xref: /openbmc/linux/arch/um/drivers/virtio_uml.c (revision dc6a81c3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Virtio vhost-user driver
4  *
5  * Copyright(c) 2019 Intel Corporation
6  *
7  * This driver allows virtio devices to be used over a vhost-user socket.
8  *
9  * Guest devices can be instantiated by kernel module or command line
10  * parameters. One device will be created for each parameter. Syntax:
11  *
12  *		virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]
13  * where:
14  *		<socket>	:= vhost-user socket path to connect
15  *		<virtio_id>	:= virtio device id (as in virtio_ids.h)
16  *		<platform_id>	:= (optional) platform device id
17  *
18  * example:
19  *		virtio_uml.device=/var/uml.socket:1
20  *
21  * Based on Virtio MMIO driver by Pawel Moll, copyright 2011-2014, ARM Ltd.
22  */
23 #include <linux/module.h>
24 #include <linux/platform_device.h>
25 #include <linux/slab.h>
26 #include <linux/virtio.h>
27 #include <linux/virtio_config.h>
28 #include <linux/virtio_ring.h>
29 #include <shared/as-layout.h>
30 #include <irq_kern.h>
31 #include <init.h>
32 #include <os.h>
33 #include "vhost_user.h"
34 
35 /* Workaround due to a conflict between irq_user.h and irqreturn.h */
36 #ifdef IRQ_NONE
37 #undef IRQ_NONE
38 #endif
39 
40 #define MAX_SUPPORTED_QUEUE_SIZE	256
41 
42 #define to_virtio_uml_device(_vdev) \
43 	container_of(_vdev, struct virtio_uml_device, vdev)
44 
45 struct virtio_uml_platform_data {
46 	u32 virtio_device_id;
47 	const char *socket_path;
48 	struct work_struct conn_broken_wk;
49 	struct platform_device *pdev;
50 };
51 
52 struct virtio_uml_device {
53 	struct virtio_device vdev;
54 	struct platform_device *pdev;
55 
56 	int sock, req_fd;
57 	u64 features;
58 	u64 protocol_features;
59 	u8 status;
60 	u8 registered:1;
61 };
62 
63 struct virtio_uml_vq_info {
64 	int kick_fd, call_fd;
65 	char name[32];
66 };
67 
68 extern unsigned long long physmem_size, highmem;
69 
70 #define vu_err(vu_dev, ...)	dev_err(&(vu_dev)->pdev->dev, __VA_ARGS__)
71 
72 /* Vhost-user protocol */
73 
74 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
75 			    const int *fds, unsigned int fds_num)
76 {
77 	int rc;
78 
79 	do {
80 		rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
81 		if (rc > 0) {
82 			buf += rc;
83 			len -= rc;
84 			fds = NULL;
85 			fds_num = 0;
86 		}
87 	} while (len && (rc >= 0 || rc == -EINTR));
88 
89 	if (rc < 0)
90 		return rc;
91 	return 0;
92 }
93 
94 static int full_read(int fd, void *buf, int len, bool abortable)
95 {
96 	int rc;
97 
98 	do {
99 		rc = os_read_file(fd, buf, len);
100 		if (rc > 0) {
101 			buf += rc;
102 			len -= rc;
103 		}
104 	} while (len && (rc > 0 || rc == -EINTR || (!abortable && rc == -EAGAIN)));
105 
106 	if (rc < 0)
107 		return rc;
108 	if (rc == 0)
109 		return -ECONNRESET;
110 	return 0;
111 }
112 
113 static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
114 {
115 	return full_read(fd, msg, sizeof(msg->header), true);
116 }
117 
118 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
119 			   int fd, struct vhost_user_msg *msg,
120 			   size_t max_payload_size)
121 {
122 	size_t size;
123 	int rc = vhost_user_recv_header(fd, msg);
124 
125 	if (rc == -ECONNRESET && vu_dev->registered) {
126 		struct virtio_uml_platform_data *pdata;
127 
128 		pdata = vu_dev->pdev->dev.platform_data;
129 
130 		virtio_break_device(&vu_dev->vdev);
131 		schedule_work(&pdata->conn_broken_wk);
132 	}
133 	if (rc)
134 		return rc;
135 	size = msg->header.size;
136 	if (size > max_payload_size)
137 		return -EPROTO;
138 	return full_read(fd, &msg->payload, size, false);
139 }
140 
141 static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
142 				struct vhost_user_msg *msg,
143 				size_t max_payload_size)
144 {
145 	int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg, max_payload_size);
146 
147 	if (rc)
148 		return rc;
149 
150 	if (msg->header.flags != (VHOST_USER_FLAG_REPLY | VHOST_USER_VERSION))
151 		return -EPROTO;
152 
153 	return 0;
154 }
155 
156 static int vhost_user_recv_u64(struct virtio_uml_device *vu_dev,
157 			       u64 *value)
158 {
159 	struct vhost_user_msg msg;
160 	int rc = vhost_user_recv_resp(vu_dev, &msg,
