xref: /freebsd/usr.sbin/bhyve/block_if.c (revision ec0e626bafb335b30c499d06066997f54b10c092)
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 #include <sys/queue.h>
34 #include <sys/errno.h>
35 #include <sys/stat.h>
36 #include <sys/ioctl.h>
37 #include <sys/disk.h>
38 
39 #include <assert.h>
40 #include <fcntl.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <pthread.h>
45 #include <pthread_np.h>
46 #include <signal.h>
47 #include <unistd.h>
48 
49 #include <machine/atomic.h>
50 
51 #include "bhyverun.h"
52 #include "mevent.h"
53 #include "block_if.h"
54 
55 #define BLOCKIF_SIG	0xb109b109
56 
57 #define BLOCKIF_NUMTHR	8
58 #define BLOCKIF_MAXREQ	(64 + BLOCKIF_NUMTHR)
59 
60 enum blockop {
61 	BOP_READ,
62 	BOP_WRITE,
63 	BOP_FLUSH,
64 	BOP_DELETE
65 };
66 
67 enum blockstat {
68 	BST_FREE,
69 	BST_BLOCK,
70 	BST_PEND,
71 	BST_BUSY,
72 	BST_DONE
73 };
74 
75 struct blockif_elem {
76 	TAILQ_ENTRY(blockif_elem) be_link;
77 	struct blockif_req  *be_req;
78 	enum blockop	     be_op;
79 	enum blockstat	     be_status;
80 	pthread_t            be_tid;
81 	off_t		     be_block;
82 };
83 
84 struct blockif_ctxt {
85 	int			bc_magic;
86 	int			bc_fd;
87 	int			bc_ischr;
88 	int			bc_candelete;
89 	int			bc_rdonly;
90 	off_t			bc_size;
91 	int			bc_sectsz;
92 	int			bc_psectsz;
93 	int			bc_psectoff;
94 	int			bc_closing;
95 	pthread_t		bc_btid[BLOCKIF_NUMTHR];
96         pthread_mutex_t		bc_mtx;
97         pthread_cond_t		bc_cond;
98 
99 	/* Request elements and free/pending/busy queues */
100 	TAILQ_HEAD(, blockif_elem) bc_freeq;
101 	TAILQ_HEAD(, blockif_elem) bc_pendq;
102 	TAILQ_HEAD(, blockif_elem) bc_busyq;
103 	struct blockif_elem	bc_reqs[BLOCKIF_MAXREQ];
104 };
105 
106 static pthread_once_t blockif_once = PTHREAD_ONCE_INIT;
107 
108 struct blockif_sig_elem {
109 	pthread_mutex_t			bse_mtx;
110 	pthread_cond_t			bse_cond;
111 	int				bse_pending;
112 	struct blockif_sig_elem		*bse_next;
113 };
114 
115 static struct blockif_sig_elem *blockif_bse_head;
116 
117 static int
118 blockif_enqueue(struct blockif_ctxt *bc, struct blockif_req *breq,
119 		enum blockop op)
120 {
121 	struct blockif_elem *be, *tbe;
122 	off_t off;
123 	int i;
124 
125 	be = TAILQ_FIRST(&bc->bc_freeq);
126 	assert(be != NULL);
127 	assert(be->be_status == BST_FREE);
128 	TAILQ_REMOVE(&bc->bc_freeq, be, be_link);
129 	be->be_req = breq;
130 	be->be_op = op;
131 	switch (op) {
132 	case BOP_READ:
133 	case BOP_WRITE:
134 	case BOP_DELETE:
135 		off = breq->br_offset;
136 		for (i = 0; i < breq->br_iovcnt; i++)
137 			off += breq->br_iov[i].iov_len;
138 		break;
139 	default:
140 		off = OFF_MAX;
141 	}
142 	be->be_block = off;
143 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
144 		if (tbe->be_block == breq->br_offset)
145 			break;
146 	}
147 	if (tbe == NULL) {
148 		TAILQ_FOREACH(tbe, &bc->bc_busyq, be_link) {
149 			if (tbe->be_block == breq->br_offset)
150 				break;
151 		}
152 	}
153 	if (tbe == NULL)
154 		be->be_status = BST_PEND;
155 	else
156 		be->be_status = BST_BLOCK;
157 	TAILQ_INSERT_TAIL(&bc->bc_pendq, be, be_link);
158 	return (be->be_status == BST_PEND);
159 }
160 
161 static int
162 blockif_dequeue(struct blockif_ctxt *bc, pthread_t t, struct blockif_elem **bep)
