xref: /freebsd/usr.sbin/bhyve/pci_virtio_scsi.c (revision ce6a89e27cd190313be39bb479880aeda4778436)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2016 Jakub Klama <jceel@FreeBSD.org>.
5  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>.
6  * All rights reserved.
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  *    in this position and unchanged.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/linker_set.h>
36 #include <sys/types.h>
37 #include <sys/uio.h>
38 #include <sys/time.h>
39 #include <sys/queue.h>
40 #include <sys/sbuf.h>
41 
42 #include <errno.h>
43 #include <fcntl.h>
44 #include <stdio.h>
45 #include <stdlib.h>
46 #include <stdbool.h>
47 #include <string.h>
48 #include <unistd.h>
49 #include <assert.h>
50 #include <pthread.h>
51 #include <pthread_np.h>
52 
53 #include <cam/scsi/scsi_all.h>
54 #include <cam/scsi/scsi_message.h>
55 #include <cam/ctl/ctl.h>
56 #include <cam/ctl/ctl_io.h>
57 #include <cam/ctl/ctl_backend.h>
58 #include <cam/ctl/ctl_ioctl.h>
59 #include <cam/ctl/ctl_util.h>
60 #include <cam/ctl/ctl_scsi_all.h>
61 #include <camlib.h>
62 
63 #include "bhyverun.h"
64 #include "debug.h"
65 #include "pci_emul.h"
66 #include "virtio.h"
67 #include "iov.h"
68 
69 #define VTSCSI_RINGSZ		64
70 #define	VTSCSI_REQUESTQ		1
71 #define	VTSCSI_THR_PER_Q	16
72 #define	VTSCSI_MAXQ		(VTSCSI_REQUESTQ + 2)
73 #define	VTSCSI_MAXSEG		64
74 
75 #define	VTSCSI_IN_HEADER_LEN(_sc)	\
76 	(sizeof(struct pci_vtscsi_req_cmd_rd) + _sc->vss_config.cdb_size)
77 
78 #define	VTSCSI_OUT_HEADER_LEN(_sc) 	\
79 	(sizeof(struct pci_vtscsi_req_cmd_wr) + _sc->vss_config.sense_size)
80 
81 #define	VIRTIO_SCSI_MAX_CHANNEL	0
82 #define	VIRTIO_SCSI_MAX_TARGET	0
83 #define	VIRTIO_SCSI_MAX_LUN	16383
84 
85 #define	VIRTIO_SCSI_F_INOUT	(1 << 0)
86 #define	VIRTIO_SCSI_F_HOTPLUG	(1 << 1)
87 #define	VIRTIO_SCSI_F_CHANGE	(1 << 2)
88 
89 static int pci_vtscsi_debug = 0;
90 #define	DPRINTF(params) if (pci_vtscsi_debug) PRINTLN params
91 #define	WPRINTF(params) PRINTLN params
92 
93 struct pci_vtscsi_config {
94 	uint32_t num_queues;
95 	uint32_t seg_max;
96 	uint32_t max_sectors;
97 	uint32_t cmd_per_lun;
98 	uint32_t event_info_size;
99 	uint32_t sense_size;
100 	uint32_t cdb_size;
101 	uint16_t max_channel;
102 	uint16_t max_target;
103 	uint32_t max_lun;
104 } __attribute__((packed));
105 
106 struct pci_vtscsi_queue {
107 	struct pci_vtscsi_softc *         vsq_sc;
108 	struct vqueue_info *              vsq_vq;
109 	pthread_mutex_t                   vsq_mtx;
110 	pthread_mutex_t                   vsq_qmtx;
111 	pthread_cond_t                    vsq_cv;
112 	STAILQ_HEAD(, pci_vtscsi_request) vsq_requests;
113 	LIST_HEAD(, pci_vtscsi_worker)    vsq_workers;
114 };
115 
116 struct pci_vtscsi_worker {
117 	struct pci_vtscsi_queue *     vsw_queue;
