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 /* 33 * Copyright 2020 Joyent, Inc. 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/param.h> 40 #ifndef WITHOUT_CAPSICUM 41 #include <sys/capsicum.h> 42 #endif 43 #include <sys/queue.h> 44 #include <sys/errno.h> 45 #include <sys/stat.h> 46 #include <sys/ioctl.h> 47 #include <sys/disk.h> 48 #ifndef __FreeBSD__ 49 #include <sys/limits.h> 50 #include <sys/uio.h> 51 #include <sys/dkio.h> 52 #endif 53 54 #include <assert.h> 55 #ifndef WITHOUT_CAPSICUM 56 #include <capsicum_helpers.h> 57 #endif 58 #include <err.h> 59 #include <fcntl.h> 60 #include <stdio.h> 61 #include <stdlib.h> 62 #include <string.h> 63 #include <pthread.h> 64 #include <pthread_np.h> 65 #include <signal.h> 66 #include <sysexits.h> 67 #include <unistd.h> 68 69 #include <machine/atomic.h> 70 71 #include "bhyverun.h" 72 #include "config.h" 73 #include "debug.h" 74 #include "mevent.h" 75 #include "pci_emul.h" 76 #include "block_if.h" 77 78 #define BLOCKIF_SIG 0xb109b109 79 80 #ifdef __FreeBSD__ 81 #define BLOCKIF_NUMTHR 8 82 #else 83 /* Enlarge to keep pace with the virtio-block ring size */ 84 #define BLOCKIF_NUMTHR 16 85 #endif 86 #define BLOCKIF_MAXREQ (BLOCKIF_RING_MAX + BLOCKIF_NUMTHR) 87 88 enum blockop { 89 BOP_READ, 90 BOP_WRITE, 91 #ifndef __FreeBSD__ 92 BOP_WRITE_SYNC, 93 #endif 94 BOP_FLUSH, 95 BOP_DELETE 96 }; 97 98 enum blockstat { 99 BST_FREE, 100 BST_BLOCK, 101 BST_PEND, 102 BST_BUSY, 103 BST_DONE 104 }; 105 106 struct blockif_elem { 107 TAILQ_ENTRY(blockif_elem) be_link; 108 struct blockif_req *be_req; 109 enum blockop be_op; 110 enum blockstat be_status; 111 pthread_t be_tid; 112 off_t be_block; 113 }; 114 115 #ifndef __FreeBSD__ 116 enum blockif_wce { 117 WCE_NONE = 0, 118 WCE_IOCTL, 119 WCE_FCNTL 120 }; 121 #endif 122 123 struct blockif_ctxt { 124 int bc_magic; 125 int bc_fd; 126 int bc_ischr; 127 int bc_isgeom; 128 int bc_candelete; 129 #ifndef __FreeBSD__ 130 enum blockif_wce bc_wce; 131 #endif 132 int bc_rdonly; 133 off_t bc_size; 134 int bc_sectsz; 135 int bc_psectsz; 136 int bc_psectoff; 137 int bc_closing; 138 pthread_t bc_btid[BLOCKIF_NUMTHR]; 139 pthread_mutex_t bc_mtx; 140 pthread_cond_t bc_cond; 141 blockif_resize_cb *bc_resize_cb; 142 void *bc_resize_cb_arg; 143 struct mevent *bc_resize_event; 144 145 /* Request elements and free/pending/busy queues */ 146 TAILQ_HEAD(, blockif_elem) bc_freeq; 147 TAILQ_HEAD(, blockif_elem) bc_pendq; 148 TAILQ_HEAD(, blockif_elem) bc_busyq; 149 struct blockif_elem bc_reqs[BLOCKIF_MAXREQ]; 150 }; 151 152 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT; 153 154 struct blockif_sig_elem { 155 pthread_mutex_t bse_mtx; 156 pthread_cond_t bse_cond; 157 int bse_pending; 158 struct blockif_sig_elem *bse_next; 159 }; 160 161 static struct blockif_sig_elem *blockif_bse_head; 162 163 static int 164 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq, 165 enum blockop op) 166 { 167 struct blockif_elem *be, *tbe; 168 off_t off; 169 int i; 170 171 be = TAILQ_FIRST(&bc->bc_freeq); 172 assert(be != NULL); 173 assert(be->be_status == BST_FREE); 174 TAILQ_REMOVE(&bc->bc_freeq, be, be_link); 175 be->be_req = breq; 176 be->be_op = op; 177 switch (op) { 178 case BOP_READ: 179 case BOP_WRITE: 180 #ifndef __FreeBSD__ 181 case BOP_WRITE_SYNC: 182 #endif 183 case BOP_DELETE: 184 off = breq->br_offset; 185 for (i = 0; i < breq->br_iovcnt; i++) 186 off += breq->br_iov[i].iov_len; 187 break; 188 default: 189 off = OFF_MAX; 190 } 191 be->be_block = off; 192 TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) { 193 if (tbe->be_block == breq->br_offset) 194 break; 195 } 196 if (tbe == NULL) { 197 TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) { 198 if (tbe->be_block == breq->br_offset) 199 break; 200 } 201 } 202 if (tbe == NULL) 203 be->be_status = BST_PEND; 204 else 205 be->be_status = BST_BLOCK; 206 TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link); 207 return (be->be_status == BST_PEND); 208 } 209 210 static int 211 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep) 212 { 213 struct blockif_elem *be; 214 215 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) { 216 if (be->be_status == BST_PEND) 217 break; 218 assert(be->be_status == BST_BLOCK); 219 } 220 if (be == NULL) 221 return (0); 222 TAILQ_REMOVE(&bc->bc_pendq, be, be_link); 223 be->be_status = BST_BUSY; 224 be->be_tid = t; 225 TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link); 226 *bep = be; 227 return (1); 228 } 229 230 static void 231 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be) 232 { 233 struct blockif_elem *tbe; 234 235 if (be->be_status == BST_DONE || be->be_status == BST_BUSY) 236 TAILQ_REMOVE(&bc->bc_busyq, be, be_link); 237 else 238 TAILQ_REMOVE(&bc->bc_pendq, be, be_link); 239 TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) { 240 if (tbe->be_req->br_offset == be->be_block) 241 tbe->be_status = BST_PEND; 242 } 243 be->be_tid = 0; 244 be->be_status = BST_FREE; 245 be->be_req = NULL; 246 TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link); 247 } 248 249 static int 250 blockif_flush_bc(struct blockif_ctxt *bc) 251 { 252 #ifdef __FreeBSD__ 253 if (bc->bc_ischr) { 254 if (ioctl(bc->bc_fd, DIOCGFLUSH)) 255 return (errno); 256 } else if (fsync(bc->bc_fd)) 257 return (errno); 258 #else 259 /* 260 * This fsync() should be adequate to flush the cache of a file 261 * or device. In VFS, the VOP_SYNC operation is converted to 262 * the appropriate ioctl in both sdev (for real devices) and 263 * zfs (for zvols). 264 */ 265 if (fsync(bc->bc_fd)) 266 return (errno); 267 #endif 268 269 return (0); 270 } 271 272 static void 273 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf) 274 { 275 struct blockif_req *br; 276 #ifdef __FreeBSD__ 277 off_t arg[2]; 278 #endif 279 ssize_t clen, len, off, boff, voff; 280 int i, err; 281 #ifdef __FreeBSD__ 282 struct spacectl_range range; 283 #endif 284 285 br = be->be_req; 286 if (br->br_iovcnt <= 1) 287 buf = NULL; 288 err = 0; 289 switch (be->be_op) { 290 case BOP_READ: 291 if (buf == NULL) { 292 if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt, 293 br->br_offset)) < 0) 294 err = errno; 295 else 296 br->br_resid -= len; 297 break; 298 } 299 i = 0; 300 off = voff = 0; 301 while (br->br_resid > 0) { 302 len = MIN(br->br_resid, MAXPHYS); 303 if (pread(bc->bc_fd, buf, len, br->br_offset + 304 off) < 0) { 305 err = errno; 306 break; 307 } 308 boff = 0; 309 do { 310 clen = MIN(len - boff, br->br_iov[i].iov_len - 311 voff); 312 memcpy(br->br_iov[i].iov_base + voff, 313 buf + boff, clen); 314 if (clen < br->br_iov[i].iov_len - voff) 315 voff += clen; 316 else { 317 i++; 318 voff = 0; 319 } 320 boff += clen; 321 } while (boff < len); 322 off += len; 323 br->br_resid -= len; 324 } 325 break; 326 case BOP_WRITE: 327 if (bc->bc_rdonly) { 328 err = EROFS; 329 break; 330 } 331 if (buf == NULL) { 332 if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt, 