xref: /illumos-gate/usr/src/cmd/bhyve/common/block_if.c (revision 5c4a5fe1)
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 /*
31  * Copyright 2020 Joyent, Inc.
32  */
33 
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 #ifndef __FreeBSD__
45 #include <sys/limits.h>
46 #include <sys/uio.h>
47 #include <sys/dkio.h>
48 #endif
49 
50 #include <assert.h>
51 #ifndef WITHOUT_CAPSICUM
52 #include <capsicum_helpers.h>
53 #endif
54 #include <err.h>
55 #include <fcntl.h>
56 #include <stdio.h>
57 #include <stdlib.h>
58 #include <string.h>
59 #include <pthread.h>
60 #include <pthread_np.h>
61 #include <signal.h>
62 #include <sysexits.h>
63 #include <unistd.h>
64 
65 #include <machine/atomic.h>
66 
67 #include "bhyverun.h"
68 #include "config.h"
69 #include "debug.h"
70 #include "mevent.h"
71 #include "pci_emul.h"
72 #include "block_if.h"
73 
74 #define BLOCKIF_SIG	0xb109b109
75 
76 #ifdef __FreeBSD__
77 #define BLOCKIF_NUMTHR	8
78 #else
79 /* Enlarge to keep pace with the virtio-block ring size */
80 #define BLOCKIF_NUMTHR	16
81 #endif
82 #define BLOCKIF_MAXREQ	(BLOCKIF_RING_MAX + BLOCKIF_NUMTHR)
83 
84 enum blockop {
85 	BOP_READ,
86 	BOP_WRITE,
87 #ifndef __FreeBSD__
88 	BOP_WRITE_SYNC,
89 #endif
90 	BOP_FLUSH,
91 	BOP_DELETE
92 };
93 
94 enum blockstat {
95 	BST_FREE,
96 	BST_BLOCK,
97 	BST_PEND,
98 	BST_BUSY,
99 	BST_DONE
100 };
101 
102 struct blockif_elem {
103 	TAILQ_ENTRY(blockif_elem) be_link;
104 	struct blockif_req  *be_req;
105 	enum blockop	     be_op;
106 	enum blockstat	     be_status;
107 	pthread_t            be_tid;
108 	off_t		     be_block;
109 };
110 
111 #ifndef __FreeBSD__
112 enum blockif_wce {
113 	WCE_NONE = 0,
114 	WCE_IOCTL,
115 	WCE_FCNTL
116 };
117 #endif
118 
119 struct blockif_ctxt {
120 	unsigned int		bc_magic;
121 	int			bc_fd;
122 	int			bc_ischr;
123 	int			bc_isgeom;
124 	int			bc_candelete;
125 #ifndef __FreeBSD__
126 	enum blockif_wce	bc_wce;
127 #endif
128 	int			bc_rdonly;
129 	off_t			bc_size;
130 	int			bc_sectsz;
131 	int			bc_psectsz;
132 	int			bc_psectoff;
133 	int			bc_closing;
134 	pthread_t		bc_btid[BLOCKIF_NUMTHR];
135 	pthread_mutex_t		bc_mtx;
136 	pthread_cond_t		bc_cond;
137 	blockif_resize_cb	*bc_resize_cb;
138 	void			*bc_resize_cb_arg;
139 	struct mevent		*bc_resize_event;
140 
141 	/* Request elements and free/pending/busy queues */
142 	TAILQ_HEAD(, blockif_elem) bc_freeq;
143 	TAILQ_HEAD(, blockif_elem) bc_pendq;
144 	TAILQ_HEAD(, blockif_elem) bc_busyq;
145 	struct blockif_elem	bc_reqs[BLOCKIF_MAXREQ];
146 	int			bc_bootindex;
147 };
148 
149 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
150 
151 struct blockif_sig_elem {
152 	pthread_mutex_t			bse_mtx;
153 	pthread_cond_t			bse_cond;
154 	int				bse_pending;
155 	struct blockif_sig_elem		*bse_next;
156 };
157 
158 static struct blockif_sig_elem *blockif_bse_head;
159 
160 static int
blockif_enqueue(struct blockif_ctxt * bc,struct blockif_req * breq,enum blockop op)161 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
162 		enum blockop op)
163 {
164 	struct blockif_elem *be, *tbe;
165 	off_t off;
166 	int i;
167 
168 	be = TAILQ_FIRST(&bc->bc_freeq);
169 	assert(be != NULL);
170 	assert(be->be_status == BST_FREE);
171 	TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
172 	be->be_req = breq;
173 	be->be_op = op;
174 	switch (op) {
175 	case BOP_READ:
176 	case BOP_WRITE:
177 #ifndef __FreeBSD__
178 	case BOP_WRITE_SYNC:
179 #endif
180 	case BOP_DELETE:
181 		off = breq->br_offset;
182 		for (i = 0; i < breq->br_iovcnt; i++)
183 			off += breq->br_iov[i].iov_len;
184 		break;
185 	default:
186 		off = OFF_MAX;
187 	}
188 	be->be_block = off;
189 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
190 		if (tbe->be_block == breq->br_offset)
191 			break;
192 	}
193 	if (tbe == NULL) {
194 		TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
