]>
Commit | Line | Data |
---|---|---|
1e0a84ea CL |
1 | /* |
2 | * vhost-vdpa.c | |
3 | * | |
4 | * Copyright(c) 2017-2018 Intel Corporation. | |
5 | * Copyright(c) 2020 Red Hat, Inc. | |
6 | * | |
7 | * This work is licensed under the terms of the GNU GPL, version 2 or later. | |
8 | * See the COPYING file in the top-level directory. | |
9 | * | |
10 | */ | |
11 | ||
12 | #include "qemu/osdep.h" | |
13 | #include "clients.h" | |
bd907ae4 | 14 | #include "hw/virtio/virtio-net.h" |
1e0a84ea CL |
15 | #include "net/vhost_net.h" |
16 | #include "net/vhost-vdpa.h" | |
17 | #include "hw/virtio/vhost-vdpa.h" | |
18 | #include "qemu/config-file.h" | |
19 | #include "qemu/error-report.h" | |
bd907ae4 EP |
20 | #include "qemu/log.h" |
21 | #include "qemu/memalign.h" | |
1e0a84ea CL |
22 | #include "qemu/option.h" |
23 | #include "qapi/error.h" | |
40237840 | 24 | #include <linux/vhost.h> |
1e0a84ea CL |
25 | #include <sys/ioctl.h> |
26 | #include <err.h> | |
27 | #include "standard-headers/linux/virtio_net.h" | |
28 | #include "monitor/monitor.h" | |
29 | #include "hw/virtio/vhost.h" | |
30 | ||
31 | /* Todo:need to add the multiqueue support here */ | |
32 | typedef struct VhostVDPAState { | |
33 | NetClientState nc; | |
34 | struct vhost_vdpa vhost_vdpa; | |
35 | VHostNetState *vhost_net; | |
2df4dd31 EP |
36 | |
37 | /* Control commands shadow buffers */ | |
17fb889f EP |
38 | void *cvq_cmd_out_buffer; |
39 | virtio_net_ctrl_ack *status; | |
40 | ||
1e0a84ea CL |
41 | bool started; |
42 | } VhostVDPAState; | |
43 | ||
44 | const int vdpa_feature_bits[] = { | |
45 | VIRTIO_F_NOTIFY_ON_EMPTY, | |
46 | VIRTIO_RING_F_INDIRECT_DESC, | |
47 | VIRTIO_RING_F_EVENT_IDX, | |
48 | VIRTIO_F_ANY_LAYOUT, | |
49 | VIRTIO_F_VERSION_1, | |
50 | VIRTIO_NET_F_CSUM, | |
51 | VIRTIO_NET_F_GUEST_CSUM, | |
52 | VIRTIO_NET_F_GSO, | |
53 | VIRTIO_NET_F_GUEST_TSO4, | |
54 | VIRTIO_NET_F_GUEST_TSO6, | |
55 | VIRTIO_NET_F_GUEST_ECN, | |
56 | VIRTIO_NET_F_GUEST_UFO, | |
57 | VIRTIO_NET_F_HOST_TSO4, | |
58 | VIRTIO_NET_F_HOST_TSO6, | |
59 | VIRTIO_NET_F_HOST_ECN, | |
60 | VIRTIO_NET_F_HOST_UFO, | |
61 | VIRTIO_NET_F_MRG_RXBUF, | |
62 | VIRTIO_NET_F_MTU, | |
40237840 JW |
63 | VIRTIO_NET_F_CTRL_RX, |
64 | VIRTIO_NET_F_CTRL_RX_EXTRA, | |
65 | VIRTIO_NET_F_CTRL_VLAN, | |
66 | VIRTIO_NET_F_GUEST_ANNOUNCE, | |
67 | VIRTIO_NET_F_CTRL_MAC_ADDR, | |
68 | VIRTIO_NET_F_RSS, | |
69 | VIRTIO_NET_F_MQ, | |
70 | VIRTIO_NET_F_CTRL_VQ, | |
1e0a84ea CL |
71 | VIRTIO_F_IOMMU_PLATFORM, |
72 | VIRTIO_F_RING_PACKED, | |
0145c393 AM |
73 | VIRTIO_NET_F_RSS, |
74 | VIRTIO_NET_F_HASH_REPORT, | |
1e0a84ea | 75 | VIRTIO_NET_F_GUEST_ANNOUNCE, |
9aa47edd | 76 | VIRTIO_NET_F_STATUS, |
1e0a84ea CL |
77 | VHOST_INVALID_FEATURE_BIT |
78 | }; | |
79 | ||
1576dbb5 EP |
80 | /** Supported device specific feature bits with SVQ */ |
81 | static const uint64_t vdpa_svq_device_features = | |
82 | BIT_ULL(VIRTIO_NET_F_CSUM) | | |
83 | BIT_ULL(VIRTIO_NET_F_GUEST_CSUM) | | |
84 | BIT_ULL(VIRTIO_NET_F_MTU) | | |
85 | BIT_ULL(VIRTIO_NET_F_MAC) | | |
86 | BIT_ULL(VIRTIO_NET_F_GUEST_TSO4) | | |
87 | BIT_ULL(VIRTIO_NET_F_GUEST_TSO6) | | |
88 | BIT_ULL(VIRTIO_NET_F_GUEST_ECN) | | |
89 | BIT_ULL(VIRTIO_NET_F_GUEST_UFO) | | |
90 | BIT_ULL(VIRTIO_NET_F_HOST_TSO4) | | |
91 | BIT_ULL(VIRTIO_NET_F_HOST_TSO6) | | |
92 | BIT_ULL(VIRTIO_NET_F_HOST_ECN) | | |