333 br->br_offset)) < 0) 334 err = errno; 335 else 336 br->br_resid -= len; 337 break; 338 } 339 i = 0; 340 off = voff = 0; 341 while (br->br_resid > 0) { 342 len = MIN(br->br_resid, MAXPHYS); 343 boff = 0; 344 do { 345 clen = MIN(len - boff, br->br_iov[i].iov_len - 346 voff); 347 memcpy(buf + boff, 348 br->br_iov[i].iov_base + voff, clen); 349 if (clen < br->br_iov[i].iov_len - voff) 350 voff += clen; 351 else { 352 i++; 353 voff = 0; 354 } 355 boff += clen; 356 } while (boff < len); 357 if (pwrite(bc->bc_fd, buf, len, br->br_offset + 358 off) < 0) { 359 err = errno; 360 break; 361 } 362 off += len; 363 br->br_resid -= len; 364 } 365 break; 366 case BOP_FLUSH: 367 err = blockif_flush_bc(bc); 368 break; 369 case BOP_DELETE: 370 if (!bc->bc_candelete) 371 err = EOPNOTSUPP; 372 else if (bc->bc_rdonly) 373 err = EROFS; 374 #ifdef __FreeBSD__ 375 else if (bc->bc_ischr) { 376 arg[0] = br->br_offset; 377 arg[1] = br->br_resid; 378 if (ioctl(bc->bc_fd, DIOCGDELETE, arg)) 379 err = errno; 380 else 381 br->br_resid = 0; 382 } else { 383 range.r_offset = br->br_offset; 384 range.r_len = br->br_resid; 385 386 while (range.r_len > 0) { 387 if (fspacectl(bc->bc_fd, SPACECTL_DEALLOC, 388 &range, 0, &range) != 0) { 389 err = errno; 390 break; 391 } 392 } 393 if (err == 0) 394 br->br_resid = 0; 395 } 396 #else 397 else if (bc->bc_ischr) { 398 dkioc_free_list_t dfl = { 399 .dfl_num_exts = 1, 400 .dfl_offset = 0, 401 .dfl_flags = 0, 402 .dfl_exts = { 403 { 404 .dfle_start = br->br_offset, 405 .dfle_length = br->br_resid 406 } 407 } 408 }; 409 410 if (ioctl(bc->bc_fd, DKIOCFREE, &dfl)) 411 err = errno; 412 else 413 br->br_resid = 0; 414 } else { 415 struct flock fl = { 416 .l_whence = 0, 417 .l_type = F_WRLCK, 418 .l_start = br->br_offset, 419 .l_len = br->br_resid 420 }; 421 422 if (fcntl(bc->bc_fd, F_FREESP, &fl)) 423 err = errno; 424 else 425 br->br_resid = 0; 426 } 427 #endif 428 break; 429 default: 430 err = EINVAL; 431 break; 432 } 433 434 be->be_status = BST_DONE; 435 436 (*br->br_callback)(br, err); 437 } 438 439 static inline bool 440 blockif_empty(const struct blockif_ctxt *bc) 441 { 442 return (TAILQ_EMPTY(&bc->bc_pendq) && TAILQ_EMPTY(&bc->bc_busyq)); 443 } 444 445 static void * 446 blockif_thr(void *arg) 447 { 448 struct blockif_ctxt *bc; 449 struct blockif_elem *be; 450 pthread_t t; 451 uint8_t *buf; 452 453 bc = arg; 454 if (bc->bc_isgeom) 455 buf = malloc(MAXPHYS); 456 else 457 buf = NULL; 458 t = pthread_self(); 459 460 pthread_mutex_lock(&bc->bc_mtx); 461 for (;;) { 462 while (blockif_dequeue(bc, t, &be)) { 463 pthread_mutex_unlock(&bc->bc_mtx); 464 blockif_proc(bc, be, buf); 465 pthread_mutex_lock(&bc->bc_mtx); 466 blockif_complete(bc, be); 467 } 468 /* Check ctxt status here to see if exit requested */ 469 if (bc->bc_closing) 470 break; 471 472 pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx); 473 } 474 pthread_mutex_unlock(&bc->bc_mtx); 475 476 if (buf) 477 free(buf); 478 pthread_exit(NULL); 479 return (NULL); 480 } 481 482 #ifdef __FreeBSD__ 483 static void 484 blockif_sigcont_handler(int signal, enum ev_type type, void *arg) 485 #else 486 static void 487 blockif_sigcont_handler(int signal) 488 #endif 489 { 490 struct blockif_sig_elem *bse; 491 492 for (;;) { 493 /* 494 * Process the entire list even if not intended for 495 * this thread. 