xref: /freebsd/usr.sbin/bhyve/block_if.c (revision 4d65a7c6951cea0333f1a0c1b32c38489cdfa6c5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
30 #include <sys/param.h>
31 #ifndef WITHOUT_CAPSICUM
32 #include <sys/capsicum.h>
33 #endif
34 #include <sys/queue.h>
35 #include <sys/errno.h>
36 #include <sys/stat.h>
37 #include <sys/ioctl.h>
38 #include <sys/disk.h>
39 
40 #include <assert.h>
41 #ifndef WITHOUT_CAPSICUM
42 #include <capsicum_helpers.h>
43 #endif
44 #include <err.h>
45 #include <fcntl.h>
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <string.h>
49 #include <pthread.h>
50 #include <pthread_np.h>
51 #include <signal.h>
52 #include <sysexits.h>
53 #include <unistd.h>
54 
55 #include <machine/atomic.h>
56 #include <machine/vmm_snapshot.h>
57 
58 #include "bhyverun.h"
59 #include "config.h"
60 #include "debug.h"
61 #include "mevent.h"
62 #include "pci_emul.h"
63 #include "block_if.h"
64 
65 #define BLOCKIF_SIG	0xb109b109
66 
67 #define BLOCKIF_NUMTHR	8
68 #define BLOCKIF_MAXREQ	(BLOCKIF_RING_MAX + BLOCKIF_NUMTHR)
69 
70 enum blockop {
71 	BOP_READ,
72 	BOP_WRITE,
73 	BOP_FLUSH,
74 	BOP_DELETE
75 };
76 
77 enum blockstat {
78 	BST_FREE,
79 	BST_BLOCK,
80 	BST_PEND,
81 	BST_BUSY,
82 	BST_DONE
83 };
84 
85 struct blockif_elem {
86 	TAILQ_ENTRY(blockif_elem) be_link;
87 	struct blockif_req  *be_req;
88 	enum blockop	     be_op;
89 	enum blockstat	     be_status;
90 	pthread_t            be_tid;
91 	off_t		     be_block;
92 };
93 
94 struct blockif_ctxt {
95 	unsigned int		bc_magic;
96 	int			bc_fd;
97 	int			bc_ischr;
98 	int			bc_isgeom;
99 	int			bc_candelete;
100 	int			bc_rdonly;
101 	off_t			bc_size;
102 	int			bc_sectsz;
103 	int			bc_psectsz;
104 	int			bc_psectoff;
105 	int			bc_closing;
106 	int			bc_paused;
107 	pthread_t		bc_btid[BLOCKIF_NUMTHR];
108 	pthread_mutex_t		bc_mtx;
109 	pthread_cond_t		bc_cond;
110 	pthread_cond_t		bc_work_done_cond;
111 	blockif_resize_cb	*bc_resize_cb;
112 	void			*bc_resize_cb_arg;
113 	struct mevent		*bc_resize_event;
114 
115 	/* Request elements and free/pending/busy queues */
116 	TAILQ_HEAD(, blockif_elem) bc_freeq;
117 	TAILQ_HEAD(, blockif_elem) bc_pendq;
118 	TAILQ_HEAD(, blockif_elem) bc_busyq;
119 	struct blockif_elem	bc_reqs[BLOCKIF_MAXREQ];
120 	int			bc_bootindex;
121 };
122 
123 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
124 
125 struct blockif_sig_elem {
126 	pthread_mutex_t			bse_mtx;
127 	pthread_cond_t			bse_cond;
128 	int				bse_pending;
129 	struct blockif_sig_elem		*bse_next;
130 };
131 
132 static struct blockif_sig_elem *blockif_bse_head;
133 
134 static int
blockif_enqueue(struct blockif_ctxt * bc,struct blockif_req * breq,enum blockop op)135 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
136 		enum blockop op)
137 {
138 	struct blockif_elem *be, *tbe;
139 	off_t off;
140 	int i;
141 
142 	be = TAILQ_FIRST(&bc->bc_freeq);
143 	assert(be != NULL);
144 	assert(be->be_status == BST_FREE);
145 	TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
146 	be->be_req = breq;
147 	be->be_op = op;
148 	switch (op) {
149 	case BOP_READ:
150 	case BOP_WRITE:
151 	case BOP_DELETE:
152 		off = breq->br_offset;
153 		for (i = 0; i < breq->br_iovcnt; i++)