195 			if (tbe->be_block == breq->br_offset)
196 				break;
197 		}
198 	}
199 	if (tbe == NULL)
200 		be->be_status = BST_PEND;
201 	else
202 		be->be_status = BST_BLOCK;
203 	TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
204 	return (be->be_status == BST_PEND);
205 }
206 
207 static int
blockif_dequeue(struct blockif_ctxt * bc,pthread_t t,struct blockif_elem ** bep)208 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
209 {
210 	struct blockif_elem *be;
211 
212 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
213 		if (be->be_status == BST_PEND)
214 			break;
215 		assert(be->be_status == BST_BLOCK);
216 	}
217 	if (be == NULL)
218 		return (0);
219 	TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
220 	be->be_status = BST_BUSY;
221 	be->be_tid = t;
222 	TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
223 	*bep = be;
224 	return (1);
225 }
226 
227 static void
blockif_complete(struct blockif_ctxt * bc,struct blockif_elem * be)228 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
229 {
230 	struct blockif_elem *tbe;
231 
232 	if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
233 		TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
234 	else
235 		TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
236 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
237 		if (tbe->be_req->br_offset == be->be_block)
238 			tbe->be_status = BST_PEND;
239 	}
240 	be->be_tid = 0;
241 	be->be_status = BST_FREE;
242 	be->be_req = NULL;
243 	TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
244 }
245 
246 static int
blockif_flush_bc(struct blockif_ctxt * bc)247 blockif_flush_bc(struct blockif_ctxt *bc)
248 {
249 #ifdef	__FreeBSD__
250 	if (bc->bc_ischr) {
251 		if (ioctl(bc->bc_fd, DIOCGFLUSH))
252 			return (errno);
253 	} else if (fsync(bc->bc_fd))
254 		return (errno);
255 #else
256 	/*
257 	 * This fsync() should be adequate to flush the cache of a file
258 	 * or device.  In VFS, the VOP_SYNC operation is converted to
259 	 * the appropriate ioctl in both sdev (for real devices) and
260 	 * zfs (for zvols).
261 	 */
262 	if (fsync(bc->bc_fd))
263 		return (errno);
264 #endif
265 
266 	return (0);
267 }
268 
269 static void
blockif_proc(struct blockif_ctxt * bc,struct blockif_elem * be,uint8_t * buf)270 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be, uint8_t *buf)
271 {
272 #ifdef	__FreeBSD__
273 	struct spacectl_range range;
274 #endif
275 	struct blockif_req *br;
276 #ifdef	__FreeBSD__
277 	off_t arg[2];
278 #endif
279 	ssize_t n;
280 	size_t clen, len, off, boff, voff;
281 	int i, err;
282 
283 	br = be->be_req;
284 	assert(br->br_resid >= 0);
285 
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 ((n = preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
293 			    br->br_offset)) < 0)
294 				err = errno;
295 			else
296 				br->br_resid -= n;
297 			break;
298 		}
299 		i = 0;
300 		off = voff = 0;
301 		while (br->br_resid > 0) {
302 			len = MIN(br->br_resid, MAXPHYS);
303 			n = pread(bc->bc_fd, buf, len, br->br_offset + off);
304 			if (n < 0) {
305 				err = errno;
306 				break;
307 			}
308 			len = (size_t)n;
309 			boff = 0;
310 			do {
311 				clen = MIN(len - boff, br->br_iov[i].iov_len -
312 				    voff);
313 				memcpy((uint8_t *)br->br_iov[i].iov_base + voff,
314 				    buf + boff, clen);
315 				if (clen < br->br_iov[i].iov_len - voff)
316 					voff += clen;
317 				else {
318 					i++;
319 					voff = 0;
320 				}
321 				boff += clen;
322 			} while (boff < len);
323 			off += len;
324 			br->br_resid -= len;
325 		}
326 		break;
327 	case BOP_WRITE:
328 		if (bc->bc_rdonly) {
329 			err = EROFS;
330 			break;
331 		}
332 		if (buf == NULL) {
333 			if ((n = pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
334 			    br->br_offset)) < 0)
335 				err = errno;
336 			else
337 				br->br_resid -= n;
