]>
Commit | Line | Data |
---|---|---|
11fdf7f2 TL |
1 | /*- |
2 | * BSD LICENSE | |
3 | * | |
4 | * Copyright(c) 2010-2016 Intel Corporation. All rights reserved. | |
5 | * All rights reserved. | |
6 | * | |
7 | * Redistribution and use in source and binary forms, with or without | |
8 | * modification, are permitted provided that the following conditions | |
9 | * are met: | |
10 | * | |
11 | * * Redistributions of source code must retain the above copyright | |
12 | * notice, this list of conditions and the following disclaimer. | |
13 | * * Redistributions in binary form must reproduce the above copyright | |
14 | * notice, this list of conditions and the following disclaimer in | |
15 | * the documentation and/or other materials provided with the | |
16 | * distribution. | |
17 | * * Neither the name of Intel Corporation nor the names of its | |
18 | * contributors may be used to endorse or promote products derived | |
19 | * from this software without specific prior written permission. | |
20 | * | |
21 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
22 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
23 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
24 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
25 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
26 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
27 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
28 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
29 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
30 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
31 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
32 | */ | |
33 | ||
34 | #include "spdk/stdinc.h" | |
35 | ||
36 | #include "vhost.h" | |
37 | ||
38 | #include "spdk/string.h" | |
9f95a23c | 39 | #include "spdk_internal/vhost_user.h" |
11fdf7f2 TL |
40 | |
41 | /* The version of the protocol we support */ | |
42 | #define VHOST_USER_VERSION 0x1 | |
43 | ||
11fdf7f2 TL |
44 | static int |
45 | vhost_user_write(int fd, void *buf, int len, int *fds, int fd_num) | |
46 | { | |
47 | int r; | |
48 | struct msghdr msgh; | |
49 | struct iovec iov; | |
50 | size_t fd_size = fd_num * sizeof(int); | |
51 | char control[CMSG_SPACE(fd_size)]; | |
52 | struct cmsghdr *cmsg; | |
53 | ||
54 | memset(&msgh, 0, sizeof(msgh)); | |
55 | memset(control, 0, sizeof(control)); | |
56 | ||
57 | iov.iov_base = (uint8_t *)buf; | |
58 | iov.iov_len = len; | |
59 | ||
60 | msgh.msg_iov = &iov; | |
61 | msgh.msg_iovlen = 1; | |
62 | ||
63 | if (fds && fd_num > 0) { | |
64 | msgh.msg_control = control; | |
65 | msgh.msg_controllen = sizeof(control); | |
66 | cmsg = CMSG_FIRSTHDR(&msgh); | |
67 | cmsg->cmsg_len = CMSG_LEN(fd_size); | |
68 | cmsg->cmsg_level = SOL_SOCKET; | |
69 | cmsg->cmsg_type = SCM_RIGHTS; | |
70 | memcpy(CMSG_DATA(cmsg), fds, fd_size); | |
71 | } else { | |
72 | msgh.msg_control = NULL; | |
73 | msgh.msg_controllen = 0; | |
74 | } | |
75 | ||
76 | do { | |
77 | r = sendmsg(fd, &msgh, 0); | |
78 | } while (r < 0 && errno == EINTR); | |
79 | ||
80 | if (r == -1) { | |
81 | return -errno; | |
82 | } | |
83 | ||
84 | return 0; | |
85 | } | |
86 | ||
87 | static int | |
88 | vhost_user_read(int fd, struct vhost_user_msg *msg) | |
89 | { | |
90 | uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION; | |
91 | ssize_t ret; | |