154 			off += breq->br_iov[i].iov_len;
155 		break;
156 	default:
157 		off = OFF_MAX;
158 	}
159 	be->be_block = off;
160 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
161 		if (tbe->be_block == breq->br_offset)
162 			break;
163 	}
164 	if (tbe == NULL) {
165 		TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
166 			if (tbe->be_block == breq->br_offset)
167 				break;
168 		}
169 	}
170 	if (tbe == NULL)
171 		be->be_status = BST_PEND;
172 	else
173 		be->be_status = BST_BLOCK;
174 	TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
175 	return (be->be_status == BST_PEND);
176 }
177 
178 static int
blockif_dequeue(struct blockif_ctxt * bc,pthread_t t,struct blockif_elem ** bep)179 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
180 {
181 	struct blockif_elem *be;
182 
183 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
184 		if (be->be_status == BST_PEND)
185 			break;
186 		assert(be->be_status == BST_BLOCK);
187 	}
188 	if (be == NULL)
189 		return (0);
190 	TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
191 	be->be_status = BST_BUSY;
192 	be->be_tid = t;
193 	TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
194 	*bep = be;
195 	return (1);
196 }
197 
198 static void
blockif_complete(struct blockif_ctxt * bc,struct blockif_elem * be)199 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
200 {
201 	struct blockif_elem *tbe;
202 
203 	if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
204 		TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
205 	else
206 		TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
207 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
208 		if (tbe->be_req->br_offset == be->be_block)
209 			tbe->be_status = BST_PEND;
210 	}
211 	be->be_tid = 0;
212 	be->be_status = BST_FREE;
213 	be->be_req = NULL;
214 	TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
215 }
216 
217 static int
blockif_flush_bc(struct blockif_ctxt * bc)218 blockif_flush_bc(struct blockif_ctxt *bc)
219 {
220 	if (bc->bc_ischr) {
221 		if (ioctl(bc->bc_fd, DIOCGFLUSH))
222 			return (errno);
223 	} else if (fsync(bc->bc_fd))
224 		return (errno);
225 
226 	return (0);
227 }
228 
229 static void
blockif_proc(struct blockif_ctxt * bc,struct blockif_elem * be,uint8_t * buf)230 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
231 {
232 	struct spacectl_range range;
233 	struct blockif_req *br;
234 	off_t arg[2];
235 	ssize_t n;
236 	size_t clen, len, off, boff, voff;
237 	int i, err;
238 
239 	br = be->be_req;
240 	assert(br->br_resid >= 0);
241 
242 	if (br->br_iovcnt <= 1)
243 		buf = NULL;
244 	err = 0;
245 	switch (be->be_op) {
246 	case BOP_READ:
247 		if (buf == NULL) {
248 			if ((n = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
249 			    br->br_offset)) < 0)
250 				err = errno;
251 			else
252 				br->br_resid -= n;
253 			break;
254 		}
255 		i = 0;
256 		off = voff = 0;
257 		while (br->br_resid > 0) {
258 			len = MIN(br->br_resid, MAXPHYS);
259 			n = pread(bc->bc_fd, buf, len, br->br_offset + off);
260 			if (n < 0) {
261 				err = errno;
262 				break;
263 			}
264 			len = (size_t)n;
265 			boff = 0;
266 			do {
267 				clen = MIN(len - boff, br->br_iov[i].iov_len -
268 				    voff);