338 			break;
339 		}
340 		i = 0;
341 		off = voff = 0;
342 		while (br->br_resid > 0) {
343 			len = MIN(br->br_resid, MAXPHYS);
344 			boff = 0;
345 			do {
346 				clen = MIN(len - boff, br->br_iov[i].iov_len -
347 				    voff);
348 				memcpy(buf + boff,
349 				    (uint8_t *)br->br_iov[i].iov_base + voff,
350 				    clen);
351 				if (clen < br->br_iov[i].iov_len - voff)
352 					voff += clen;
353 				else {
354 					i++;
355 					voff = 0;
356 				}
357 				boff += clen;
358 			} while (boff < len);
359 
360 			n = pwrite(bc->bc_fd, buf, len, br->br_offset + off);
361 			if (n < 0) {
362 				err = errno;
363 				break;
364 			}
365 			off += n;
366 			br->br_resid -= n;
367 		}
368 		break;
369 	case BOP_FLUSH:
370 		err = blockif_flush_bc(bc);
371 		break;
372 	case BOP_DELETE:
373 		if (!bc->bc_candelete)
374 			err = EOPNOTSUPP;
375 		else if (bc->bc_rdonly)
376 			err = EROFS;
377 #ifdef	__FreeBSD__
378 		else if (bc->bc_ischr) {
379 			arg[0] = br->br_offset;
380 			arg[1] = br->br_resid;
381 			if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
382 				err = errno;
383 			else
384 				br->br_resid = 0;
385 		} else {
386 			range.r_offset = br->br_offset;
387 			range.r_len = br->br_resid;
388 
389 			while (range.r_len > 0) {
390 				if (fspacectl(bc->bc_fd, SPACECTL_DEALLOC,
391 				    &range, 0, &range) != 0) {
392 					err = errno;
393 					break;
394 				}
395 			}
396 			if (err == 0)
397 				br->br_resid = 0;
398 		}
399 #else
400 		else if (bc->bc_ischr) {
401 			dkioc_free_list_t dfl = {
402 				.dfl_num_exts = 1,
403 				.dfl_offset = 0,
404 				.dfl_flags = 0,
405 				.dfl_exts = {
406 					{
407 						.dfle_start = br->br_offset,
408 						.dfle_length = br->br_resid
409 					}
410 				}
411 			};
412 
413 			if (ioctl(bc->bc_fd, DKIOCFREE, &dfl))
414 				err = errno;
415 			else
416 				br->br_resid = 0;
417 		} else {
418 			struct flock fl = {
419 				.l_whence = 0,
420 				.l_type = F_WRLCK,
421 				.l_start = br->br_offset,
422 				.l_len = br->br_resid
423 			};
424 
425 			if (fcntl(bc->bc_fd, F_FREESP, &fl))
426 				err = errno;
427 			else
428 				br->br_resid = 0;
429 		}
430 #endif
431 		break;
432 	default:
433 		err = EINVAL;
434 		break;
435 	}
436 
437 	be->be_status = BST_DONE;
438 
439 	(*br->br_callback)(br, err);
440 }
441 
442 static inline bool
blockif_empty(const struct blockif_ctxt * bc)443 blockif_empty(const struct blockif_ctxt *bc)
444 {
445 	return (TAILQ_EMPTY(&bc->bc_pendq) && TAILQ_EMPTY(&bc->bc_busyq));
446 }
447 
448 static void *
blockif_thr(void * arg)449 blockif_thr(void *arg)
450 {
451 	struct blockif_ctxt *bc;
452 	struct blockif_elem *be;
453 	pthread_t t;
454 	uint8_t *buf;
455 
456 	bc = arg;
457 	if (bc->bc_isgeom)
458 		buf = malloc(MAXPHYS);
459 	else
460 		buf = NULL;
461 	t = pthread_self();
462 
463 	pthread_mutex_lock(&bc->bc_mtx);
464 	for (;;) {
465 		while (blockif_dequeue(bc, t, &be)) {
466 			pthread_mutex_unlock(&bc->bc_mtx);
467 			blockif_proc(bc, be, buf);
468 			pthread_mutex_lock(&bc->bc_mtx);
469 			blockif_complete(bc, be);
470 		}
471 		/* Check ctxt status here to see if exit requested */
472 		if (bc->bc_closing)
473 			break;
474 
475 		pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
476 	}
477 	pthread_mutex_unlock(&bc->bc_mtx);
478 
479 	if (buf)
480 		free(buf);
481 	pthread_exit(NULL);
482 	return (NULL);
483 }
484 
485 #ifdef	__FreeBSD__
486 static void
blockif_sigcont_handler(int signal __unused,enum ev_type type __unused,void * arg __unused)487 blockif_sigcont_handler(int signal __unused, enum ev_type type __unused,
488     void *arg __unused)
489 #else
490 static void
491 blockif_sigcont_handler(int signal __unused)
492 #endif
493 {
494 	struct blockif_sig_elem *bse;
495 
496 	for (;;) {
497 		/*
498 		 * Process the entire list even if not intended for
499 		 * this thread.