93 | BIT_ULL(VIRTIO_NET_F_HOST_UFO) | | |
94 | BIT_ULL(VIRTIO_NET_F_MRG_RXBUF) | | |
95 | BIT_ULL(VIRTIO_NET_F_STATUS) | | |
96 | BIT_ULL(VIRTIO_NET_F_CTRL_VQ) | | |
97 | BIT_ULL(VIRTIO_F_ANY_LAYOUT) | | |
98 | BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR) | | |
99 | BIT_ULL(VIRTIO_NET_F_RSC_EXT) | | |
100 | BIT_ULL(VIRTIO_NET_F_STANDBY); | |
101 | ||
1e0a84ea CL |
102 | VHostNetState *vhost_vdpa_get_vhost_net(NetClientState *nc) |
103 | { | |
104 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
105 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
106 | return s->vhost_net; | |
107 | } | |
108 | ||
1e0a84ea CL |
109 | static int vhost_vdpa_net_check_device_id(struct vhost_net *net) |
110 | { | |
111 | uint32_t device_id; | |
112 | int ret; | |
113 | struct vhost_dev *hdev; | |
114 | ||
115 | hdev = (struct vhost_dev *)&net->dev; | |
116 | ret = hdev->vhost_ops->vhost_get_device_id(hdev, &device_id); | |
117 | if (device_id != VIRTIO_ID_NET) { | |
118 | return -ENOTSUP; | |
119 | } | |
120 | return ret; | |
121 | } | |
122 | ||
40237840 JW |
123 | static int vhost_vdpa_add(NetClientState *ncs, void *be, |
124 | int queue_pair_index, int nvqs) | |
1e0a84ea CL |
125 | { |
126 | VhostNetOptions options; | |
127 | struct vhost_net *net = NULL; | |
128 | VhostVDPAState *s; | |
129 | int ret; | |
130 | ||
131 | options.backend_type = VHOST_BACKEND_TYPE_VDPA; | |
132 | assert(ncs->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
133 | s = DO_UPCAST(VhostVDPAState, nc, ncs); | |
134 | options.net_backend = ncs; | |
135 | options.opaque = be; | |
136 | options.busyloop_timeout = 0; | |
40237840 | 137 | options.nvqs = nvqs; |
1e0a84ea CL |
138 | |
139 | net = vhost_net_init(&options); | |
140 | if (!net) { | |
141 | error_report("failed to init vhost_net for queue"); | |
a97ef87a | 142 | goto err_init; |
1e0a84ea | 143 | } |
1e0a84ea CL |
144 | s->vhost_net = net; |
145 | ret = vhost_vdpa_net_check_device_id(net); | |
146 | if (ret) { | |
a97ef87a | 147 | goto err_check; |
1e0a84ea CL |
148 | } |
149 | return 0; | |
a97ef87a JW |
150 | err_check: |
151 | vhost_net_cleanup(net); | |
152 | g_free(net); | |
153 | err_init: | |
1e0a84ea CL |
154 | return -1; |
155 | } | |
156 | ||
157 | static void vhost_vdpa_cleanup(NetClientState *nc) | |
158 | { | |
159 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
1576dbb5 | 160 | struct vhost_dev *dev = &s->vhost_net->dev; |
1e0a84ea | 161 | |
2df4dd31 | 162 | qemu_vfree(s->cvq_cmd_out_buffer); |
17fb889f | 163 | qemu_vfree(s->status); |
1576dbb5 EP |
164 | if (dev->vq_index + dev->nvqs == dev->vq_index_end) { |
165 | g_clear_pointer(&s->vhost_vdpa.iova_tree, vhost_iova_tree_delete); | |
166 | } | |
1e0a84ea CL |
167 | if (s->vhost_net) { |
168 | vhost_net_cleanup(s->vhost_net); | |
169 | g_free(s->vhost_net); | |
170 | s->vhost_net = NULL; | |
171 | } | |
57b3a7d8 CL |
172 | if (s->vhost_vdpa.device_fd >= 0) { |
173 | qemu_close(s->vhost_vdpa.device_fd); | |
174 | s->vhost_vdpa.device_fd = -1; | |
175 | } | |
1e0a84ea CL |
176 | } |
177 | ||
178 | static bool vhost_vdpa_has_vnet_hdr(NetClientState *nc) | |
179 | { | |
180 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
181 | ||
182 | return true; | |
183 | } | |
184 | ||
185 | static bool vhost_vdpa_has_ufo(NetClientState *nc) | |
186 | { | |
