1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Intel Corporation
3 */
4
5 #include <sys/types.h>
6 #include <sys/stat.h>
7 #include <fcntl.h>
8 #include <unistd.h>
9 #include <errno.h>
10
11 #include <rte_memory.h>
12
13 #include "vhost.h"
14 #include "virtio_user_dev.h"
15 #include "vhost_kernel_tap.h"
16
17 struct vhost_kernel_data {
18 int *vhostfds;
19 int *tapfds;
20 };
21
22 struct vhost_memory_kernel {
23 uint32_t nregions;
24 uint32_t padding;
25 struct vhost_memory_region regions[0];
26 };
27
28 /* vhost kernel ioctls */
29 #define VHOST_VIRTIO 0xAF
30 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
31 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
32 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
33 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
34 #define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
35 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
36 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
37 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
38 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
39 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
40 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
41 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
42 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
43 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
44 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
45
46 /* with below features, vhost kernel does not need to do the checksum and TSO,
47 * these info will be passed to virtio_user through virtio net header.
48 */
49 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
50 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
51 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
52 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
53 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
54 (1ULL << VIRTIO_NET_F_GUEST_UFO))
55
56 /* with below features, when flows from virtio_user to vhost kernel
57 * (1) if flows goes up through the kernel networking stack, it does not need
58 * to verify checksum, which can save CPU cycles;
59 * (2) if flows goes through a Linux bridge and outside from an interface
60 * (kernel driver), checksum and TSO will be done by GSO in kernel or even
61 * offloaded into real physical device.
62 */
63 #define VHOST_KERNEL_HOST_OFFLOADS_MASK \
64 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
65 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
66 (1ULL << VIRTIO_NET_F_CSUM))
67
68 static uint64_t max_regions = 64;
69
70 static void
get_vhost_kernel_max_regions(void)71 get_vhost_kernel_max_regions(void)
72 {
73 int fd;
74 char buf[20] = {'\0'};
75
76 fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
77 if (fd < 0)
78 return;
79
80 if (read(fd, buf, sizeof(buf) - 1) > 0)
81 max_regions = strtoull(buf, NULL, 10);
82
83 close(fd);
84 }
85
86 static int
vhost_kernel_ioctl(int fd,uint64_t request,void * arg)87 vhost_kernel_ioctl(int fd, uint64_t request, void *arg)
88 {
89 int ret;
90
91 ret = ioctl(fd, request, arg);
92 if (ret) {
93 PMD_DRV_LOG(ERR, "Vhost-kernel ioctl %"PRIu64" failed (%s)",
94 request, strerror(errno));
95 return -1;
96 }
97
98 return 0;
99 }
100
101 static int
vhost_kernel_set_owner(struct virtio_user_dev * dev)102 vhost_kernel_set_owner(struct virtio_user_dev *dev)
103 {
104 int ret;
105 uint32_t i;
106 struct vhost_kernel_data *data = dev->backend_data;
107
108 for (i = 0; i < dev->max_queue_pairs; ++i) {
109 if (data->vhostfds[i] < 0)
110 continue;
111
112 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_OWNER, NULL);
113 if (ret < 0)
114 return ret;
115 }
116
117 return 0;
118 }
119
120 static int
vhost_kernel_get_features(struct virtio_user_dev * dev,uint64_t * features)121 vhost_kernel_get_features(struct virtio_user_dev *dev, uint64_t *features)
122 {
123 struct vhost_kernel_data *data = dev->backend_data;
124 unsigned int tap_flags;
125 int ret;
126
127 ret = vhost_kernel_ioctl(data->vhostfds[0], VHOST_GET_FEATURES, features);
128 if (ret < 0) {
129 PMD_DRV_LOG(ERR, "Failed to get features");
130 return -1;
131 }
132
133 ret = tap_get_flags(data->tapfds[0], &tap_flags);
134 if (ret < 0) {
135 PMD_DRV_LOG(ERR, "Failed to get TAP features");
136 return -1;
137 }
138
139 /* with tap as the backend, all these features are supported
140 * but not claimed by vhost-net, so we add them back when
141 * reporting to upper layer.