496 */ 497 do { 498 bse = blockif_bse_head; 499 if (bse == NULL) 500 return; 501 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head, 502 (uintptr_t)bse, 503 (uintptr_t)bse->bse_next)); 504 505 pthread_mutex_lock(&bse->bse_mtx); 506 bse->bse_pending = 0; 507 pthread_cond_signal(&bse->bse_cond); 508 pthread_mutex_unlock(&bse->bse_mtx); 509 } 510 } 511 512 static void 513 blockif_init(void) 514 { 515 #ifdef __FreeBSD__ 516 mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL); 517 (void) signal(SIGCONT, SIG_IGN); 518 #else 519 (void) sigset(SIGCONT, blockif_sigcont_handler); 520 #endif 521 } 522 523 int 524 blockif_legacy_config(nvlist_t *nvl, const char *opts) 525 { 526 char *cp, *path; 527 528 if (opts == NULL) 529 return (0); 530 531 cp = strchr(opts, ','); 532 if (cp == NULL) { 533 set_config_value_node(nvl, "path", opts); 534 return (0); 535 } 536 path = strndup(opts, cp - opts); 537 set_config_value_node(nvl, "path", path); 538 free(path); 539 return (pci_parse_legacy_config(nvl, cp + 1)); 540 } 541 542 struct blockif_ctxt * 543 blockif_open(nvlist_t *nvl, const char *ident) 544 { 545 char tname[MAXCOMLEN + 1]; 546 #ifdef __FreeBSD__ 547 char name[MAXPATHLEN]; 548 #endif 549 const char *path, *pssval, *ssval; 550 char *cp; 551 struct blockif_ctxt *bc; 552 struct stat sbuf; 553 #ifdef __FreeBSD__ 554 struct diocgattr_arg arg; 555 #else 556 enum blockif_wce wce = WCE_NONE; 557 #endif 558 off_t size, psectsz, psectoff; 559 int extra, fd, i, sectsz; 560 int ro, candelete, geom, ssopt, pssopt; 561 int nodelete; 562 563 #ifndef WITHOUT_CAPSICUM 564 cap_rights_t rights; 565 cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE, DIOCGMEDIASIZE }; 566 #endif 567 568 pthread_once(&blockif_once, blockif_init); 569 570 fd = -1; 571 extra = 0; 572 ssopt = 0; 573 #ifndef __FreeBSD__ 574 pssopt = 0; 575 #endif 576 ro = 0; 577 nodelete = 0; 578 579 if (get_config_bool_node_default(nvl, "nocache", false)) 580 extra |= O_DIRECT; 581 if (get_config_bool_node_default(nvl, "nodelete", false)) 582 nodelete = 1; 583 if (get_config_bool_node_default(nvl, "sync", false) || 584 get_config_bool_node_default(nvl, "direct", false)) 585 extra |= O_SYNC; 586 if (get_config_bool_node_default(nvl, "ro", false)) 587 ro = 1; 588 ssval = get_config_value_node(nvl, "sectorsize"); 589 if (ssval != NULL) { 590 ssopt = strtol(ssval, &cp, 10); 591 if (cp == ssval) { 592 EPRINTLN("Invalid sector size \"%s\"", ssval); 593 goto err; 594 } 595 if (*cp == '\0') { 596 pssopt = ssopt; 597 } else if (*cp == '/') { 598 pssval = cp + 1; 599 pssopt = strtol(pssval, &cp, 10); 600 if (cp == pssval || *cp != '\0') { 601 EPRINTLN("Invalid sector size \"%s\"", ssval); 602 goto err; 603 } 604 } else { 605 EPRINTLN("Invalid sector size \"%s\"", ssval); 606 goto err; 607 } 608 } 609 610 path = get_config_value_node(nvl, "path"); 611 if (path == NULL) { 612 EPRINTLN("Missing \"path\" for block device."); 613 goto err; 614 } 615 616 fd = open(path, (ro ? O_RDONLY : O_RDWR) | extra); 617 if (fd < 0 && !ro) { 618 /* Attempt a r/w fail with a r/o open */ 619 fd = open(path, O_RDONLY | extra); 620 ro = 1; 621 } 622 623 if (fd < 0) { 624 warn("Could not open backing file: %s", path); 625 goto err; 626 } 627 628 if (fstat(fd, &sbuf) < 0) { 629 warn("Could not stat backing file %s", path); 630 goto err; 631 } 632 633 #ifndef WITHOUT_CAPSICUM 634 cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK, 635 CAP_WRITE, CAP_FSTAT, CAP_EVENT, CAP_FPATHCONF); 636 if (ro) 637 cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE); 