187 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
188 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
189 | uint64_t features = 0; | |
190 | features |= (1ULL << VIRTIO_NET_F_HOST_UFO); | |
191 | features = vhost_net_get_features(s->vhost_net, features); | |
192 | return !!(features & (1ULL << VIRTIO_NET_F_HOST_UFO)); | |
193 | ||
194 | } | |
195 | ||
ee8a1c63 KW |
196 | static bool vhost_vdpa_check_peer_type(NetClientState *nc, ObjectClass *oc, |
197 | Error **errp) | |
198 | { | |
199 | const char *driver = object_class_get_name(oc); | |
200 | ||
201 | if (!g_str_has_prefix(driver, "virtio-net-")) { | |
202 | error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*"); | |
203 | return false; | |
204 | } | |
205 | ||
206 | return true; | |
207 | } | |
208 | ||
846a1e85 EP |
209 | /** Dummy receive in case qemu falls back to userland tap networking */ |
210 | static ssize_t vhost_vdpa_receive(NetClientState *nc, const uint8_t *buf, | |
211 | size_t size) | |
212 | { | |
213 | return 0; | |
214 | } | |
215 | ||
1e0a84ea CL |
216 | static NetClientInfo net_vhost_vdpa_info = { |
217 | .type = NET_CLIENT_DRIVER_VHOST_VDPA, | |
218 | .size = sizeof(VhostVDPAState), | |
846a1e85 | 219 | .receive = vhost_vdpa_receive, |
1e0a84ea CL |
220 | .cleanup = vhost_vdpa_cleanup, |
221 | .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, | |
222 | .has_ufo = vhost_vdpa_has_ufo, | |
ee8a1c63 | 223 | .check_peer_type = vhost_vdpa_check_peer_type, |
1e0a84ea CL |
224 | }; |
225 | ||
2df4dd31 EP |
226 | static void vhost_vdpa_cvq_unmap_buf(struct vhost_vdpa *v, void *addr) |
227 | { | |
228 | VhostIOVATree *tree = v->iova_tree; | |
229 | DMAMap needle = { | |
230 | /* | |
231 | * No need to specify size or to look for more translations since | |
232 | * this contiguous chunk was allocated by us. | |
233 | */ | |
234 | .translated_addr = (hwaddr)(uintptr_t)addr, | |
235 | }; | |
236 | const DMAMap *map = vhost_iova_tree_find_iova(tree, &needle); | |
237 | int r; | |
238 | ||
239 | if (unlikely(!map)) { | |
240 | error_report("Cannot locate expected map"); | |
241 | return; | |
242 | } | |
243 | ||
244 | r = vhost_vdpa_dma_unmap(v, map->iova, map->size + 1); | |
245 | if (unlikely(r != 0)) { | |
246 | error_report("Device cannot unmap: %s(%d)", g_strerror(r), r); | |
247 | } | |
248 | ||
69292a8e | 249 | vhost_iova_tree_remove(tree, *map); |
2df4dd31 EP |
250 | } |
251 | ||
252 | static size_t vhost_vdpa_net_cvq_cmd_len(void) | |
253 | { | |
254 | /* | |
255 | * MAC_TABLE_SET is the ctrl command that produces the longer out buffer. | |
256 | * In buffer is always 1 byte, so it should fit here | |
257 | */ | |
258 | return sizeof(struct virtio_net_ctrl_hdr) + | |
259 | 2 * sizeof(struct virtio_net_ctrl_mac) + | |
260 | MAC_TABLE_ENTRIES * ETH_ALEN; | |
261 | } | |
262 | ||
263 | static size_t vhost_vdpa_net_cvq_cmd_page_len(void) | |
264 | { | |
265 | return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size()); | |
266 | } | |
267 | ||
7a7f87e9 EP |
268 | /** Map CVQ buffer. */ |
269 | static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size, | |
270 | bool write) | |
2df4dd31 EP |
271 | { |
272 | DMAMap map = {}; | |
273 | int r; | |
274 | ||
2df4dd31 | 275 | map.translated_addr = (hwaddr)(uintptr_t)buf; |
7a7f87e9 | 276 | map.size = size - 1; |
