FD.io VPP  v17.10-9-gd594711
Vector Packet Processing
vhost-user.c
Go to the documentation of this file.
1 /*
2  *------------------------------------------------------------------
3  * vhost.c - vhost-user
4  *
5  * Copyright (c) 2014 Cisco and/or its affiliates.
6  * Licensed under the Apache License, Version 2.0 (the "License");
7  * you may not use this file except in compliance with the License.
8  * You may obtain a copy of the License at:
9  *
10  * http://www.apache.org/licenses/LICENSE-2.0
11  *
12  * Unless required by applicable law or agreed to in writing, software
13  * distributed under the License is distributed on an "AS IS" BASIS,
14  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15  * See the License for the specific language governing permissions and
16  * limitations under the License.
17  *------------------------------------------------------------------
18  */
19 
20 #include <fcntl.h> /* for open */
21 #include <sys/ioctl.h>
22 #include <sys/socket.h>
23 #include <sys/un.h>
24 #include <sys/stat.h>
25 #include <sys/types.h>
26 #include <sys/uio.h> /* for iovec */
27 #include <netinet/in.h>
28 #include <sys/vfs.h>
29 
30 #include <linux/if_arp.h>
31 #include <linux/if_tun.h>
32 
33 #include <vlib/vlib.h>
34 #include <vlib/unix/unix.h>
35 
36 #include <vnet/ip/ip.h>
37 
38 #include <vnet/ethernet/ethernet.h>
39 #include <vnet/devices/devices.h>
40 #include <vnet/feature/feature.h>
41 
43 
44 /**
45  * @file
46  * @brief vHost User Device Driver.
47  *
48  * This file contains the source code for vHost User interface.
49  */
50 
51 
52 #define VHOST_DEBUG_VQ 0
53 
54 #define DBG_SOCK(args...) \
55  { \
56  vhost_user_main_t *_vum = &vhost_user_main; \
57  if (_vum->debug) \
58  clib_warning(args); \
59  };
60 
61 #if VHOST_DEBUG_VQ == 1
62 #define DBG_VQ(args...) clib_warning(args);
63 #else
64 #define DBG_VQ(args...)
65 #endif
66 
67 /*
68  * When an RX queue is down but active, received packets
69  * must be discarded. This value controls up to how many
70  * packets will be discarded during each round.
71  */
72 #define VHOST_USER_DOWN_DISCARD_COUNT 256
73 
74 /*
75  * When the number of available buffers gets under this threshold,
76  * RX node will start discarding packets.
77  */
78 #define VHOST_USER_RX_BUFFER_STARVATION 32
79 
80 /*
81  * On the receive side, the host should free descriptors as soon
82  * as possible in order to avoid TX drop in the VM.
83  * This value controls the number of copy operations that are stacked
84  * before copy is done for all and descriptors are given back to
85  * the guest.
86  * The value 64 was obtained by testing (48 and 128 were not as good).
87  */
88 #define VHOST_USER_RX_COPY_THRESHOLD 64
89 /*
90  * On the transmit side, we keep processing the buffers from vlib in the while
91  * loop and prepare the copy order to be executed later. However, the static
92  * array which we keep the copy order is limited to VHOST_USER_COPY_ARRAY_N
93  * entries. In order to not corrupt memory, we have to do the copy when the
94  * static array reaches the copy threshold. We subtract 40 in case the code
95  * goes into the inner loop for a maximum of 64k frames which may require
96  * more array entries.
97  */
98 #define VHOST_USER_TX_COPY_THRESHOLD (VHOST_USER_COPY_ARRAY_N - 40)
99 
100 #define UNIX_GET_FD(unixfd_idx) \
101  (unixfd_idx != ~0) ? \
102  pool_elt_at_index (file_main.file_pool, \
103  unixfd_idx)->file_descriptor : -1;
104 
105 #define foreach_virtio_trace_flags \
106  _ (SIMPLE_CHAINED, 0, "Simple descriptor chaining") \
107  _ (SINGLE_DESC, 1, "Single descriptor packet") \
108  _ (INDIRECT, 2, "Indirect descriptor") \
109  _ (MAP_ERROR, 4, "Memory mapping error")
110 
111 typedef enum
112 {
113 #define _(n,i,s) VIRTIO_TRACE_F_##n,
115 #undef _
117 
119 
120 #define foreach_vhost_user_tx_func_error \
121  _(NONE, "no error") \
122  _(NOT_READY, "vhost vring not ready") \
123  _(DOWN, "vhost interface is down") \
124  _(PKT_DROP_NOBUF, "tx packet drops (no available descriptors)") \
125  _(PKT_DROP_NOMRG, "tx packet drops (cannot merge descriptors)") \
126  _(MMAP_FAIL, "mmap failure") \
127  _(INDIRECT_OVERFLOW, "indirect descriptor table overflow")
128 
129 typedef enum
130 {
131 #define _(f,s) VHOST_USER_TX_FUNC_ERROR_##f,
133 #undef _
136 
138 #define _(n,s) s,
140 #undef _
141 };
142 
143 #define foreach_vhost_user_input_func_error \
144  _(NO_ERROR, "no error") \
145  _(NO_BUFFER, "no available buffer") \
146  _(MMAP_FAIL, "mmap failure") \
147  _(INDIRECT_OVERFLOW, "indirect descriptor overflows table") \