161 				      sizeof(msg.payload.integer));
162 
163 	if (rc)
164 		return rc;
165 	if (msg.header.size != sizeof(msg.payload.integer))
166 		return -EPROTO;
167 	*value = msg.payload.integer;
168 	return 0;
169 }
170 
171 static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
172 			       struct vhost_user_msg *msg,
173 			       size_t max_payload_size)
174 {
175 	int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg, max_payload_size);
176 
177 	if (rc)
178 		return rc;
179 
180 	if ((msg->header.flags & ~VHOST_USER_FLAG_NEED_REPLY) !=
181 			VHOST_USER_VERSION)
182 		return -EPROTO;
183 
184 	return 0;
185 }
186 
187 static int vhost_user_send(struct virtio_uml_device *vu_dev,
188 			   bool need_response, struct vhost_user_msg *msg,
189 			   int *fds, size_t num_fds)
190 {
191 	size_t size = sizeof(msg->header) + msg->header.size;
192 	bool request_ack;
193 	int rc;
194 
195 	msg->header.flags |= VHOST_USER_VERSION;
196 
197 	/*
198 	 * The need_response flag indicates that we already need a response,
199 	 * e.g. to read the features. In these cases, don't request an ACK as
200 	 * it is meaningless. Also request an ACK only if supported.
201 	 */
202 	request_ack = !need_response;
203 	if (!(vu_dev->protocol_features &
204 			BIT_ULL(VHOST_USER_PROTOCOL_F_REPLY_ACK)))
205 		request_ack = false;
206 
207 	if (request_ack)
208 		msg->header.flags |= VHOST_USER_FLAG_NEED_REPLY;
209 
210 	rc = full_sendmsg_fds(vu_dev->sock, msg, size, fds, num_fds);
211 	if (rc < 0)
212 		return rc;
213 
214 	if (request_ack) {
215 		uint64_t status;
216 
217 		rc = vhost_user_recv_u64(vu_dev, &status);
218 		if (rc)
219 			return rc;
220 
221 		if (status) {
222 			vu_err(vu_dev, "slave reports error: %llu\n", status);
223 			return -EIO;
224 		}
225 	}
226 
227 	return 0;
228 }
229 
230 static int vhost_user_send_no_payload(struct virtio_uml_device *vu_dev,
231 				      bool need_response, u32 request)
232 {
233 	struct vhost_user_msg msg = {
234 		.header.request = request,
235 	};
236 
237 	return vhost_user_send(vu_dev, need_response, &msg, NULL, 0);
238 }
239 
240 static int vhost_user_send_no_payload_fd(struct virtio_uml_device *vu_dev,
241 					 u32 request, int fd)
242 {
243 	struct vhost_user_msg msg = {
244 		.header.request = request,
245 	};
246 
247 	return vhost_user_send(vu_dev, false, &msg, &fd, 1);
248 }
249 
250 static int vhost_user_send_u64(struct virtio_uml_device *vu_dev,
251 			       u32 request, u64 value)
252 {
253 	struct vhost_user_msg msg = {
254 		.header.request = request,
255 		.header.size = sizeof(msg.payload.integer),
256 		.payload.integer = value,
257 	};
258 
259 	return vhost_user_send(vu_dev, false, &msg, NULL, 0);
260 }
261 
262 static int vhost_user_set_owner(struct virtio_uml_device *vu_dev)
263 {
264 	return vhost_user_send_no_payload(vu_dev, false, VHOST_USER_SET_OWNER);
265 }
266 
267 static int vhost_user_get_features(struct virtio_uml_device *vu_dev,
268 				   u64 *features)
269 {
270 	int rc = vhost_user_send_no_payload(vu_dev, true,
271 					    VHOST_USER_GET_FEATURES);
272 
273 	if (rc)
274 		return rc;
275 	return vhost_user_recv_u64(vu_dev, features);
276 }
277 
278 static int vhost_user_set_features(struct virtio_uml_device *vu_dev,
279 				   u64 features)
280 {
281 	return vhost_user_send_u64(vu_dev, VHOST_USER_SET_FEATURES, features);
282 }
283 
284 static int vhost_user_get_protocol_features(struct virtio_uml_device *vu_dev,
285 					    u64 *protocol_features)
286 {
287 	int rc = vhost_user_send_no_payload(vu_dev, true,
288 			VHOST_USER_GET_PROTOCOL_FEATURES);
289 
290 	if (rc)
291 		return rc;
292 	return vhost_user_recv_u64(vu_dev, protocol_features);
293 }
294 
295 static int vhost_user_set_protocol_features(struct virtio_uml_device *vu_dev,
296 					    u64 protocol_features)
297 {
298 	return vhost_user_send_u64(vu_dev, VHOST_USER_SET_PROTOCOL_FEATURES,
299 				   protocol_features);
300 }
301 
302 static void vhost_user_reply(struct virtio_uml_device *vu_dev,
303 			     struct vhost_user_msg *msg, int response)
304 {
305 	struct vhost_user_msg reply = {
306 		.payload.integer = response,
307 	};
308 	size_t size = sizeof(reply.header) + sizeof(reply.payload.integer);
309 	int rc;
310 
311 	reply.header = msg->header;