500 		 */
501 		do {
502 			bse = blockif_bse_head;
503 			if (bse == NULL)
504 				return;
505 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
506 					    (uintptr_t)bse,
507 					    (uintptr_t)bse->bse_next));
508 
509 		pthread_mutex_lock(&bse->bse_mtx);
510 		bse->bse_pending = 0;
511 		pthread_cond_signal(&bse->bse_cond);
512 		pthread_mutex_unlock(&bse->bse_mtx);
513 	}
514 }
515 
516 static void
blockif_init(void)517 blockif_init(void)
518 {
519 #ifdef	__FreeBSD__
520 	mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
521 	(void) signal(SIGCONT, SIG_IGN);
522 #else
523 	(void) sigset(SIGCONT, blockif_sigcont_handler);
524 #endif
525 }
526 
527 int
blockif_legacy_config(nvlist_t * nvl,const char * opts)528 blockif_legacy_config(nvlist_t *nvl, const char *opts)
529 {
530 	char *cp, *path;
531 
532 	if (opts == NULL)
533 		return (0);
534 
535 	cp = strchr(opts, ',');
536 	if (cp == NULL) {
537 		set_config_value_node(nvl, "path", opts);
538 		return (0);
539 	}
540 	path = strndup(opts, cp - opts);
541 	set_config_value_node(nvl, "path", path);
542 	free(path);
543 	return (pci_parse_legacy_config(nvl, cp + 1));
544 }
545 
546 int
blockif_add_boot_device(struct pci_devinst * const pi,struct blockif_ctxt * const bc)547 blockif_add_boot_device(struct pci_devinst *const pi,
548     struct blockif_ctxt *const bc)
549 {
550 	if (bc->bc_bootindex < 0)
551 		return (0);
552 
553 	return (pci_emul_add_boot_device(pi, bc->bc_bootindex));
554 }
555 
556 struct blockif_ctxt *
blockif_open(nvlist_t * nvl,const char * ident)557 blockif_open(nvlist_t *nvl, const char *ident)
558 {
559 	char tname[MAXCOMLEN + 1];
560 #ifdef	__FreeBSD__
561 	char name[MAXPATHLEN];
562 #endif
563 	const char *path, *pssval, *ssval, *bootindex_val;
564 	char *cp;
565 	struct blockif_ctxt *bc;
566 	struct stat sbuf;
567 #ifdef	__FreeBSD__
568 	struct diocgattr_arg arg;
569 #else
570 	enum blockif_wce wce = WCE_NONE;
571 #endif
572 	off_t size, psectsz, psectoff;
573 	int extra, fd, i, sectsz;
574 	int ro, candelete, geom, ssopt, pssopt;
575 	int nodelete;
576 	int bootindex;
577 
578 #ifndef WITHOUT_CAPSICUM
579 	cap_rights_t rights;
580 	cap_ioctl_t cmds[] = { DIOCGFLUSH, DIOCGDELETE, DIOCGMEDIASIZE };
581 #endif
582 
583 	pthread_once(&blockif_once, blockif_init);
584 
585 	fd = -1;
586 	extra = 0;
587 	ssopt = 0;
588 #ifndef __FreeBSD__
589 	pssopt = 0;
590 #endif
591 	ro = 0;
592 	nodelete = 0;
593 	bootindex = -1;
594 
595 	if (get_config_bool_node_default(nvl, "nocache", false))
596 		extra |= O_DIRECT;
597 	if (get_config_bool_node_default(nvl, "nodelete", false))
598 		nodelete = 1;
599 	if (get_config_bool_node_default(nvl, "sync", false) ||
600 	    get_config_bool_node_default(nvl, "direct", false))
601 		extra |= O_SYNC;
602 	if (get_config_bool_node_default(nvl, "ro", false))
603 		ro = 1;
604 	ssval = get_config_value_node(nvl, "sectorsize");
605 	if (ssval != NULL) {
606 		ssopt = strtol(ssval, &cp, 10);
607 		if (cp == ssval) {
608 			EPRINTLN("Invalid sector size \"%s\"", ssval);
609 			goto err;
610 		}
611 		if (*cp == '\0') {
612 			pssopt = ssopt;
613 		} else if (*cp == '/') {
614 			pssval = cp + 1;
615 			pssopt = strtol(pssval, &cp, 10);
616 			if (cp == pssval || *cp != '\0') {
617 				EPRINTLN("Invalid sector size \"%s\"", ssval);
618 				goto err;
619 			}
620 		} else {
621 			EPRINTLN("Invalid sector size \"%s\"", ssval);
622 			goto err;
623 		}
624 	}
625 
626 	bootindex_val = get_config_value_node(nvl, "bootindex");
627 	if (bootindex_val != NULL) {
628 		bootindex = atoi(bootindex_val);
629 	}
630 
631 	path = get_config_value_node(nvl, "path");
632 	if (path == NULL) {
633 		EPRINTLN("Missing \"path\" for block device.");
634 		goto err;
635 	}
636 
637 	fd = open(path, (ro ? O_RDONLY : O_RDWR) | extra);
638 	if (fd < 0 && !ro) {
639 		/* Attempt a r/w fail with a r/o open */
640 		fd = open(path, O_RDONLY | extra);
641 		ro = 1;
642 	}
643 
644 	if (fd < 0) {
645 		warn("Could not open backing file: %s", path);
646 		goto err;
647 	}
648 
649         if (fstat(fd, &sbuf) < 0) {
650 		warn("Could not stat backing file %s", path);
651 		goto err;
652         }
653 
654 #ifndef WITHOUT_CAPSICUM