118 	pthread_t                     vsw_thread;
119 	bool                          vsw_exiting;
120 	LIST_ENTRY(pci_vtscsi_worker) vsw_link;
121 };
122 
123 struct pci_vtscsi_request {
124 	struct pci_vtscsi_queue * vsr_queue;
125 	struct iovec              vsr_iov_in[VTSCSI_MAXSEG];
126 	int                       vsr_niov_in;
127 	struct iovec              vsr_iov_out[VTSCSI_MAXSEG];
128 	int                       vsr_niov_out;
129 	uint32_t                  vsr_idx;
130 	STAILQ_ENTRY(pci_vtscsi_request) vsr_link;
131 };
132 
133 /*
134  * Per-device softc
135  */
136 struct pci_vtscsi_softc {
137 	struct virtio_softc      vss_vs;
138 	struct vqueue_info       vss_vq[VTSCSI_MAXQ];
139 	struct pci_vtscsi_queue  vss_queues[VTSCSI_REQUESTQ];
140 	pthread_mutex_t          vss_mtx;
141 	int                      vss_iid;
142 	int                      vss_ctl_fd;
143 	uint32_t                 vss_features;
144 	struct pci_vtscsi_config vss_config;
145 };
146 
147 #define	VIRTIO_SCSI_T_TMF			0
148 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK		0
149 #define	VIRTIO_SCSI_T_TMF_ABORT_TASK_SET	1
150 #define	VIRTIO_SCSI_T_TMF_CLEAR_ACA		2
151 #define	VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET	3
152 #define	VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET	4
153 #define	VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET	5
154 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK		6
155 #define	VIRTIO_SCSI_T_TMF_QUERY_TASK_SET 	7
156 
157 /* command-specific response values */
158 #define	VIRTIO_SCSI_S_FUNCTION_COMPLETE		0
159 #define	VIRTIO_SCSI_S_FUNCTION_SUCCEEDED	10
160 #define	VIRTIO_SCSI_S_FUNCTION_REJECTED		11
161 
162 struct pci_vtscsi_ctrl_tmf {
163 	uint32_t type;
164 	uint32_t subtype;
165 	uint8_t lun[8];
166 	uint64_t id;
167 	uint8_t response;
168 } __attribute__((packed));
169 
170 #define	VIRTIO_SCSI_T_AN_QUERY			1
171 #define	VIRTIO_SCSI_EVT_ASYNC_OPERATIONAL_CHANGE 2
172 #define	VIRTIO_SCSI_EVT_ASYNC_POWER_MGMT	4
173 #define	VIRTIO_SCSI_EVT_ASYNC_EXTERNAL_REQUEST	8
174 #define	VIRTIO_SCSI_EVT_ASYNC_MEDIA_CHANGE	16
175 #define	VIRTIO_SCSI_EVT_ASYNC_MULTI_HOST	32
176 #define	VIRTIO_SCSI_EVT_ASYNC_DEVICE_BUSY	64
177 
178 struct pci_vtscsi_ctrl_an {
179 	uint32_t type;
180 	uint8_t lun[8];
181 	uint32_t event_requested;
182 	uint32_t event_actual;
183 	uint8_t response;
184 } __attribute__((packed));
185 
186 /* command-specific response values */
187 #define	VIRTIO_SCSI_S_OK 			0
188 #define	VIRTIO_SCSI_S_OVERRUN			1
189 #define	VIRTIO_SCSI_S_ABORTED			2
190 #define	VIRTIO_SCSI_S_BAD_TARGET		3
191 #define	VIRTIO_SCSI_S_RESET			4
192 #define	VIRTIO_SCSI_S_BUSY			5
193 #define	VIRTIO_SCSI_S_TRANSPORT_FAILURE		6
194 #define	VIRTIO_SCSI_S_TARGET_FAILURE		7