163 {
164 	struct blockif_elem *be;
165 
166 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
167 		if (be->be_status == BST_PEND)
168 			break;
169 		assert(be->be_status == BST_BLOCK);
170 	}
171 	if (be == NULL)
172 		return (0);
173 	TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
174 	be->be_status = BST_BUSY;
175 	be->be_tid = t;
176 	TAILQ_INSERT_TAIL(&bc->bc_busyq, be, be_link);
177 	*bep = be;
178 	return (1);
179 }
180 
181 static void
182 blockif_complete(struct blockif_ctxt *bc, struct blockif_elem *be)
183 {
184 	struct blockif_elem *tbe;
185 
186 	if (be->be_status == BST_DONE || be->be_status == BST_BUSY)
187 		TAILQ_REMOVE(&bc->bc_busyq, be, be_link);
188 	else
189 		TAILQ_REMOVE(&bc->bc_pendq, be, be_link);
190 	TAILQ_FOREACH(tbe, &bc->bc_pendq, be_link) {
191 		if (tbe->be_req->br_offset == be->be_block)
192 			tbe->be_status = BST_PEND;
193 	}
194 	be->be_tid = 0;
195 	be->be_status = BST_FREE;
196 	be->be_req = NULL;
197 	TAILQ_INSERT_TAIL(&bc->bc_freeq, be, be_link);
198 }
199 
200 static void
201 blockif_proc(struct blockif_ctxt *bc, struct blockif_elem *be)
202 {
203 	struct blockif_req *br;
204 	off_t arg[2];
205 	int err;
206 
207 	br = be->be_req;
208 	err = 0;
209 
210 	switch (be->be_op) {
211 	case BOP_READ:
212 		if (preadv(bc->bc_fd, br->br_iov, br->br_iovcnt,
213 			   br->br_offset) < 0)
214 			err = errno;
215 		break;
216 	case BOP_WRITE:
217 		if (bc->bc_rdonly)
218 			err = EROFS;
219 		else if (pwritev(bc->bc_fd, br->br_iov, br->br_iovcnt,
220 			     br->br_offset) < 0)
221 			err = errno;
222 		break;
223 	case BOP_FLUSH:
224 		if (bc->bc_ischr) {
225 			if (ioctl(bc->bc_fd, DIOCGFLUSH))
226 				err = errno;
227 		} else if (fsync(bc->bc_fd))
228 			err = errno;
229 		break;
230 	case BOP_DELETE:
231 		if (!bc->bc_candelete)
232 			err = EOPNOTSUPP;
233 		else if (bc->bc_rdonly)
234 			err = EROFS;
235 		else if (bc->bc_ischr) {
236 			arg[0] = br->br_offset;
237 			arg[1] = br->br_iov[0].iov_len;
238 			if (ioctl(bc->bc_fd, DIOCGDELETE, arg))
239 				err = errno;
240 		} else
241 			err = EOPNOTSUPP;
242 		break;
243 	default:
244 		err = EINVAL;
245 		break;
246 	}
247 
248 	be->be_status = BST_DONE;
249 
250 	(*br->br_callback)(br, err);
251 }
252 
253 static void *
254 blockif_thr(void *arg)
255 {
256 	struct blockif_ctxt *bc;
257 	struct blockif_elem *be;
258 	pthread_t t;
259 
260 	bc = arg;
261 	t = pthread_self();
262 
263 	pthread_mutex_lock(&bc->bc_mtx);
264 	for (;;) {
265 		while (blockif_dequeue(bc, t, &be)) {
266 			pthread_mutex_unlock(&bc->bc_mtx);
267 			blockif_proc(bc, be);
268 			pthread_mutex_lock(&bc->bc_mtx);
269 			blockif_complete(bc, be);
270 		}
271 		/* Check ctxt status here to see if exit requested */
272 		if (bc->bc_closing)
273 			break;
274 		pthread_cond_wait(&bc->bc_cond, &bc->bc_mtx);
275 	}
276 	pthread_mutex_unlock(&bc->bc_mtx);
277 
278 	pthread_exit(NULL);
279 	return (NULL);
280 }
281 
282 static void
283 blockif_sigcont_handler(int signal, enum ev_type type, void *arg)
284 {
285 	struct blockif_sig_elem *bse;
286 
287 	for (;;) {
288 		/*
289 		 * Process the entire list even if not intended for
290 		 * this thread.