638 639 if (caph_rights_limit(fd, &rights) == -1) 640 errx(EX_OSERR, "Unable to apply rights for sandbox"); 641 #endif 642 643 /* 644 * Deal with raw devices 645 */ 646 size = sbuf.st_size; 647 sectsz = DEV_BSIZE; 648 psectsz = psectoff = 0; 649 candelete = geom = 0; 650 #ifdef __FreeBSD__ 651 if (S_ISCHR(sbuf.st_mode)) { 652 if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 || 653 ioctl(fd, DIOCGSECTORSIZE, §sz)) { 654 perror("Could not fetch dev blk/sector size"); 655 goto err; 656 } 657 assert(size != 0); 658 assert(sectsz != 0); 659 if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0) 660 ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff); 661 strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); 662 arg.len = sizeof(arg.value.i); 663 if (nodelete == 0 && ioctl(fd, DIOCGATTR, &arg) == 0) 664 candelete = arg.value.i; 665 if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0) 666 geom = 1; 667 } else { 668 psectsz = sbuf.st_blksize; 669 /* Avoid fallback implementation */ 670 candelete = fpathconf(fd, _PC_DEALLOC_PRESENT) == 1; 671 } 672 #else 673 psectsz = sbuf.st_blksize; 674 if (S_ISCHR(sbuf.st_mode)) { 675 struct dk_minfo_ext dkmext; 676 int wce_val; 677 678 /* Look for a more accurate physical block/media size */ 679 if (ioctl(fd, DKIOCGMEDIAINFOEXT, &dkmext) == 0) { 680 psectsz = dkmext.dki_pbsize; 681 size = dkmext.dki_lbsize * dkmext.dki_capacity; 682 } 683 /* See if a configurable write cache is present and working */ 684 if (ioctl(fd, DKIOCGETWCE, &wce_val) == 0) { 685 /* 686 * If WCE is already active, disable it until the 687 * specific device driver calls for its return. If it 688 * is not active, toggle it on and off to verify that 689 * such actions are possible. 690 */ 691 if (wce_val != 0) { 692 wce_val = 0; 693 /* 694 * Inability to disable the cache is a threat 695 * to data durability. 696 */ 697 assert(ioctl(fd, DKIOCSETWCE, &wce_val) == 0); 698 wce = WCE_IOCTL; 699 } else { 700 int r1, r2; 701 702 wce_val = 1; 703 r1 = ioctl(fd, DKIOCSETWCE, &wce_val); 704 wce_val = 0; 705 r2 = ioctl(fd, DKIOCSETWCE, &wce_val); 706 707 if (r1 == 0 && r2 == 0) { 708 wce = WCE_IOCTL; 709 } else { 710 /* 711 * If the cache cache toggle was not 712 * successful, ensure that the cache 713 * was not left enabled. 714 */ 715 assert(r1 != 0); 716 } 717 } 718 } 719 720 if (nodelete == 0 && ioctl(fd, DKIOC_CANFREE, &candelete)) 721 candelete = 0; 722 723 } else { 724 int flags; 725 726 if ((flags = fcntl(fd, F_GETFL)) >= 0) { 727 flags |= O_DSYNC; 728 if (fcntl(fd, F_SETFL, flags) != -1) { 729 wce = WCE_FCNTL; 730 } 731 } 732 733 /* 734 * We don't have a way to discover if a file supports the 735 * FREESP fcntl cmd (other than trying it). However, 736 * zfs, ufs, tmpfs, and udfs all support the FREESP fcntl cmd. 737 * Nfsv4 and nfsv4 also forward the FREESP request 738 * to the server, so we always enable it for file based 739 * volumes. Anyone trying to run volumes on an unsupported 740 * configuration is on their own, and should be prepared 741 * for the requests to fail. 742 */ 743 if (nodelete == 0) 744 candelete = 1; 745 } 746 #endif 747 748 #ifndef WITHOUT_CAPSICUM 749 if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1) 750 errx(EX_OSERR, "Unable to apply rights for sandbox"); 751 #endif 752 753 if (ssopt != 0) { 754 if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 || 755 ssopt > pssopt) { 756 EPRINTLN("Invalid sector size %d/%d", 757 ssopt, pssopt); 758 goto err; 759 } 760 761 /* 762 * Some backend drivers (e.g. cd0, ada0) require that the I/O 763 * size be a multiple of the device's sector size. 764 * 765 * Validate that the emulated sector size complies with this 766 * requirement. 