92 | size_t sz_hdr = VHOST_USER_HDR_SIZE, sz_payload; | |
93 | ||
94 | ret = recv(fd, (void *)msg, sz_hdr, 0); | |
95 | if ((size_t)ret != sz_hdr) { | |
96 | SPDK_WARNLOG("Failed to recv msg hdr: %zd instead of %zu.\n", | |
97 | ret, sz_hdr); | |
98 | if (ret == -1) { | |
99 | return -errno; | |
100 | } else { | |
101 | return -EBUSY; | |
102 | } | |
103 | } | |
104 | ||
105 | /* validate msg flags */ | |
106 | if (msg->flags != (valid_flags)) { | |
107 | SPDK_WARNLOG("Failed to recv msg: flags %"PRIx32" instead of %"PRIx32".\n", | |
108 | msg->flags, valid_flags); | |
109 | return -EIO; | |
110 | } | |
111 | ||
112 | sz_payload = msg->size; | |
113 | ||
9f95a23c | 114 | if (sz_payload > VHOST_USER_PAYLOAD_SIZE) { |
11fdf7f2 | 115 | SPDK_WARNLOG("Received oversized msg: payload size %zu > available space %zu\n", |
9f95a23c | 116 | sz_payload, VHOST_USER_PAYLOAD_SIZE); |
11fdf7f2 TL |
117 | return -EIO; |
118 | } | |
119 | ||
120 | if (sz_payload) { | |
121 | ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0); | |
122 | if ((size_t)ret != sz_payload) { | |
123 | SPDK_WARNLOG("Failed to recv msg payload: %zd instead of %"PRIu32".\n", | |
124 | ret, msg->size); | |
125 | if (ret == -1) { | |
126 | return -errno; | |
127 | } else { | |
128 | return -EBUSY; | |
129 | } | |
130 | } | |
131 | } | |
132 | ||
133 | return 0; | |
134 | } | |
135 | ||
136 | struct hugepage_file_info { | |
137 | uint64_t addr; /**< virtual addr */ | |
138 | size_t size; /**< the file size */ | |
139 | char path[PATH_MAX]; /**< path to backing file */ | |
140 | }; | |
141 | ||
142 | /* Two possible options: | |
143 | * 1. Match HUGEPAGE_INFO_FMT to find the file storing struct hugepage_file | |
144 | * array. This is simple but cannot be used in secondary process because | |
145 | * secondary process will close and munmap that file. | |
146 | * 2. Match HUGEFILE_FMT to find hugepage files directly. | |
147 | * | |
148 | * We choose option 2. | |
149 | */ | |
150 | static int | |
151 | get_hugepage_file_info(struct hugepage_file_info huges[], int max) | |
152 | { | |
153 | int idx, rc; | |
154 | FILE *f; | |
155 | char buf[BUFSIZ], *tmp, *tail; | |
156 | char *str_underline, *str_start; | |
157 | int huge_index; | |
158 | uint64_t v_start, v_end; | |
159 | ||
160 | f = fopen("/proc/self/maps", "r"); | |
161 | if (!f) { | |
162 | SPDK_ERRLOG("cannot open /proc/self/maps\n"); | |
163 | rc = -errno; | |
164 | assert(rc < 0); /* scan-build hack */ | |
165 | return rc; | |
166 | } | |
167 | ||
168 | idx = 0; | |
169 | while (fgets(buf, sizeof(buf), f) != NULL) { | |
170 | if (sscanf(buf, "%" PRIx64 "-%" PRIx64, &v_start, &v_end) < 2) { | |
171 | SPDK_ERRLOG("Failed to parse address\n"); | |
172 | rc = -EIO; | |
173 | goto out; | |
174 | } | |
175 | ||
176 | tmp = strchr(buf, ' ') + 1; /** skip address */ | |
177 | tmp = strchr(tmp, ' ') + 1; /** skip perm */ | |
178 | tmp = strchr(tmp, ' ') + 1; /** skip offset */ | |
179 | tmp = strchr(tmp, ' ') + 1; /** skip dev */ | |
180 | tmp = strchr(tmp, ' ') + 1; /** skip inode */ | |
181 | while (*tmp == ' ') { /** skip spaces */ | |
182 | tmp++; | |
183 | } | |
184 | tail = strrchr(tmp, '\n'); /** remove newline if exists */ | |
185 | if (tail) { | |
186 | *tail = '\0'; | |
187 | } | |
188 | ||
189 | /* Match HUGEFILE_FMT, aka "%s/%smap_%d", | |