142 */
143 if (tap_flags & IFF_VNET_HDR) {
144 *features |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
145 *features |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
146 }
147
148 /* vhost_kernel will not declare this feature, but it does
149 * support multi-queue.
150 */
151 if (tap_flags & IFF_MULTI_QUEUE)
152 *features |= (1ull << VIRTIO_NET_F_MQ);
153
154 return 0;
155 }
156
157 static int
vhost_kernel_set_features(struct virtio_user_dev * dev,uint64_t features)158 vhost_kernel_set_features(struct virtio_user_dev *dev, uint64_t features)
159 {
160 struct vhost_kernel_data *data = dev->backend_data;
161 uint32_t i;
162 int ret;
163
164 /* We don't need memory protection here */
165 features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
166 /* VHOST kernel does not know about below flags */
167 features &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
168 features &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
169 features &= ~(1ULL << VIRTIO_NET_F_MQ);
170
171 for (i = 0; i < dev->max_queue_pairs; ++i) {
172 if (data->vhostfds[i] < 0)
173 continue;
174
175 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_FEATURES, &features);
176 if (ret < 0)
177 return ret;
178 }
179
180 return 0;
181 }
182
183 static int
add_memseg_list(const struct rte_memseg_list * msl,void * arg)184 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
185 {
186 struct vhost_memory_kernel *vm = arg;
187 struct vhost_memory_region *mr;
188 void *start_addr;
189 uint64_t len;
190
191 if (msl->external)
192 return 0;
193
194 if (vm->nregions >= max_regions)
195 return -1;
196
197 start_addr = msl->base_va;
198 len = msl->page_sz * msl->memseg_arr.len;
199
200 mr = &vm->regions[vm->nregions++];
201
202 mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
203 mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
204 mr->memory_size = len;
205 mr->mmap_offset = 0; /* flags_padding */
206
207 PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
208 vm->nregions - 1, start_addr, len);
209
210 return 0;
211 }
212
213 /* By default, vhost kernel module allows 64 regions, but DPDK may
214 * have much more memory regions. Below function will treat each
215 * contiguous memory space reserved by DPDK as one region.
216 */
217 static int
vhost_kernel_set_memory_table(struct virtio_user_dev * dev)218 vhost_kernel_set_memory_table(struct virtio_user_dev *dev)
219 {
220 uint32_t i;
221 struct vhost_kernel_data *data = dev->backend_data;
222 struct vhost_memory_kernel *vm;
223 int ret;
224
225 vm = malloc(sizeof(struct vhost_memory_kernel) +
226 max_regions *
227 sizeof(struct vhost_memory_region));
228 if (!vm)
229 goto err;
230
231 vm->nregions = 0;
232 vm->padding = 0;
233
234 /*
235 * The memory lock has already been taken by memory subsystem
236 * or virtio_user_start_device().