195 #define	VIRTIO_SCSI_S_NEXUS_FAILURE		8
196 #define	VIRTIO_SCSI_S_FAILURE			9
197 #define	VIRTIO_SCSI_S_INCORRECT_LUN		12
198 
199 /* task_attr */
200 #define	VIRTIO_SCSI_S_SIMPLE			0
201 #define	VIRTIO_SCSI_S_ORDERED			1
202 #define	VIRTIO_SCSI_S_HEAD			2
203 #define	VIRTIO_SCSI_S_ACA			3
204 
205 struct pci_vtscsi_event {
206 	uint32_t event;
207 	uint8_t lun[8];
208 	uint32_t reason;
209 } __attribute__((packed));
210 
211 struct pci_vtscsi_req_cmd_rd {
212 	uint8_t lun[8];
213 	uint64_t id;
214 	uint8_t task_attr;
215 	uint8_t prio;
216 	uint8_t crn;
217 	uint8_t cdb[];
218 } __attribute__((packed));
219 
220 struct pci_vtscsi_req_cmd_wr {
221 	uint32_t sense_len;
222 	uint32_t residual;
223 	uint16_t status_qualifier;
224 	uint8_t status;
225 	uint8_t response;
226 	uint8_t sense[];
227 } __attribute__((packed));
228 
229 static void *pci_vtscsi_proc(void *);
230 static void pci_vtscsi_reset(void *);
231 static void pci_vtscsi_neg_features(void *, uint64_t);
232 static int pci_vtscsi_cfgread(void *, int, int, uint32_t *);
233 static int pci_vtscsi_cfgwrite(void *, int, int, uint32_t);
234 static inline int pci_vtscsi_get_lun(uint8_t *);
235 static int pci_vtscsi_control_handle(struct pci_vtscsi_softc *, void *, size_t);
236 static int pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *,
237     struct pci_vtscsi_ctrl_tmf *);
238 static int pci_vtscsi_an_handle(struct pci_vtscsi_softc *,
239     struct pci_vtscsi_ctrl_an *);
240 static int pci_vtscsi_request_handle(struct pci_vtscsi_queue *, struct iovec *,
241     int, struct iovec *, int);
242 static void pci_vtscsi_controlq_notify(void *, struct vqueue_info *);
243 static void pci_vtscsi_eventq_notify(void *, struct vqueue_info *);
244 static void pci_vtscsi_requestq_notify(void *, struct vqueue_info *);
245 static int  pci_vtscsi_init_queue(struct pci_vtscsi_softc *,
246     struct pci_vtscsi_queue *, int);
247 static int pci_vtscsi_init(struct vmctx *, struct pci_devinst *, char *);
248 
249 static struct virtio_consts vtscsi_vi_consts = {
250 	"vtscsi",				/* our name */
251 	VTSCSI_MAXQ,				/* we support 2+n virtqueues */
252 	sizeof(struct pci_vtscsi_config),	/* config reg size */
253 	pci_vtscsi_reset,			/* reset */
254 	NULL,					/* device-wide qnotify */
255 	pci_vtscsi_cfgread,			/* read virtio config */
256 	pci_vtscsi_cfgwrite,			/* write virtio config */
257 	pci_vtscsi_neg_features,		/* apply negotiated features */
258 	0,					/* our capabilities */
259 };
260 
261 static void *
262 pci_vtscsi_proc(void *arg)
263 {
264 	struct pci_vtscsi_worker *worker = (struct pci_vtscsi_worker *)arg;
265 	struct pci_vtscsi_queue *q = worker->vsw_queue;
266 	struct pci_vtscsi_request *req;
267 	int iolen;
268 
269 	for (;;) {
270 		pthread_mutex_lock(&q->vsq_mtx);
