xref: /freebsd/usr.sbin/bhyve/block_if.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*-
2  * Copyright (c) 2013  Peter Grehan <grehan@freebsd.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #ifndef WITHOUT_CAPSICUM
34 #include <sys/capsicum.h>
35 #endif
36 #include <sys/queue.h>
37 #include <sys/errno.h>
38 #include <sys/stat.h>
39 #include <sys/ioctl.h>
40 #include <sys/disk.h>
41 
42 #include <assert.h>
43 #include <err.h>
44 #include <fcntl.h>
45 #include <stdio.h>
46 #include <stdlib.h>
47 #include <string.h>
48 #include <pthread.h>
49 #include <pthread_np.h>
50 #include <signal.h>
51 #include <sysexits.h>
52 #include <unistd.h>
53 
54 #include <machine/atomic.h>
55 
56 #include "bhyverun.h"
57 #include "mevent.h"
58 #include "block_if.h"
59 
60 #define BLOCKIF_SIG	0xb109b109
61 
62 #define BLOCKIF_NUMTHR	8
63 #define BLOCKIF_MAXREQ	(64 + BLOCKIF_NUMTHR)
64 
65 enum blockop {
66 	BOP_READ,
67 	BOP_WRITE,
68 	BOP_FLUSH,
69 	BOP_DELETE
70 };
71 
72 enum blockstat {
73 	BST_FREE,
74 	BST_BLOCK,
75 	BST_PEND,
76 	BST_BUSY,
77 	BST_DONE
78 };
79 
80 struct blockif_elem {
81 	TAILQ_ENTRY(blockif_elem) be_link;
82 	struct blockif_req  *be_req;
83 	enum blockop	     be_op;
84 	enum blockstat	     be_status;
85 	pthread_t            be_tid;
86 	off_t		     be_block;
87 };
88 
89 struct blockif_ctxt {
90 	int			bc_magic;
91 	int			bc_fd;
92 	int			bc_ischr;
93 	int			bc_isgeom;
94 	int			bc_candelete;
95 	int			bc_rdonly;
96 	off_t			bc_size;
97 	int			bc_sectsz;
98 	int			bc_psectsz;
99 	int			bc_psectoff;
100 	int			bc_closing;
101 	pthread_t		bc_btid[BLOCKIF_NUMTHR];
102         pthread_mutex_t		bc_mtx;
103         pthread_cond_t		bc_cond;
104 
105 	/* Request elements and free/pending/busy queues */
106 	TAILQ_HEAD(, blockif_elem) bc_freeq;
107 	TAILQ_HEAD(, blockif_elem) bc_pendq;
108 	TAILQ_HEAD(, blockif_elem) bc_busyq;
109 	struct blockif_elem	bc_reqs[BLOCKIF_MAXREQ];
110 };
111 
112 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
113 
114 struct blockif_sig_elem {
115 	pthread_mutex_t			bse_mtx;
116 	pthread_cond_t			bse_cond;
117 	int				bse_pending;
118 	struct blockif_sig_elem		*bse_next;
119 };
120 
121 static struct blockif_sig_elem *blockif_bse_head;
122 
123 static int
124 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
125 		enum blockop op)
126 {
127 	struct blockif_elem *be, *tbe;
128 	off_t off;
129 	int i;
130 
131 	be = TAILQ_FIRST(&bc->bc_freeq);
132 	assert(be != NULL);
133 	assert(be->be_status == BST_FREE);
134 	TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
135 	be->be_req = breq;
136 	be->be_op = op;
137 	switch (op) {
138 	case BOP_READ:
139 	case BOP_WRITE:
140 	case BOP_DELETE:
141 		off = breq->br_offset;
142 		for (i = 0; i < breq->br_iovcnt; i++)
143 			off += breq->br_iov[i].iov_len;
144 		break;
145 	default:
146 		off = OFF_MAX;
147 	}
148 	be->be_block = off;
149 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
150 		if (tbe->be_block == breq->br_offset)
151 			break;
152 	}
153 	if (tbe == NULL) {
154 		TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
155 			if (tbe->be_block == breq->br_offset)
156 				break;
157 		}
158 	}
159 	if (tbe == NULL)
160 		be->be_status = BST_PEND;
161 	else
162 		be->be_status = BST_BLOCK;