291 		 */
292 		do {
293 			bse = blockif_bse_head;
294 			if (bse == NULL)
295 				return;
296 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
297 					    (uintptr_t)bse,
298 					    (uintptr_t)bse->bse_next));
299 
300 		pthread_mutex_lock(&bse->bse_mtx);
301 		bse->bse_pending = 0;
302 		pthread_cond_signal(&bse->bse_cond);
303 		pthread_mutex_unlock(&bse->bse_mtx);
304 	}
305 }
306 
307 static void
308 blockif_init(void)
309 {
310 	mevent_add(SIGCONT, EVF_SIGNAL, blockif_sigcont_handler, NULL);
311 	(void) signal(SIGCONT, SIG_IGN);
312 }
313 
314 struct blockif_ctxt *
315 blockif_open(const char *optstr, const char *ident)
316 {
317 	char tname[MAXCOMLEN + 1];
318 	char *nopt, *xopts;
319 	struct blockif_ctxt *bc;
320 	struct stat sbuf;
321 	struct diocgattr_arg arg;
322 	off_t size, psectsz, psectoff;
323 	int extra, fd, i, sectsz;
324 	int nocache, sync, ro, candelete;
325 
326 	pthread_once(&blockif_once, blockif_init);
327 
328 	nocache = 0;
329 	sync = 0;
330 	ro = 0;
331 
332 	/*
333 	 * The first element in the optstring is always a pathname.
334 	 * Optional elements follow
335 	 */
336 	nopt = strdup(optstr);
337 	for (xopts = strtok(nopt, ",");
338 	     xopts != NULL;
339 	     xopts = strtok(NULL, ",")) {
340 		if (!strcmp(xopts, "nocache"))
341 			nocache = 1;
342 		else if (!strcmp(xopts, "sync"))
343 			sync = 1;
344 		else if (!strcmp(xopts, "ro"))
345 			ro = 1;
346 	}
347 
348 	extra = 0;
349 	if (nocache)
350 		extra |= O_DIRECT;
351 	if (sync)
352 		extra |= O_SYNC;
353 
354 	fd = open(nopt, (ro ? O_RDONLY : O_RDWR) | extra);
355 	if (fd < 0 && !ro) {
356 		/* Attempt a r/w fail with a r/o open */
357 		fd = open(nopt, O_RDONLY | extra);
358 		ro = 1;
359 	}
360 
361 	if (fd < 0) {
362 		perror("Could not open backing file");
363 		return (NULL);
364 	}
365 
366         if (fstat(fd, &sbuf) < 0) {
367                 perror("Could not stat backing file");
368                 close(fd);
369                 return (NULL);
370         }
371 
372         /*
373 	 * Deal with raw devices
374 	 */
375         size = sbuf.st_size;
376 	sectsz = DEV_BSIZE;
377 	psectsz = psectoff = 0;
378 	candelete = 0;
379 	if (S_ISCHR(sbuf.st_mode)) {
380 		if (ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
381 		    ioctl(fd, DIOCGSECTORSIZE, &sectsz)) {
382 			perror("Could not fetch dev blk/sector size");
383 			close(fd);
384 			return (NULL);
385 		}
386 		assert(size != 0);
387 		assert(sectsz != 0);
388 		if (ioctl(fd, DIOCGSTRIPESIZE, &psectsz) == 0 && psectsz > 0)
389 			ioctl(fd, DIOCGSTRIPEOFFSET, &psectoff);
390 		strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name));
391 		arg.len = sizeof(arg.value.i);
392 		if (ioctl(fd, DIOCGATTR, &arg) == 0)
393 			candelete = arg.value.i;
394 	} else
395 		psectsz = sbuf.st_blksize;
396 
397 	bc = calloc(1, sizeof(struct blockif_ctxt));
398 	if (bc == NULL) {
399 		close(fd);
400 		return (NULL);
401 	}
402 
403 	bc->bc_magic = BLOCKIF_SIG;
404 	bc->bc_fd = fd;
405 	bc->bc_ischr = S_ISCHR(sbuf.st_mode);
406 	bc->bc_candelete = candelete;
407 	bc->bc_rdonly = ro;
408 	bc->bc_size = size;
409 	bc->bc_sectsz = sectsz;
410 	bc->bc_psectsz = psectsz;
411 	bc->bc_psectoff = psectoff;
412 	pthread_mutex_init(&bc->bc_mtx, NULL);
413 	pthread_cond_init(&bc->bc_cond, NULL);
414 	TAILQ_INIT(&bc->bc_freeq);
415 	TAILQ_INIT(&bc->bc_pendq);
416 	TAILQ_INIT(&bc->bc_busyq);
417 	for (i = 0; i < BLOCKIF_MAXREQ; i++) {
418 		bc->bc_reqs[i].be_status = BST_FREE;
419 		TAILQ_INSERT_HEAD(&bc->bc_freeq, &bc->bc_reqs[i], be_link);
420 	}
421 
422 	for (i = 0; i < BLOCKIF_NUMTHR; i++) {
423 		pthread_create(&bc->bc_btid[i], NULL, blockif_thr, bc);
424 		snprintf(tname, sizeof(tname), "blk-%s-%d", ident, i);
425 		pthread_set_name_np(bc->bc_btid[i], tname);
426 	}
427 
428 	return (bc);
429 }
430 
431 static int
432 blockif_request(struct blockif_ctxt *bc, struct blockif_req *breq,
433 		enum blockop op)
434 {
435 	int err;
436 
437 	err = 0;
438 
439 	pthread_mutex_lock(&bc->bc_mtx);
440 	if (!TAILQ_EMPTY(&bc->bc_freeq)) {
441 		/*
442 		 * Enqueue and inform the block i/o thread
443 		 * that there is work available
444 		 */
445 		if (blockif_enqueue(bc, breq, op))
446 			pthread_cond_signal(&bc->bc_cond);
447 	} else {
448 		/*
449 		 * Callers are not allowed to enqueue more than
450 		 * the specified blockif queue limit. Return an
451 		 * error to indicate that the queue length has been
452 		 * exceeded.
