]> git.proxmox.com Git - ceph.git/blame - ceph/src/dpdk/lib/librte_vhost/vhost_user.c
projects / ceph.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bump version to 12.2.12-pve1
[ceph.git] / ceph / src / dpdk / lib / librte_vhost / vhost_user.c
CommitLineData
7c673cae
FG		 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 <stdint.h>
35#include <stdio.h>
36#include <stdlib.h>
37#include <string.h>
38#include <unistd.h>
39#include <sys/mman.h>
40#include <sys/types.h>
41#include <sys/stat.h>
42#include <assert.h>
43#ifdef RTE_LIBRTE_VHOST_NUMA
44#include <numaif.h>
45#endif
46
47#include <rte_common.h>
48#include <rte_malloc.h>
49#include <rte_log.h>
50
51#include "vhost.h"
52#include "vhost_user.h"
53
54static const char *vhost_message_str[VHOST_USER_MAX] = {
55 [VHOST_USER_NONE] = "VHOST_USER_NONE",
56 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
57 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
58 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
59 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
60 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
61 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
62 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
63 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
64 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
65 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
66 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
67 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
68 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
69 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
70 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
71 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
72 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
73 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
74 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
75};
76
77static uint64_t
78get_blk_size(int fd)
79{
80 struct stat stat;
81 int ret;
82
83 ret = fstat(fd, &stat);
84 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
85}
86
87static void
88free_mem_region(struct virtio_net *dev)
89{
90 uint32_t i;
91 struct virtio_memory_region *reg;
92
93 if (!dev || !dev->mem)
94 return;
95
96 for (i = 0; i < dev->mem->nregions; i++) {
97 reg = &dev->mem->regions[i];
98 if (reg->host_user_addr) {
99 munmap(reg->mmap_addr, reg->mmap_size);
100 close(reg->fd);
101 }
102 }
103}
104
105void
106vhost_backend_cleanup(struct virtio_net *dev)
107{
108 if (dev->mem) {
109 free_mem_region(dev);
110 rte_free(dev->mem);
111 dev->mem = NULL;
112 }
113 if (dev->log_addr) {
114 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
115 dev->log_addr = 0;
116 }
117}
118
119/*
120 * This function just returns success at the moment unless
121 * the device hasn't been initialised.
122 */
123static int
124vhost_user_set_owner(void)
125{
126 return 0;
127}
128
129static int
130vhost_user_reset_owner(struct virtio_net *dev)
131{
132 if (dev->flags & VIRTIO_DEV_RUNNING) {
133 dev->flags &= ~VIRTIO_DEV_RUNNING;
134 notify_ops->destroy_device(dev->vid);
135 }
136
137 cleanup_device(dev, 0);
138 reset_device(dev);
139 return 0;
140}
141
142/*
143 * The features that we support are requested.
144 */
145static uint64_t
146vhost_user_get_features(void)
147{
148 return VHOST_FEATURES;
149}
150
151/*
152 * We receive the negotiated features supported by us and the virtio device.
153 */
154static int
155vhost_user_set_features(struct virtio_net *dev, uint64_t features)
156{
157 if (features & ~VHOST_FEATURES)
158 return -1;
159
160 dev->features = features;
161 if (dev->features &
162 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
163 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
164 } else {
165 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
166 }
167 LOG_DEBUG(VHOST_CONFIG,
168 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
169 dev->vid,
170 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
171 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
172
173 return 0;
174}
175
176/*
177 * The virtio device sends us the size of the descriptor ring.
178 */
179static int
180vhost_user_set_vring_num(struct virtio_net *dev,
181 struct vhost_vring_state *state)
182{
183 struct vhost_virtqueue *vq = dev->virtqueue[state->index];
184
185 vq->size = state->num;
186
187 if (dev->dequeue_zero_copy) {
188 vq->nr_zmbuf = 0;
189 vq->last_zmbuf_idx = 0;
190 vq->zmbuf_size = vq->size;
191 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
192 sizeof(struct zcopy_mbuf), 0);
193 if (vq->zmbufs == NULL) {
194 RTE_LOG(WARNING, VHOST_CONFIG,
195 "failed to allocate mem for zero copy; "
196 "zero copy is force disabled\n");
197 dev->dequeue_zero_copy = 0;
198 }
199 }
200
201 vq->shadow_used_ring = rte_malloc(NULL,
202 vq->size * sizeof(struct vring_used_elem),
203 RTE_CACHE_LINE_SIZE);
204 if (!vq->shadow_used_ring) {
205 RTE_LOG(ERR, VHOST_CONFIG,
206 "failed to allocate memory for shadow used ring.\n");
207 return -1;
208 }
209
210 return 0;
211}
212
213/*
214 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
215 * same numa node as the memory of vring descriptor.