271 
272 		while (STAILQ_EMPTY(&q->vsq_requests)
273 		    && !worker->vsw_exiting)
274 			pthread_cond_wait(&q->vsq_cv, &q->vsq_mtx);
275 
276 		if (worker->vsw_exiting)
277 			break;
278 
279 		req = STAILQ_FIRST(&q->vsq_requests);
280 		STAILQ_REMOVE_HEAD(&q->vsq_requests, vsr_link);
281 
282 		pthread_mutex_unlock(&q->vsq_mtx);
283 		iolen = pci_vtscsi_request_handle(q, req->vsr_iov_in,
284 		    req->vsr_niov_in, req->vsr_iov_out, req->vsr_niov_out);
285 
286 		pthread_mutex_lock(&q->vsq_qmtx);
287 		vq_relchain(q->vsq_vq, req->vsr_idx, iolen);
288 		vq_endchains(q->vsq_vq, 0);
289 		pthread_mutex_unlock(&q->vsq_qmtx);
290 
291 		DPRINTF(("virtio-scsi: request <idx=%d> completed",
292 		    req->vsr_idx));
293 		free(req);
294 	}
295 
296 	pthread_mutex_unlock(&q->vsq_mtx);
297 	return (NULL);
298 }
299 
300 static void
301 pci_vtscsi_reset(void *vsc)
302 {
303 	struct pci_vtscsi_softc *sc;
304 
305 	sc = vsc;
306 
307 	DPRINTF(("vtscsi: device reset requested"));
308 	vi_reset_dev(&sc->vss_vs);
309 
310 	/* initialize config structure */
311 	sc->vss_config = (struct pci_vtscsi_config){
312 		.num_queues = VTSCSI_REQUESTQ,
313 		/* Leave room for the request and the response. */
314 		.seg_max = VTSCSI_MAXSEG - 2,
315 		.max_sectors = 2,
316 		.cmd_per_lun = 1,
317 		.event_info_size = sizeof(struct pci_vtscsi_event),
318 		.sense_size = 96,
319 		.cdb_size = 32,
320 		.max_channel = VIRTIO_SCSI_MAX_CHANNEL,
321 		.max_target = VIRTIO_SCSI_MAX_TARGET,
322 		.max_lun = VIRTIO_SCSI_MAX_LUN
323 	};
324 }
325 
326 static void
327 pci_vtscsi_neg_features(void *vsc, uint64_t negotiated_features)
328 {
329 	struct pci_vtscsi_softc *sc = vsc;
330 
331 	sc->vss_features = negotiated_features;
332 }
333 
334 static int
335 pci_vtscsi_cfgread(void *vsc, int offset, int size, uint32_t *retval)
336 {
337 	struct pci_vtscsi_softc *sc = vsc;
338 	void *ptr;
339 
340 	ptr = (uint8_t *)&sc->vss_config + offset;
341 	memcpy(retval, ptr, size);
342 	return (0);
343 }
344 
345 static int
346 pci_vtscsi_cfgwrite(void *vsc, int offset, int size, uint32_t val)
347 {
348 
349 	return (0);
350 }
351 
352 static inline int
353 pci_vtscsi_get_lun(uint8_t *lun)
354 {
355 
356 	return (((lun[2] << 8) | lun[3]) & 0x3fff);
357 }
358 
359 static int
360 pci_vtscsi_control_handle(struct pci_vtscsi_softc *sc, void *buf,
361     size_t bufsize)
362 {
363 	struct pci_vtscsi_ctrl_tmf *tmf;
364 	struct pci_vtscsi_ctrl_an *an;
365 	uint32_t type;
366 
367 	type = *(uint32_t *)buf;
368 
369 	if (type == VIRTIO_SCSI_T_TMF) {
370 		tmf = (struct pci_vtscsi_ctrl_tmf *)buf;
371 		return (pci_vtscsi_tmf_handle(sc, tmf));
372 	}
373 
374 	if (type == VIRTIO_SCSI_T_AN_QUERY) {
375 		an = (struct pci_vtscsi_ctrl_an *)buf;
376 		return (pci_vtscsi_an_handle(sc, an));