767 */ 768 if (S_ISCHR(sbuf.st_mode)) { 769 if (ssopt < sectsz || (ssopt % sectsz) != 0) { 770 EPRINTLN("Sector size %d incompatible " 771 "with underlying device sector size %d", 772 ssopt, sectsz); 773 goto err; 774 } 775 } 776 777 sectsz = ssopt; 778 psectsz = pssopt; 779 psectoff = 0; 780 } 781 782 bc = calloc(1, sizeof(struct blockif_ctxt)); 783 if (bc == NULL) { 784 perror("calloc"); 785 goto err; 786 } 787 788 bc->bc_magic = BLOCKIF_SIG; 789 bc->bc_fd = fd; 790 bc->bc_ischr = S_ISCHR(sbuf.st_mode); 791 bc->bc_isgeom = geom; 792 bc->bc_candelete = candelete; 793 #ifndef __FreeBSD__ 794 bc->bc_wce = wce; 795 #endif 796 bc->bc_rdonly = ro; 797 bc->bc_size = size; 798 bc->bc_sectsz = sectsz; 799 bc->bc_psectsz = psectsz; 800 bc->bc_psectoff = psectoff; 801 pthread_mutex_init(&bc->bc_mtx, NULL); 802 pthread_cond_init(&bc->bc_cond, NULL); 803 TAILQ_INIT(&bc->bc_freeq); 804 TAILQ_INIT(&bc->bc_pendq); 805 TAILQ_INIT(&bc->bc_busyq); 806 for (i = 0; i < BLOCKIF_MAXREQ; i++) { 807 bc->bc_reqs[i].be_status = BST_FREE; 808 TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link); 809 } 810 811 for (i = 0; i < BLOCKIF_NUMTHR; i++) { 812 pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc); 813 snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i); 814 pthread_set_name_np(bc->bc_btid[i], tname); 815 } 816 817 return (bc); 818 err: 819 if (fd >= 0) 820 close(fd); 821 return (NULL); 822 } 823 824 static void 825 blockif_resized(int fd, enum ev_type type, void *arg) 826 { 827 struct blockif_ctxt *bc; 828 struct stat sb; 829 off_t mediasize; 830 831 if (fstat(fd, &sb) != 0) 832 return; 833 834 #ifdef __FreeBSD__ 835 if (S_ISCHR(sb.st_mode)) { 836 if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) < 0) { 837 EPRINTLN("blockif_resized: get mediasize failed: %s", 838 strerror(errno)); 839 return; 840 } 841 } else 842 mediasize = sb.st_size; 843 #else 844 mediasize = sb.st_size; 845 if (S_ISCHR(sb.st_mode)) { 846 struct dk_minfo dkm; 847 848 if (ioctl(fd, DKIOCGMEDIAINFO, &dkm) == 0) 849 mediasize = dkm.dki_lbsize * dkm.dki_capacity; 850 } 851 #endif 852 853 bc = arg; 854 pthread_mutex_lock(&bc->bc_mtx); 855 if (mediasize != bc->bc_size) { 856 bc->bc_size = mediasize; 857 bc->bc_resize_cb(bc, bc->bc_resize_cb_arg, bc->bc_size); 858 } 859 pthread_mutex_unlock(&bc->bc_mtx); 860 } 861 862 int 863 blockif_register_resize_callback(struct blockif_ctxt *bc, blockif_resize_cb *cb, 864 void *cb_arg) 865 { 866 struct stat sb; 867 int err; 868 #ifndef __FreeBSD__ 869 err = 0; 870 #endif 871 872 if (cb == NULL) 873 return (EINVAL); 874 875 pthread_mutex_lock(&bc->bc_mtx); 876 if (bc->bc_resize_cb != NULL) { 877 err = EBUSY; 878 goto out; 879 } 880 881 assert(bc->bc_closing == 0); 882 883 if (fstat(bc->bc_fd, &sb) != 0) { 884 err = errno; 885 goto out; 886 } 887 888 bc->bc_resize_event = mevent_add_flags(bc->bc_fd, EVF_VNODE, 889 EVFF_ATTRIB, blockif_resized, bc); 890 if (bc->bc_resize_event == NULL) { 891 err = ENXIO; 892 goto out; 893 } 894 895 bc->bc_resize_cb = cb; 896 bc->bc_resize_cb_arg = cb_arg; 897 out: 898 pthread_mutex_unlock(&bc->bc_mtx); 899 900 return (err); 901 } 902 903 static int 904 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq, 