2df4dd31 EP |
277 | map.perm = write ? IOMMU_RW : IOMMU_RO, |
278 | r = vhost_iova_tree_map_alloc(v->iova_tree, &map); | |
279 | if (unlikely(r != IOVA_OK)) { | |
280 | error_report("Cannot map injected element"); | |
7a7f87e9 | 281 | return r; |
2df4dd31 EP |
282 | } |
283 | ||
284 | r = vhost_vdpa_dma_map(v, map.iova, vhost_vdpa_net_cvq_cmd_page_len(), buf, | |
285 | !write); | |
286 | if (unlikely(r < 0)) { | |
287 | goto dma_map_err; | |
288 | } | |
289 | ||
7a7f87e9 | 290 | return 0; |
2df4dd31 EP |
291 | |
292 | dma_map_err: | |
69292a8e | 293 | vhost_iova_tree_remove(v->iova_tree, map); |
7a7f87e9 | 294 | return r; |
2df4dd31 EP |
295 | } |
296 | ||
7a7f87e9 | 297 | static int vhost_vdpa_net_cvq_start(NetClientState *nc) |
2df4dd31 | 298 | { |
7a7f87e9 EP |
299 | VhostVDPAState *s; |
300 | int r; | |
2df4dd31 | 301 | |
7a7f87e9 EP |
302 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); |
303 | ||
304 | s = DO_UPCAST(VhostVDPAState, nc, nc); | |
305 | if (!s->vhost_vdpa.shadow_vqs_enabled) { | |
306 | return 0; | |
2df4dd31 EP |
307 | } |
308 | ||
7a7f87e9 EP |
309 | r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer, |
310 | vhost_vdpa_net_cvq_cmd_page_len(), false); | |
311 | if (unlikely(r < 0)) { | |
312 | return r; | |
313 | } | |
314 | ||
17fb889f | 315 | r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->status, |
7a7f87e9 EP |
316 | vhost_vdpa_net_cvq_cmd_page_len(), true); |
317 | if (unlikely(r < 0)) { | |
2df4dd31 | 318 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); |
2df4dd31 EP |
319 | } |
320 | ||
7a7f87e9 EP |
321 | return r; |
322 | } | |
323 | ||
324 | static void vhost_vdpa_net_cvq_stop(NetClientState *nc) | |
325 | { | |
326 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
327 | ||
328 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
329 | ||
330 | if (s->vhost_vdpa.shadow_vqs_enabled) { | |
331 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); | |
17fb889f | 332 | vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->status); |
7a7f87e9 | 333 | } |
2df4dd31 EP |
334 | } |
335 | ||
be4278b6 EP |
336 | static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, |
337 | size_t in_len) | |
338 | { | |
339 | /* Buffers for the device */ | |
340 | const struct iovec out = { | |
341 | .iov_base = s->cvq_cmd_out_buffer, | |
342 | .iov_len = out_len, | |
343 | }; | |
344 | const struct iovec in = { | |
17fb889f | 345 | .iov_base = s->status, |
be4278b6 EP |
346 | .iov_len = sizeof(virtio_net_ctrl_ack), |
347 | }; | |
348 | VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0); | |
349 | int r; | |
350 | ||
351 | r = vhost_svq_add(svq, &out, 1, &in, 1, NULL); | |
352 | if (unlikely(r != 0)) { | |
353 | if (unlikely(r == -ENOSPC)) { | |
354 | qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n", | |
355 | __func__); | |
356 | } | |
357 | return r; | |
358 | } | |
359 | ||
360 | /* | |
361 | * We can poll here since we've had BQL from the time we sent the | |
362 | * descriptor. Also, we need to take the answer before SVQ pulls by itself, | |
363 | * when BQL is released | |
364 | */ | |
365 | return vhost_svq_poll(svq); | |
366 | } | |
367 | ||
f73c0c43 EP |
368 | static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class, |
369 | uint8_t cmd, const void *data, | |
370 | size_t data_size) | |