163 	TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
164 	return (be->be_status == BST_PEND);
165 }
166 
167 static int
168 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
169 {
170 	struct blockif_elem *be;
171 
172 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
173 		if (be->be_status == BST_PEND)
174 			break;
175 		assert(be->be_status == BST_BLOCK);
176 	}
177 	if (be == NULL)
178 		return (0);
179 	TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
180 	be->be_status = BST_BUSY;
181 	be->be_tid = t;
182 	TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
183 	*bep = be;
184 	return (1);
185 }
186 
187 static void
188 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
189 {
190 	struct blockif_elem *tbe;
191 
192 	if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
193 		TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
194 	else
195 		TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
196 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
197 		if (tbe->be_req->br_offset == be->be_block)
198 			tbe->be_status = BST_PEND;
199 	}
200 	be->be_tid = 0;
201 	be->be_status = BST_FREE;
202 	be->be_req = NULL;
203 	TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
204 }
205 
206 static void
207 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
208 {
209 	struct blockif_req *br;
210 	off_t arg[2];
211 	ssize_t clen, len, off, boff, voff;
212 	int i, err;
213 
214 	br = be->be_req;
215 	if (br->br_iovcnt <= 1)
216 		buf = NULL;
217 	err = 0;
218 	switch (be->be_op) {
219 	case BOP_READ:
220 		if (buf == NULL) {
221 			if ((len = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
222 				   br->br_offset)) < 0)
223 				err = errno;
224 			else
225 				br->br_resid -= len;
226 			break;
227 		}
228 		i = 0;
229 		off = voff = 0;
230 		while (br->br_resid > 0) {
231 			len = MIN(br->br_resid, MAXPHYS);
232 			if (pread(bc->bc_fd, buf, len, br->br_offset +
233 			    off) < 0) {
234 				err = errno;
235 				break;
236 			}
237 			boff = 0;
238 			do {
239 				clen = MIN(len - boff, br->br_iov[i].iov_len -
240 				    voff);
241 				memcpy(br->br_iov[i].iov_base + voff,
242 				    buf + boff, clen);
243 				if (clen < br->br_iov[i].iov_len - voff)
244 					voff += clen;
245 				else {
246 					i++;
247 					voff = 0;
248 				}
249 				boff += clen;
250 			} while (boff < len);
251 			off += len;
252 			br->br_resid -= len;
253 		}
254 		break;
255 	case BOP_WRITE:
256 		if (bc->bc_rdonly) {
257 			err = EROFS;
258 			break;
259 		}
260 		if (buf == NULL) {
261 			if ((len = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
262 				    br->br_offset)) < 0)
263 				err = errno;
264 			else
265 				br->br_resid -= len;
266 			break;
267 		}
268 		i = 0;
269 		off = voff = 0;
270 		while (br->br_resid > 0) {
271 			len = MIN(br->br_resid, MAXPHYS);
272 			boff = 0;
273 			do {
274 				clen = MIN(len - boff, br->br_iov[i].iov_len -
275 				    voff);
276 				memcpy(buf + boff,
277 				    br->br_iov[i].iov_base + voff, clen);
278 				if (clen < br->br_iov[i].iov_len - voff)
279 					voff += clen;
280 				else {
281 					i++;
282 					voff = 0;
283 				}
284 				boff += clen;
285 			} while (boff < len);