377 	}
378 
379 	return (0);
380 }
381 
382 static int
383 pci_vtscsi_tmf_handle(struct pci_vtscsi_softc *sc,
384     struct pci_vtscsi_ctrl_tmf *tmf)
385 {
386 	union ctl_io *io;
387 	int err;
388 
389 	io = ctl_scsi_alloc_io(sc->vss_iid);
390 	ctl_scsi_zero_io(io);
391 
392 	io->io_hdr.io_type = CTL_IO_TASK;
393 	io->io_hdr.nexus.initid = sc->vss_iid;
394 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(tmf->lun);
395 	io->taskio.tag_type = CTL_TAG_SIMPLE;
396 	io->taskio.tag_num = (uint32_t)tmf->id;
397 
398 	switch (tmf->subtype) {
399 	case VIRTIO_SCSI_T_TMF_ABORT_TASK:
400 		io->taskio.task_action = CTL_TASK_ABORT_TASK;
401 		break;
402 
403 	case VIRTIO_SCSI_T_TMF_ABORT_TASK_SET:
404 		io->taskio.task_action = CTL_TASK_ABORT_TASK_SET;
405 		break;
406 
407 	case VIRTIO_SCSI_T_TMF_CLEAR_ACA:
408 		io->taskio.task_action = CTL_TASK_CLEAR_ACA;
409 		break;
410 
411 	case VIRTIO_SCSI_T_TMF_CLEAR_TASK_SET:
412 		io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET;
413 		break;
414 
415 	case VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET:
416 		io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET;
417 		break;
418 
419 	case VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET:
420 		io->taskio.task_action = CTL_TASK_LUN_RESET;
421 		break;
422 
423 	case VIRTIO_SCSI_T_TMF_QUERY_TASK:
424 		io->taskio.task_action = CTL_TASK_QUERY_TASK;
425 		break;
426 
427 	case VIRTIO_SCSI_T_TMF_QUERY_TASK_SET:
428 		io->taskio.task_action = CTL_TASK_QUERY_TASK_SET;
429 		break;
430 	}
431 
432 	if (pci_vtscsi_debug) {
433 		struct sbuf *sb = sbuf_new_auto();
434 		ctl_io_sbuf(io, sb);
435 		sbuf_finish(sb);
436 		DPRINTF(("pci_virtio_scsi: %s", sbuf_data(sb)));
437 		sbuf_delete(sb);
438 	}
439 
440 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
441 	if (err != 0)
442 		WPRINTF(("CTL_IO: err=%d (%s)", errno, strerror(errno)));
443 
444 	tmf->response = io->taskio.task_status;
445 	ctl_scsi_free_io(io);
446 	return (1);
447 }
448 
449 static int
450 pci_vtscsi_an_handle(struct pci_vtscsi_softc *sc,
451     struct pci_vtscsi_ctrl_an *an)
452 {
453 
454 	return (0);
455 }
456 
457 static int
458 pci_vtscsi_request_handle(struct pci_vtscsi_queue *q, struct iovec *iov_in,
459     int niov_in, struct iovec *iov_out, int niov_out)
460 {
461 	struct pci_vtscsi_softc *sc = q->vsq_sc;
462 	struct pci_vtscsi_req_cmd_rd *cmd_rd = NULL;
463 	struct pci_vtscsi_req_cmd_wr *cmd_wr;
464 	struct iovec data_iov_in[VTSCSI_MAXSEG], data_iov_out[VTSCSI_MAXSEG];
465 	union ctl_io *io;
466 	int data_niov_in, data_niov_out;
467 	void *ext_data_ptr = NULL;
468 	uint32_t ext_data_len = 0, ext_sg_entries = 0;
469 	int err, nxferred;
470 
471 	seek_iov(iov_in, niov_in, data_iov_in, &data_niov_in,
472 	    VTSCSI_IN_HEADER_LEN(sc));