371 | { | |
372 | const struct virtio_net_ctrl_hdr ctrl = { | |
373 | .class = class, | |
374 | .cmd = cmd, | |
375 | }; | |
376 | ||
377 | assert(data_size < vhost_vdpa_net_cvq_cmd_page_len() - sizeof(ctrl)); | |
378 | ||
379 | memcpy(s->cvq_cmd_out_buffer, &ctrl, sizeof(ctrl)); | |
380 | memcpy(s->cvq_cmd_out_buffer + sizeof(ctrl), data, data_size); | |
381 | ||
382 | return vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + data_size, | |
383 | sizeof(virtio_net_ctrl_ack)); | |
384 | } | |
385 | ||
386 | static int vhost_vdpa_net_load_mac(VhostVDPAState *s, const VirtIONet *n) | |
387 | { | |
388 | uint64_t features = n->parent_obj.guest_features; | |
389 | if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) { | |
390 | ssize_t dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MAC, | |
391 | VIRTIO_NET_CTRL_MAC_ADDR_SET, | |
392 | n->mac, sizeof(n->mac)); | |
393 | if (unlikely(dev_written < 0)) { | |
394 | return dev_written; | |
395 | } | |
396 | ||
397 | return *s->status != VIRTIO_NET_OK; | |
398 | } | |
399 | ||
400 | return 0; | |
401 | } | |
402 | ||
dd036d8d EP |
403 | static int vhost_vdpa_net_load(NetClientState *nc) |
404 | { | |
405 | VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); | |
f73c0c43 | 406 | struct vhost_vdpa *v = &s->vhost_vdpa; |
dd036d8d | 407 | const VirtIONet *n; |
f73c0c43 | 408 | int r; |
dd036d8d EP |
409 | |
410 | assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
411 | ||
412 | if (!v->shadow_vqs_enabled) { | |
413 | return 0; | |
414 | } | |
415 | ||
416 | n = VIRTIO_NET(v->dev->vdev); | |
f73c0c43 EP |
417 | r = vhost_vdpa_net_load_mac(s, n); |
418 | if (unlikely(r < 0)) { | |
419 | return r; | |
dd036d8d EP |
420 | } |
421 | ||
422 | return 0; | |
423 | } | |
424 | ||
f8972b56 EP |
425 | static NetClientInfo net_vhost_vdpa_cvq_info = { |
426 | .type = NET_CLIENT_DRIVER_VHOST_VDPA, | |
427 | .size = sizeof(VhostVDPAState), | |
428 | .receive = vhost_vdpa_receive, | |
7a7f87e9 | 429 | .start = vhost_vdpa_net_cvq_start, |
dd036d8d | 430 | .load = vhost_vdpa_net_load, |
7a7f87e9 | 431 | .stop = vhost_vdpa_net_cvq_stop, |
f8972b56 EP |
432 | .cleanup = vhost_vdpa_cleanup, |
433 | .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, | |
434 | .has_ufo = vhost_vdpa_has_ufo, | |
435 | .check_peer_type = vhost_vdpa_check_peer_type, | |
436 | }; | |
437 | ||
2df4dd31 EP |
438 | /** |
439 | * Do not forward commands not supported by SVQ. Otherwise, the device could | |
440 | * accept it and qemu would not know how to update the device model. | |
441 | */ | |
7a7f87e9 | 442 | static bool vhost_vdpa_net_cvq_validate_cmd(const void *out_buf, size_t len) |
2df4dd31 EP |
443 | { |
444 | struct virtio_net_ctrl_hdr ctrl; | |
2df4dd31 | 445 | |
7a7f87e9 | 446 | if (unlikely(len < sizeof(ctrl))) { |
2df4dd31 | 447 | qemu_log_mask(LOG_GUEST_ERROR, |
7a7f87e9 | 448 | "%s: invalid legnth of out buffer %zu\n", __func__, len); |
2df4dd31 EP |
449 | return false; |
450 | } | |
451 | ||
7a7f87e9 | 452 | memcpy(&ctrl, out_buf, sizeof(ctrl)); |
2df4dd31 EP |
453 | switch (ctrl.class) { |
454 | case VIRTIO_NET_CTRL_MAC: | |
455 | switch (ctrl.cmd) { | |
456 | case VIRTIO_NET_CTRL_MAC_ADDR_SET: | |
457 | return true; | |
458 | default: | |
459 | qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid mac cmd %u\n", | |