312 	reply.header.flags &= ~VHOST_USER_FLAG_NEED_REPLY;
313 	reply.header.flags |= VHOST_USER_FLAG_REPLY;
314 	reply.header.size = sizeof(reply.payload.integer);
315 
316 	rc = full_sendmsg_fds(vu_dev->req_fd, &reply, size, NULL, 0);
317 
318 	if (rc)
319 		vu_err(vu_dev,
320 		       "sending reply to slave request failed: %d (size %zu)\n",
321 		       rc, size);
322 }
323 
324 static irqreturn_t vu_req_interrupt(int irq, void *data)
325 {
326 	struct virtio_uml_device *vu_dev = data;
327 	int response = 1;
328 	struct {
329 		struct vhost_user_msg msg;
330 		u8 extra_payload[512];
331 	} msg;
332 	int rc;
333 
334 	rc = vhost_user_recv_req(vu_dev, &msg.msg,
335 				 sizeof(msg.msg.payload) +
336 				 sizeof(msg.extra_payload));
337 
338 	if (rc)
339 		return IRQ_NONE;
340 
341 	switch (msg.msg.header.request) {
342 	case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG:
343 		virtio_config_changed(&vu_dev->vdev);
344 		response = 0;
345 		break;
346 	case VHOST_USER_SLAVE_IOTLB_MSG:
347 		/* not supported - VIRTIO_F_IOMMU_PLATFORM */
348 	case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
349 		/* not supported - VHOST_USER_PROTOCOL_F_HOST_NOTIFIER */
350 	default:
351 		vu_err(vu_dev, "unexpected slave request %d\n",
352 		       msg.msg.header.request);
353 	}
354 
355 	if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
356 		vhost_user_reply(vu_dev, &msg.msg, response);
357 
358 	return IRQ_HANDLED;
359 }
360 
361 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
362 {
363 	int rc, req_fds[2];
364 
365 	/* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
366 	rc = os_pipe(req_fds, true, true);
367 	if (rc < 0)
368 		return rc;
369 	vu_dev->req_fd = req_fds[0];
370 
371 	rc = um_request_irq(VIRTIO_IRQ, vu_dev->req_fd, IRQ_READ,
372 			    vu_req_interrupt, IRQF_SHARED,
373 			    vu_dev->pdev->name, vu_dev);
374 	if (rc)
375 		goto err_close;
376 
377 	rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
378 					   req_fds[1]);
379 	if (rc)
380 		goto err_free_irq;
381 
382 	goto out;
383 
384 err_free_irq:
385 	um_free_irq(VIRTIO_IRQ, vu_dev);
386 err_close:
387 	os_close_file(req_fds[0]);
388 out:
389 	/* Close unused write end of request fds */
390 	os_close_file(req_fds[1]);
391 	return rc;
392 }
393 
394 static int vhost_user_init(struct virtio_uml_device *vu_dev)
395 {
396 	int rc = vhost_user_set_owner(vu_dev);
397 
398 	if (rc)
399 		return rc;
400 	rc = vhost_user_get_features(vu_dev, &vu_dev->features);
401 	if (rc)
402 		return rc;
403 
404 	if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
405 		rc = vhost_user_get_protocol_features(vu_dev,
406 				&vu_dev->protocol_features);
407 		if (rc)
408 			return rc;
409 		vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
410 		rc = vhost_user_set_protocol_features(vu_dev,
411 				vu_dev->protocol_features);
412 		if (rc)
413 			return rc;
414 	}
415 
416 	if (vu_dev->protocol_features &
417 			BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
418 		rc = vhost_user_init_slave_req(vu_dev);
419 		if (rc)
420 			return rc;
421 	}
422 
423 	return 0;
424 }
425 
426 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
427 				  u32 offset, void *buf, u32 len)
428 {
429 	u32 cfg_size = offset + len;
430 	struct vhost_user_msg *msg;
431 	size_t payload_size = sizeof(msg->payload.config) + cfg_size;
432 	size_t msg_size = sizeof(msg->header) + payload_size;
433 	int rc;
434 
435 	if (!(vu_dev->protocol_features &
436 	      BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
437 		return;
438 
439 	msg = kzalloc(msg_size, GFP_KERNEL);
440 	if (!msg)
441 		return;
442 	msg->header.request = VHOST_USER_GET_CONFIG;
443 	msg->header.size = payload_size;
444 	msg->payload.config.offset = 0;
445 	msg->payload.config.size = cfg_size;
446 
447 	rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
448 	if (rc) {
449 		vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
450 		       rc);
451 		goto free;
452 	}
453 
454 	rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
455 	if (rc) {
456 		vu_err(vu_dev,
457 		       "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
458 		       rc);
459 		goto free;
460 	}
461 
462 	if (msg->header.size != payload_size ||