216 */
217#ifdef RTE_LIBRTE_VHOST_NUMA
218static struct virtio_net*
219numa_realloc(struct virtio_net *dev, int index)
220{
221 int oldnode, newnode;
222 struct virtio_net *old_dev;
223 struct vhost_virtqueue *old_vq, *vq;
224 int ret;
225
226 /*
227 * vq is allocated on pairs, we should try to do realloc
228 * on first queue of one queue pair only.
229 */
230 if (index % VIRTIO_QNUM != 0)
231 return dev;
232
233 old_dev = dev;
234 vq = old_vq = dev->virtqueue[index];
235
236 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
237 MPOL_F_NODE | MPOL_F_ADDR);
238
239 /* check if we need to reallocate vq */
240 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
241 MPOL_F_NODE | MPOL_F_ADDR);
242 if (ret) {
243 RTE_LOG(ERR, VHOST_CONFIG,
244 "Unable to get vq numa information.\n");
245 return dev;
246 }
247 if (oldnode != newnode) {
248 RTE_LOG(INFO, VHOST_CONFIG,
249 "reallocate vq from %d to %d node\n", oldnode, newnode);
250 vq = rte_malloc_socket(NULL, sizeof(*vq) * VIRTIO_QNUM, 0,
251 newnode);
252 if (!vq)
253 return dev;
254
255 memcpy(vq, old_vq, sizeof(*vq) * VIRTIO_QNUM);
256 rte_free(old_vq);
257 }
258
259 /* check if we need to reallocate dev */
260 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
261 MPOL_F_NODE | MPOL_F_ADDR);
262 if (ret) {
263 RTE_LOG(ERR, VHOST_CONFIG,
264 "Unable to get dev numa information.\n");
265 goto out;
266 }
267 if (oldnode != newnode) {
268 RTE_LOG(INFO, VHOST_CONFIG,
269 "reallocate dev from %d to %d node\n",
270 oldnode, newnode);
271 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
272 if (!dev) {
273 dev = old_dev;
274 goto out;
275 }
276
277 memcpy(dev, old_dev, sizeof(*dev));
278 rte_free(old_dev);
279 }
280
281out:
282 dev->virtqueue[index] = vq;
283 dev->virtqueue[index + 1] = vq + 1;
284 vhost_devices[dev->vid] = dev;
285
286 return dev;
287}
288#else
289static struct virtio_net*
290numa_realloc(struct virtio_net *dev, int index __rte_unused)
291{
292 return dev;
293}
294#endif
295
296/*
297 * Converts QEMU virtual address to Vhost virtual address. This function is
298 * used to convert the ring addresses to our address space.
299 */
300static uint64_t
301qva_to_vva(struct virtio_net *dev, uint64_t qva)
302{
303 struct virtio_memory_region *reg;
304 uint32_t i;
305
306 /* Find the region where the address lives. */
307 for (i = 0; i < dev->mem->nregions; i++) {
308 reg = &dev->mem->regions[i];
309
310 if (qva >= reg->guest_user_addr &&
311 qva < reg->guest_user_addr + reg->size) {
312 return qva - reg->guest_user_addr +
313 reg->host_user_addr;
314 }
315 }
316
317 return 0;
318}
319
320/*
321 * The virtio device sends us the desc, used and avail ring addresses.
322 * This function then converts these to our address space.