473 	seek_iov(iov_out, niov_out, data_iov_out, &data_niov_out,
474 	    VTSCSI_OUT_HEADER_LEN(sc));
475 
476 	truncate_iov(iov_in, &niov_in, VTSCSI_IN_HEADER_LEN(sc));
477 	truncate_iov(iov_out, &niov_out, VTSCSI_OUT_HEADER_LEN(sc));
478 	iov_to_buf(iov_in, niov_in, (void **)&cmd_rd);
479 
480 	cmd_wr = malloc(VTSCSI_OUT_HEADER_LEN(sc));
481 	io = ctl_scsi_alloc_io(sc->vss_iid);
482 	ctl_scsi_zero_io(io);
483 
484 	io->io_hdr.nexus.initid = sc->vss_iid;
485 	io->io_hdr.nexus.targ_lun = pci_vtscsi_get_lun(cmd_rd->lun);
486 
487 	io->io_hdr.io_type = CTL_IO_SCSI;
488 
489 	if (data_niov_in > 0) {
490 		ext_data_ptr = (void *)data_iov_in;
491 		ext_sg_entries = data_niov_in;
492 		ext_data_len = count_iov(data_iov_in, data_niov_in);
493 		io->io_hdr.flags |= CTL_FLAG_DATA_OUT;
494 	} else if (data_niov_out > 0) {
495 		ext_data_ptr = (void *)data_iov_out;
496 		ext_sg_entries = data_niov_out;
497 		ext_data_len = count_iov(data_iov_out, data_niov_out);
498 		io->io_hdr.flags |= CTL_FLAG_DATA_IN;
499 	}
500 
501 	io->scsiio.sense_len = sc->vss_config.sense_size;
502 	io->scsiio.tag_num = (uint32_t)cmd_rd->id;
503 	switch (cmd_rd->task_attr) {
504 	case VIRTIO_SCSI_S_ORDERED:
505 		io->scsiio.tag_type = CTL_TAG_ORDERED;
506 		break;
507 	case VIRTIO_SCSI_S_HEAD:
508 		io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE;
509 		break;
510 	case VIRTIO_SCSI_S_ACA:
511 		io->scsiio.tag_type = CTL_TAG_ACA;
512 		break;
513 	case VIRTIO_SCSI_S_SIMPLE:
514 	default:
515 		io->scsiio.tag_type = CTL_TAG_SIMPLE;
516 		break;
517 	}
518 	io->scsiio.ext_sg_entries = ext_sg_entries;
519 	io->scsiio.ext_data_ptr = ext_data_ptr;
520 	io->scsiio.ext_data_len = ext_data_len;
521 	io->scsiio.ext_data_filled = 0;
522 	io->scsiio.cdb_len = sc->vss_config.cdb_size;
523 	memcpy(io->scsiio.cdb, cmd_rd->cdb, sc->vss_config.cdb_size);
524 
525 	if (pci_vtscsi_debug) {
526 		struct sbuf *sb = sbuf_new_auto();
527 		ctl_io_sbuf(io, sb);
528 		sbuf_finish(sb);
529 		DPRINTF(("pci_virtio_scsi: %s", sbuf_data(sb)));
530 		sbuf_delete(sb);
531 	}
532 
533 	err = ioctl(sc->vss_ctl_fd, CTL_IO, io);
534 	if (err != 0) {
535 		WPRINTF(("CTL_IO: err=%d (%s)", errno, strerror(errno)));
536 		cmd_wr->response = VIRTIO_SCSI_S_FAILURE;
537 	} else {
538 		cmd_wr->sense_len = MIN(io->scsiio.sense_len,
539 		    sc->vss_config.sense_size);
540 		cmd_wr->residual = io->scsiio.residual;
541 		cmd_wr->status = io->scsiio.scsi_status;
542 		cmd_wr->response = VIRTIO_SCSI_S_OK;
543 		memcpy(&cmd_wr->sense, &io->scsiio.sense_data,
544 		    cmd_wr->sense_len);
545 	}
546 
547 	buf_to_iov(cmd_wr, VTSCSI_OUT_HEADER_LEN(sc), iov_out, niov_out, 0);
548 	nxferred = VTSCSI_OUT_HEADER_LEN(sc) + io->scsiio.ext_data_filled;
549 	free(cmd_rd);
550 	free(cmd_wr);