463 	    msg->payload.config.size != cfg_size) {
464 		rc = -EPROTO;
465 		vu_err(vu_dev,
466 		       "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
467 		       msg->header.size, payload_size,
468 		       msg->payload.config.size, cfg_size);
469 		goto free;
470 	}
471 	memcpy(buf, msg->payload.config.payload + offset, len);
472 
473 free:
474 	kfree(msg);
475 }
476 
477 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
478 				  u32 offset, const void *buf, u32 len)
479 {
480 	struct vhost_user_msg *msg;
481 	size_t payload_size = sizeof(msg->payload.config) + len;
482 	size_t msg_size = sizeof(msg->header) + payload_size;
483 	int rc;
484 
485 	if (!(vu_dev->protocol_features &
486 	      BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
487 		return;
488 
489 	msg = kzalloc(msg_size, GFP_KERNEL);
490 	if (!msg)
491 		return;
492 	msg->header.request = VHOST_USER_SET_CONFIG;
493 	msg->header.size = payload_size;
494 	msg->payload.config.offset = offset;
495 	msg->payload.config.size = len;
496 	memcpy(msg->payload.config.payload, buf, len);
497 
498 	rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
499 	if (rc)
500 		vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
501 		       rc);
502 
503 	kfree(msg);
504 }
505 
506 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
507 				      struct vhost_user_mem_region *region_out)
508 {
509 	unsigned long long mem_offset;
510 	int rc = phys_mapping(addr, &mem_offset);
511 
512 	if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
513 		return -EFAULT;
514 	*fd_out = rc;
515 	region_out->guest_addr = addr;
516 	region_out->user_addr = addr;
517 	region_out->size = size;
518 	region_out->mmap_offset = mem_offset;
519 
520 	/* Ensure mapping is valid for the entire region */
521 	rc = phys_mapping(addr + size - 1, &mem_offset);
522 	if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
523 		 addr + size - 1, rc, *fd_out))
524 		return -EFAULT;
525 	return 0;
526 }
527 
528 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
529 {
530 	struct vhost_user_msg msg = {
531 		.header.request = VHOST_USER_SET_MEM_TABLE,
532 		.header.size = sizeof(msg.payload.mem_regions),
533 		.payload.mem_regions.num = 1,
534 	};
535 	unsigned long reserved = uml_reserved - uml_physmem;
536 	int fds[2];
537 	int rc;
538 
539 	/*
540 	 * This is a bit tricky, see also the comment with setup_physmem().
541 	 *
542 	 * Essentially, setup_physmem() uses a file to mmap() our physmem,
543 	 * but the code and data we *already* have is omitted. To us, this
544 	 * is no difference, since they both become part of our address
545 	 * space and memory consumption. To somebody looking in from the
546 	 * outside, however, it is different because the part of our memory
547 	 * consumption that's already part of the binary (code/data) is not
548 	 * mapped from the file, so it's not visible to another mmap from
549 	 * the file descriptor.
550 	 *
551 	 * Thus, don't advertise this space to the vhost-user slave. This
552 	 * means that the slave will likely abort or similar when we give
553 	 * it an address from the hidden range, since it's not marked as
554 	 * a valid address, but at least that way we detect the issue and
555 	 * don't just have the slave read an all-zeroes buffer from the
556 	 * shared memory file, or write something there that we can never
557 	 * see (depending on the direction of the virtqueue traffic.)
558 	 *
559 	 * Since we usually don't want to use .text for virtio buffers,
560 	 * this effectively means that you cannot use
561 	 *  1) global variables, which are in the .bss and not in the shm
562 	 *     file-backed memory
563 	 *  2) the stack in some processes, depending on where they have
564 	 *     their stack (or maybe only no interrupt stack?)
565 	 *
566 	 * The stack is already not typically valid for DMA, so this isn't
567 	 * much of a restriction, but global variables might be encountered.
568 	 *
569 	 * It might be possible to fix it by copying around the data that's
570 	 * between bss_start and where we map the file now, but it's not
571 	 * something that you typically encounter with virtio drivers, so
572 	 * it didn't seem worthwhile.