269 				memcpy((uint8_t *)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 ((n = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
290 			    br->br_offset)) < 0)
291 				err = errno;
292 			else
293 				br->br_resid -= n;
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 				    (uint8_t *)br->br_iov[i].iov_base + voff,
306 				    clen);
307 				if (clen < br->br_iov[i].iov_len - voff)
308 					voff += clen;
309 				else {
310 					i++;
311 					voff = 0;
312 				}
313 				boff += clen;
314 			} while (boff < len);
315 
316 			n = pwrite(bc->bc_fd, buf, len, br->br_offset + off);
317 			if (n < 0) {
318 				err = errno;
319 				break;
320 			}
321 			off += n;
322 			br->br_resid -= n;
323 		}
324 		break;
325 	case BOP_FLUSH:
326 		err = blockif_flush_bc(bc);
327 		break;
328 	case BOP_DELETE:
329 		if (!bc->bc_candelete)
330 			err = EOPNOTSUPP;
331 		else if (bc->bc_rdonly)
332 			err = EROFS;
333 		else if (bc->bc_ischr) {
334 			arg[0] = br->br_offset;
335 			arg[1] = br->br_resid;
336 			if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
337 				err = errno;
338 			else
339 				br->br_resid = 0;
340 		} else {
341 			range.r_offset = br->br_offset;
342 			range.r_len = br->br_resid;
343 
344 			while (range.r_len > 0) {
345 				if (fspacectl(bc->bc_fd, SPACECTL_DEALLOC,
346 				    &range, 0, &range) != 0) {
347 					err = errno;
348 					break;
349 				}
350 			}
351 			if (err == 0)
352 				br->br_resid = 0;
353 		}
354 		break;
355 	default:
356 		err = EINVAL;
357 		break;
358 	}
359 
360 	be->be_status = BST_DONE;
361 
362 	(*br->br_callback)(br, err);
363 }
364 
365 static inline bool
blockif_empty(const struct blockif_ctxt * bc)366 blockif_empty(const struct blockif_ctxt *bc)
367 {
368 	return (TAILQ_EMPTY(&bc->bc_pendq) && TAILQ_EMPTY(&bc->bc_busyq));
369 }
370 
371 static void *
blockif_thr(void * arg)372 blockif_thr(void *arg)
373 {
374 	struct blockif_ctxt *bc;
375 	struct blockif_elem *be;
376 	pthread_t t;
377 	uint8_t *buf;
378 
379 	bc = arg;
380 	if (bc->bc_isgeom)
381 		buf = malloc(MAXPHYS);
382 	else
383 		buf = NULL;
384 	t = pthread_self();
385 
386 	pthread_mutex_lock(&bc->bc_mtx);
387 	for (;;) {
388 		while (blockif_dequeue(bc, t, &be)) {
389 			pthread_mutex_unlock(&bc->bc_mtx);
390 			blockif_proc(bc, be, buf);
391 			pthread_mutex_lock(&bc->bc_mtx);
392 			blockif_complete(bc, be);
393 		}
394 
395 		/* If none to work, notify the main thread */
396 		if (blockif_empty(bc))
397 			pthread_cond_broadcast(&bc->bc_work_done_cond);
398 
399 		/* Check ctxt status here to see if exit requested */
400 		if (bc->bc_closing)
401 			break;
402 
403 		pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
404 	}
405 	pthread_mutex_unlock(&bc->bc_mtx);
406 
407 	if (buf)
408 		free(buf);
409 	pthread_exit(NULL);
410 	return (NULL);
411 }
412 
413 static void
blockif_sigcont_handler(int signal __unused,enum ev_type type __unused,void * arg __unused)414 blockif_sigcont_handler(int signal __unused, enum ev_type type __unused,
415     void *arg __unused)
416 {
417 	struct blockif_sig_elem *bse;
418 
419 	for (;;) {
420 		/*
421 		 * Process the entire list even if not intended for
422 		 * this thread.