453 		 */
454 		err = E2BIG;
455 	}
456 	pthread_mutex_unlock(&bc->bc_mtx);
457 
458 	return (err);
459 }
460 
461 int
462 blockif_read(struct blockif_ctxt *bc, struct blockif_req *breq)
463 {
464 
465 	assert(bc->bc_magic == BLOCKIF_SIG);
466 	return (blockif_request(bc, breq, BOP_READ));
467 }
468 
469 int
470 blockif_write(struct blockif_ctxt *bc, struct blockif_req *breq)
471 {
472 
473 	assert(bc->bc_magic == BLOCKIF_SIG);
474 	return (blockif_request(bc, breq, BOP_WRITE));
475 }
476 
477 int
478 blockif_flush(struct blockif_ctxt *bc, struct blockif_req *breq)
479 {
480 
481 	assert(bc->bc_magic == BLOCKIF_SIG);
482 	return (blockif_request(bc, breq, BOP_FLUSH));
483 }
484 
485 int
486 blockif_delete(struct blockif_ctxt *bc, struct blockif_req *breq)
487 {
488 
489 	assert(bc->bc_magic == BLOCKIF_SIG);
490 	return (blockif_request(bc, breq, BOP_DELETE));
491 }
492 
493 int
494 blockif_cancel(struct blockif_ctxt *bc, struct blockif_req *breq)
495 {
496 	struct blockif_elem *be;
497 
498 	assert(bc->bc_magic == BLOCKIF_SIG);
499 
500 	pthread_mutex_lock(&bc->bc_mtx);
501 	/*
502 	 * Check pending requests.
503 	 */
504 	TAILQ_FOREACH(be, &bc->bc_pendq, be_link) {
505 		if (be->be_req == breq)
506 			break;
507 	}
508 	if (be != NULL) {
509 		/*
510 		 * Found it.
511 		 */
512 		blockif_complete(bc, be);
513 		pthread_mutex_unlock(&bc->bc_mtx);
514 
515 		return (0);
516 	}
517 
518 	/*
519 	 * Check in-flight requests.
520 	 */
521 	TAILQ_FOREACH(be, &bc->bc_busyq, be_link) {
522 		if (be->be_req == breq)
523 			break;
524 	}
525 	if (be == NULL) {
526 		/*
527 		 * Didn't find it.
528 		 */
529 		pthread_mutex_unlock(&bc->bc_mtx);
530 		return (EINVAL);
531 	}
532 
533 	/*
534 	 * Interrupt the processing thread to force it return
535 	 * prematurely via it's normal callback path.
536 	 */
537 	while (be->be_status == BST_BUSY) {
538 		struct blockif_sig_elem bse, *old_head;
539 
540 		pthread_mutex_init(&bse.bse_mtx, NULL);
541 		pthread_cond_init(&bse.bse_cond, NULL);
542 
543 		bse.bse_pending = 1;
544 
545 		do {
546 			old_head = blockif_bse_head;
547 			bse.bse_next = old_head;
548 		} while (!atomic_cmpset_ptr((uintptr_t *)&blockif_bse_head,
549 					    (uintptr_t)old_head,
550 					    (uintptr_t)&bse));
551 
552 		pthread_kill(be->be_tid, SIGCONT);
553 
554 		pthread_mutex_lock(&bse.bse_mtx);
555 		while (bse.bse_pending)
556 			pthread_cond_wait(&bse.bse_cond, &bse.bse_mtx);
557 		pthread_mutex_unlock(&bse.bse_mtx);
558 	}
559 
560 	pthread_mutex_unlock(&bc->bc_mtx);
561 
562 	/*
563 	 * The processing thread has been interrupted.  Since it's not
564 	 * clear if the callback has been invoked yet, return EBUSY.