323 */
324static int
325vhost_user_set_vring_addr(struct virtio_net *dev, struct vhost_vring_addr *addr)
326{
327 struct vhost_virtqueue *vq;
328
329 if (dev->mem == NULL)
330 return -1;
331
332 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
333 vq = dev->virtqueue[addr->index];
334
335 /* The addresses are converted from QEMU virtual to Vhost virtual. */
336 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
337 addr->desc_user_addr);
338 if (vq->desc == 0) {
339 RTE_LOG(ERR, VHOST_CONFIG,
340 "(%d) failed to find desc ring address.\n",
341 dev->vid);
342 return -1;
343 }
344
345 dev = numa_realloc(dev, addr->index);
346 vq = dev->virtqueue[addr->index];
347
348 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
349 addr->avail_user_addr);
350 if (vq->avail == 0) {
351 RTE_LOG(ERR, VHOST_CONFIG,
352 "(%d) failed to find avail ring address.\n",
353 dev->vid);
354 return -1;
355 }
356
357 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
358 addr->used_user_addr);
359 if (vq->used == 0) {
360 RTE_LOG(ERR, VHOST_CONFIG,
361 "(%d) failed to find used ring address.\n",
362 dev->vid);
363 return -1;
364 }
365
366 if (vq->last_used_idx != vq->used->idx) {
367 RTE_LOG(WARNING, VHOST_CONFIG,
368 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
369 "some packets maybe resent for Tx and dropped for Rx\n",
370 vq->last_used_idx, vq->used->idx);
371 vq->last_used_idx = vq->used->idx;
372 vq->last_avail_idx = vq->used->idx;
373 }
374
375 vq->log_guest_addr = addr->log_guest_addr;
376
377 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
378 dev->vid, vq->desc);
379 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
380 dev->vid, vq->avail);
381 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
382 dev->vid, vq->used);
383 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
384 dev->vid, vq->log_guest_addr);
385
386 return 0;
387}
388
389/*
390 * The virtio device sends us the available ring last used index.
391 */
392static int
393vhost_user_set_vring_base(struct virtio_net *dev,
394 struct vhost_vring_state *state)
395{
396 dev->virtqueue[state->index]->last_used_idx = state->num;
397 dev->virtqueue[state->index]->last_avail_idx = state->num;
398
399 return 0;
400}
401
402static void
403add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
404 uint64_t host_phys_addr, uint64_t size)
405{
406 struct guest_page *page, *last_page;
407
408 if (dev->nr_guest_pages == dev->max_guest_pages) {
409 dev->max_guest_pages *= 2;
410 dev->guest_pages = realloc(dev->guest_pages,
411 dev->max_guest_pages * sizeof(*page));
412 }
413
414 if (dev->nr_guest_pages > 0) {
415 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
416 /* merge if the two pages are continuous */
417 if (host_phys_addr == last_page->host_phys_addr +
418 last_page->size) {
419 last_page->size += size;
420 return;
421 }
422 }
423
424 page = &dev->guest_pages[dev->nr_guest_pages++];
425 page->guest_phys_addr = guest_phys_addr;
426 page->host_phys_addr = host_phys_addr;
427 page->size = size;
428}
429
430static void
431add_guest_pages(struct virtio_net *dev, struct virtio_memory_region *reg,
432 uint64_t page_size)
433{
434 uint64_t reg_size = reg->size;
435 uint64_t host_user_addr = reg->host_user_addr;
436 uint64_t guest_phys_addr = reg->guest_phys_addr;
437 uint64_t host_phys_addr;
438 uint64_t size;
439
440 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
441 size = page_size - (guest_phys_addr & (page_size - 1));
442 size = RTE_MIN(size, reg_size);
443
444 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
445 host_user_addr += size;
446 guest_phys_addr += size;
447 reg_size -= size;
448
449 while (reg_size > 0) {
450 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
451 host_user_addr);
452 add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
453 page_size);
454
455 host_user_addr += page_size;
456 guest_phys_addr += page_size;
457 reg_size -= page_size;
458 }
459}
460
461#ifdef RTE_LIBRTE_VHOST_DEBUG
462/* TODO: enable it only in debug mode? */
463static void
464dump_guest_pages(struct virtio_net *dev)
465{
466 uint32_t i;
467 struct guest_page *page;
468
469 for (i = 0; i < dev->nr_guest_pages; i++) {
470 page = &dev->guest_pages[i];
471
472 RTE_LOG(INFO, VHOST_CONFIG,
473 "guest physical page region %u\n"
474 "\t guest_phys_addr: %" PRIx64 "\n"
475 "\t host_phys_addr : %" PRIx64 "\n"
476 "\t size : %" PRIx64 "\n",
477 i,
478 page->guest_phys_addr,
479 page->host_phys_addr,
480 page->size);
481 }
482}
483#else
484#define dump_guest_pages(dev)