905 enum blockop op) 906 { 907 int err; 908 909 err = 0; 910 911 pthread_mutex_lock(&bc->bc_mtx); 912 if (!TAILQ_EMPTY(&bc->bc_freeq)) { 913 /* 914 * Enqueue and inform the block i/o thread 915 * that there is work available 916 */ 917 if (blockif_enqueue(bc, breq, op)) 918 pthread_cond_signal(&bc->bc_cond); 919 } else { 920 /* 921 * Callers are not allowed to enqueue more than 922 * the specified blockif queue limit. Return an 923 * error to indicate that the queue length has been 924 * exceeded. 925 */ 926 err = E2BIG; 927 } 928 pthread_mutex_unlock(&bc->bc_mtx); 929 930 return (err); 931 } 932 933 int 934 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq) 935 { 936 assert(bc->bc_magic == BLOCKIF_SIG); 937 return (blockif_request(bc, breq, BOP_READ)); 938 } 939 940 int 941 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq) 942 { 943 assert(bc->bc_magic == BLOCKIF_SIG); 944 return (blockif_request(bc, breq, BOP_WRITE)); 945 } 946 947 int 948 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq) 949 { 950 assert(bc->bc_magic == BLOCKIF_SIG); 951 return (blockif_request(bc, breq, BOP_FLUSH)); 952 } 953 954 int 955 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq) 956 { 957 assert(bc->bc_magic == BLOCKIF_SIG); 958 return (blockif_request(bc, breq, BOP_DELETE)); 959 } 960 961 int 962 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq) 963 { 964 struct blockif_elem *be; 965 966 assert(bc->bc_magic == BLOCKIF_SIG); 967 968 pthread_mutex_lock(&bc->bc_mtx); 969 /* 970 * Check pending requests. 971 */ 972 TAILQ_FOREACH(be, &bc->bc_pendq, be_link) { 973 if (be->be_req == breq) 974 break; 975 } 976 if (be != NULL) { 977 /* 978 * Found it. 979 */ 980 blockif_complete(bc, be); 981 pthread_mutex_unlock(&bc->bc_mtx); 982 983 return (0); 984 } 985 986 /* 987 * Check in-flight requests. 988 */ 989 TAILQ_FOREACH(be, &bc->bc_busyq, be_link) { 990 if (be->be_req == breq) 991 break; 992 } 993 if (be == NULL) { 994 /* 995 * Didn't find it. 996 */ 997 pthread_mutex_unlock(&bc->bc_mtx); 998 return (EINVAL); 999 } 1000 1001 /* 1002 * Interrupt the processing thread to force it return 1003 * prematurely via it's normal callback path. 1004 */ 1005 while (be->be_status == BST_BUSY) { 1006 struct blockif_sig_elem bse, *old_head; 1007 1008 pthread_mutex_init(&bse.bse_mtx, NULL); 1009 pthread_cond_init(&bse.bse_cond, NULL); 1010 1011 bse.bse_pending = 1; 1012 1013 do { 1014 old_head = blockif_bse_head; 1015 bse.bse_next = old_head; 1016 } while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head, 1017 (uintptr_t)old_head, 1018 (uintptr_t)&bse)); 1019 1020 pthread_kill(be->be_tid, SIGCONT); 1021 1022 pthread_mutex_lock(&bse.bse_mtx); 1023 while (bse.bse_pending) 1024 pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx); 1025 pthread_mutex_unlock(&bse.bse_mtx); 1026 } 1027 1028 pthread_mutex_unlock(&bc->bc_mtx); 1029 1030 /* 1031 * The processing thread has been interrupted. Since it's not 1032 * clear if the callback has been invoked yet, return EBUSY. 