid > 0) { 299 len = MIN(br->br_resid, MAXPHYS); 300 boff = 0; 301 do { 302 clen = MIN(len - boff, br->br_iov[i].iov_len - 303 voff); 304 memcpy(buf + boff, 305 br->br_iov[i].iov_base + voff, clen); 306 if (clen < br->br_iov[i].iov_len - voff) 307 voff += clen; 308 else { 309 i++; 310 voff = 0; 311 } 312 boff += clen; 313 } while (boff < len); 314 if (pwrite(bc->bc_fd, buf, len, br->br_offset + 315 off) < 0) { 316 err = errno; 317 break; 318 } 319 off += len; 320 br->br_resid -= len; 321 } 322 break; 323 case BOP_FLUSH: 324 err = blockif_flush_bc(bc); 325 break; 326 case BOP_DELETE: 327 if (!bc->bc_candelete) 328 err = EOPNOTSUPP; 329 else if (bc->bc_rdonly) 330 err = EROFS; 331 else if (bc->bc_ischr) { 332 arg[0] = br->br_offset; 333 arg[1] = br->br_resid; 334 if (ioctl(bc->bc_fd, DIOCGDELETE, arg)) 335 err = errno; 336 else 337 br->br_resid = 0; 338 } else { 339 range.r_offset = br->br_offset; 340 range.r_len = br->br_resid; 341 342 while (range.r_len > 0) { 343 if (fspacectl(bc->bc_fd, SPACECTL_DEALLOC, 344 &range, 0, &range) != 0) { 345 err = errno; 346 break; 347 } 348 } 349 if (err == 0) 350 br->br_resid = 0; 351 } 352 break; 353 default: 354 err = EINVAL; 355 break; 356 } 357 358 be->be_status = BST_DONE; 359 360 (*br->br_callback)(br, err); 361 } 362 363 static inline bool 364 blockif_empty(const struct blockif_ctxt *bc) 365 { 366 return (TAILQ_EMPTY(&bc->bc_pendq) && TAILQ_EMPTY(&bc->bc_busyq)); 367 } 368 369 static void * 370 blockif_thr(void *arg) 371 { 372 struct blockif_ctxt *bc; 373 struct blockif_elem *be; 374 pthread_t t; 375 uint8_t *buf; 376 377 bc = arg; 378 if (bc->bc_isgeom) 379 buf = malloc(MAXPHYS); 380 else 381 buf = NULL; 382 t = pthread_self(); 383 384 pthread_mutex_lock(&bc->bc_mtx); 385 for (;;) { 386 while (blockif_dequeue(bc, t, &be)) { 387 pthread_mutex_unlock(&bc->bc_mtx); 388 blockif_proc(bc, be, buf); 389 pthread_mutex_lock(&bc->bc_mtx); 390 blockif_complete(bc, be); 391 } 392 393 /* If none to work, notify the main thread */ 394 if (blockif_empty(bc)) 395 pthread_cond_broadcast(&bc->bc_work_done_cond); 396 397 /* Check ctxt status here to see if exit requested */ 398 if (bc->bc_closing) 399 break; 400 401 pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx); 402 } 403 pthread_mutex_unlock(&bc->bc_mtx); 404 405 if (buf) 406 free(buf); 407 pthread_exit(NULL); 408 return (NULL); 409 } 410 411 static void 412 blockif_sigcont_handler(int signal __unused, enum ev_type type __unused, 413 void *arg __unused) 414 { 415 struct blockif_sig_elem *bse; 416 417 for (;;) { 418 /* 419 * Process the entire list even if not intended for 420 * this thread. 421 */ 422 do { 423 bse = blockif_bse_head; 424 if (bse == NULL) 425 return; 426 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head, 427 (uintptr_t)bse, 428 (uintptr_t)bse->bse_next)); 429 430 pthread_mutex_lock(&bse->bse_mtx); 431 bse->bse_pending = 0; 432 pthread_cond_signal(&bse->bse_cond); 433 pthread_mutex_unlock(&bse->bse_mtx); 434 } 435 } 436 437 static void 438 blockif_init(void) 439 { 440 mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL); 441 (void) signal(SIGCONT, SIG_IGN); 442 } 443 444 int 445 blockif_legacy_config(nvlist_t *nvl, const char *opts) 446 { 447 char *cp, *path; 448 449 if (opts == NULL) 450 return (0); 451 452 cp = strchr(opts, ','); 453 if (cp == NULL) { 454 set_config_value_node(nvl, "path", opts); 455 return (0); 456 } 457 path = strndup(opts, cp - opts); 458 set_config_value_node(nvl, "path", path); 459 free(path); 460 return (pci_parse_legacy_config(nvl, cp + 1)); 461 } 462 463 struct blockif_ctxt * 464 blockif_open(nvlist_t *nvl, const char *ident) 465 { 466 char tname[MAXCOMLEN + 1]; 467 char name[MAXPATHLEN]; 468 const char *path, *pssval, *ssval; 469 char *cp; 470 struct blockif_ctxt *bc; 471 struct stat sbuf; 472 struct diocgattr_arg arg; 473 off_t size, psectsz, psectoff; 474 int extra, fd, i, sectsz; 475 int ro, candelete, geom, ssopt, pssopt; 476 int nodelete; 477 478 #ifndef WITHOUT_CAPSICUM 479 cap_rights_t rights; 480 cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE, DIOCGMEDIASIZE }; 481 #endif 482 483 pthread_once(&blockif_once, blockif_init); 484 485 fd = -1; 486 extra = 0; 487 ssopt = 0; 488 ro = 0; 489 nodelete = 0; 490 491 if (get_config_bool_node_default(nvl, "nocache", false)) 492 extra |= O_DIRECT; 493 if (get_config_bool_node_default(nvl, "nodelete", false)) 494 nodelete = 1; 495 if (get_config_bool_node_default(nvl, "sync", false) || 496 get_config_bool_node_default(nvl, "direct", false)) 497 extra |= O_SYNC; 498 if (get_config_bool_node_default(nvl, "ro", false)) 499 ro = 1; 500 ssval = get_config_value_node(nvl, "sectorsize"); 501 if (ssval != NULL) { 502 ssopt = strtol(ssval, &cp, 10); 503 if (cp == ssval) { 504 EPRINTLN("Invalid sector size \"%s\"", ssval); 505 goto err; 506 } 507 if (*cp == '\0') { 508 pssopt = ssopt; 509 } else if (*cp == '/') { 510 pssval = cp + 1; 511 pssopt = strtol(pssval, &cp, 10); 512 if (cp == pssval || *cp != '\0') { 513 EPRINTLN("Invalid sector size \"%s\"", ssval); 514 goto err; 515 } 516 } else { 517 EPRINTLN("Invalid sector size \"%s\"", ssval); 518 goto err; 519 } 520 } 521 522 path = get_config_value_node(nvl, "path"); 523 if (path == NULL) { 524 EPRINTLN("Missing \"path\" for block device."); 525 goto err; 526 } 527 528 fd = open(path, (ro ? O_RDONLY : O_RDWR) | extra); 529 if (fd < 0 && !ro) { 530 /* Attempt a r/w fail with a r/o open */ 531 fd = open(path, O_RDONLY | extra); 532 ro = 1; 533 } 534 535 if (fd < 0) { 536 warn("Could not open backing file: %s", path); 537 goto err; 538 } 539 540 if (fstat(fd, &sbuf) < 0) { 541 warn("Could not stat backing file %s", path); 542 goto err; 543 } 544 545 #ifndef WITHOUT_CAPSICUM 546 cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK, 547 CAP_WRITE, CAP_FSTAT, CAP_EVENT, CAP_FPATHCONF); 548 if (ro) 549 cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE); 550 551 if (caph_rights_limit(fd, &rights) == -1) 552 errx(EX_OSERR, "Unable to apply rights for sandbox"); 553 #endif 554 555 /* 556 * Deal with raw devices 557 */ 558 size = sbuf.st_size; 559 sectsz = DEV_BSIZE; 560 psectsz = psectoff = 0; 561 candelete = geom = 0; 562 if (S_ISCHR(sbuf.st_mode)) { 563 if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 || 564 ioctl(fd, DIOCGSECTORSIZE, §sz)) { 565 perror("Could not fetch dev blk/sector size"); 566 goto err; 567 } 568 assert(size != 0); 569 assert(sectsz != 0); 570 if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0) 571 ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff); 572 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 573 arg.len = sizeof(arg.value.i); 574 if (nodelete == 0 && ioctl(fd, DIOCGATTR, &arg) == 0) 575 candelete = arg.value.i; 576 if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0) 577 geom = 1; 578 } else { 579 psectsz = sbuf.st_blksize; 