237 */
238 ret = rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm);
239 if (ret < 0)
240 goto err_free;
241
242 for (i = 0; i < dev->max_queue_pairs; ++i) {
243 if (data->vhostfds[i] < 0)
244 continue;
245
246 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_MEM_TABLE, vm);
247 if (ret < 0)
248 goto err_free;
249 }
250
251 free(vm);
252
253 return 0;
254 err_free:
255 free(vm);
256 err:
257 PMD_DRV_LOG(ERR, "Failed to set memory table");
258 return -1;
259 }
260
261 static int
vhost_kernel_set_vring(struct virtio_user_dev * dev,uint64_t req,struct vhost_vring_state * state)262 vhost_kernel_set_vring(struct virtio_user_dev *dev, uint64_t req, struct vhost_vring_state *state)
263 {
264 int ret, fd;
265 unsigned int index = state->index;
266 struct vhost_kernel_data *data = dev->backend_data;
267
268 /* Convert from queue index to queue-pair & offset */
269 fd = data->vhostfds[state->index / 2];
270 state->index %= 2;
271
272 ret = vhost_kernel_ioctl(fd, req, state);
273 if (ret < 0) {
274 PMD_DRV_LOG(ERR, "Failed to set vring (request %" PRIu64 ")", req);
275 return -1;
276 }
277
278 /* restore index back to queue index */
279 state->index = index;
280
281 return 0;
282 }
283
284 static int
vhost_kernel_set_vring_num(struct virtio_user_dev * dev,struct vhost_vring_state * state)285 vhost_kernel_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
286 {
287 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_NUM, state);
288 }
289
290 static int
vhost_kernel_set_vring_base(struct virtio_user_dev * dev,struct vhost_vring_state * state)291 vhost_kernel_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
292 {
293 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_BASE, state);
294 }
295
296 static int
vhost_kernel_get_vring_base(struct virtio_user_dev * dev,struct vhost_vring_state * state)297 vhost_kernel_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
298 {
299 return vhost_kernel_set_vring(dev, VHOST_GET_VRING_BASE, state);
300 }
301
302 static int
vhost_kernel_set_vring_file(struct virtio_user_dev * dev,uint64_t req,struct vhost_vring_file * file)303 vhost_kernel_set_vring_file(struct virtio_user_dev *dev, uint64_t req,
304 struct vhost_vring_file *file)
305 {
306 int ret, fd;
307 unsigned int index = file->index;
308 struct vhost_kernel_data *data = dev->backend_data;
309
310 /* Convert from queue index to queue-pair & offset */
311 fd = data->vhostfds[file->index / 2];
312 file->index %= 2;
313
314 ret = vhost_kernel_ioctl(fd, req, file);
315 if (ret < 0) {
316 PMD_DRV_LOG(ERR, "Failed to set vring file (request %" PRIu64 ")", req);
317 return -1;
318 }
319
320 /* restore index back to queue index */
321 file->index = index;
322
323 return 0;
324 }
325
326 static int
vhost_kernel_set_vring_kick(struct virtio_user_dev * dev,struct vhost_vring_file * file)327 vhost_kernel_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
328 {
329 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_KICK, file);
330 }
331
332 static int
vhost_kernel_set_vring_call(struct virtio_user_dev * dev,struct vhost_vring_file * file)333 vhost_kernel_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
334 {
335 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_CALL, file);
336 }
337
338 static int
vhost_kernel_set_vring_addr(struct virtio_user_dev * dev,struct vhost_vring_addr * addr)339 vhost_kernel_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
340 {
341 int ret, fd;
342 unsigned int index = addr->index;
343 struct vhost_kernel_data *data = dev->backend_data;
344
345 /* Convert from queue index to queue-pair & offset */
346 fd = data->vhostfds[addr->index / 2];
347 addr->index %= 2;
348
349 ret = vhost_kernel_ioctl(fd, VHOST_SET_VRING_ADDR, addr);
350 if (ret < 0) {
351 PMD_DRV_LOG(ERR, "Failed to set vring address");
352 return -1;
353 }
354
355 /* restore index back to queue index */
356 addr->index = index;
357
358 return 0;
359 }
360
361 static int
vhost_kernel_get_status(struct virtio_user_dev * dev __rte_unused,uint8_t * status __rte_unused)362 vhost_kernel_get_status(struct virtio_user_dev *dev __rte_unused, uint8_t *status __rte_unused)
363 {
364 return -ENOTSUP;
365 }
366
367 static int
vhost_kernel_set_status(struct virtio_user_dev * dev __rte_unused,uint8_t status __rte_unused)368 vhost_kernel_set_status(struct virtio_user_dev *dev __rte_unused, uint8_t status __rte_unused)
369 {
370 return -ENOTSUP;
371 }
372
373 /**
374 * Set up environment to talk with a vhost kernel backend.
375 *
376 * @return
377 * - (-1) if fail to set up;
378 * - (>=0) if successful.