1033 */ 1034 return (EBUSY); 1035 } 1036 1037 int 1038 blockif_close(struct blockif_ctxt *bc) 1039 { 1040 void *jval; 1041 int i; 1042 1043 assert(bc->bc_magic == BLOCKIF_SIG); 1044 1045 /* 1046 * Stop the block i/o thread 1047 */ 1048 pthread_mutex_lock(&bc->bc_mtx); 1049 bc->bc_closing = 1; 1050 if (bc->bc_resize_event != NULL) 1051 mevent_disable(bc->bc_resize_event); 1052 pthread_mutex_unlock(&bc->bc_mtx); 1053 pthread_cond_broadcast(&bc->bc_cond); 1054 for (i = 0; i < BLOCKIF_NUMTHR; i++) 1055 pthread_join(bc->bc_btid[i], &jval); 1056 1057 /* XXX Cancel queued i/o's ??? */ 1058 1059 /* 1060 * Release resources 1061 */ 1062 bc->bc_magic = 0; 1063 close(bc->bc_fd); 1064 free(bc); 1065 1066 return (0); 1067 } 1068 1069 /* 1070 * Return virtual C/H/S values for a given block. Use the algorithm 1071 * outlined in the VHD specification to calculate values. 1072 */ 1073 void 1074 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s) 1075 { 1076 off_t sectors; /* total sectors of the block dev */ 1077 off_t hcyl; /* cylinders times heads */ 1078 uint16_t secpt; /* sectors per track */ 1079 uint8_t heads; 1080 1081 assert(bc->bc_magic == BLOCKIF_SIG); 1082 1083 sectors = bc->bc_size / bc->bc_sectsz; 1084 1085 /* Clamp the size to the largest possible with CHS */ 1086 if (sectors > 65535UL*16*255) 1087 sectors = 65535UL*16*255; 1088 1089 if (sectors >= 65536UL*16*63) { 1090 secpt = 255; 1091 heads = 16; 1092 hcyl = sectors / secpt; 1093 } else { 1094 secpt = 17; 1095 hcyl = sectors / secpt; 1096 heads = (hcyl + 1023) / 1024; 1097 1098 if (heads < 4) 1099 heads = 4; 1100 1101 if (hcyl >= (heads * 1024) || heads > 16) { 1102 secpt = 31; 1103 heads = 16; 1104 hcyl = sectors / secpt; 1105 } 1106 if (hcyl >= (heads * 1024)) { 1107 secpt = 63; 1108 heads = 16; 1109 hcyl = sectors / secpt; 1110 } 1111 } 1112 1113 *c = hcyl / heads; 1114 *h = heads; 1115 *s = secpt; 1116 } 1117 1118 /* 1119 * Accessors 1120 */ 1121 off_t 1122 blockif_size(struct blockif_ctxt *bc) 1123 { 1124 assert(bc->bc_magic == BLOCKIF_SIG); 1125 return (bc->bc_size); 1126 } 1127 1128 int 1129 blockif_sectsz(struct blockif_ctxt *bc) 1130 { 1131 assert(bc->bc_magic == BLOCKIF_SIG); 1132 return (bc->bc_sectsz); 1133 } 1134 1135 void 1136 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off) 1137 { 1138 assert(bc->bc_magic == BLOCKIF_SIG); 1139 *size = bc->bc_psectsz; 1140 *off = bc->bc_psectoff; 1141 } 1142 1143 int 1144 blockif_queuesz(struct blockif_ctxt *bc) 1145 { 1146 assert(bc->bc_magic == BLOCKIF_SIG); 1147 return (BLOCKIF_MAXREQ - 1); 1148 } 1149 1150 int 1151 blockif_is_ro(struct blockif_ctxt *bc) 1152 { 1153 assert(bc->bc_magic == BLOCKIF_SIG); 1154 return (bc->bc_rdonly); 1155 } 1156 1157 int 1158 blockif_candelete(struct blockif_ctxt *bc) 1159 { 1160 assert(bc->bc_magic == BLOCKIF_SIG); 1161 return (bc->bc_candelete); 1162 } 1163 1164 #ifndef __FreeBSD__ 1165 int 1166 blockif_set_wce(struct blockif_ctxt *bc, int wc_enable) 1167 { 1168 int res = 0, flags; 1169 int clean_val = (wc_enable != 0) ? 1 : 0; 1170 1171 (void) pthread_mutex_lock(&bc->bc_mtx); 1172 switch (bc->bc_wce) { 1173 case WCE_IOCTL: 1174 res = ioctl(bc->bc_fd, DKIOCSETWCE, &clean_val); 1175 break; 1176 case WCE_FCNTL: 1177 if ((flags = fcntl(bc->bc_fd, F_GETFL)) >= 0) { 1178 if (wc_enable == 0) { 1179 flags |= O_DSYNC; 1180 } else { 1181 flags &= ~O_DSYNC; 1182 } 1183 if (fcntl(bc->bc_fd, F_SETFL, flags) == -1) { 1184 res = -1; 1185 } 1186 } else { 1187 res = -1; 1188 } 1189 break; 1190 default: 1191 break; 1192 } 1193 1194 /* 1195 * After a successful disable of the write cache, ensure that any 1196 * lingering data in the cache is synced out. 1197 */ 1198 if (res == 0 && wc_enable == 0) { 1199 res = fsync(bc->bc_fd); 1200 } 1201 (void) pthread_mutex_unlock(&bc->bc_mtx); 1202 1203 return (res); 1204 } 1205 #endif /* __FreeBSD__ */ 1206