423 		 */
424 		do {
425 			bse = blockif_bse_head;
426 			if (bse == NULL)
427 				return;
428 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
429 					    (uintptr_t)bse,
430 					    (uintptr_t)bse->bse_next));
431 
432 		pthread_mutex_lock(&bse->bse_mtx);
433 		bse->bse_pending = 0;
434 		pthread_cond_signal(&bse->bse_cond);
435 		pthread_mutex_unlock(&bse->bse_mtx);
436 	}
437 }
438 
439 static void
blockif_init(void)440 blockif_init(void)
441 {
442 	mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
443 	(void) signal(SIGCONT, SIG_IGN);
444 }
445 
446 int
blockif_legacy_config(nvlist_t * nvl,const char * opts)447 blockif_legacy_config(nvlist_t *nvl, const char *opts)
448 {
449 	char *cp, *path;
450 
451 	if (opts == NULL)
452 		return (0);
453 
454 	cp = strchr(opts, ',');
455 	if (cp == NULL) {
456 		set_config_value_node(nvl, "path", opts);
457 		return (0);
458 	}
459 	path = strndup(opts, cp - opts);
460 	set_config_value_node(nvl, "path", path);
461 	free(path);
462 	return (pci_parse_legacy_config(nvl, cp + 1));
463 }
464 
465 int
blockif_add_boot_device(struct pci_devinst * const pi,struct blockif_ctxt * const bc)466 blockif_add_boot_device(struct pci_devinst *const pi,
467     struct blockif_ctxt *const bc)
468 {
469 	if (bc->bc_bootindex < 0)
470 		return (0);
471 
472 	return (pci_emul_add_boot_device(pi, bc->bc_bootindex));
473 }
474 
475 struct blockif_ctxt *
blockif_open(nvlist_t * nvl,const char * ident)476 blockif_open(nvlist_t *nvl, const char *ident)
477 {
478 	char tname[MAXCOMLEN + 1];
479 	char name[MAXPATHLEN];
480 	const char *path, *pssval, *ssval, *bootindex_val;
481 	char *cp;
482 	struct blockif_ctxt *bc;
483 	struct stat sbuf;
484 	struct diocgattr_arg arg;
485 	off_t size, psectsz, psectoff;
486 	int extra, fd, i, sectsz;
487 	int ro, candelete, geom, ssopt, pssopt;
488 	int nodelete;
489 	int bootindex;
490 
491 #ifndef WITHOUT_CAPSICUM
492 	cap_rights_t rights;
493 	cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE, DIOCGMEDIASIZE };
494 #endif
495 
496 	pthread_once(&blockif_once, blockif_init);
497 
498 	fd = -1;
499 	extra = 0;
500 	ssopt = 0;
501 	ro = 0;
502 	nodelete = 0;
503 	bootindex = -1;
504 
505 	if (get_config_bool_node_default(nvl, "nocache", false))
506 		extra |= O_DIRECT;
507 	if (get_config_bool_node_default(nvl, "nodelete", false))
508 		nodelete = 1;
509 	if (get_config_bool_node_default(nvl, "sync", false) ||
510 	    get_config_bool_node_default(nvl, "direct", false))
511 		extra |= O_SYNC;
512 	if (get_config_bool_node_default(nvl, "ro", false))
513 		ro = 1;
514 	ssval = get_config_value_node(nvl, "sectorsize");
515 	if (ssval != NULL) {
516 		ssopt = strtol(ssval, &cp, 10);
517 		if (cp == ssval) {
518 			EPRINTLN("Invalid sector size \"%s\"", ssval);
519 			goto err;
520 		}
521 		if (*cp == '\0') {
522 			pssopt = ssopt;
523 		} else if (*cp == '/') {
524 			pssval = cp + 1;
525 			pssopt = strtol(pssval, &cp, 10);
526 			if (cp == pssval || *cp != '\0') {
527 				EPRINTLN("Invalid sector size \"%s\"", ssval);
528 				goto err;
529 			}
530 		} else {
531 			EPRINTLN("Invalid sector size \"%s\"", ssval);
532 			goto err;
533 		}
534 	}
535 
536 	bootindex_val = get_config_value_node(nvl, "bootindex");
537 	if (bootindex_val != NULL) {
538 		bootindex = atoi(bootindex_val);
539 	}
540 
541 	path = get_config_value_node(nvl, "path");
542 	if (path == NULL) {
543 		EPRINTLN("Missing \"path\" for block device.");
544 		goto err;
545 	}
546 
547 	fd = open(path, (ro ? O_RDONLY : O_RDWR) | extra);
548 	if (fd < 0 && !ro) {
549 		/* Attempt a r/w fail with a r/o open */
550 		fd = open(path, O_RDONLY | extra);
551 		ro = 1;
552 	}
553 
554 	if (fd < 0) {
555 		warn("Could not open backing file: %s", path);
556 		goto err;