190 | * which is defined in eal_filesystem.h | |
191 | */ | |
192 | str_underline = strrchr(tmp, '_'); | |
193 | if (!str_underline) { | |
194 | continue; | |
195 | } | |
196 | ||
197 | str_start = str_underline - strlen("map"); | |
198 | if (str_start < tmp) { | |
199 | continue; | |
200 | } | |
201 | ||
202 | if (sscanf(str_start, "map_%d", &huge_index) != 1) { | |
203 | continue; | |
204 | } | |
205 | ||
206 | if (idx >= max) { | |
207 | SPDK_ERRLOG("Exceed maximum of %d\n", max); | |
208 | rc = -ENOSPC; | |
209 | goto out; | |
210 | } | |
211 | ||
212 | if (idx > 0 && | |
213 | strncmp(tmp, huges[idx - 1].path, PATH_MAX) == 0 && | |
214 | v_start == huges[idx - 1].addr + huges[idx - 1].size) { | |
215 | huges[idx - 1].size += (v_end - v_start); | |
216 | continue; | |
217 | } | |
218 | ||
219 | huges[idx].addr = v_start; | |
220 | huges[idx].size = v_end - v_start; | |
221 | snprintf(huges[idx].path, PATH_MAX, "%s", tmp); | |
222 | idx++; | |
223 | } | |
224 | ||
225 | rc = idx; | |
226 | out: | |
227 | fclose(f); | |
228 | return rc; | |
229 | } | |
230 | ||
231 | static int | |
232 | prepare_vhost_memory_user(struct vhost_user_msg *msg, int fds[]) | |
233 | { | |
234 | int i, num; | |
9f95a23c | 235 | struct hugepage_file_info huges[VHOST_USER_MEMORY_MAX_NREGIONS]; |
11fdf7f2 | 236 | |
9f95a23c | 237 | num = get_hugepage_file_info(huges, VHOST_USER_MEMORY_MAX_NREGIONS); |
11fdf7f2 TL |
238 | if (num < 0) { |
239 | SPDK_ERRLOG("Failed to prepare memory for vhost-user\n"); | |
240 | return num; | |
241 | } | |
242 | ||
243 | for (i = 0; i < num; ++i) { | |
244 | /* the memory regions are unaligned */ | |
245 | msg->payload.memory.regions[i].guest_phys_addr = huges[i].addr; /* use vaddr! */ | |
246 | msg->payload.memory.regions[i].userspace_addr = huges[i].addr; | |
247 | msg->payload.memory.regions[i].memory_size = huges[i].size; | |
248 | msg->payload.memory.regions[i].flags_padding = 0; | |
249 | fds[i] = open(huges[i].path, O_RDWR); | |
250 | } | |
251 | ||
252 | msg->payload.memory.nregions = num; | |
253 | msg->payload.memory.padding = 0; | |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
258 | static const char *const vhost_msg_strings[VHOST_USER_MAX] = { | |
259 | [VHOST_USER_SET_OWNER] = "VHOST_SET_OWNER", | |
260 | [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER", | |
261 | [VHOST_USER_SET_FEATURES] = "VHOST_SET_FEATURES", | |
262 | [VHOST_USER_GET_FEATURES] = "VHOST_GET_FEATURES", | |
263 | [VHOST_USER_SET_VRING_CALL] = "VHOST_SET_VRING_CALL", | |
264 | [VHOST_USER_SET_VRING_NUM] = "VHOST_SET_VRING_NUM", | |
265 | [VHOST_USER_SET_VRING_BASE] = "VHOST_SET_VRING_BASE", | |
266 | [VHOST_USER_GET_VRING_BASE] = "VHOST_GET_VRING_BASE", | |
267 | [VHOST_USER_SET_VRING_ADDR] = "VHOST_SET_VRING_ADDR", | |
268 | [VHOST_USER_SET_VRING_KICK] = "VHOST_SET_VRING_KICK", | |
269 | [VHOST_USER_SET_MEM_TABLE] = "VHOST_SET_MEM_TABLE", | |
270 | [VHOST_USER_SET_VRING_ENABLE] = "VHOST_SET_VRING_ENABLE", | |
271 | [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM", | |
272 | [VHOST_USER_GET_CONFIG] = "VHOST_USER_GET_CONFIG", | |
273 | [VHOST_USER_SET_CONFIG] = "VHOST_USER_SET_CONFIG", | |
274 | }; | |
275 | ||
276 | static int | |
277 | vhost_user_sock(struct virtio_user_dev *dev, | |
278 | enum vhost_user_request req, | |
279 | void *arg) | |
280 | { | |