655 	cap_rights_init(&rights, CAP_FSYNC, CAP_IOCTL, CAP_READ, CAP_SEEK,
656 	    CAP_WRITE, CAP_FSTAT, CAP_EVENT, CAP_FPATHCONF);
657 	if (ro)
658 		cap_rights_clear(&rights, CAP_FSYNC, CAP_WRITE);
659 
660 	if (caph_rights_limit(fd, &rights) == -1)
661 		errx(EX_OSERR, "Unable to apply rights for sandbox");
662 #endif
663 
664         /*
665 	 * Deal with raw devices
666 	 */
667         size = sbuf.st_size;
668 	sectsz = DEV_BSIZE;
669 	psectsz = psectoff = 0;
670 	candelete = geom = 0;
671 #ifdef	__FreeBSD__
672 	if (S_ISCHR(sbuf.st_mode)) {
673 		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
674 		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
675 			perror("Could not fetch dev blk/sector size");
676 			goto err;
677 		}
678 		assert(size != 0);
679 		assert(sectsz != 0);
680 		if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
681 			ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
682 		strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
683 		arg.len = sizeof(arg.value.i);
684 		if (nodelete == 0 && ioctl(fd, DIOCGATTR, &arg) == 0)
685 			candelete = arg.value.i;
686 		if (ioctl(fd, DIOCGPROVIDERNAME, name) == 0)
687 			geom = 1;
688 	} else {
689 		psectsz = sbuf.st_blksize;
690 		/* Avoid fallback implementation */
691 		candelete = fpathconf(fd, _PC_DEALLOC_PRESENT) == 1;
692 	}
693 #else
694 	psectsz = sbuf.st_blksize;
695 	if (S_ISCHR(sbuf.st_mode)) {
696 		struct dk_minfo_ext dkmext;
697 		int wce_val;
698 
699 		/* Look for a more accurate physical block/media size */
700 		if (ioctl(fd, DKIOCGMEDIAINFOEXT, &dkmext) == 0) {
701 			psectsz = dkmext.dki_pbsize;
702 			size = dkmext.dki_lbsize * dkmext.dki_capacity;
703 		}
704 		/* See if a configurable write cache is present and working */
705 		if (ioctl(fd, DKIOCGETWCE, &wce_val) == 0) {
706 			/*
707 			 * If WCE is already active, disable it until the
708 			 * specific device driver calls for its return.  If it
709 			 * is not active, toggle it on and off to verify that
710 			 * such actions are possible.
711 			 */
712 			if (wce_val != 0) {
713 				wce_val = 0;
714 				/*
715 				 * Inability to disable the cache is a threat
716 				 * to data durability.
717 				 */
718 				assert(ioctl(fd, DKIOCSETWCE, &wce_val) == 0);
719 				wce = WCE_IOCTL;
720 			} else {
721 				int r1, r2;
722 
723 				wce_val = 1;
724 				r1 = ioctl(fd, DKIOCSETWCE, &wce_val);
725 				wce_val = 0;
726 				r2 = ioctl(fd, DKIOCSETWCE, &wce_val);
727 
728 				if (r1 == 0 && r2 == 0) {
729 					wce = WCE_IOCTL;
730 				} else {
731 					/*
732 					 * If the cache cache toggle was not
733 					 * successful, ensure that the cache
734 					 * was not left enabled.
735 					 */
736 					assert(r1 != 0);
737 				}
738 			}
739 		}
740 
741 		if (nodelete == 0 && ioctl(fd, DKIOC_CANFREE, &candelete))
742 			candelete = 0;
743 
744 	} else {
745 		int flags;
746 
747 		if ((flags = fcntl(fd, F_GETFL)) >= 0) {
748 			flags |= O_DSYNC;
749 			if (fcntl(fd, F_SETFL, flags) != -1) {
750 				wce = WCE_FCNTL;
751 			}
752 		}
753 
754 		/*
755 		 * We don't have a way to discover if a file supports the
756 		 * FREESP fcntl cmd (other than trying it).  However,
757 		 * zfs, ufs, tmpfs, and udfs all support the FREESP fcntl cmd.
758 		 * Nfsv4 and nfsv4 also forward the FREESP request
759 		 * to the server, so we always enable it for file based
760 		 * volumes. Anyone trying to run volumes on an unsupported
761 		 * configuration is on their own, and should be prepared
762 		 * for the requests to fail.
763 		 */
764 		if (nodelete == 0)
765 			candelete = 1;
766 	}
767 #endif
768 
769 #ifndef WITHOUT_CAPSICUM
770 	if (caph_ioctls_limit(fd, cmds, nitems(cmds)) == -1)
771 		errx(EX_OSERR, "Unable to apply rights for sandbox");
772 #endif
773 
774 	if (ssopt != 0) {
775 		if (!powerof2(ssopt) || !powerof2(pssopt) || ssopt < 512 ||
776 		    ssopt > pssopt) {
777 			EPRINTLN("Invalid sector size %d/%d",
778 			    ssopt, pssopt);
779 			goto err;
780 		}
781 
782 		/*
783 		 * Some backend drivers (e.g. cd0, ada0) require that the I/O
784 		 * size be a multiple of the device's sector size.
785 		 *
786 		 * Validate that the emulated sector size complies with this
787 		 * requirement.