573 	 */
574 	rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
575 					&fds[0],
576 					&msg.payload.mem_regions.regions[0]);
577 
578 	if (rc < 0)
579 		return rc;
580 	if (highmem) {
581 		msg.payload.mem_regions.num++;
582 		rc = vhost_user_init_mem_region(__pa(end_iomem), highmem,
583 				&fds[1], &msg.payload.mem_regions.regions[1]);
584 		if (rc < 0)
585 			return rc;
586 	}
587 
588 	return vhost_user_send(vu_dev, false, &msg, fds,
589 			       msg.payload.mem_regions.num);
590 }
591 
592 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
593 				      u32 request, u32 index, u32 num)
594 {
595 	struct vhost_user_msg msg = {
596 		.header.request = request,
597 		.header.size = sizeof(msg.payload.vring_state),
598 		.payload.vring_state.index = index,
599 		.payload.vring_state.num = num,
600 	};
601 
602 	return vhost_user_send(vu_dev, false, &msg, NULL, 0);
603 }
604 
605 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
606 				    u32 index, u32 num)
607 {
608 	return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
609 					  index, num);
610 }
611 
612 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
613 				     u32 index, u32 offset)
614 {
615 	return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
616 					  index, offset);
617 }
618 
619 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
620 				     u32 index, u64 desc, u64 used, u64 avail,
621 				     u64 log)
622 {
623 	struct vhost_user_msg msg = {
624 		.header.request = VHOST_USER_SET_VRING_ADDR,
625 		.header.size = sizeof(msg.payload.vring_addr),
626 		.payload.vring_addr.index = index,
627 		.payload.vring_addr.desc = desc,
628 		.payload.vring_addr.used = used,
629 		.payload.vring_addr.avail = avail,
630 		.payload.vring_addr.log = log,
631 	};
632 
633 	return vhost_user_send(vu_dev, false, &msg, NULL, 0);
634 }
635 
636 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
637 				   u32 request, int index, int fd)
638 {
639 	struct vhost_user_msg msg = {
640 		.header.request = request,
641 		.header.size = sizeof(msg.payload.integer),
642 		.payload.integer = index,
643 	};
644 
645 	if (index & ~VHOST_USER_VRING_INDEX_MASK)
646 		return -EINVAL;
647 	if (fd < 0) {
648 		msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
649 		return vhost_user_send(vu_dev, false, &msg, NULL, 0);
650 	}
651 	return vhost_user_send(vu_dev, false, &msg, &fd, 1);
652 }
653 
654 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
655 				     int index, int fd)
656 {
657 	return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
658 				       index, fd);
659 }
660 
661 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
662 				     int index, int fd)
663 {
664 	return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
665 				       index, fd);
666 }
667 
668 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
669 				       u32 index, bool enable)
670 {
671 	if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
672 		return 0;
673 
674 	return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
675 					  index, enable);
676 }
677 
678 
679 /* Virtio interface */
680 
681 static bool vu_notify(struct virtqueue *vq)
682 {
683 	struct virtio_uml_vq_info *info = vq->priv;
684 	const uint64_t n = 1;
685 	int rc;
686 
687 	do {
688 		rc = os_write_file(info->kick_fd, &n, sizeof(n));
689 	} while (rc == -EINTR);
690 	return !WARN(rc != sizeof(n), "write returned %d\n", rc);
691 }
692 
693 static irqreturn_t vu_interrupt(int irq, void *opaque)
694 {
695 	struct virtqueue *vq = opaque;
696 	struct virtio_uml_vq_info *info = vq->priv;
697 	uint64_t n;
698 	int rc;
699 	irqreturn_t ret = IRQ_NONE;
700 
701 	do {
702 		rc = os_read_file(info->call_fd, &n, sizeof(n));
703 		if (rc == sizeof(n))
704 			ret |= vring_interrupt(irq, vq);
705 	} while (rc == sizeof(n) || rc == -EINTR);
706 	WARN(rc != -EAGAIN, "read returned %d\n", rc);
707 	return ret;
708 }
709 
710 
711 static void vu_get(struct virtio_device *vdev, unsigned offset,
712 		   void *buf, unsigned len)
713 {
714 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
715 
716 	vhost_user_get_config(vu_dev, offset, buf, len);
717 }
718 
719 static void vu_set(struct virtio_device *vdev, unsigned offset,
720 		   const void *buf, unsigned len)
721 {
722 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
723 
724 	vhost_user_set_config(vu_dev, offset, buf, len);