485#endif
486
487static int
488vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
489{
490 struct VhostUserMemory memory = pmsg->payload.memory;
491 struct virtio_memory_region *reg;
492 void *mmap_addr;
493 uint64_t mmap_size;
494 uint64_t mmap_offset;
495 uint64_t alignment;
496 uint32_t i;
497 int fd;
498
499 /* Remove from the data plane. */
500 if (dev->flags & VIRTIO_DEV_RUNNING) {
501 dev->flags &= ~VIRTIO_DEV_RUNNING;
502 notify_ops->destroy_device(dev->vid);
503 }
504
505 if (dev->mem) {
506 free_mem_region(dev);
507 rte_free(dev->mem);
508 dev->mem = NULL;
509 }
510
511 dev->nr_guest_pages = 0;
512 if (!dev->guest_pages) {
513 dev->max_guest_pages = 8;
514 dev->guest_pages = malloc(dev->max_guest_pages *
515 sizeof(struct guest_page));
516 }
517
518 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct virtio_memory) +
519 sizeof(struct virtio_memory_region) * memory.nregions, 0);
520 if (dev->mem == NULL) {
521 RTE_LOG(ERR, VHOST_CONFIG,
522 "(%d) failed to allocate memory for dev->mem\n",
523 dev->vid);
524 return -1;
525 }
526 dev->mem->nregions = memory.nregions;
527
528 for (i = 0; i < memory.nregions; i++) {
529 fd = pmsg->fds[i];
530 reg = &dev->mem->regions[i];
531
532 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
533 reg->guest_user_addr = memory.regions[i].userspace_addr;
534 reg->size = memory.regions[i].memory_size;
535 reg->fd = fd;
536
537 mmap_offset = memory.regions[i].mmap_offset;
538 mmap_size = reg->size + mmap_offset;
539
540 /* mmap() without flag of MAP_ANONYMOUS, should be called
541 * with length argument aligned with hugepagesz at older
542 * longterm version Linux, like 2.6.32 and 3.2.72, or
543 * mmap() will fail with EINVAL.
544 *
545 * to avoid failure, make sure in caller to keep length
546 * aligned.
547 */
548 alignment = get_blk_size(fd);
549 if (alignment == (uint64_t)-1) {
550 RTE_LOG(ERR, VHOST_CONFIG,
551 "couldn't get hugepage size through fstat\n");
552 goto err_mmap;
553 }
554 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
555
556 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
557 MAP_SHARED | MAP_POPULATE, fd, 0);
558
559 if (mmap_addr == MAP_FAILED) {
560 RTE_LOG(ERR, VHOST_CONFIG,
561 "mmap region %u failed.\n", i);
562 goto err_mmap;
563 }
564
565 reg->mmap_addr = mmap_addr;
566 reg->mmap_size = mmap_size;
567 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
568 mmap_offset;
569
570 add_guest_pages(dev, reg, alignment);
571
572 RTE_LOG(INFO, VHOST_CONFIG,
573 "guest memory region %u, size: 0x%" PRIx64 "\n"
574 "\t guest physical addr: 0x%" PRIx64 "\n"
575 "\t guest virtual addr: 0x%" PRIx64 "\n"
576 "\t host virtual addr: 0x%" PRIx64 "\n"
577 "\t mmap addr : 0x%" PRIx64 "\n"
578 "\t mmap size : 0x%" PRIx64 "\n"
579 "\t mmap align: 0x%" PRIx64 "\n"
580 "\t mmap off : 0x%" PRIx64 "\n",
581 i, reg->size,
582 reg->guest_phys_addr,
583 reg->guest_user_addr,
584 reg->host_user_addr,
585 (uint64_t)(uintptr_t)mmap_addr,
586 mmap_size,
587 alignment,
588 mmap_offset);
589 }
590
591 dump_guest_pages(dev);
592
593 return 0;
594
595err_mmap:
596 free_mem_region(dev);
597 rte_free(dev->mem);
598 dev->mem = NULL;
599 return -1;
600}
601
602static int
603vq_is_ready(struct vhost_virtqueue *vq)
604{
605 return vq && vq->desc &&
606 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
607 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
608}
609
610static int
611virtio_is_ready(struct virtio_net *dev)
612{
613 struct vhost_virtqueue *rvq, *tvq;
614 uint32_t i;
615
616 for (i = 0; i < dev->virt_qp_nb; i++) {
617 rvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
618 tvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
619
620 if (!vq_is_ready(rvq) || !vq_is_ready(tvq)) {
621 RTE_LOG(INFO, VHOST_CONFIG,
622 "virtio is not ready for processing.\n");
623 return 0;
624 }
625 }
626
627 RTE_LOG(INFO, VHOST_CONFIG,
628 "virtio is now ready for processing.\n");
629 return 1;
630}
631
632static void
633vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
634{
635 struct vhost_vring_file file;
636 struct vhost_virtqueue *vq;
637 uint32_t cur_qp_idx;
638
639 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
640 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
641 file.fd = VIRTIO_INVALID_EVENTFD;
642 else
643 file.fd = pmsg->fds[0];
644 RTE_LOG(INFO, VHOST_CONFIG,
645 "vring call idx:%d file:%d\n", file.index, file.fd);
646
647 /*
648 * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
649 * we get, so we do vring queue pair allocation here.