286 			if (pwrite(bc->bc_fd, buf, len, br->br_offset +
287 			    off) < 0) {
288 				err = errno;
289 				break;
290 			}
291 			off += len;
292 			br->br_resid -= len;
293 		}
294 		break;
295 	case BOP_FLUSH:
296 		if (bc->bc_ischr) {
297 			if (ioctl(bc->bc_fd, DIOCGFLUSH))
298 				err = errno;
299 		} else if (fsync(bc->bc_fd))
300 			err = errno;
301 		break;
302 	case BOP_DELETE:
303 		if (!bc->bc_candelete)
304 			err = EOPNOTSUPP;
305 		else if (bc->bc_rdonly)
306 			err = EROFS;
307 		else if (bc->bc_ischr) {
308 			arg[0] = br->br_offset;
309 			arg[1] = br->br_resid;
310 			if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
311 				err = errno;
312 			else
313 				br->br_resid = 0;
314 		} else
315 			err = EOPNOTSUPP;
316 		break;
317 	default:
318 		err = EINVAL;
319 		break;
320 	}
321 
322 	be->be_status = BST_DONE;
323 
324 	(*br->br_callback)(br, err);
325 }
326 
327 static void *
328 blockif_thr(void *arg)
329 {
330 	struct blockif_ctxt *bc;
331 	struct blockif_elem *be;
332 	pthread_t t;
333 	uint8_t *buf;
334 
335 	bc = arg;
336 	if (bc->bc_isgeom)
337 		buf = malloc(MAXPHYS);
338 	else
339 		buf = NULL;
340 	t = pthread_self();
341 
342 	pthread_mutex_lock(&bc->bc_mtx);
343 	for (;;) {
344 		while (blockif_dequeue(bc, t, &be)) {
345 			pthread_mutex_unlock(&bc->bc_mtx);
346 			blockif_proc(bc, be, buf);
347 			pthread_mutex_lock(&bc->bc_mtx);
348 			blockif_complete(bc, be);
349 		}
350 		/* Check ctxt status here to see if exit requested */
351 		if (bc->bc_closing)
352 			break;
353 		pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
354 	}
355 	pthread_mutex_unlock(&bc->bc_mtx);
356 
357 	if (buf)
358 		free(buf);
359 	pthread_exit(NULL);
360 	return (NULL);
361 }
362 
363 static void
364 blockif_sigcont_handler(int signal, enum ev_type type, void *arg)
365 {
366 	struct blockif_sig_elem *bse;
367 
368 	for (;;) {
369 		/*
370 		 * Process the entire list even if not intended for
371 		 * this thread.
372 		 */
373 		do {
374 			bse = blockif_bse_head;
375 			if (bse == NULL)
376 				return;
377 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
378 					    (uintptr_t)bse,
379 					    (uintptr_t)bse->bse_next));
380 
381 		pthread_mutex_lock(&bse->bse_mtx);
382 		bse->bse_pending = 0;
383 		pthread_cond_signal(&bse->bse_cond);
384 		pthread_mutex_unlock(&bse->bse_mtx);
385 	}
386 }
387 
388 static void
389 blockif_init(void)
390 {
391 	mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
392 	(void) signal(SIGCONT, SIG_IGN);
393 }
394 
395 struct blockif_ctxt *
396 blockif_open(const char *optstr, const char *ident)
397 {
398 	char tname[MAXCOMLEN + 1];
399 	char name[MAXPATHLEN];
400 	char *nopt, *xopts, *cp;
401 	struct blockif_ctxt *bc;
402 	struct stat sbuf;
403 	struct diocgattr_arg arg;
404 	off_t size, psectsz, psectoff;
405 	int extra, fd, i, sectsz;
406 	int nocache, sync, ro, candelete, geom, ssopt, pssopt;
407 #ifndef WITHOUT_CAPSICUM
408 	cap_rights_t rights;
409 	cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE };
410 #endif
411 
412 	pthread_once(&blockif_once, blockif_init);
413 
414 	fd = -1;
415 	ssopt = 0;
416 	nocache = 0;
417 	sync = 0;
418 	ro = 0;
419 
420 	/*
421 	 * The first element in the optstring is always a pathname.