580 /* Avoid fallback implementation */ 581 candelete = fpathconf(fd, _PC_DEALLOC_PRESENT) == 1; 582 } 583 584 #ifndef WITHOUT_CAPSICUM 585 if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1) 586 errx(EX_OSERR, "Unable to apply rights for sandbox"); 587 #endif 588 589 if (ssopt != 0) { 590 if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 || 591 ssopt > pssopt) { 592 EPRINTLN("Invalid sector size %d/%d", 593 ssopt, pssopt); 594 goto err; 595 } 596 597 /* 598 * Some backend drivers (e.g. cd0, ada0) require that the I/O 599 * size be a multiple of the device's sector size. 600 * 601 * Validate that the emulated sector size complies with this 602 * requirement. 603 */ 604 if (S_ISCHR(sbuf.st_mode)) { 605 if (ssopt < sectsz || (ssopt % sectsz) != 0) { 606 EPRINTLN("Sector size %d incompatible " 607 "with underlying device sector size %d", 608 ssopt, sectsz); 609 goto err; 610 } 611 } 612 613 sectsz = ssopt; 614 psectsz = pssopt; 615 psectoff = 0; 616 } 617 618 bc = calloc(1, sizeof(struct blockif_ctxt)); 619 if (bc == NULL) { 620 perror("calloc"); 621 goto err; 622 } 623 624 bc->bc_magic = BLOCKIF_SIG; 625 bc->bc_fd = fd; 626 bc->bc_ischr = S_ISCHR(sbuf.st_mode); 627 bc->bc_isgeom = geom; 628 bc->bc_candelete = candelete; 629 bc->bc_rdonly = ro; 630 bc->bc_size = size; 631 bc->bc_sectsz = sectsz; 632 bc->bc_psectsz = psectsz; 633 bc->bc_psectoff = psectoff; 634 pthread_mutex_init(&bc->bc_mtx, NULL); 635 pthread_cond_init(&bc->bc_cond, NULL); 636 bc->bc_paused = 0; 637 pthread_cond_init(&bc->bc_work_done_cond, NULL); 638 TAILQ_INIT(&bc->bc_freeq); 639 TAILQ_INIT(&bc->bc_pendq); 640 TAILQ_INIT(&bc->bc_busyq); 641 for (i = 0; i < BLOCKIF_MAXREQ; i++) { 642 bc->bc_reqs[i].be_status = BST_FREE; 643 TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link); 644 } 645 646 for (i = 0; i < BLOCKIF_NUMTHR; i++) { 647 pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc); 648 snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i); 649 pthread_set_name_np(bc->bc_btid[i], tname); 650 } 651 652 return (bc); 653 err: 654 if (fd >= 0) 655 close(fd); 656 return (NULL); 657 } 658 659 static void 660 blockif_resized(int fd, enum ev_type type __unused, void *arg) 661 { 662 struct blockif_ctxt *bc; 663 struct stat sb; 664 off_t mediasize; 665 666 if (fstat(fd, &sb) != 0) 667 return; 668 669 if (S_ISCHR(sb.st_mode)) { 670 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) < 0) { 671 EPRINTLN("blockif_resized: get mediasize failed: %s", 672 strerror(errno)); 673 return; 674 } 675 } else 676 mediasize = sb.st_size; 677 678 bc = arg; 679 pthread_mutex_lock(&bc->bc_mtx); 680 if (mediasize != bc->bc_size) { 681 bc->bc_size = mediasize; 682 bc->bc_resize_cb(bc, bc->bc_resize_cb_arg, bc->bc_size); 683 } 684 pthread_mutex_unlock(&bc->bc_mtx); 685 } 686 687 int 688 blockif_register_resize_callback(struct blockif_ctxt *bc, blockif_resize_cb *cb, 689 void *cb_arg) 690 { 691 struct stat sb; 692 int err; 693 694 if (cb == NULL) 695 return (EINVAL); 696 697 err = 0; 698 699 pthread_mutex_lock(&bc->bc_mtx); 700 if (bc->bc_resize_cb != NULL) { 701 err = EBUSY; 702 goto out; 703 } 704 705 assert(bc->bc_closing == 0); 706 707 if (fstat(bc->bc_fd, &sb) != 0) { 708 err = errno; 709 goto out; 710 } 711 712 bc->bc_resize_event = mevent_add_flags(bc->bc_fd, EVF_VNODE, 713 EVFF_ATTRIB, blockif_resized, bc); 714 if (bc->bc_resize_event == NULL) { 715 err = ENXIO; 716 goto out; 