379 */
380 static int
vhost_kernel_setup(struct virtio_user_dev * dev)381 vhost_kernel_setup(struct virtio_user_dev *dev)
382 {
383 struct vhost_kernel_data *data;
384 unsigned int tap_features;
385 unsigned int tap_flags;
386 unsigned int r_flags;
387 const char *ifname;
388 uint32_t q, i;
389 int vhostfd;
390
391 if (tap_support_features(&tap_features) < 0)
392 return -1;
393
394 if ((tap_features & IFF_VNET_HDR) == 0) {
395 PMD_INIT_LOG(ERR, "TAP does not support IFF_VNET_HDR");
396 return -1;
397 }
398 r_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
399
400 if (tap_features & IFF_NAPI)
401 r_flags |= IFF_NAPI;
402
403 data = malloc(sizeof(*data));
404 if (!data) {
405 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost-kernel data", dev->path);
406 return -1;
407 }
408
409 data->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
410 if (!data->vhostfds) {
411 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost FDs", dev->path);
412 goto err_data;
413 }
414 data->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
415 if (!data->tapfds) {
416 PMD_INIT_LOG(ERR, "(%s) Failed to allocate TAP FDs", dev->path);
417 goto err_vhostfds;
418 }
419
420 for (q = 0; q < dev->max_queue_pairs; ++q) {
421 data->vhostfds[q] = -1;
422 data->tapfds[q] = -1;
423 }
424
425 get_vhost_kernel_max_regions();
426
427 for (i = 0; i < dev->max_queue_pairs; ++i) {
428 vhostfd = open(dev->path, O_RDWR);
429 if (vhostfd < 0) {
430 PMD_DRV_LOG(ERR, "fail to open %s, %s", dev->path, strerror(errno));
431 goto err_tapfds;
432 }
433 data->vhostfds[i] = vhostfd;
434 }
435
436 ifname = dev->ifname != NULL ? dev->ifname : "tap%d";
437 data->tapfds[0] = tap_open(ifname, r_flags, (tap_features & IFF_MULTI_QUEUE) != 0);
438 if (data->tapfds[0] < 0)
439 goto err_tapfds;
440 if (dev->ifname == NULL && tap_get_name(data->tapfds[0], &dev->ifname) < 0) {
441 PMD_DRV_LOG(ERR, "fail to get tap name (%d)", data->tapfds[0]);
442 goto err_tapfds;
443 }
444 if (tap_get_flags(data->tapfds[0], &tap_flags) < 0) {
445 PMD_DRV_LOG(ERR, "fail to get tap flags for tap %s", dev->ifname);
446 goto err_tapfds;
447 }
448 if ((tap_flags & IFF_MULTI_QUEUE) == 0 && dev->max_queue_pairs > 1) {
449 PMD_DRV_LOG(ERR, "tap %s does not support multi queue", dev->ifname);
450 goto err_tapfds;
451 }
452
453 for (i = 1; i < dev->max_queue_pairs; i++) {
454 data->tapfds[i] = tap_open(dev->ifname, r_flags, true);
455 if (data->tapfds[i] < 0)
456 goto err_tapfds;
457 }
458
459 dev->backend_data = data;
460
461 return 0;
462
463 err_tapfds:
464 for (i = 0; i < dev->max_queue_pairs; i++) {
465 if (data->vhostfds[i] >= 0)
466 close(data->vhostfds[i]);
467 if (data->tapfds[i] >= 0)
468 close(data->tapfds[i]);
469 }
470
471 free(data->tapfds);
472 err_vhostfds:
473 free(data->vhostfds);
474 err_data:
475 free(data);
476
477 return -1;
478 }
479
480 static int
vhost_kernel_destroy(struct virtio_user_dev * dev)481 vhost_kernel_destroy(struct virtio_user_dev *dev)
482 {
483 struct vhost_kernel_data *data = dev->backend_data;
484 uint32_t i;
485
486 if (!data)
487 return 0;
488
489 for (i = 0; i < dev->max_queue_pairs; ++i) {
490 if (data->vhostfds[i] >= 0)
491 close(data->vhostfds[i]);
492 if (data->tapfds[i] >= 0)
493 close(data->tapfds[i]);
494 }
495
496 free(data->vhostfds);
497 free(data->tapfds);
498 free(data);
499 dev->backend_data = NULL;
500
501 return 0;
502 }
503
504 static int
vhost_kernel_set_backend(int vhostfd,int tapfd)505 vhost_kernel_set_backend(int vhostfd, int tapfd)