788 		 */
789 		if (S_ISCHR(sbuf.st_mode)) {
790 			if (ssopt < sectsz || (ssopt % sectsz) != 0) {
791 				EPRINTLN("Sector size %d incompatible "
792 				    "with underlying device sector size %d",
793 				    ssopt, sectsz);
794 				goto err;
795 			}
796 		}
797 
798 		sectsz = ssopt;
799 		psectsz = pssopt;
800 		psectoff = 0;
801 	}
802 
803 	bc = calloc(1, sizeof(struct blockif_ctxt));
804 	if (bc == NULL) {
805 		perror("calloc");
806 		goto err;
807 	}
808 
809 	bc->bc_magic = BLOCKIF_SIG;
810 	bc->bc_fd = fd;
811 	bc->bc_ischr = S_ISCHR(sbuf.st_mode);
812 	bc->bc_isgeom = geom;
813 	bc->bc_candelete = candelete;
814 #ifndef __FreeBSD__
815 	bc->bc_wce = wce;
816 #endif
817 	bc->bc_rdonly = ro;
818 	bc->bc_size = size;
819 	bc->bc_sectsz = sectsz;
820 	bc->bc_psectsz = psectsz;
821 	bc->bc_psectoff = psectoff;
822 	pthread_mutex_init(&bc->bc_mtx, NULL);
823 	pthread_cond_init(&bc->bc_cond, NULL);
824 	TAILQ_INIT(&bc->bc_freeq);
825 	TAILQ_INIT(&bc->bc_pendq);
826 	TAILQ_INIT(&bc->bc_busyq);
827 	bc->bc_bootindex = bootindex;
828 	for (i = 0; i < BLOCKIF_MAXREQ; i++) {
829 		bc->bc_reqs[i].be_status = BST_FREE;
830 		TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
831 	}
832 
833 	for (i = 0; i < BLOCKIF_NUMTHR; i++) {
834 		pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
835 		snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
836 		pthread_set_name_np(bc->bc_btid[i], tname);
837 	}
838 
839 	return (bc);
840 err:
841 	if (fd >= 0)
842 		close(fd);
843 	return (NULL);
844 }
845 
846 static void
blockif_resized(int fd,enum ev_type type __unused,void * arg)847 blockif_resized(int fd, enum ev_type type __unused, void *arg)
848 {
849 	struct blockif_ctxt *bc;
850 	struct stat sb;
851 	off_t mediasize;
852 
853 	if (fstat(fd, &sb) != 0)
854 		return;
855 
856 #ifdef __FreeBSD__
857 	if (S_ISCHR(sb.st_mode)) {
858 		if (ioctl(fd, DIOCGMEDIASIZE, &mediasize) < 0) {
859 			EPRINTLN("blockif_resized: get mediasize failed: %s",
860 			    strerror(errno));
861 			return;
862 		}
863 	} else
864 		mediasize = sb.st_size;
865 #else
866 	mediasize = sb.st_size;
867 	if (S_ISCHR(sb.st_mode)) {
868 		struct dk_minfo dkm;
869 
870 		if (ioctl(fd, DKIOCGMEDIAINFO, &dkm) == 0)
871 			mediasize = dkm.dki_lbsize * dkm.dki_capacity;
872 	}
873 #endif
874 
875 	bc = arg;
876 	pthread_mutex_lock(&bc->bc_mtx);
877 	if (mediasize != bc->bc_size) {
878 		bc->bc_size = mediasize;
879 		bc->bc_resize_cb(bc, bc->bc_resize_cb_arg, bc->bc_size);
880 	}
881 	pthread_mutex_unlock(&bc->bc_mtx);
882 }
883 
884 int
blockif_register_resize_callback(struct blockif_ctxt * bc,blockif_resize_cb * cb,void * cb_arg)885 blockif_register_resize_callback(struct blockif_ctxt *bc, blockif_resize_cb *cb,
886     void *cb_arg)
887 {
888 	struct stat sb;
889 	int err;
890 
891 	if (cb == NULL)
892 		return (EINVAL);
893 
894 	err = 0;
895 
896 	pthread_mutex_lock(&bc->bc_mtx);
897 	if (bc->bc_resize_cb != NULL) {
898 		err = EBUSY;
899 		goto out;
900 	}
901 
902 	assert(bc->bc_closing == 0);
903 
904 	if (fstat(bc->bc_fd, &sb) != 0) {
905 		err = errno;
906 		goto out;
907 	}
908 
909 	bc->bc_resize_event = mevent_add_flags(bc->bc_fd, EVF_VNODE,
910 	    EVFF_ATTRIB, blockif_resized, bc);
911 	if (bc->bc_resize_event == NULL) {
912 		err = ENXIO;
913 		goto out;
914 	}
915 
916 	bc->bc_resize_cb = cb;
917 	bc->bc_resize_cb_arg = cb_arg;
918 out:
919 	pthread_mutex_unlock(&bc->bc_mtx);
920 
921 	return (err);
922 }
923 
924 static int
blockif_request(struct blockif_ctxt * bc,struct blockif_req * breq,enum blockop op)925 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
926 		enum blockop op)
927 {
928 	int err;
929 
930 	err = 0;
931 
932 	pthread_mutex_lock(&bc->bc_mtx);
933 	if (!TAILQ_EMPTY(&bc->bc_freeq)) {
934 		/*
935 		 * Enqueue and inform the block i/o thread
936 		 * that there is work available
937 		 */
938 		if (blockif_enqueue(bc, breq, op))
939 			pthread_cond_signal(&bc->bc_cond);
940 	} else {
941 		/*
942 		 * Callers are not allowed to enqueue more than
943 		 * the specified blockif queue limit. Return an
944 		 * error to indicate that the queue length has been
945 		 * exceeded.