460 | __func__, ctrl.cmd); | |
461 | }; | |
462 | break; | |
463 | default: | |
464 | qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid control class %u\n", | |
465 | __func__, ctrl.class); | |
466 | }; | |
467 | ||
468 | return false; | |
469 | } | |
470 | ||
471 | /** | |
472 | * Validate and copy control virtqueue commands. | |
473 | * | |
474 | * Following QEMU guidelines, we offer a copy of the buffers to the device to | |
475 | * prevent TOCTOU bugs. | |
bd907ae4 EP |
476 | */ |
477 | static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, | |
478 | VirtQueueElement *elem, | |
479 | void *opaque) | |
480 | { | |
2df4dd31 | 481 | VhostVDPAState *s = opaque; |
be4278b6 | 482 | size_t in_len; |
bd907ae4 | 483 | virtio_net_ctrl_ack status = VIRTIO_NET_ERR; |
7a7f87e9 EP |
484 | /* Out buffer sent to both the vdpa device and the device model */ |
485 | struct iovec out = { | |
486 | .iov_base = s->cvq_cmd_out_buffer, | |
487 | }; | |
2df4dd31 EP |
488 | /* in buffer used for device model */ |
489 | const struct iovec in = { | |
490 | .iov_base = &status, | |
491 | .iov_len = sizeof(status), | |
492 | }; | |
be4278b6 | 493 | ssize_t dev_written = -EINVAL; |
2df4dd31 EP |
494 | bool ok; |
495 | ||
7a7f87e9 EP |
496 | out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0, |
497 | s->cvq_cmd_out_buffer, | |
498 | vhost_vdpa_net_cvq_cmd_len()); | |
499 | ok = vhost_vdpa_net_cvq_validate_cmd(s->cvq_cmd_out_buffer, out.iov_len); | |
2df4dd31 EP |
500 | if (unlikely(!ok)) { |
501 | goto out; | |
502 | } | |
bd907ae4 | 503 | |
be4278b6 EP |
504 | dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status)); |
505 | if (unlikely(dev_written < 0)) { | |
bd907ae4 EP |
506 | goto out; |
507 | } | |
508 | ||
bd907ae4 EP |
509 | if (unlikely(dev_written < sizeof(status))) { |
510 | error_report("Insufficient written data (%zu)", dev_written); | |
2df4dd31 EP |
511 | goto out; |
512 | } | |
513 | ||
17fb889f | 514 | if (*s->status != VIRTIO_NET_OK) { |
be4278b6 | 515 | return VIRTIO_NET_ERR; |
2df4dd31 EP |
516 | } |
517 | ||
518 | status = VIRTIO_NET_ERR; | |
7a7f87e9 | 519 | virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1); |
2df4dd31 EP |
520 | if (status != VIRTIO_NET_OK) { |
521 | error_report("Bad CVQ processing in model"); | |
bd907ae4 EP |
522 | } |
523 | ||
524 | out: | |
525 | in_len = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, | |
526 | sizeof(status)); | |
527 | if (unlikely(in_len < sizeof(status))) { | |
528 | error_report("Bad device CVQ written length"); | |
529 | } | |
530 | vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status))); | |
531 | g_free(elem); | |
be4278b6 | 532 | return dev_written < 0 ? dev_written : 0; |
bd907ae4 EP |
533 | } |
534 | ||
535 | static const VhostShadowVirtqueueOps vhost_vdpa_net_svq_ops = { | |
536 | .avail_handler = vhost_vdpa_net_handle_ctrl_avail, | |
537 | }; | |
538 | ||
654790b6 JW |
539 | static NetClientState *net_vhost_vdpa_init(NetClientState *peer, |
540 | const char *device, | |
541 | const char *name, | |
40237840 JW |
542 | int vdpa_device_fd, |
543 | int queue_pair_index, | |
544 | int nvqs, | |
1576dbb5 EP |
545 | bool is_datapath, |
546 | bool svq, | |
547 | VhostIOVATree *iova_tree) | |
1e0a84ea CL |
548 | { |
549 | NetClientState *nc = NULL; | |
550 | VhostVDPAState *s; | |