650 */
651 cur_qp_idx = file.index / VIRTIO_QNUM;
652 if (cur_qp_idx + 1 > dev->virt_qp_nb) {
653 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
654 return;
655 }
656
657 vq = dev->virtqueue[file.index];
658 assert(vq != NULL);
659
660 if (vq->callfd >= 0)
661 close(vq->callfd);
662
663 vq->callfd = file.fd;
664}
665
666/*
667 * In vhost-user, when we receive kick message, will test whether virtio
668 * device is ready for packet processing.
669 */
670static void
671vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
672{
673 struct vhost_vring_file file;
674 struct vhost_virtqueue *vq;
675
676 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
677 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
678 file.fd = VIRTIO_INVALID_EVENTFD;
679 else
680 file.fd = pmsg->fds[0];
681 RTE_LOG(INFO, VHOST_CONFIG,
682 "vring kick idx:%d file:%d\n", file.index, file.fd);
683
684 vq = dev->virtqueue[file.index];
685 if (vq->kickfd >= 0)
686 close(vq->kickfd);
687 vq->kickfd = file.fd;
688
689 if (virtio_is_ready(dev) && !(dev->flags & VIRTIO_DEV_RUNNING)) {
690 if (dev->dequeue_zero_copy) {
691 RTE_LOG(INFO, VHOST_CONFIG,
692 "dequeue zero copy is enabled\n");
693 }
694
695 if (notify_ops->new_device(dev->vid) == 0)
696 dev->flags |= VIRTIO_DEV_RUNNING;
697 }
698}
699
700static void
701free_zmbufs(struct vhost_virtqueue *vq)
702{
703 struct zcopy_mbuf *zmbuf, *next;
704
705 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
706 zmbuf != NULL; zmbuf = next) {
707 next = TAILQ_NEXT(zmbuf, next);
708
709 rte_pktmbuf_free(zmbuf->mbuf);
710 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
711 }
712
713 rte_free(vq->zmbufs);
714}
715
716/*
717 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
718 */
719static int
720vhost_user_get_vring_base(struct virtio_net *dev,
721 struct vhost_vring_state *state)
722{
723 struct vhost_virtqueue *vq = dev->virtqueue[state->index];
724
725 /* We have to stop the queue (virtio) if it is running. */
726 if (dev->flags & VIRTIO_DEV_RUNNING) {
727 dev->flags &= ~VIRTIO_DEV_RUNNING;
728 notify_ops->destroy_device(dev->vid);
729 }
730
731 /* Here we are safe to get the last used index */
732 state->num = vq->last_used_idx;
733
734 RTE_LOG(INFO, VHOST_CONFIG,
735 "vring base idx:%d file:%d\n", state->index, state->num);
736 /*
737 * Based on current qemu vhost-user implementation, this message is
738 * sent and only sent in vhost_vring_stop.
739 * TODO: cleanup the vring, it isn't usable since here.