422 	 * Optional elements follow
423 	 */
424 	nopt = xopts = strdup(optstr);
425 	while (xopts != NULL) {
426 		cp = strsep(&xopts, ",");
427 		if (cp == nopt)		/* file or device pathname */
428 			continue;
429 		else if (!strcmp(cp, "nocache"))
430 			nocache = 1;
431 		else if (!strcmp(cp, "sync") || !strcmp(cp, "direct"))
432 			sync = 1;
433 		else if (!strcmp(cp, "ro"))
434 			ro = 1;
435 		else if (sscanf(cp, "sectorsize=%d/%d", &ssopt, &pssopt) == 2)
436 			;
437 		else if (sscanf(cp, "sectorsize=%d", &ssopt) == 1)
438 			pssopt = ssopt;
439 		else {
440 			fprintf(stderr, "Invalid device option \"%s\"\n", cp);
441 			goto err;
442 		}
443 	}
444 
445 	extra = 0;
446 	if (nocache)
447 		extra |= O_DIRECT;
448 	if (sync)
449 		extra |= O_SYNC;
450 
451 	fd = open(nopt, (ro ? O_RDONLY : O_RDWR) | extra);
452 	if (fd < 0 && !ro) {
453 		/* Attempt a r/w fail with a r/o open */
454 		fd = open(nopt, O_RDONLY | extra);
455 		ro = 1;
456 	}
457 
458 	if (fd < 0) {
459 		warn("Could not open backing file: %s", nopt);
460 		goto err;
461 	}
462 
463         if (fstat(fd, &sbuf) < 0) {
464 		warn("Could not stat backing file %s", nopt);
465 		goto err;
466         }
467 
468 #ifndef WITHOUT_CAPSICUM
469 	cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK,
470 	    CAP_WRITE);
471 	if (ro)
472 		cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE);
473 
474 	if (cap_rights_limit(fd, &rights) == -1 && errno != ENOSYS)
475 		errx(EX_OSERR, "Unable to apply rights for sandbox");
476 #endif
477 
478         /*
479 	 * Deal with raw devices
480 	 */
481         size = sbuf.st_size;
482 	sectsz = DEV_BSIZE;
483 	psectsz = psectoff = 0;
484 	candelete = geom = 0;
485 	if (S_ISCHR(sbuf.st_mode)) {
486 		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
487 		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
488 			perror("Could not fetch dev blk/sector size");
489 			goto err;
490 		}
491 		assert(size != 0);
492 		assert(sectsz != 0);
493 		if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
494 			ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
495 		strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
496 		arg.len = sizeof(arg.value.i);
497 		if (ioctl(fd, DIOCGATTR, &arg) == 0)
498 			candelete = arg.value.i;
499 		if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0)
500 			geom = 1;
501 	} else
502 		psectsz = sbuf.st_blksize;
503 
504 #ifndef WITHOUT_CAPSICUM
505 	if (cap_ioctls_limit(fd, cmds, nitems(cmds)) == -1 && errno != ENOSYS)
506 		errx(EX_OSERR, "Unable to apply rights for sandbox");
507 #endif
508 
509 	if (ssopt != 0) {
510 		if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
511 		    ssopt > pssopt) {
512 			fprintf(stderr, "Invalid sector size %d/%d\n",
513 			    ssopt, pssopt);
514 			goto err;
515 		}
516 
517 		/*
518 		 * Some backend drivers (e.g. cd0, ada0) require that the I/O
519 		 * size be a multiple of the device's sector size.
520 		 *
521 		 * Validate that the emulated sector size complies with this
522 		 * requirement.
523 		 */
524 		if (S_ISCHR(sbuf.st_mode)) {
525 			if (ssopt < sectsz || (ssopt % sectsz) != 0) {
526 				fprintf(stderr, "Sector size %d incompatible "
527 				    "with underlying device sector size %d\n",
528 				    ssopt, sectsz);
529 				goto err;
530 			}
531 		}
532 
533 		sectsz = ssopt;
534 		psectsz = pssopt;
535 		psectoff = 0;
536 	}
537 
538 	bc = calloc(1, sizeof(struct blockif_ctxt));
539 	if (bc == NULL) {
540 		perror("calloc");
541 		goto err;
542 	}
543 
544 	bc->bc_magic = BLOCKIF_SIG;
545 	bc->bc_fd = fd;
546 	bc->bc_ischr = S_ISCHR(sbuf.st_mode);
547 	bc->bc_isgeom = geom;
548 	bc->bc_candelete = candelete;
549 	bc->bc_rdonly = ro;
550 	bc->bc_size = size;
551 	bc->bc_sectsz = sectsz;
552 	bc->bc_psectsz = psectsz;
553 	bc->bc_psectoff = psectoff;
554 	pthread_mutex_init(&bc->bc_mtx, NULL);