506 {
507 struct vhost_vring_file f;
508
509 f.fd = tapfd;
510 f.index = 0;
511 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
512 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
513 strerror(errno));
514 return -1;
515 }
516
517 f.index = 1;
518 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
519 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
520 strerror(errno));
521 return -1;
522 }
523
524 return 0;
525 }
526
527 static int
vhost_kernel_enable_queue_pair(struct virtio_user_dev * dev,uint16_t pair_idx,int enable)528 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
529 uint16_t pair_idx,
530 int enable)
531 {
532 struct vhost_kernel_data *data = dev->backend_data;
533 int hdr_size;
534 int vhostfd;
535 int tapfd;
536
537 if (dev->qp_enabled[pair_idx] == enable)
538 return 0;
539
540 vhostfd = data->vhostfds[pair_idx];
541 tapfd = data->tapfds[pair_idx];
542
543 if (!enable) {
544 if (vhost_kernel_set_backend(vhostfd, -1) < 0) {
545 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
546 return -1;
547 }
548 dev->qp_enabled[pair_idx] = false;
549 return 0;
550 }
551
552 if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
553 (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
554 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
555 else
556 hdr_size = sizeof(struct virtio_net_hdr);
557
558 /* Set mac on tap only once when starting */
559 if (!dev->started && pair_idx == 0 &&
560 tap_set_mac(data->tapfds[pair_idx], dev->mac_addr) < 0)
561 return -1;
562
563 if (vhost_kernel_tap_setup(tapfd, hdr_size, dev->features) < 0) {
564 PMD_DRV_LOG(ERR, "fail to setup tap for vhost kernel");
565 return -1;
566 }
567
568 if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
569 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
570 return -1;
571 }
572
573 dev->qp_enabled[pair_idx] = true;
574 return 0;
575 }
576
577 static int
vhost_kernel_get_backend_features(uint64_t * features)578 vhost_kernel_get_backend_features(uint64_t *features)
579 {
580 *features = 0;
581
582 return 0;
583 }
584
585 static int
vhost_kernel_update_link_state(struct virtio_user_dev * dev __rte_unused)586 vhost_kernel_update_link_state(struct virtio_user_dev *dev __rte_unused)
587 {
588 /* Nothing to update (Maybe get TAP interface link state?) */
589 return 0;
590 }
591
592 static int
vhost_kernel_get_intr_fd(struct virtio_user_dev * dev __rte_unused)593 vhost_kernel_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
594 {
595 /* No link state interrupt with Vhost-kernel */
596 return -1;
597 }
598
599 struct virtio_user_backend_ops virtio_ops_kernel = {
600 .setup = vhost_kernel_setup,
601 .destroy = vhost_kernel_destroy,
602 .get_backend_features = vhost_kernel_get_backend_features,
603 .set_owner = vhost_kernel_set_owner,
604 .get_features = vhost_kernel_get_features,
605 .set_features = vhost_kernel_set_features,
606 .set_memory_table = vhost_kernel_set_memory_table,
607 .set_vring_num = vhost_kernel_set_vring_num,
608 .set_vring_base = vhost_kernel_set_vring_base,
609 .get_vring_base = vhost_kernel_get_vring_base,
610 .set_vring_call = vhost_kernel_set_vring_call,
611 .set_vring_kick = vhost_kernel_set_vring_kick,
612 .set_vring_addr = vhost_kernel_set_vring_addr,
613 .get_status = vhost_kernel_get_status,
614 .set_status = vhost_kernel_set_status,
615 .enable_qp = vhost_kernel_enable_queue_pair,
616 .update_link_state = vhost_kernel_update_link_state,
617 .get_intr_fd = vhost_kernel_get_intr_fd,
618 };
619