740 */
741 if (vq->kickfd >= 0)
742 close(vq->kickfd);
743
744 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
745
746 if (dev->dequeue_zero_copy)
747 free_zmbufs(vq);
748 rte_free(vq->shadow_used_ring);
749 vq->shadow_used_ring = NULL;
750
751 return 0;
752}
753
754/*
755 * when virtio queues are ready to work, qemu will send us to
756 * enable the virtio queue pair.
757 */
758static int
759vhost_user_set_vring_enable(struct virtio_net *dev,
760 struct vhost_vring_state *state)
761{
762 int enable = (int)state->num;
763
764 RTE_LOG(INFO, VHOST_CONFIG,
765 "set queue enable: %d to qp idx: %d\n",
766 enable, state->index);
767
768 if (notify_ops->vring_state_changed)
769 notify_ops->vring_state_changed(dev->vid, state->index, enable);
770
771 dev->virtqueue[state->index]->enabled = enable;
772
773 return 0;
774}
775
776static void
777vhost_user_set_protocol_features(struct virtio_net *dev,
778 uint64_t protocol_features)
779{
780 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
781 return;
782
783 dev->protocol_features = protocol_features;
784}
785
786static int
787vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
788{
789 int fd = msg->fds[0];
790 uint64_t size, off;
791 void *addr;
792
793 if (fd < 0) {
794 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
795 return -1;
796 }
797
798 if (msg->size != sizeof(VhostUserLog)) {
799 RTE_LOG(ERR, VHOST_CONFIG,
800 "invalid log base msg size: %"PRId32" != %d\n",
801 msg->size, (int)sizeof(VhostUserLog));
802 return -1;
803 }
804
805 size = msg->payload.log.mmap_size;
806 off = msg->payload.log.mmap_offset;
807 RTE_LOG(INFO, VHOST_CONFIG,
808 "log mmap size: %"PRId64", offset: %"PRId64"\n",
809 size, off);
810
811 /*
812 * mmap from 0 to workaround a hugepage mmap bug: mmap will
813 * fail when offset is not page size aligned.
814 */
815 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
816 close(fd);
817 if (addr == MAP_FAILED) {
818 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
819 return -1;
820 }
821
822 /*
823 * Free previously mapped log memory on occasionally
824 * multiple VHOST_USER_SET_LOG_BASE.
825 */
826 if (dev->log_addr) {
827 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
828 }
829 dev->log_addr = (uint64_t)(uintptr_t)addr;
830 dev->log_base = dev->log_addr + off;
831 dev->log_size = size;
832
833 return 0;
834}
835
836/*
837 * An rarp packet is constructed and broadcasted to notify switches about
838 * the new location of the migrated VM, so that packets from outside will
839 * not be lost after migration.
840 *
841 * However, we don't actually "send" a rarp packet here, instead, we set
842 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
843 */
844static int
845vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
846{
847 uint8_t *mac = (uint8_t *)&msg->payload.u64;
848
849 RTE_LOG(DEBUG, VHOST_CONFIG,
850 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
851 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
852 memcpy(dev->mac.addr_bytes, mac, 6);
853
854 /*
855 * Set the flag to inject a RARP broadcast packet at
856 * rte_vhost_dequeue_burst().
857 *
858 * rte_smp_wmb() is for making sure the mac is copied
859 * before the flag is set.