1e0a84ea CL |
551 | int ret = 0; |
552 | assert(name); | |
40237840 JW |
553 | if (is_datapath) { |
554 | nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, | |
555 | name); | |
556 | } else { | |
f8972b56 | 557 | nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer, |
40237840 JW |
558 | device, name); |
559 | } | |
56e6f594 | 560 | snprintf(nc->info_str, sizeof(nc->info_str), TYPE_VHOST_VDPA); |
1e0a84ea | 561 | s = DO_UPCAST(VhostVDPAState, nc, nc); |
7327813d | 562 | |
1e0a84ea | 563 | s->vhost_vdpa.device_fd = vdpa_device_fd; |
40237840 | 564 | s->vhost_vdpa.index = queue_pair_index; |
1576dbb5 EP |
565 | s->vhost_vdpa.shadow_vqs_enabled = svq; |
566 | s->vhost_vdpa.iova_tree = iova_tree; | |
bd907ae4 | 567 | if (!is_datapath) { |
2df4dd31 EP |
568 | s->cvq_cmd_out_buffer = qemu_memalign(qemu_real_host_page_size(), |
569 | vhost_vdpa_net_cvq_cmd_page_len()); | |
570 | memset(s->cvq_cmd_out_buffer, 0, vhost_vdpa_net_cvq_cmd_page_len()); | |
17fb889f EP |
571 | s->status = qemu_memalign(qemu_real_host_page_size(), |
572 | vhost_vdpa_net_cvq_cmd_page_len()); | |
573 | memset(s->status, 0, vhost_vdpa_net_cvq_cmd_page_len()); | |
2df4dd31 | 574 | |
bd907ae4 EP |
575 | s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops; |
576 | s->vhost_vdpa.shadow_vq_ops_opaque = s; | |
577 | } | |
40237840 | 578 | ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs); |
74af5eec | 579 | if (ret) { |
74af5eec | 580 | qemu_del_net_client(nc); |
654790b6 | 581 | return NULL; |
74af5eec | 582 | } |
654790b6 | 583 | return nc; |
1e0a84ea CL |
584 | } |
585 | ||
1576dbb5 EP |
586 | static int vhost_vdpa_get_iova_range(int fd, |
587 | struct vhost_vdpa_iova_range *iova_range) | |
588 | { | |
589 | int ret = ioctl(fd, VHOST_VDPA_GET_IOVA_RANGE, iova_range); | |
590 | ||
591 | return ret < 0 ? -errno : 0; | |
592 | } | |
593 | ||
8170ab3f EP |
594 | static int vhost_vdpa_get_features(int fd, uint64_t *features, Error **errp) |
595 | { | |
596 | int ret = ioctl(fd, VHOST_GET_FEATURES, features); | |
597 | if (unlikely(ret < 0)) { | |
598 | error_setg_errno(errp, errno, | |
599 | "Fail to query features from vhost-vDPA device"); | |
600 | } | |
601 | return ret; | |
602 | } | |
603 | ||
604 | static int vhost_vdpa_get_max_queue_pairs(int fd, uint64_t features, | |
605 | int *has_cvq, Error **errp) | |
40237840 JW |
606 | { |
607 | unsigned long config_size = offsetof(struct vhost_vdpa_config, buf); | |
cd523a41 | 608 | g_autofree struct vhost_vdpa_config *config = NULL; |
40237840 | 609 | __virtio16 *max_queue_pairs; |
40237840 JW |
610 | int ret; |
611 | ||
40237840 JW |
612 | if (features & (1 << VIRTIO_NET_F_CTRL_VQ)) { |
613 | *has_cvq = 1; | |
614 | } else { | |
615 | *has_cvq = 0; | |
616 | } | |
617 | ||
618 | if (features & (1 << VIRTIO_NET_F_MQ)) { | |
619 | config = g_malloc0(config_size + sizeof(*max_queue_pairs)); | |
620 | config->off = offsetof(struct virtio_net_config, max_virtqueue_pairs); | |
621 | config->len = sizeof(*max_queue_pairs); | |
622 | ||
623 | ret = ioctl(fd, VHOST_VDPA_GET_CONFIG, config); | |
624 | if (ret) { | |
625 | error_setg(errp, "Fail to get config from vhost-vDPA device"); | |
626 | return -ret; | |
627 | } | |
628 | ||
629 | max_queue_pairs = (__virtio16 *)&config->buf; | |
630 | ||
631 | return lduw_le_p(max_queue_pairs); | |