555 	pthread_cond_init(&bc->bc_cond, NULL);
556 	TAILQ_INIT(&bc->bc_freeq);
557 	TAILQ_INIT(&bc->bc_pendq);
558 	TAILQ_INIT(&bc->bc_busyq);
559 	for (i = 0; i < BLOCKIF_MAXREQ; i++) {
560 		bc->bc_reqs[i].be_status = BST_FREE;
561 		TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
562 	}
563 
564 	for (i = 0; i < BLOCKIF_NUMTHR; i++) {
565 		pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
566 		snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
567 		pthread_set_name_np(bc->bc_btid[i], tname);
568 	}
569 
570 	return (bc);
571 err:
572 	if (fd >= 0)
573 		close(fd);
574 	return (NULL);
575 }
576 
577 static int
578 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
579 		enum blockop op)
580 {
581 	int err;
582 
583 	err = 0;
584 
585 	pthread_mutex_lock(&bc->bc_mtx);
586 	if (!TAILQ_EMPTY(&bc->bc_freeq)) {
587 		/*
588 		 * Enqueue and inform the block i/o thread
589 		 * that there is work available
590 		 */
591 		if (blockif_enqueue(bc, breq, op))
592 			pthread_cond_signal(&bc->bc_cond);
593 	} else {
594 		/*
595 		 * Callers are not allowed to enqueue more than
596 		 * the specified blockif queue limit. Return an
597 		 * error to indicate that the queue length has been
598 		 * exceeded.
599 		 */
600 		err = E2BIG;
601 	}
602 	pthread_mutex_unlock(&bc->bc_mtx);
603 
604 	return (err);
605 }
606 
607 int
608 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
609 {
610 
611 	assert(bc->bc_magic == BLOCKIF_SIG);
612 	return (blockif_request(bc, breq, BOP_READ));
613 }
614 
615 int
616 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
617 {
618 
619 	assert(bc->bc_magic == BLOCKIF_SIG);
620 	return (blockif_request(bc, breq, BOP_WRITE));
621 }
622 
623 int
624 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
625 {
626 
627 	assert(bc->bc_magic == BLOCKIF_SIG);
628 	return (blockif_request(bc, breq, BOP_FLUSH));
629 }
630 
631 int
632 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
633 {
634 
635 	assert(bc->bc_magic == BLOCKIF_SIG);
636 	return (blockif_request(bc, breq, BOP_DELETE));
637 }
638 
639 int
640 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
641 {
642 	struct blockif_elem *be;
643 
644 	assert(bc->bc_magic == BLOCKIF_SIG);
645 
646 	pthread_mutex_lock(&bc->bc_mtx);
647 	/*
648 	 * Check pending requests.
649 	 */
650 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
651 		if (be->be_req == breq)
652 			break;
653 	}
654 	if (be != NULL) {
655 		/*
656 		 * Found it.
657 		 */
658 		blockif_complete(bc, be);
659 		pthread_mutex_unlock(&bc->bc_mtx);
660 
661 		return (0);
662 	}
663 
664 	/*
665 	 * Check in-flight requests.
666 	 */
667 	TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
668 		if (be->be_req == breq)
669 			break;
670 	}
671 	if (be == NULL) {
672 		/*
673 		 * Didn't find it.
674 		 */
675 		pthread_mutex_unlock(&bc->bc_mtx);
676 		return (EINVAL);
677 	}
678 
679 	/*
680 	 * Interrupt the processing thread to force it return
681 	 * prematurely via it's normal callback path.
682 	 */
683 	while (be->be_status == BST_BUSY) {
684 		struct blockif_sig_elem bse, *old_head;
685 
686 		pthread_mutex_init(&bse.bse_mtx, NULL);
687 		pthread_cond_init(&bse.bse_cond, NULL);
688 
689 		bse.bse_pending = 1;
690 
691 		do {
692 			old_head = blockif_bse_head;
693 			bse.bse_next = old_head;
694 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
695 					    (uintptr_t)old_head,
696 					    (uintptr_t)&bse));
697 
698 		pthread_kill(be->be_tid, SIGCONT);
699 
700 		pthread_mutex_lock(&bse.bse_mtx);
701 		while (bse.bse_pending)
702 			pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
703 		pthread_mutex_unlock(&bse.bse_mtx);
704 	}
705 
706 	pthread_mutex_unlock(&bc->bc_mtx);
707 
708 	/*
709 	 * The processing thread has been interrupted.  Since it's not
710 	 * clear if the callback has been invoked yet, return EBUSY.