725 }
726 
727 static u8 vu_get_status(struct virtio_device *vdev)
728 {
729 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
730 
731 	return vu_dev->status;
732 }
733 
734 static void vu_set_status(struct virtio_device *vdev, u8 status)
735 {
736 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
737 
738 	vu_dev->status = status;
739 }
740 
741 static void vu_reset(struct virtio_device *vdev)
742 {
743 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
744 
745 	vu_dev->status = 0;
746 }
747 
748 static void vu_del_vq(struct virtqueue *vq)
749 {
750 	struct virtio_uml_vq_info *info = vq->priv;
751 
752 	um_free_irq(VIRTIO_IRQ, vq);
753 
754 	os_close_file(info->call_fd);
755 	os_close_file(info->kick_fd);
756 
757 	vring_del_virtqueue(vq);
758 	kfree(info);
759 }
760 
761 static void vu_del_vqs(struct virtio_device *vdev)
762 {
763 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
764 	struct virtqueue *vq, *n;
765 	u64 features;
766 
767 	/* Note: reverse order as a workaround to a decoding bug in snabb */
768 	list_for_each_entry_reverse(vq, &vdev->vqs, list)
769 		WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
770 
771 	/* Ensure previous messages have been processed */
772 	WARN_ON(vhost_user_get_features(vu_dev, &features));
773 
774 	list_for_each_entry_safe(vq, n, &vdev->vqs, list)
775 		vu_del_vq(vq);
776 }
777 
778 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
779 			       struct virtqueue *vq)
780 {
781 	struct virtio_uml_vq_info *info = vq->priv;
782 	int call_fds[2];
783 	int rc;
784 
785 	/* Use a pipe for call fd, since SIGIO is not supported for eventfd */
786 	rc = os_pipe(call_fds, true, true);
787 	if (rc < 0)
788 		return rc;
789 
790 	info->call_fd = call_fds[0];
791 	rc = um_request_irq(VIRTIO_IRQ, info->call_fd, IRQ_READ,
792 			    vu_interrupt, IRQF_SHARED, info->name, vq);
793 	if (rc)
794 		goto close_both;
795 
796 	rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
797 	if (rc)
798 		goto release_irq;
799 
800 	goto out;
801 
802 release_irq:
803 	um_free_irq(VIRTIO_IRQ, vq);
804 close_both:
805 	os_close_file(call_fds[0]);
806 out:
807 	/* Close (unused) write end of call fds */
808 	os_close_file(call_fds[1]);
809 
810 	return rc;
811 }
812 
813 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
814 				     unsigned index, vq_callback_t *callback,
815 				     const char *name, bool ctx)
816 {
817 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
818 	struct platform_device *pdev = vu_dev->pdev;
819 	struct virtio_uml_vq_info *info;
820 	struct virtqueue *vq;
821 	int num = MAX_SUPPORTED_QUEUE_SIZE;
822 	int rc;
823 
824 	info = kzalloc(sizeof(*info), GFP_KERNEL);
825 	if (!info) {
826 		rc = -ENOMEM;
827 		goto error_kzalloc;
828 	}
829 	snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
830 		 pdev->id, name);
831 
832 	vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
833 				    ctx, vu_notify, callback, info->name);
834 	if (!vq) {
835 		rc = -ENOMEM;
836 		goto error_create;
837 	}
838 	vq->priv = info;
839 	num = virtqueue_get_vring_size(vq);
840 
841 	rc = os_eventfd(0, 0);
842 	if (rc < 0)
843 		goto error_kick;
844 	info->kick_fd = rc;
845 
846 	rc = vu_setup_vq_call_fd(vu_dev, vq);
847 	if (rc)
848 		goto error_call;
849 
850 	rc = vhost_user_set_vring_num(vu_dev, index, num);
851 	if (rc)
852 		goto error_setup;
853 
854 	rc = vhost_user_set_vring_base(vu_dev, index, 0);
855 	if (rc)
856 		goto error_setup;
857 
858 	rc = vhost_user_set_vring_addr(vu_dev, index,
859 				       virtqueue_get_desc_addr(vq),
860 				       virtqueue_get_used_addr(vq),
861 				       virtqueue_get_avail_addr(vq),
862 				       (u64) -1);
863 	if (rc)
864 		goto error_setup;
865 
866 	return vq;
867 
868 error_setup:
869 	um_free_irq(VIRTIO_IRQ, vq);
870 	os_close_file(info->call_fd);
871 error_call:
872 	os_close_file(info->kick_fd);
873 error_kick:
874 	vring_del_virtqueue(vq);
875 error_create:
876 	kfree(info);
877 error_kzalloc:
878 	return ERR_PTR(rc);
879 }
880 
881 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
882 		       struct virtqueue *vqs[], vq_callback_t *callbacks[],
883 		       const char * const names[], const bool *ctx,
884 		       struct irq_affinity *desc)
885 {
886 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
887 	int i, queue_idx = 0, rc;
888 	struct virtqueue *vq;
889 