557 	}
558 
559         if (fstat(fd, &sbuf) < 0) {
560 		warn("Could not stat backing file %s", path);
561 		goto err;
562         }
563 
564 #ifndef WITHOUT_CAPSICUM
565 	cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK,
566 	    CAP_WRITE, CAP_FSTAT, CAP_EVENT, CAP_FPATHCONF);
567 	if (ro)
568 		cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE);
569 
570 	if (caph_rights_limit(fd, &rights) == -1)
571 		errx(EX_OSERR, "Unable to apply rights for sandbox");
572 #endif
573 
574         /*
575 	 * Deal with raw devices
576 	 */
577         size = sbuf.st_size;
578 	sectsz = DEV_BSIZE;
579 	psectsz = psectoff = 0;
580 	candelete = geom = 0;
581 	if (S_ISCHR(sbuf.st_mode)) {
582 		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
583 		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
584 			perror("Could not fetch dev blk/sector size");
585 			goto err;
586 		}
587 		assert(size != 0);
588 		assert(sectsz != 0);
589 		if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
590 			ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
591 		strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
592 		arg.len = sizeof(arg.value.i);
593 		if (nodelete == 0 && ioctl(fd, DIOCGATTR, &arg) == 0)
594 			candelete = arg.value.i;
595 		if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0)
596 			geom = 1;
597 	} else {
598 		psectsz = sbuf.st_blksize;
599 		/* Avoid fallback implementation */
600 		candelete = fpathconf(fd, _PC_DEALLOC_PRESENT) == 1;
601 	}
602 
603 #ifndef WITHOUT_CAPSICUM
604 	if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1)
605 		errx(EX_OSERR, "Unable to apply rights for sandbox");
606 #endif
607 
608 	if (ssopt != 0) {
609 		if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
610 		    ssopt > pssopt) {
611 			EPRINTLN("Invalid sector size %d/%d",
612 			    ssopt, pssopt);
613 			goto err;
614 		}
615 
616 		/*
617 		 * Some backend drivers (e.g. cd0, ada0) require that the I/O
618 		 * size be a multiple of the device's sector size.
619 		 *
620 		 * Validate that the emulated sector size complies with this
621 		 * requirement.
622 		 */
623 		if (S_ISCHR(sbuf.st_mode)) {
624 			if (ssopt < sectsz || (ssopt % sectsz) != 0) {
625 				EPRINTLN("Sector size %d incompatible "
626 				    "with underlying device sector size %d",
627 				    ssopt, sectsz);
628 				goto err;
629 			}
630 		}
631 
632 		sectsz = ssopt;
633 		psectsz = pssopt;
634 		psectoff = 0;
635 	}
636 
637 	bc = calloc(1, sizeof(struct blockif_ctxt));
638 	if (bc == NULL) {
639 		perror("calloc");
640 		goto err;
641 	}
642 
643 	bc->bc_magic = BLOCKIF_SIG;
644 	bc->bc_fd = fd;
645 	bc->bc_ischr = S_ISCHR(sbuf.st_mode);
646 	bc->bc_isgeom = geom;
647 	bc->bc_candelete = candelete;
648 	bc->bc_rdonly = ro;
649 	bc->bc_size = size;
650 	bc->bc_sectsz = sectsz;
651 	bc->bc_psectsz = psectsz;
652 	bc->bc_psectoff = psectoff;
653 	pthread_mutex_init(&bc->bc_mtx, NULL);
654 	pthread_cond_init(&bc->bc_cond, NULL);
655 	bc->bc_paused = 0;
656 	pthread_cond_init(&bc->bc_work_done_cond, NULL);
657 	TAILQ_INIT(&bc->bc_freeq);
658 	TAILQ_INIT(&bc->bc_pendq);
659 	TAILQ_INIT(&bc->bc_busyq);
660 	bc->bc_bootindex = bootindex;
661 	for (i = 0; i < BLOCKIF_MAXREQ; i++) {
662 		bc->bc_reqs[i].be_status = BST_FREE;
663 		TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
664 	}
665 
666 	for (i = 0; i < BLOCKIF_NUMTHR; i++) {
667 		pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
668 		snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
669 		pthread_set_name_np(bc->bc_btid[i], tname);
670 	}
671 
672 	return (bc);
673 err:
674 	if (fd >= 0)
675 		close(fd);
676 	return (NULL);
677 }
678 
679 static void
blockif_resized(int fd,enum ev_type type __unused,void * arg)680 blockif_resized(int fd, enum ev_type type __unused, void *arg)