632 | } | |
633 | ||
634 | return 1; | |
635 | } | |
636 | ||
1e0a84ea CL |
637 | int net_init_vhost_vdpa(const Netdev *netdev, const char *name, |
638 | NetClientState *peer, Error **errp) | |
639 | { | |
640 | const NetdevVhostVDPAOptions *opts; | |
8170ab3f | 641 | uint64_t features; |
654790b6 | 642 | int vdpa_device_fd; |
eb3cb751 | 643 | g_autofree NetClientState **ncs = NULL; |
1576dbb5 | 644 | g_autoptr(VhostIOVATree) iova_tree = NULL; |
eb3cb751 | 645 | NetClientState *nc; |
aed5da45 | 646 | int queue_pairs, r, i = 0, has_cvq = 0; |
1e0a84ea CL |
647 | |
648 | assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA); | |
649 | opts = &netdev->u.vhost_vdpa; | |
c8295404 EP |
650 | if (!opts->vhostdev) { |
651 | error_setg(errp, "vdpa character device not specified with vhostdev"); | |
652 | return -1; | |
653 | } | |
7327813d | 654 | |
0351152b | 655 | vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR, errp); |
7327813d JW |
656 | if (vdpa_device_fd == -1) { |
657 | return -errno; | |
658 | } | |
659 | ||
8170ab3f EP |
660 | r = vhost_vdpa_get_features(vdpa_device_fd, &features, errp); |
661 | if (unlikely(r < 0)) { | |
aed5da45 | 662 | goto err; |
8170ab3f EP |
663 | } |
664 | ||
665 | queue_pairs = vhost_vdpa_get_max_queue_pairs(vdpa_device_fd, features, | |
40237840 JW |
666 | &has_cvq, errp); |
667 | if (queue_pairs < 0) { | |
7327813d | 668 | qemu_close(vdpa_device_fd); |
40237840 JW |
669 | return queue_pairs; |
670 | } | |
671 | ||
1576dbb5 EP |
672 | if (opts->x_svq) { |
673 | struct vhost_vdpa_iova_range iova_range; | |
674 | ||
675 | uint64_t invalid_dev_features = | |
676 | features & ~vdpa_svq_device_features & | |
677 | /* Transport are all accepted at this point */ | |
678 | ~MAKE_64BIT_MASK(VIRTIO_TRANSPORT_F_START, | |
679 | VIRTIO_TRANSPORT_F_END - VIRTIO_TRANSPORT_F_START); | |
680 | ||
681 | if (invalid_dev_features) { | |
682 | error_setg(errp, "vdpa svq does not work with features 0x%" PRIx64, | |
683 | invalid_dev_features); | |
684 | goto err_svq; | |
685 | } | |
686 | ||
687 | vhost_vdpa_get_iova_range(vdpa_device_fd, &iova_range); | |
688 | iova_tree = vhost_iova_tree_new(iova_range.first, iova_range.last); | |
689 | } | |
690 | ||
40237840 JW |
691 | ncs = g_malloc0(sizeof(*ncs) * queue_pairs); |
692 | ||
693 | for (i = 0; i < queue_pairs; i++) { | |
694 | ncs[i] = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, | |
1576dbb5 EP |
695 | vdpa_device_fd, i, 2, true, opts->x_svq, |
696 | iova_tree); | |
40237840 JW |
697 | if (!ncs[i]) |
698 | goto err; | |
7327813d JW |
699 | } |
700 | ||
40237840 JW |
701 | if (has_cvq) { |
702 | nc = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, | |
1576dbb5 EP |
703 | vdpa_device_fd, i, 1, false, |
704 | opts->x_svq, iova_tree); | |
40237840 JW |
705 | if (!nc) |
706 | goto err; | |
707 | } | |
708 | ||
1576dbb5 EP |
709 | /* iova_tree ownership belongs to last NetClientState */ |
710 | g_steal_pointer(&iova_tree); | |
654790b6 | 711 | return 0; |
40237840 JW |
712 | |
713 | err: | |
714 | if (i) { | |
9bd05507 SWL |
715 | for (i--; i >= 0; i--) { |
716 | qemu_del_net_client(ncs[i]); | |
717 | } | |
40237840 | 718 | } |
1576dbb5 EP |
719 | |
720 | err_svq: | |
40237840 | 721 | qemu_close(vdpa_device_fd); |
40237840 JW |
722 | |
723 | return -1; | |
1e0a84ea | 724 | } |