281 | struct vhost_user_msg msg; | |
282 | struct vhost_vring_file *file = 0; | |
283 | int need_reply = 0; | |
9f95a23c | 284 | int fds[VHOST_USER_MEMORY_MAX_NREGIONS]; |
11fdf7f2 TL |
285 | int fd_num = 0; |
286 | int i, len, rc; | |
287 | int vhostfd = dev->vhostfd; | |
288 | ||
289 | SPDK_DEBUGLOG(SPDK_LOG_VIRTIO_USER, "sent message %d = %s\n", req, vhost_msg_strings[req]); | |
290 | ||
291 | msg.request = req; | |
292 | msg.flags = VHOST_USER_VERSION; | |
293 | msg.size = 0; | |
294 | ||
295 | switch (req) { | |
296 | case VHOST_USER_GET_FEATURES: | |
297 | case VHOST_USER_GET_PROTOCOL_FEATURES: | |
298 | case VHOST_USER_GET_QUEUE_NUM: | |
299 | need_reply = 1; | |
300 | break; | |
301 | ||
302 | case VHOST_USER_SET_FEATURES: | |
303 | case VHOST_USER_SET_LOG_BASE: | |
304 | case VHOST_USER_SET_PROTOCOL_FEATURES: | |
305 | msg.payload.u64 = *((__u64 *)arg); | |
306 | msg.size = sizeof(msg.payload.u64); | |
307 | break; | |
308 | ||
309 | case VHOST_USER_SET_OWNER: | |
310 | case VHOST_USER_RESET_OWNER: | |
311 | break; | |
312 | ||
313 | case VHOST_USER_SET_MEM_TABLE: | |
314 | rc = prepare_vhost_memory_user(&msg, fds); | |
315 | if (rc < 0) { | |
316 | return rc; | |
317 | } | |
318 | fd_num = msg.payload.memory.nregions; | |
319 | msg.size = sizeof(msg.payload.memory.nregions); | |
320 | msg.size += sizeof(msg.payload.memory.padding); | |
321 | msg.size += fd_num * sizeof(struct vhost_memory_region); | |
322 | break; | |
323 | ||
324 | case VHOST_USER_SET_LOG_FD: | |
325 | fds[fd_num++] = *((int *)arg); | |
326 | break; | |
327 | ||
328 | case VHOST_USER_SET_VRING_NUM: | |
329 | case VHOST_USER_SET_VRING_BASE: | |
330 | case VHOST_USER_SET_VRING_ENABLE: | |
331 | memcpy(&msg.payload.state, arg, sizeof(msg.payload.state)); | |
332 | msg.size = sizeof(msg.payload.state); | |
333 | break; | |
334 | ||
335 | case VHOST_USER_GET_VRING_BASE: | |
336 | memcpy(&msg.payload.state, arg, sizeof(msg.payload.state)); | |
337 | msg.size = sizeof(msg.payload.state); | |
338 | need_reply = 1; | |
339 | break; | |
340 | ||
341 | case VHOST_USER_SET_VRING_ADDR: | |
342 | memcpy(&msg.payload.addr, arg, sizeof(msg.payload.addr)); | |
343 | msg.size = sizeof(msg.payload.addr); | |
344 | break; | |
345 | ||
346 | case VHOST_USER_SET_VRING_KICK: | |
347 | case VHOST_USER_SET_VRING_CALL: | |
348 | case VHOST_USER_SET_VRING_ERR: | |
349 | file = arg; | |
350 | msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK; | |
351 | msg.size = sizeof(msg.payload.u64); | |
352 | if (file->fd > 0) { | |
353 | fds[fd_num++] = file->fd; | |
354 | } else { | |
355 | msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK; | |
356 | } | |
357 | break; | |
358 | ||
359 | case VHOST_USER_GET_CONFIG: | |
360 | memcpy(&msg.payload.cfg, arg, sizeof(msg.payload.cfg)); | |
361 | msg.size = sizeof(msg.payload.cfg); | |
362 | need_reply = 1; | |
363 | break; | |
364 | ||
365 | case VHOST_USER_SET_CONFIG: | |
366 | memcpy(&msg.payload.cfg, arg, sizeof(msg.payload.cfg)); | |
367 | msg.size = sizeof(msg.payload.cfg); | |
368 | break; | |
369 | ||
370 | default: | |
371 | SPDK_ERRLOG("trying to send unknown msg\n"); | |
372 | return -EINVAL; | |
373 | } | |
374 | ||
375 | len = VHOST_USER_HDR_SIZE + msg.size; | |
376 | rc = vhost_user_write(vhostfd, &msg, len, fds, fd_num); | |
377 | if (rc < 0) { | |
378 | SPDK_ERRLOG("%s failed: %s\n", | |