565 	 */
566 	return (EBUSY);
567 }
568 
569 int
570 blockif_close(struct blockif_ctxt *bc)
571 {
572 	void *jval;
573 	int err, i;
574 
575 	err = 0;
576 
577 	assert(bc->bc_magic == BLOCKIF_SIG);
578 
579 	/*
580 	 * Stop the block i/o thread
581 	 */
582 	pthread_mutex_lock(&bc->bc_mtx);
583 	bc->bc_closing = 1;
584 	pthread_mutex_unlock(&bc->bc_mtx);
585 	pthread_cond_broadcast(&bc->bc_cond);
586 	for (i = 0; i < BLOCKIF_NUMTHR; i++)
587 		pthread_join(bc->bc_btid[i], &jval);
588 
589 	/* XXX Cancel queued i/o's ??? */
590 
591 	/*
592 	 * Release resources
593 	 */
594 	bc->bc_magic = 0;
595 	close(bc->bc_fd);
596 	free(bc);
597 
598 	return (0);
599 }
600 
601 /*
602  * Return virtual C/H/S values for a given block. Use the algorithm
603  * outlined in the VHD specification to calculate values.
604  */
605 void
606 blockif_chs(struct blockif_ctxt *bc, uint16_t *c, uint8_t *h, uint8_t *s)
607 {
608 	off_t sectors;		/* total sectors of the block dev */
609 	off_t hcyl;		/* cylinders times heads */
610 	uint16_t secpt;		/* sectors per track */
611 	uint8_t heads;
612 
613 	assert(bc->bc_magic == BLOCKIF_SIG);
614 
615 	sectors = bc->bc_size / bc->bc_sectsz;
616 
617 	/* Clamp the size to the largest possible with CHS */
618 	if (sectors > 65535UL*16*255)
619 		sectors = 65535UL*16*255;
620 
621 	if (sectors >= 65536UL*16*63) {
622 		secpt = 255;
623 		heads = 16;
624 		hcyl = sectors / secpt;
625 	} else {
626 		secpt = 17;
627 		hcyl = sectors / secpt;
628 		heads = (hcyl + 1023) / 1024;
629 
630 		if (heads < 4)
631 			heads = 4;
632 
633 		if (hcyl >= (heads * 1024) || heads > 16) {
634 			secpt = 31;
635 			heads = 16;
636 			hcyl = sectors / secpt;
637 		}
638 		if (hcyl >= (heads * 1024)) {
639 			secpt = 63;
640 			heads = 16;
641 			hcyl = sectors / secpt;
642 		}
643 	}
644 
645 	*c = hcyl / heads;
646 	*h = heads;
647 	*s = secpt;
648 }
649 
650 /*
651  * Accessors
652  */
653 off_t
654 blockif_size(struct blockif_ctxt *bc)
655 {
656 
657 	assert(bc->bc_magic == BLOCKIF_SIG);
658 	return (bc->bc_size);
659 }
660 
661 int
662 blockif_sectsz(struct blockif_ctxt *bc)
663 {
664 
665 	assert(bc->bc_magic == BLOCKIF_SIG);
666 	return (bc->bc_sectsz);
667 }
668 
669 void
670 blockif_psectsz(struct blockif_ctxt *bc, int *size, int *off)
671 {
672 
673 	assert(bc->bc_magic == BLOCKIF_SIG);
674 	*size = bc->bc_psectsz;
675 	*off = bc->bc_psectoff;
676 }
677 
678 int
679 blockif_queuesz(struct blockif_ctxt *bc)
680 {
681 
682 	assert(bc->bc_magic == BLOCKIF_SIG);
683 	return (BLOCKIF_MAXREQ - 1);
684 }
685 
686 int
687 blockif_is_ro(struct blockif_ctxt *bc)
688 {
689 
690 	assert(bc->bc_magic == BLOCKIF_SIG);
691 	return (bc->bc_rdonly);
692 }
693 
694 int
695 blockif_candelete(struct blockif_ctxt *bc)
696 {
697 
698 	assert(bc->bc_magic == BLOCKIF_SIG);
699 	return (bc->bc_candelete);
700 }
701