551 	ctl_scsi_free_io(io);
552 	return (nxferred);
553 }
554 
555 static void
556 pci_vtscsi_controlq_notify(void *vsc, struct vqueue_info *vq)
557 {
558 	struct pci_vtscsi_softc *sc;
559 	struct iovec iov[VTSCSI_MAXSEG];
560 	uint16_t idx, n;
561 	void *buf = NULL;
562 	size_t bufsize;
563 	int iolen;
564 
565 	sc = vsc;
566 
567 	while (vq_has_descs(vq)) {
568 		n = vq_getchain(vq, &idx, iov, VTSCSI_MAXSEG, NULL);
569 		bufsize = iov_to_buf(iov, n, &buf);
570 		iolen = pci_vtscsi_control_handle(sc, buf, bufsize);
571 		buf_to_iov(buf + bufsize - iolen, iolen, iov, n,
572 		    bufsize - iolen);
573 
574 		/*
575 		 * Release this chain and handle more
576 		 */
577 		vq_relchain(vq, idx, iolen);
578 	}
579 	vq_endchains(vq, 1);	/* Generate interrupt if appropriate. */
580 	free(buf);
581 }
582 
583 static void
584 pci_vtscsi_eventq_notify(void *vsc, struct vqueue_info *vq)
585 {
586 
587 	vq_kick_disable(vq);
588 }
589 
590 static void
591 pci_vtscsi_requestq_notify(void *vsc, struct vqueue_info *vq)
592 {
593 	struct pci_vtscsi_softc *sc;
594 	struct pci_vtscsi_queue *q;
595 	struct pci_vtscsi_request *req;
596 	struct iovec iov[VTSCSI_MAXSEG];
597 	uint16_t flags[VTSCSI_MAXSEG];
598 	uint16_t idx, n, i;
599 	int readable;
600 
601 	sc = vsc;
602 	q = &sc->vss_queues[vq->vq_num - 2];
603 
604 	while (vq_has_descs(vq)) {
605 		readable = 0;
606 		n = vq_getchain(vq, &idx, iov, VTSCSI_MAXSEG, flags);
607 
608 		/* Count readable descriptors */
609 		for (i = 0; i < n; i++) {
610 			if (flags[i] & VRING_DESC_F_WRITE)
611 				break;
612 
613 			readable++;
614 		}
615 
616 		req = calloc(1, sizeof(struct pci_vtscsi_request));
617 		req->vsr_idx = idx;
618 		req->vsr_queue = q;
619 		req->vsr_niov_in = readable;
620 		req->vsr_niov_out = n - readable;
621 		memcpy(req->vsr_iov_in, iov,
622 		    req->vsr_niov_in * sizeof(struct iovec));
623 		memcpy(req->vsr_iov_out, iov + readable,
624 		    req->vsr_niov_out * sizeof(struct iovec));
625 
626 		pthread_mutex_lock(&q->vsq_mtx);
627 		STAILQ_INSERT_TAIL(&q->vsq_requests, req, vsr_link);
628 		pthread_cond_signal(&q->vsq_cv);
629 		pthread_mutex_unlock(&q->vsq_mtx);
630 
631 		DPRINTF(("virtio-scsi: request <idx=%d> enqueued", idx));
632 	}
633 }
634 
635 static int
636 pci_vtscsi_init_queue(struct pci_vtscsi_softc *sc,
637     struct pci_vtscsi_queue *queue, int num)
638 {
639 	struct pci_vtscsi_worker *worker;
640 	char tname[MAXCOMLEN + 1];
641 	int i;
642 
643 	queue->vsq_sc = sc;
644 	queue->vsq_vq = &sc->vss_vq[num + 2];
645 
646 	pthread_mutex_init(&queue->vsq_mtx, NULL);
647 	pthread_mutex_init(&queue->vsq_qmtx, NULL);
648 	pthread_cond_init(&queue->vsq_cv, NULL);
649 	STAILQ_INIT(&queue->vsq_requests);
650 	LIST_INIT(&queue->vsq_workers);
651 