860 */
861 rte_smp_wmb();
862 rte_atomic16_set(&dev->broadcast_rarp, 1);
863
864 return 0;
865}
866
867/* return bytes# of read on success or negative val on failure. */
868static int
869read_vhost_message(int sockfd, struct VhostUserMsg *msg)
870{
871 int ret;
872
873 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
874 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
875 if (ret <= 0)
876 return ret;
877
878 if (msg && msg->size) {
879 if (msg->size > sizeof(msg->payload)) {
880 RTE_LOG(ERR, VHOST_CONFIG,
881 "invalid msg size: %d\n", msg->size);
882 return -1;
883 }
884 ret = read(sockfd, &msg->payload, msg->size);
885 if (ret <= 0)
886 return ret;
887 if (ret != (int)msg->size) {
888 RTE_LOG(ERR, VHOST_CONFIG,
889 "read control message failed\n");
890 return -1;
891 }
892 }
893
894 return ret;
895}
896
897static int
898send_vhost_message(int sockfd, struct VhostUserMsg *msg)
899{
900 int ret;
901
902 if (!msg)
903 return 0;
904
905 msg->flags &= ~VHOST_USER_VERSION_MASK;
906 msg->flags |= VHOST_USER_VERSION;
907 msg->flags |= VHOST_USER_REPLY_MASK;
908
909 ret = send_fd_message(sockfd, (char *)msg,
910 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
911
912 return ret;
913}
914
915int
916vhost_user_msg_handler(int vid, int fd)
917{
918 struct virtio_net *dev;
919 struct VhostUserMsg msg;
920 int ret;
921
922 dev = get_device(vid);
923 if (dev == NULL)
924 return -1;
925
926 ret = read_vhost_message(fd, &msg);
927 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
928 if (ret < 0)
929 RTE_LOG(ERR, VHOST_CONFIG,
930 "vhost read message failed\n");
931 else if (ret == 0)
932 RTE_LOG(INFO, VHOST_CONFIG,
933 "vhost peer closed\n");
934 else
935 RTE_LOG(ERR, VHOST_CONFIG,
936 "vhost read incorrect message\n");
937
938 return -1;
939 }
940
941 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
942 vhost_message_str[msg.request]);
943 switch (msg.request) {
944 case VHOST_USER_GET_FEATURES:
945 msg.payload.u64 = vhost_user_get_features();
946 msg.size = sizeof(msg.payload.u64);
947 send_vhost_message(fd, &msg);
948 break;
949 case VHOST_USER_SET_FEATURES:
950 vhost_user_set_features(dev, msg.payload.u64);
951 break;
952
953 case VHOST_USER_GET_PROTOCOL_FEATURES:
954 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
955 msg.size = sizeof(msg.payload.u64);
956 send_vhost_message(fd, &msg);
957 break;
958 case VHOST_USER_SET_PROTOCOL_FEATURES:
959 vhost_user_set_protocol_features(dev, msg.payload.u64);
960 break;
961
962 case VHOST_USER_SET_OWNER:
963 vhost_user_set_owner();
964 break;
965 case VHOST_USER_RESET_OWNER:
966 vhost_user_reset_owner(dev);
967 break;
968
969 case VHOST_USER_SET_MEM_TABLE:
970 vhost_user_set_mem_table(dev, &msg);
971 break;
972
973 case VHOST_USER_SET_LOG_BASE:
974 vhost_user_set_log_base(dev, &msg);
975
976 /* it needs a reply */
977 msg.size = sizeof(msg.payload.u64);
978 send_vhost_message(fd, &msg);
979 break;
980 case VHOST_USER_SET_LOG_FD:
981 close(msg.fds[0]);
982 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
983 break;
984
985 case VHOST_USER_SET_VRING_NUM:
986 vhost_user_set_vring_num(dev, &msg.payload.state);
987 break;
988 case VHOST_USER_SET_VRING_ADDR:
989 vhost_user_set_vring_addr(dev, &msg.payload.addr);
990 break;
991 case VHOST_USER_SET_VRING_BASE:
992 vhost_user_set_vring_base(dev, &msg.payload.state);
993 break;
994
995 case VHOST_USER_GET_VRING_BASE:
996 ret = vhost_user_get_vring_base(dev, &msg.payload.state);
997 msg.size = sizeof(msg.payload.state);
998 send_vhost_message(fd, &msg);
999 break;
1000
1001 case VHOST_USER_SET_VRING_KICK:
1002 vhost_user_set_vring_kick(dev, &msg);
1003 break;
1004 case VHOST_USER_SET_VRING_CALL:
1005 vhost_user_set_vring_call(dev, &msg);
1006 break;
1007
1008 case VHOST_USER_SET_VRING_ERR:
1009 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1010 close(msg.fds[0]);
1011 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1012 break;
1013
1014 case VHOST_USER_GET_QUEUE_NUM:
1015 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1016 msg.size = sizeof(msg.payload.u64);
1017 send_vhost_message(fd, &msg);
1018 break;
1019
1020 case VHOST_USER_SET_VRING_ENABLE:
1021 vhost_user_set_vring_enable(dev, &msg.payload.state);
1022 break;
1023 case VHOST_USER_SEND_RARP:
1024 vhost_user_send_rarp(dev, &msg);
1025 break;
1026
1027 default:
1028 break;
1029
1030 }
1031
1032 return 0;
1033}
Ceph