379 | vhost_msg_strings[req], spdk_strerror(-rc)); | |
380 | return rc; | |
381 | } | |
382 | ||
383 | if (req == VHOST_USER_SET_MEM_TABLE) | |
384 | for (i = 0; i < fd_num; ++i) { | |
385 | close(fds[i]); | |
386 | } | |
387 | ||
388 | if (need_reply) { | |
389 | rc = vhost_user_read(vhostfd, &msg); | |
390 | if (rc < 0) { | |
391 | SPDK_WARNLOG("Received msg failed: %s\n", spdk_strerror(-rc)); | |
392 | return rc; | |
393 | } | |
394 | ||
395 | if (req != msg.request) { | |
396 | SPDK_WARNLOG("Received unexpected msg type\n"); | |
397 | return -EIO; | |
398 | } | |
399 | ||
400 | switch (req) { | |
401 | case VHOST_USER_GET_FEATURES: | |
402 | case VHOST_USER_GET_PROTOCOL_FEATURES: | |
403 | case VHOST_USER_GET_QUEUE_NUM: | |
404 | if (msg.size != sizeof(msg.payload.u64)) { | |
405 | SPDK_WARNLOG("Received bad msg size\n"); | |
406 | return -EIO; | |
407 | } | |
408 | *((__u64 *)arg) = msg.payload.u64; | |
409 | break; | |
410 | case VHOST_USER_GET_VRING_BASE: | |
411 | if (msg.size != sizeof(msg.payload.state)) { | |
412 | SPDK_WARNLOG("Received bad msg size\n"); | |
413 | return -EIO; | |
414 | } | |
415 | memcpy(arg, &msg.payload.state, | |
416 | sizeof(struct vhost_vring_state)); | |
417 | break; | |
418 | case VHOST_USER_GET_CONFIG: | |
419 | if (msg.size != sizeof(msg.payload.cfg)) { | |
420 | SPDK_WARNLOG("Received bad msg size\n"); | |
421 | return -EIO; | |
422 | } | |
423 | memcpy(arg, &msg.payload.cfg, sizeof(msg.payload.cfg)); | |
424 | break; | |
425 | default: | |
426 | SPDK_WARNLOG("Received unexpected msg type\n"); | |
427 | return -EBADMSG; | |
428 | } | |
429 | } | |
430 | ||
431 | return 0; | |
432 | } | |
433 | ||
434 | /** | |
435 | * Set up environment to talk with a vhost user backend. | |
436 | * | |
437 | * @return | |
438 | * - (-1) if fail; | |
439 | * - (0) if succeed. | |
440 | */ | |
441 | static int | |
442 | vhost_user_setup(struct virtio_user_dev *dev) | |
443 | { | |
444 | int fd; | |
445 | int flag; | |
446 | struct sockaddr_un un; | |
447 | ssize_t rc; | |
448 | ||
449 | fd = socket(AF_UNIX, SOCK_STREAM, 0); | |
450 | if (fd < 0) { | |
451 | SPDK_ERRLOG("socket() error, %s\n", spdk_strerror(errno)); | |
452 | return -errno; | |
453 | } | |
454 | ||
455 | flag = fcntl(fd, F_GETFD); | |
456 | if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0) { | |
457 | SPDK_ERRLOG("fcntl failed, %s\n", spdk_strerror(errno)); | |
458 | } | |
459 | ||
460 | memset(&un, 0, sizeof(un)); | |
461 | un.sun_family = AF_UNIX; | |
462 | rc = snprintf(un.sun_path, sizeof(un.sun_path), "%s", dev->path); | |
463 | if (rc < 0 || (size_t)rc >= sizeof(un.sun_path)) { | |
464 | SPDK_ERRLOG("socket path too long\n"); | |
465 | close(fd); | |
466 | if (rc < 0) { | |
467 | return -errno; | |
468 | } else { | |
469 | return -EINVAL; | |
470 | } | |
471 | } | |
472 | if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) { | |
473 | SPDK_ERRLOG("connect error, %s\n", spdk_strerror(errno)); | |
474 | close(fd); | |
475 | return -errno; | |
476 | } | |
477 | ||
478 | dev->vhostfd = fd; | |
479 | return 0; | |
480 | } | |
481 | ||
482 | struct virtio_user_backend_ops ops_user = { | |
483 | .setup = vhost_user_setup, | |
484 | .send_request = vhost_user_sock, | |
485 | }; | |
486 | ||
487 | SPDK_LOG_REGISTER_COMPONENT("virtio_user", SPDK_LOG_VIRTIO_USER) |