681 {
682 	struct blockif_ctxt *bc;
683 	struct stat sb;
684 	off_t mediasize;
685 
686 	if (fstat(fd, &sb) != 0)
687 		return;
688 
689 	if (S_ISCHR(sb.st_mode)) {
690 		if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) < 0) {
691 			EPRINTLN("blockif_resized: get mediasize failed: %s",
692 			    strerror(errno));
693 			return;
694 		}
695 	} else
696 		mediasize = sb.st_size;
697 
698 	bc = arg;
699 	pthread_mutex_lock(&bc->bc_mtx);
700 	if (mediasize != bc->bc_size) {
701 		bc->bc_size = mediasize;
702 		bc->bc_resize_cb(bc, bc->bc_resize_cb_arg, bc->bc_size);
703 	}
704 	pthread_mutex_unlock(&bc->bc_mtx);
705 }
706 
707 int
blockif_register_resize_callback(struct blockif_ctxt * bc,blockif_resize_cb * cb,void * cb_arg)708 blockif_register_resize_callback(struct blockif_ctxt *bc, blockif_resize_cb *cb,
709     void *cb_arg)
710 {
711 	struct stat sb;
712 	int err;
713 
714 	if (cb == NULL)
715 		return (EINVAL);
716 
717 	err = 0;
718 
719 	pthread_mutex_lock(&bc->bc_mtx);
720 	if (bc->bc_resize_cb != NULL) {
721 		err = EBUSY;
722 		goto out;
723 	}
724 
725 	assert(bc->bc_closing == 0);
726 
727 	if (fstat(bc->bc_fd, &sb) != 0) {
728 		err = errno;
729 		goto out;
730 	}
731 
732 	bc->bc_resize_event = mevent_add_flags(bc->bc_fd, EVF_VNODE,
733 	    EVFF_ATTRIB, blockif_resized, bc);
734 	if (bc->bc_resize_event == NULL) {
735 		err = ENXIO;
736 		goto out;
737 	}
738 
739 	bc->bc_resize_cb = cb;
740 	bc->bc_resize_cb_arg = cb_arg;
741 out:
742 	pthread_mutex_unlock(&bc->bc_mtx);
743 
744 	return (err);
745 }
746 
747 static int
blockif_request(struct blockif_ctxt * bc,struct blockif_req * breq,enum blockop op)748 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
749 		enum blockop op)
750 {
751 	int err;
752 
753 	err = 0;
754 
755 	pthread_mutex_lock(&bc->bc_mtx);
756 	assert(!bc->bc_paused);
757 	if (!TAILQ_EMPTY(&bc->bc_freeq)) {
758 		/*
759 		 * Enqueue and inform the block i/o thread
760 		 * that there is work available
761 		 */
762 		if (blockif_enqueue(bc, breq, op))
763 			pthread_cond_signal(&bc->bc_cond);
764 	} else {
765 		/*
766 		 * Callers are not allowed to enqueue more than
767 		 * the specified blockif queue limit. Return an
768 		 * error to indicate that the queue length has been
769 		 * exceeded.
770 		 */
771 		err = E2BIG;
772 	}
773 	pthread_mutex_unlock(&bc->bc_mtx);
774 
775 	return (err);
776 }
777 
778 int
blockif_read(struct blockif_ctxt * bc,struct blockif_req * breq)779 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
780 {
781 	assert(bc->bc_magic == BLOCKIF_SIG);
782 	return (blockif_request(bc, breq, BOP_READ));
783 }
784 
785 int
blockif_write(struct blockif_ctxt * bc,struct blockif_req * breq)786 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
787 {
788 	assert(bc->bc_magic == BLOCKIF_SIG);
789 	return (blockif_request(bc, breq, BOP_WRITE));
790 }
791 
792 int
blockif_flush(struct blockif_ctxt * bc,struct blockif_req * breq)793 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
794 {
795 	assert(bc->bc_magic == BLOCKIF_SIG);
796 	return (blockif_request(bc, breq, BOP_FLUSH));
797 }
798 
799 int
blockif_delete(struct blockif_ctxt * bc,struct blockif_req * breq)800 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
801 {
802 	assert(bc->bc_magic == BLOCKIF_SIG);
803 	return (blockif_request(bc, breq, BOP_DELETE));
804 }
805 
806 int
blockif_cancel(struct blockif_ctxt * bc,struct blockif_req * breq)807 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
808 {
809 	struct blockif_elem *be;
810 
811 	assert(bc->bc_magic == BLOCKIF_SIG);
812 
813 	pthread_mutex_lock(&bc->bc_mtx);
814 	/* XXX: not waiting while paused */
815 
816 	/*
817 	 * Check pending requests.