717 } 718 719 bc->bc_resize_cb = cb; 720 bc->bc_resize_cb_arg = cb_arg; 721 out: 722 pthread_mutex_unlock(&bc->bc_mtx); 723 724 return (err); 725 } 726 727 static int 728 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq, 729 enum blockop op) 730 { 731 int err; 732 733 err = 0; 734 735 pthread_mutex_lock(&bc->bc_mtx); 736 assert(!bc->bc_paused); 737 if (!TAILQ_EMPTY(&bc->bc_freeq)) { 738 /* 739 * Enqueue and inform the block i/o thread 740 * that there is work available 741 */ 742 if (blockif_enqueue(bc, breq, op)) 743 pthread_cond_signal(&bc->bc_cond); 744 } else { 745 /* 746 * Callers are not allowed to enqueue more than 747 * the specified blockif queue limit. Return an 748 * error to indicate that the queue length has been 749 * exceeded. 750 */ 751 err = E2BIG; 752 } 753 pthread_mutex_unlock(&bc->bc_mtx); 754 755 return (err); 756 } 757 758 int 759 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq) 760 { 761 assert(bc->bc_magic == BLOCKIF_SIG); 762 return (blockif_request(bc, breq, BOP_READ)); 763 } 764 765 int 766 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq) 767 { 768 assert(bc->bc_magic == BLOCKIF_SIG); 769 return (blockif_request(bc, breq, BOP_WRITE)); 770 } 771 772 int 773 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq) 774 { 775 assert(bc->bc_magic == BLOCKIF_SIG); 776 return (blockif_request(bc, breq, BOP_FLUSH)); 777 } 778 779 int 780 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq) 781 { 782 assert(bc->bc_magic == BLOCKIF_SIG); 783 return (blockif_request(bc, breq, BOP_DELETE)); 784 } 785 786 int 787 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq) 788 { 789 struct blockif_elem *be; 790 791 assert(bc->bc_magic == BLOCKIF_SIG); 792 793 pthread_mutex_lock(&bc->bc_mtx); 794 /* XXX: not waiting while paused */ 795 796 /* 797 * Check pending requests. 798 */ 799 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) { 800 if (be->be_req == breq) 801 break; 802 } 803 if (be != NULL) { 804 /* 805 * Found it. 806 */ 807 blockif_complete(bc, be); 808 pthread_mutex_unlock(&bc->bc_mtx); 809 810 return (0); 811 } 812 813 /* 814 * Check in-flight requests. 815 */ 816 TAILQ_FOREACH(be, &bc->bc_busyq, be_link) { 817 if (be->be_req == breq) 818 break; 819 } 820 if (be == NULL) { 821 /* 822 * Didn't find it. 823 */ 824 pthread_mutex_unlock(&bc->bc_mtx); 825 return (EINVAL); 826 } 827 828 /* 829 * Interrupt the processing thread to force it return 830 * prematurely via it's normal callback path. 831 */ 832 while (be->be_status == BST_BUSY) { 833 struct blockif_sig_elem bse, *old_head; 834 835 pthread_mutex_init(&bse.bse_mtx, NULL); 836 pthread_cond_init(&bse.bse_cond, NULL); 837 838 bse.bse_pending = 1; 839 840 do { 841 old_head = blockif_bse_head; 842 bse.bse_next = old_head; 843 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head, 844 (uintptr_t)old_head, 845 (uintptr_t)&bse)); 846 847 pthread_kill(be->be_tid, SIGCONT); 848 849 pthread_mutex_lock(&bse.bse_mtx); 850 while (bse.bse_pending) 851 pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx); 852 pthread_mutex_unlock(&bse.bse_mtx); 853 } 854 855 pthread_mutex_unlock(&bc->bc_mtx); 856 857 /* 858 * The processing thread has been interrupted. Since it's not 859 * clear if the callback has been invoked yet, return EBUSY. 