890 	rc = vhost_user_set_mem_table(vu_dev);
891 	if (rc)
892 		return rc;
893 
894 	for (i = 0; i < nvqs; ++i) {
895 		if (!names[i]) {
896 			vqs[i] = NULL;
897 			continue;
898 		}
899 
900 		vqs[i] = vu_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
901 				     ctx ? ctx[i] : false);
902 		if (IS_ERR(vqs[i])) {
903 			rc = PTR_ERR(vqs[i]);
904 			goto error_setup;
905 		}
906 	}
907 
908 	list_for_each_entry(vq, &vdev->vqs, list) {
909 		struct virtio_uml_vq_info *info = vq->priv;
910 
911 		rc = vhost_user_set_vring_kick(vu_dev, vq->index,
912 					       info->kick_fd);
913 		if (rc)
914 			goto error_setup;
915 
916 		rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
917 		if (rc)
918 			goto error_setup;
919 	}
920 
921 	return 0;
922 
923 error_setup:
924 	vu_del_vqs(vdev);
925 	return rc;
926 }
927 
928 static u64 vu_get_features(struct virtio_device *vdev)
929 {
930 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
931 
932 	return vu_dev->features;
933 }
934 
935 static int vu_finalize_features(struct virtio_device *vdev)
936 {
937 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
938 	u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
939 
940 	vring_transport_features(vdev);
941 	vu_dev->features = vdev->features | supported;
942 
943 	return vhost_user_set_features(vu_dev, vu_dev->features);
944 }
945 
946 static const char *vu_bus_name(struct virtio_device *vdev)
947 {
948 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
949 
950 	return vu_dev->pdev->name;
951 }
952 
953 static const struct virtio_config_ops virtio_uml_config_ops = {
954 	.get = vu_get,
955 	.set = vu_set,
956 	.get_status = vu_get_status,
957 	.set_status = vu_set_status,
958 	.reset = vu_reset,
959 	.find_vqs = vu_find_vqs,
960 	.del_vqs = vu_del_vqs,
961 	.get_features = vu_get_features,
962 	.finalize_features = vu_finalize_features,
963 	.bus_name = vu_bus_name,
964 };
965 
966 static void virtio_uml_release_dev(struct device *d)
967 {
968 	struct virtio_device *vdev =
969 			container_of(d, struct virtio_device, dev);
970 	struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
971 
972 	/* might not have been opened due to not negotiating the feature */
973 	if (vu_dev->req_fd >= 0) {
974 		um_free_irq(VIRTIO_IRQ, vu_dev);
975 		os_close_file(vu_dev->req_fd);
976 	}
977 
978 	os_close_file(vu_dev->sock);
979 }
980 
981 /* Platform device */
982 
983 static int virtio_uml_probe(struct platform_device *pdev)
984 {
985 	struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
986 	struct virtio_uml_device *vu_dev;
987 	int rc;
988 
989 	if (!pdata)
990 		return -EINVAL;
991 
992 	vu_dev = devm_kzalloc(&pdev->dev, sizeof(*vu_dev), GFP_KERNEL);
993 	if (!vu_dev)
994 		return -ENOMEM;
995 
996 	vu_dev->vdev.dev.parent = &pdev->dev;
997 	vu_dev->vdev.dev.release = virtio_uml_release_dev;
998 	vu_dev->vdev.config = &virtio_uml_config_ops;
999 	vu_dev->vdev.id.device = pdata->virtio_device_id;
1000 	vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1001 	vu_dev->pdev = pdev;
1002 	vu_dev->req_fd = -1;
1003 
1004 	do {
1005 		rc = os_connect_socket(pdata->socket_path);
1006 	} while (rc == -EINTR);
1007 	if (rc < 0)
1008 		return rc;
1009 	vu_dev->sock = rc;
1010 
1011 	rc = vhost_user_init(vu_dev);
1012 	if (rc)
1013 		goto error_init;
1014 
1015 	platform_set_drvdata(pdev, vu_dev);
1016 
1017 	rc = register_virtio_device(&vu_dev->vdev);
1018 	if (rc)
1019 		put_device(&vu_dev->vdev.dev);
1020 	vu_dev->registered = 1;
1021 	return rc;
1022 
1023 error_init:
1024 	os_close_file(vu_dev->sock);
1025 	return rc;
1026 }
1027 
1028 static int virtio_uml_remove(struct platform_device *pdev)
1029 {
1030 	struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1031 
1032 	unregister_virtio_device(&vu_dev->vdev);
1033 	return 0;
1034 }
1035 
1036 /* Command line device list */
1037 
1038 static void vu_cmdline_release_dev(struct device *d)
1039 {
1040 }
1041 
1042 static struct device vu_cmdline_parent = {
1043 	.init_name = "virtio-uml-cmdline",
1044 	.release = vu_cmdline_release_dev,
1045 };
1046 
1047 static bool vu_cmdline_parent_registered;
1048 static int vu_cmdline_id;
1049 
1050 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1051 {
1052 	struct platform_device *pdev = to_platform_device(dev);
1053 	struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1054 
1055 	kfree(pdata->socket_path);