818 	 */
819 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
820 		if (be->be_req == breq)
821 			break;
822 	}
823 	if (be != NULL) {
824 		/*
825 		 * Found it.
826 		 */
827 		blockif_complete(bc, be);
828 		pthread_mutex_unlock(&bc->bc_mtx);
829 
830 		return (0);
831 	}
832 
833 	/*
834 	 * Check in-flight requests.
835 	 */
836 	TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
837 		if (be->be_req == breq)
838 			break;
839 	}
840 	if (be == NULL) {
841 		/*
842 		 * Didn't find it.
843 		 */
844 		pthread_mutex_unlock(&bc->bc_mtx);
845 		return (EINVAL);
846 	}
847 
848 	/*
849 	 * Interrupt the processing thread to force it return
850 	 * prematurely via it's normal callback path.
851 	 */
852 	while (be->be_status == BST_BUSY) {
853 		struct blockif_sig_elem bse, *old_head;
854 
855 		pthread_mutex_init(&bse.bse_mtx, NULL);
856 		pthread_cond_init(&bse.bse_cond, NULL);
857 
858 		bse.bse_pending = 1;
859 
860 		do {
861 			old_head = blockif_bse_head;
862 			bse.bse_next = old_head;
863 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
864 					    (uintptr_t)old_head,
865 					    (uintptr_t)&bse));
866 
867 		pthread_kill(be->be_tid, SIGCONT);
868 
869 		pthread_mutex_lock(&bse.bse_mtx);
870 		while (bse.bse_pending)
871 			pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
872 		pthread_mutex_unlock(&bse.bse_mtx);
873 	}
874 
875 	pthread_mutex_unlock(&bc->bc_mtx);
876 
877 	/*
878 	 * The processing thread has been interrupted.  Since it's not
879 	 * clear if the callback has been invoked yet, return EBUSY.
880 	 */
881 	return (EBUSY);
882 }
883 
884 int
blockif_close(struct blockif_ctxt * bc)885 blockif_close(struct blockif_ctxt *bc)
886 {
887 	void *jval;
888 	int i;
889 
890 	assert(bc->bc_magic == BLOCKIF_SIG);
891 
892 	/*
893 	 * Stop the block i/o thread
894 	 */
895 	pthread_mutex_lock(&bc->bc_mtx);
896 	bc->bc_closing = 1;
897 	if (bc->bc_resize_event != NULL)
898 		mevent_disable(bc->bc_resize_event);
899 	pthread_mutex_unlock(&bc->bc_mtx);
900 	pthread_cond_broadcast(&bc->bc_cond);
901 	for (i = 0; i < BLOCKIF_NUMTHR; i++)
902 		pthread_join(bc->bc_btid[i], &jval);
903 
904 	/* XXX Cancel queued i/o's ??? */
905 
906 	/*
907 	 * Release resources
908 	 */
909 	bc->bc_magic = 0;
910 	close(bc->bc_fd);
911 	free(bc);
912 
913 	return (0);
914 }
915 
916 /*
917  * Return virtual C/H/S values for a given block. Use the algorithm
918  * outlined in the VHD specification to calculate values.