711 	 */
712 	return (EBUSY);
713 }
714 
715 int
716 blockif_close(struct blockif_ctxt *bc)
717 {
718 	void *jval;
719 	int i;
720 
721 	assert(bc->bc_magic == BLOCKIF_SIG);
722 
723 	/*
724 	 * Stop the block i/o thread
725 	 */
726 	pthread_mutex_lock(&bc->bc_mtx);
727 	bc->bc_closing = 1;
728 	pthread_mutex_unlock(&bc->bc_mtx);
729 	pthread_cond_broadcast(&bc->bc_cond);
730 	for (i = 0; i < BLOCKIF_NUMTHR; i++)
731 		pthread_join(bc->bc_btid[i], &jval);
732 
733 	/* XXX Cancel queued i/o's ??? */
734 
735 	/*
736 	 * Release resources
737 	 */
738 	bc->bc_magic = 0;
739 	close(bc->bc_fd);
740 	free(bc);
741 
742 	return (0);
743 }
744 
745 /*
746  * Return virtual C/H/S values for a given block. Use the algorithm
747  * outlined in the VHD specification to calculate values.
748  */
749 void
750 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
751 {
752 	off_t sectors;		/* total sectors of the block dev */
753 	off_t hcyl;		/* cylinders times heads */
754 	uint16_t secpt;		/* sectors per track */
755 	uint8_t heads;
756 
757 	assert(bc->bc_magic == BLOCKIF_SIG);
758 
759 	sectors = bc->bc_size / bc->bc_sectsz;
760 
761 	/* Clamp the size to the largest possible with CHS */
762 	if (sectors > 65535UL*16*255)
763 		sectors = 65535UL*16*255;
764 
765 	if (sectors >= 65536UL*16*63) {
766 		secpt = 255;
767 		heads = 16;
768 		hcyl = sectors / secpt;
769 	} else {
770 		secpt = 17;
771 		hcyl = sectors / secpt;
772 		heads = (hcyl + 1023) / 1024;
773 
774 		if (heads < 4)
775 			heads = 4;
776 
777 		if (hcyl >= (heads * 1024) || heads > 16) {
778 			secpt = 31;
779 			heads = 16;
780 			hcyl = sectors / secpt;
781 		}
782 		if (hcyl >= (heads * 1024)) {
783 			secpt = 63;
784 			heads = 16;
785 			hcyl = sectors / secpt;
786 		}
787 	}
788 
789 	*c = hcyl / heads;
790 	*h = heads;
791 	*s = secpt;
792 }
793 
794 /*
795  * Accessors
796  */
797 off_t
798 blockif_size(struct blockif_ctxt *bc)
799 {
800 
801 	assert(bc->bc_magic == BLOCKIF_SIG);
802 	return (bc->bc_size);
803 }
804 
805 int
806 blockif_sectsz(struct blockif_ctxt *bc)
807 {
808 
809 	assert(bc->bc_magic == BLOCKIF_SIG);
810 	return (bc->bc_sectsz);
811 }
812 
813 void
814 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
815 {
816 
817 	assert(bc->bc_magic == BLOCKIF_SIG);
818 	*size = bc->bc_psectsz;
819 	*off = bc->bc_psectoff;
820 }
821 
822 int
823 blockif_queuesz(struct blockif_ctxt *bc)
824 {
825 
826 	assert(bc->bc_magic == BLOCKIF_SIG);
827 	return (BLOCKIF_MAXREQ - 1);
828 }
829 
830 int
831 blockif_is_ro(struct blockif_ctxt *bc)
832 {
833 
834 	assert(bc->bc_magic == BLOCKIF_SIG);
835 	return (bc->bc_rdonly);
836 }
837 
838 int
839 blockif_candelete(struct blockif_ctxt *bc)
840 {
841 
842 	assert(bc->bc_magic == BLOCKIF_SIG);
843 	return (bc->bc_candelete);
844 }
845