1056 	platform_device_unregister(pdev);
1057 	return 0;
1058 }
1059 
1060 static void vu_conn_broken(struct work_struct *wk)
1061 {
1062 	struct virtio_uml_platform_data *pdata;
1063 
1064 	pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1065 	vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1066 }
1067 
1068 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1069 {
1070 	const char *ids = strchr(device, ':');
1071 	unsigned int virtio_device_id;
1072 	int processed, consumed, err;
1073 	char *socket_path;
1074 	struct virtio_uml_platform_data pdata, *ppdata;
1075 	struct platform_device *pdev;
1076 
1077 	if (!ids || ids == device)
1078 		return -EINVAL;
1079 
1080 	processed = sscanf(ids, ":%u%n:%d%n",
1081 			   &virtio_device_id, &consumed,
1082 			   &vu_cmdline_id, &consumed);
1083 
1084 	if (processed < 1 || ids[consumed])
1085 		return -EINVAL;
1086 
1087 	if (!vu_cmdline_parent_registered) {
1088 		err = device_register(&vu_cmdline_parent);
1089 		if (err) {
1090 			pr_err("Failed to register parent device!\n");
1091 			put_device(&vu_cmdline_parent);
1092 			return err;
1093 		}
1094 		vu_cmdline_parent_registered = true;
1095 	}
1096 
1097 	socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1098 	if (!socket_path)
1099 		return -ENOMEM;
1100 
1101 	pdata.virtio_device_id = (u32) virtio_device_id;
1102 	pdata.socket_path = socket_path;
1103 
1104 	pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1105 		vu_cmdline_id, virtio_device_id, socket_path);
1106 
1107 	pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1108 					     vu_cmdline_id++, &pdata,
1109 					     sizeof(pdata));
1110 	err = PTR_ERR_OR_ZERO(pdev);
1111 	if (err)
1112 		goto free;
1113 
1114 	ppdata = pdev->dev.platform_data;
1115 	ppdata->pdev = pdev;
1116 	INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1117 
1118 	return 0;
1119 
1120 free:
1121 	kfree(socket_path);
1122 	return err;
1123 }
1124 
1125 static int vu_cmdline_get_device(struct device *dev, void *data)
1126 {
1127 	struct platform_device *pdev = to_platform_device(dev);
1128 	struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1129 	char *buffer = data;
1130 	unsigned int len = strlen(buffer);
1131 
1132 	snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1133 		 pdata->socket_path, pdata->virtio_device_id, pdev->id);
1134 	return 0;
1135 }
1136 
1137 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1138 {
1139 	buffer[0] = '\0';
1140 	if (vu_cmdline_parent_registered)
1141 		device_for_each_child(&vu_cmdline_parent, buffer,
1142 				      vu_cmdline_get_device);
1143 	return strlen(buffer) + 1;
1144 }
1145 
1146 static const struct kernel_param_ops vu_cmdline_param_ops = {
1147 	.set = vu_cmdline_set,
1148 	.get = vu_cmdline_get,
1149 };
1150 
1151 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1152 __uml_help(vu_cmdline_param_ops,
1153 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1154 "    Configure a virtio device over a vhost-user socket.\n"
1155 "    See virtio_ids.h for a list of possible virtio device id values.\n"
1156 "    Optionally use a specific platform_device id.\n\n"
1157 );
1158 
1159 
1160 static void vu_unregister_cmdline_devices(void)
1161 {
1162 	if (vu_cmdline_parent_registered) {
1163 		device_for_each_child(&vu_cmdline_parent, NULL,
1164 				      vu_unregister_cmdline_device);
1165 		device_unregister(&vu_cmdline_parent);
1166 		vu_cmdline_parent_registered = false;
1167 	}
1168 }
1169 
1170 /* Platform driver */
1171 
1172 static const struct of_device_id virtio_uml_match[] = {
1173 	{ .compatible = "virtio,uml", },
1174 	{ }
1175 };
1176 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1177 
1178 static struct platform_driver virtio_uml_driver = {
1179 	.probe = virtio_uml_probe,
1180 	.remove = virtio_uml_remove,
1181 	.driver = {
1182 		.name = "virtio-uml",
1183 		.of_match_table = virtio_uml_match,
1184 	},
1185 };
1186 
1187 static int __init virtio_uml_init(void)
1188 {
1189 	return platform_driver_register(&virtio_uml_driver);
1190 }
1191 
1192 static void __exit virtio_uml_exit(void)
1193 {
1194 	platform_driver_unregister(&virtio_uml_driver);
1195 	vu_unregister_cmdline_devices();
1196 }
1197 
1198 module_init(virtio_uml_init);
1199 module_exit(virtio_uml_exit);
1200 __uml_exitcall(virtio_uml_exit);
1201 
1202 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1203 MODULE_LICENSE("GPL");
1204