919  */
920 void
blockif_chs(struct blockif_ctxt * bc,uint16_t * c,uint8_t * h,uint8_t * s)921 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
922 {
923 	off_t sectors;		/* total sectors of the block dev */
924 	off_t hcyl;		/* cylinders times heads */
925 	uint16_t secpt;		/* sectors per track */
926 	uint8_t heads;
927 
928 	assert(bc->bc_magic == BLOCKIF_SIG);
929 
930 	sectors = bc->bc_size / bc->bc_sectsz;
931 
932 	/* Clamp the size to the largest possible with CHS */
933 	if (sectors > 65535L * 16 * 255)
934 		sectors = 65535L * 16 * 255;
935 
936 	if (sectors >= 65536L * 16 * 63) {
937 		secpt = 255;
938 		heads = 16;
939 		hcyl = sectors / secpt;
940 	} else {
941 		secpt = 17;
942 		hcyl = sectors / secpt;
943 		heads = (hcyl + 1023) / 1024;
944 
945 		if (heads < 4)
946 			heads = 4;
947 
948 		if (hcyl >= (heads * 1024) || heads > 16) {
949 			secpt = 31;
950 			heads = 16;
951 			hcyl = sectors / secpt;
952 		}
953 		if (hcyl >= (heads * 1024)) {
954 			secpt = 63;
955 			heads = 16;
956 			hcyl = sectors / secpt;
957 		}
958 	}
959 
960 	*c = hcyl / heads;
961 	*h = heads;
962 	*s = secpt;
963 }
964 
965 /*
966  * Accessors
967  */
968 off_t
blockif_size(struct blockif_ctxt * bc)969 blockif_size(struct blockif_ctxt *bc)
970 {
971 	assert(bc->bc_magic == BLOCKIF_SIG);
972 	return (bc->bc_size);
973 }
974 
975 int
blockif_sectsz(struct blockif_ctxt * bc)976 blockif_sectsz(struct blockif_ctxt *bc)
977 {
978 	assert(bc->bc_magic == BLOCKIF_SIG);
979 	return (bc->bc_sectsz);
980 }
981 
982 void
blockif_psectsz(struct blockif_ctxt * bc,int * size,int * off)983 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
984 {
985 	assert(bc->bc_magic == BLOCKIF_SIG);
986 	*size = bc->bc_psectsz;
987 	*off = bc->bc_psectoff;
988 }
989 
990 int
blockif_queuesz(struct blockif_ctxt * bc)991 blockif_queuesz(struct blockif_ctxt *bc)
992 {
993 	assert(bc->bc_magic == BLOCKIF_SIG);
994 	return (BLOCKIF_MAXREQ - 1);
995 }
996 
997 int
blockif_is_ro(struct blockif_ctxt * bc)998 blockif_is_ro(struct blockif_ctxt *bc)
999 {
1000 	assert(bc->bc_magic == BLOCKIF_SIG);
1001 	return (bc->bc_rdonly);
1002 }
1003 
1004 int
blockif_candelete(struct blockif_ctxt * bc)1005 blockif_candelete(struct blockif_ctxt *bc)
1006 {
1007 	assert(bc->bc_magic == BLOCKIF_SIG);
1008 	return (bc->bc_candelete);
1009 }
1010 
1011 #ifdef BHYVE_SNAPSHOT
1012 void
blockif_pause(struct blockif_ctxt * bc)1013 blockif_pause(struct blockif_ctxt *bc)
1014 {
1015 	assert(bc != NULL);
1016 	assert(bc->bc_magic == BLOCKIF_SIG);
1017 
1018 	pthread_mutex_lock(&bc->bc_mtx);
1019 	bc->bc_paused = 1;
1020 
1021 	/* The interface is paused. Wait for workers to finish their work */
1022 	while (!blockif_empty(bc))
1023 		pthread_cond_wait(&bc->bc_work_done_cond, &bc->bc_mtx);
1024 	pthread_mutex_unlock(&bc->bc_mtx);
1025 
1026 	if (!bc->bc_rdonly && blockif_flush_bc(bc))
1027 		EPRINTLN("%s: [WARN] failed to flush backing file.",
1028 			__func__);
1029 }
1030 
1031 void
blockif_resume(struct blockif_ctxt * bc)1032 blockif_resume(struct blockif_ctxt *bc)
1033 {
1034 	assert(bc != NULL);
1035 	assert(bc->bc_magic == BLOCKIF_SIG);
1036 
1037 	pthread_mutex_lock(&bc->bc_mtx);
1038 	bc->bc_paused = 0;
1039 	pthread_mutex_unlock(&bc->bc_mtx);
1040 }
1041 #endif	/* BHYVE_SNAPSHOT */
1042