860 */ 861 return (EBUSY); 862 } 863 864 int 865 blockif_close(struct blockif_ctxt *bc) 866 { 867 void *jval; 868 int i; 869 870 assert(bc->bc_magic == BLOCKIF_SIG); 871 872 /* 873 * Stop the block i/o thread 874 */ 875 pthread_mutex_lock(&bc->bc_mtx); 876 bc->bc_closing = 1; 877 if (bc->bc_resize_event != NULL) 878 mevent_disable(bc->bc_resize_event); 879 pthread_mutex_unlock(&bc->bc_mtx); 880 pthread_cond_broadcast(&bc->bc_cond); 881 for (i = 0; i < BLOCKIF_NUMTHR; i++) 882 pthread_join(bc->bc_btid[i], &jval); 883 884 /* XXX Cancel queued i/o's ??? */ 885 886 /* 887 * Release resources 888 */ 889 bc->bc_magic = 0; 890 close(bc->bc_fd); 891 free(bc); 892 893 return (0); 894 } 895 896 /* 897 * Return virtual C/H/S values for a given block. Use the algorithm 898 * outlined in the VHD specification to calculate values. 899 */ 900 void 901 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s) 902 { 903 off_t sectors; /* total sectors of the block dev */ 904 off_t hcyl; /* cylinders times heads */ 905 uint16_t secpt; /* sectors per track */ 906 uint8_t heads; 907 908 assert(bc->bc_magic == BLOCKIF_SIG); 909 910 sectors = bc->bc_size / bc->bc_sectsz; 911 912 /* Clamp the size to the largest possible with CHS */ 913 if (sectors > 65535UL*16*255) 914 sectors = 65535UL*16*255; 915 916 if (sectors >= 65536UL*16*63) { 917 secpt = 255; 918 heads = 16; 919 hcyl = sectors / secpt; 920 } else { 921 secpt = 17; 922 hcyl = sectors / secpt; 923 heads = (hcyl + 1023) / 1024; 924 925 if (heads < 4) 926 heads = 4; 927 928 if (hcyl >= (heads * 1024) || heads > 16) { 929 secpt = 31; 930 heads = 16; 931 hcyl = sectors / secpt; 932 } 933 if (hcyl >= (heads * 1024)) { 934 secpt = 63; 935 heads = 16; 936 hcyl = sectors / secpt; 937 } 938 } 939 940 *c = hcyl / heads; 941 *h = heads; 942 *s = secpt; 943 } 944 945 /* 946 * Accessors 947 */ 948 off_t 949 blockif_size(struct blockif_ctxt *bc) 950 { 951 assert(bc->bc_magic == BLOCKIF_SIG); 952 return (bc->bc_size); 953 } 954 955 int 956 blockif_sectsz(struct blockif_ctxt *bc) 957 { 958 assert(bc->bc_magic == BLOCKIF_SIG); 959 return (bc->bc_sectsz); 960 } 961 962 void 963 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off) 964 { 965 assert(bc->bc_magic == BLOCKIF_SIG); 966 *size = bc->bc_psectsz; 967 *off = bc->bc_psectoff; 968 } 969 970 int 971 blockif_queuesz(struct blockif_ctxt *bc) 972 { 973 assert(bc->bc_magic == BLOCKIF_SIG); 974 return (BLOCKIF_MAXREQ - 1); 975 } 976 977 int 978 blockif_is_ro(struct blockif_ctxt *bc) 979 { 980 assert(bc->bc_magic == BLOCKIF_SIG); 981 return (bc->bc_rdonly); 982 } 983 984 int 985 blockif_candelete(struct blockif_ctxt *bc) 986 { 987 assert(bc->bc_magic == BLOCKIF_SIG); 988 return (bc->bc_candelete); 989 } 990 991 #ifdef BHYVE_SNAPSHOT 992 void 993 blockif_pause(struct blockif_ctxt *bc) 994 { 995 assert(bc != NULL); 996 assert(bc->bc_magic == BLOCKIF_SIG); 997 998 pthread_mutex_lock(&bc->bc_mtx); 999 bc->bc_paused = 1; 1000 1001 /* The interface is paused. Wait for workers to finish their work */ 1002 while (!blockif_empty(bc)) 1003 pthread_cond_wait(&bc->bc_work_done_cond, &bc->bc_mtx); 1004 pthread_mutex_unlock(&bc->bc_mtx); 1005 1006 if (blockif_flush_bc(bc)) 1007 fprintf(stderr, "%s: [WARN] failed to flush backing file.\r\n", 1008 __func__); 1009 } 1010 1011 void 1012 blockif_resume(struct blockif_ctxt *bc) 1013 { 1014 assert(bc != NULL); 1015 assert(bc->bc_magic == BLOCKIF_SIG); 1016 1017 pthread_mutex_lock(&bc->bc_mtx); 1018 bc->bc_paused = 0; 1019 pthread_mutex_unlock(&bc->bc_mtx); 1020 } 1021 #endif /* BHYVE_SNAPSHOT */ 1022