946 		 */
947 		err = E2BIG;
948 	}
949 	pthread_mutex_unlock(&bc->bc_mtx);
950 
951 	return (err);
952 }
953 
954 int
blockif_read(struct blockif_ctxt * bc,struct blockif_req * breq)955 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
956 {
957 	assert(bc->bc_magic == BLOCKIF_SIG);
958 	return (blockif_request(bc, breq, BOP_READ));
959 }
960 
961 int
blockif_write(struct blockif_ctxt * bc,struct blockif_req * breq)962 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
963 {
964 	assert(bc->bc_magic == BLOCKIF_SIG);
965 	return (blockif_request(bc, breq, BOP_WRITE));
966 }
967 
968 int
blockif_flush(struct blockif_ctxt * bc,struct blockif_req * breq)969 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
970 {
971 	assert(bc->bc_magic == BLOCKIF_SIG);
972 	return (blockif_request(bc, breq, BOP_FLUSH));
973 }
974 
975 int
blockif_delete(struct blockif_ctxt * bc,struct blockif_req * breq)976 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
977 {
978 	assert(bc->bc_magic == BLOCKIF_SIG);
979 	return (blockif_request(bc, breq, BOP_DELETE));
980 }
981 
982 int
blockif_cancel(struct blockif_ctxt * bc,struct blockif_req * breq)983 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
984 {
985 	struct blockif_elem *be;
986 
987 	assert(bc->bc_magic == BLOCKIF_SIG);
988 
989 	pthread_mutex_lock(&bc->bc_mtx);
990 	/*
991 	 * Check pending requests.
992 	 */
993 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
994 		if (be->be_req == breq)
995 			break;
996 	}
997 	if (be != NULL) {
998 		/*
999 		 * Found it.
1000 		 */
1001 		blockif_complete(bc, be);
1002 		pthread_mutex_unlock(&bc->bc_mtx);
1003 
1004 		return (0);
1005 	}
1006 
1007 	/*
1008 	 * Check in-flight requests.
1009 	 */
1010 	TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
1011 		if (be->be_req == breq)
1012 			break;
1013 	}
1014 	if (be == NULL) {
1015 		/*
1016 		 * Didn't find it.
1017 		 */
1018 		pthread_mutex_unlock(&bc->bc_mtx);
1019 		return (EINVAL);
1020 	}
1021 
1022 	/*
1023 	 * Interrupt the processing thread to force it return
1024 	 * prematurely via it's normal callback path.
1025 	 */
1026 	while (be->be_status == BST_BUSY) {
1027 		struct blockif_sig_elem bse, *old_head;
1028 
1029 		pthread_mutex_init(&bse.bse_mtx, NULL);
1030 		pthread_cond_init(&bse.bse_cond, NULL);
1031 
1032 		bse.bse_pending = 1;
1033 
1034 		do {
1035 			old_head = blockif_bse_head;
1036 			bse.bse_next = old_head;
1037 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
1038 					    (uintptr_t)old_head,
1039 					    (uintptr_t)&bse));
1040 
1041 		pthread_kill(be->be_tid, SIGCONT);
1042 
1043 		pthread_mutex_lock(&bse.bse_mtx);
1044 		while (bse.bse_pending)
1045 			pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
1046 		pthread_mutex_unlock(&bse.bse_mtx);
1047 	}
1048 
1049 	pthread_mutex_unlock(&bc->bc_mtx);
1050 
1051 	/*
1052 	 * The processing thread has been interrupted.  Since it's not
1053 	 * clear if the callback has been invoked yet, return EBUSY.
1054 	 */
1055 	return (EBUSY);
1056 }
1057 
1058 int
blockif_close(struct blockif_ctxt * bc)1059 blockif_close(struct blockif_ctxt *bc)
1060 {
1061 	void *jval;
1062 	int i;
1063 
1064 	assert(bc->bc_magic == BLOCKIF_SIG);
1065 
1066 	/*
1067 	 * Stop the block i/o thread
1068 	 */
1069 	pthread_mutex_lock(&bc->bc_mtx);
1070 	bc->bc_closing = 1;
1071 	if (bc->bc_resize_event != NULL)
1072 		mevent_disable(bc->bc_resize_event);
1073 	pthread_mutex_unlock(&bc->bc_mtx);
1074 	pthread_cond_broadcast(&bc->bc_cond);
1075 	for (i = 0; i < BLOCKIF_NUMTHR; i++)
1076 		pthread_join(bc->bc_btid[i], &jval);
1077 
1078 	/* XXX Cancel queued i/o's ??? */
1079 
1080 	/*
1081 	 * Release resources
1082 	 */
1083 	bc->bc_magic = 0;
1084 	close(bc->bc_fd);
1085 	free(bc);
1086 
1087 	return (0);
1088 }
1089 
1090 /*
1091  * Return virtual C/H/S values for a given block. Use the algorithm
1092  * outlined in the VHD specification to calculate values.