652 	for (i = 0; i < VTSCSI_THR_PER_Q; i++) {
653 		worker = calloc(1, sizeof(struct pci_vtscsi_worker));
654 		worker->vsw_queue = queue;
655 
656 		pthread_create(&worker->vsw_thread, NULL, &pci_vtscsi_proc,
657 		    (void *)worker);
658 
659 		snprintf(tname, sizeof(tname), "vtscsi:%d-%d", num, i);
660 		pthread_set_name_np(worker->vsw_thread, tname);
661 		LIST_INSERT_HEAD(&queue->vsq_workers, worker, vsw_link);
662 	}
663 
664 	return (0);
665 }
666 
667 static int
668 pci_vtscsi_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
669 {
670 	struct pci_vtscsi_softc *sc;
671 	char *opt, *optname;
672 	const char *devname;
673 	int i, optidx = 0;
674 
675 	sc = calloc(1, sizeof(struct pci_vtscsi_softc));
676 	devname = "/dev/cam/ctl";
677 	while ((opt = strsep(&opts, ",")) != NULL) {
678 		optname = strsep(&opt, "=");
679 		if (opt == NULL && optidx == 0) {
680 			if (optname[0] != 0)
681 				devname = optname;
682 		} else if (strcmp(optname, "dev") == 0 && opt != NULL) {
683 			devname = opt;
684 		} else if (strcmp(optname, "iid") == 0 && opt != NULL) {
685 			sc->vss_iid = strtoul(opt, NULL, 10);
686 		} else {
687 			EPRINTLN("Invalid option %s", optname);
688 			free(sc);
689 			return (1);
690 		}
691 		optidx++;
692 	}
693 
694 	sc->vss_ctl_fd = open(devname, O_RDWR);
695 	if (sc->vss_ctl_fd < 0) {
696 		WPRINTF(("cannot open %s: %s", devname, strerror(errno)));
697 		free(sc);
698 		return (1);
699 	}
700 
701 	vi_softc_linkup(&sc->vss_vs, &vtscsi_vi_consts, sc, pi, sc->vss_vq);
702 	sc->vss_vs.vs_mtx = &sc->vss_mtx;
703 
704 	/* controlq */
705 	sc->vss_vq[0].vq_qsize = VTSCSI_RINGSZ;
706 	sc->vss_vq[0].vq_notify = pci_vtscsi_controlq_notify;
707 
708 	/* eventq */
709 	sc->vss_vq[1].vq_qsize = VTSCSI_RINGSZ;
710 	sc->vss_vq[1].vq_notify = pci_vtscsi_eventq_notify;
711 
712 	/* request queues */
713 	for (i = 2; i < VTSCSI_MAXQ; i++) {
714 		sc->vss_vq[i].vq_qsize = VTSCSI_RINGSZ;
715 		sc->vss_vq[i].vq_notify = pci_vtscsi_requestq_notify;
716 		pci_vtscsi_init_queue(sc, &sc->vss_queues[i - 2], i - 2);
717 	}
718 
719 	/* initialize config space */
720 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_SCSI);
721 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
722 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_STORAGE);
723 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_SCSI);
724 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
725 
726 	if (vi_intr_init(&sc->vss_vs, 1, fbsdrun_virtio_msix()))
727 		return (1);
728 	vi_set_io_bar(&sc->vss_vs, 0);
729 
730 	return (0);
731 }
732 
733 
734 struct pci_devemu pci_de_vscsi = {
735 	.pe_emu =	"virtio-scsi",
736 	.pe_init =	pci_vtscsi_init,
737 	.pe_barwrite =	vi_pci_write,
738 	.pe_barread =	vi_pci_read
739 };
740 PCI_EMUL_SET(pci_de_vscsi);
741