1093  */
1094 void
blockif_chs(struct blockif_ctxt * bc,uint16_t * c,uint8_t * h,uint8_t * s)1095 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
1096 {
1097 	off_t sectors;		/* total sectors of the block dev */
1098 	off_t hcyl;		/* cylinders times heads */
1099 	uint16_t secpt;		/* sectors per track */
1100 	uint8_t heads;
1101 
1102 	assert(bc->bc_magic == BLOCKIF_SIG);
1103 
1104 	sectors = bc->bc_size / bc->bc_sectsz;
1105 
1106 	/* Clamp the size to the largest possible with CHS */
1107 	if (sectors > 65535L * 16 * 255)
1108 		sectors = 65535L * 16 * 255;
1109 
1110 	if (sectors >= 65536L * 16 * 63) {
1111 		secpt = 255;
1112 		heads = 16;
1113 		hcyl = sectors / secpt;
1114 	} else {
1115 		secpt = 17;
1116 		hcyl = sectors / secpt;
1117 		heads = (hcyl + 1023) / 1024;
1118 
1119 		if (heads < 4)
1120 			heads = 4;
1121 
1122 		if (hcyl >= (heads * 1024) || heads > 16) {
1123 			secpt = 31;
1124 			heads = 16;
1125 			hcyl = sectors / secpt;
1126 		}
1127 		if (hcyl >= (heads * 1024)) {
1128 			secpt = 63;
1129 			heads = 16;
1130 			hcyl = sectors / secpt;
1131 		}
1132 	}
1133 
1134 	*c = hcyl / heads;
1135 	*h = heads;
1136 	*s = secpt;
1137 }
1138 
1139 /*
1140  * Accessors
1141  */
1142 off_t
blockif_size(struct blockif_ctxt * bc)1143 blockif_size(struct blockif_ctxt *bc)
1144 {
1145 	assert(bc->bc_magic == BLOCKIF_SIG);
1146 	return (bc->bc_size);
1147 }
1148 
1149 int
blockif_sectsz(struct blockif_ctxt * bc)1150 blockif_sectsz(struct blockif_ctxt *bc)
1151 {
1152 	assert(bc->bc_magic == BLOCKIF_SIG);
1153 	return (bc->bc_sectsz);
1154 }
1155 
1156 void
blockif_psectsz(struct blockif_ctxt * bc,int * size,int * off)1157 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
1158 {
1159 	assert(bc->bc_magic == BLOCKIF_SIG);
1160 	*size = bc->bc_psectsz;
1161 	*off = bc->bc_psectoff;
1162 }
1163 
1164 int
blockif_queuesz(struct blockif_ctxt * bc)1165 blockif_queuesz(struct blockif_ctxt *bc)
1166 {
1167 	assert(bc->bc_magic == BLOCKIF_SIG);
1168 	return (BLOCKIF_MAXREQ - 1);
1169 }
1170 
1171 int
blockif_is_ro(struct blockif_ctxt * bc)1172 blockif_is_ro(struct blockif_ctxt *bc)
1173 {
1174 	assert(bc->bc_magic == BLOCKIF_SIG);
1175 	return (bc->bc_rdonly);
1176 }
1177 
1178 int
blockif_candelete(struct blockif_ctxt * bc)1179 blockif_candelete(struct blockif_ctxt *bc)
1180 {
1181 	assert(bc->bc_magic == BLOCKIF_SIG);
1182 	return (bc->bc_candelete);
1183 }
1184 
1185 #ifndef __FreeBSD__
1186 int
blockif_set_wce(struct blockif_ctxt * bc,int wc_enable)1187 blockif_set_wce(struct blockif_ctxt *bc, int wc_enable)
1188 {
1189 	int res = 0, flags;
1190 	int clean_val = (wc_enable != 0) ? 1 : 0;
1191 
1192 	(void) pthread_mutex_lock(&bc->bc_mtx);
1193 	switch (bc->bc_wce) {
1194 	case WCE_IOCTL:
1195 		res = ioctl(bc->bc_fd, DKIOCSETWCE, &clean_val);
1196 		break;
1197 	case WCE_FCNTL:
1198 		if ((flags = fcntl(bc->bc_fd, F_GETFL)) >= 0) {
1199 			if (wc_enable == 0) {
1200 				flags |= O_DSYNC;
1201 			} else {
1202 				flags &= ~O_DSYNC;
1203 			}
1204 			if (fcntl(bc->bc_fd, F_SETFL, flags) == -1) {
1205 				res = -1;
1206 			}
1207 		} else {
1208 			res = -1;
1209 		}
1210 		break;
1211 	default:
1212 		break;
1213 	}
1214 
1215 	/*
1216 	 * After a successful disable of the write cache, ensure that any
1217 	 * lingering data in the cache is synced out.
1218 	 */
1219 	if (res == 0 && wc_enable == 0) {
1220 		res = fsync(bc->bc_fd);
1221 	}
1222 	(void) pthread_mutex_unlock(&bc->bc_mtx);
1223 
1224 	return (res);
1225 }
1226 #endif /* __FreeBSD__ */
1227