148  _(UNDERSIZED_FRAME, "undersized ethernet frame received (< 14 bytes)") \
149  _(FULL_RX_QUEUE, "full rx queue (possible driver tx drop)")
150 
151 typedef enum
152 {
153 #define _(f,s) VHOST_USER_INPUT_FUNC_ERROR_##f,
155 #undef _
158 
160 #define _(n,s) s,
162 #undef _
163 };
164 
165 /* *INDENT-OFF* */
166 static vhost_user_main_t vhost_user_main = {
167  .mtu_bytes = 1518,
168 };
169 
170 VNET_HW_INTERFACE_CLASS (vhost_interface_class, static) = {
171  .name = "vhost-user",
172 };
173 /* *INDENT-ON* */
174 
175 static u8 *
177 {
178  u32 i = va_arg (*args, u32);
179  u32 show_dev_instance = ~0;
181 
183  show_dev_instance = vum->show_dev_instance_by_real_dev_instance[i];
184 
185  if (show_dev_instance != ~0)
186  i = show_dev_instance;
187 
188  s = format (s, "VirtualEthernet0/0/%d", i);
189  return s;
190 }
191 
192 static int
194 {
195  // FIXME: check if the new dev instance is already used
198  hi->dev_instance, ~0);
199 
201  new_dev_instance;
202 
203  DBG_SOCK ("renumbered vhost-user interface dev_instance %d to %d",
204  hi->dev_instance, new_dev_instance);
205 
206  return 0;
207 }
208 
211 {
212  int i = *hint;
213  if (PREDICT_TRUE ((vui->regions[i].guest_phys_addr <= addr) &&
214  ((vui->regions[i].guest_phys_addr +
215  vui->regions[i].memory_size) > addr)))
216  {
217  return (void *) (vui->region_mmap_addr[i] + addr -
218  vui->regions[i].guest_phys_addr);
219  }
220 #if __SSE4_2__
221  __m128i rl, rh, al, ah, r;
222  al = _mm_set1_epi64x (addr + 1);
223  ah = _mm_set1_epi64x (addr);
224 
225  rl = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_lo[0]);
226  rl = _mm_cmpgt_epi64 (al, rl);
227  rh = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_hi[0]);
228  rh = _mm_cmpgt_epi64 (rh, ah);
229  r = _mm_and_si128 (rl, rh);
230 
231  rl = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_lo[2]);
232  rl = _mm_cmpgt_epi64 (al, rl);
233  rh = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_hi[2]);
234  rh = _mm_cmpgt_epi64 (rh, ah);
235  r = _mm_blend_epi16 (r, _mm_and_si128 (rl, rh), 0x22);
236 
237  rl = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_lo[4]);
238  rl = _mm_cmpgt_epi64 (al, rl);
239  rh = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_hi[4]);
240  rh = _mm_cmpgt_epi64 (rh, ah);
241  r = _mm_blend_epi16 (r, _mm_and_si128 (rl, rh), 0x44);
242 
243  rl = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_lo[6]);
244  rl = _mm_cmpgt_epi64 (al, rl);
245  rh = _mm_loadu_si128 ((__m128i *) & vui->region_guest_addr_hi[6]);
246  rh = _mm_cmpgt_epi64 (rh, ah);
247  r = _mm_blend_epi16 (r, _mm_and_si128 (rl, rh), 0x88);
248 
249  r = _mm_shuffle_epi8 (r, _mm_set_epi64x (0, 0x0e060c040a020800));
250  i = __builtin_ctzll (_mm_movemask_epi8 (r) |
252 
253  if (i < vui->nregions)
254  {
255  *hint = i;
256  return (void *) (vui->region_mmap_addr[i] + addr -
257  vui->regions[i].guest_phys_addr);
258  }
259 
260 #else
261  for (i = 0; i < vui->nregions; i++)
262  {
263  if ((vui->regions[i].guest_phys_addr <= addr) &&
264  ((vui->regions[i].guest_phys_addr + vui->regions[i].memory_size) >
265  addr))
266  {
267  *hint = i;
268  return (void *) (vui->region_mmap_addr[i] + addr -
269  vui->regions[i].guest_phys_addr);
270  }
271  }
272 #endif
273  DBG_VQ ("failed to map guest mem addr %llx", addr);
274  *hint = 0;
275  return 0;
276 }
277 
278 static inline void *
280 {
281  int i;
282  for (i = 0; i < vui->nregions; i++)
283  {
284  if ((vui->regions[i].userspace_addr <= addr) &&
285  ((vui->regions[i].userspace_addr + vui->regions[i].memory_size) >
286  addr))
287  {
288  return (void *) (vui->region_mmap_addr[i] + addr -
289  vui->regions[i].userspace_addr);
290  }
291  }
292  return 0;
293 }
294 
295 static long
297 {
298  struct statfs s;
299  fstatfs (fd, &s);
300  return s.f_bsize;
301 }
302 
303 static void
305 {
306  int i, r;
307  for (i = 0; i < vui->nregions; i++)
308  {
309  if (vui->region_mmap_addr[i] != (void *) -1)
310  {
311 
312  long page_sz = get_huge_page_size (vui->region_mmap_fd[i]);
313 
314  ssize_t map_sz = (vui->regions[i].memory_size +
315  vui->regions[i].mmap_offset +
316  page_sz - 1) & ~(page_sz - 1);
317 
318  r =
319  munmap (vui->region_mmap_addr[i] - vui->regions[i].mmap_offset,
320  map_sz);
321 
322  DBG_SOCK
323  ("unmap memory region %d addr 0x%lx len 0x%lx page_sz 0x%x", i,
324  vui->region_mmap_addr[i], map_sz, page_sz);
325 
326  vui->region_mmap_addr[i] = (void *) -1;
327 
328  if (r == -1)
329  {
330  clib_warning ("failed to unmap memory region (errno %d)",
331  errno);
332  }
333  close (vui->region_mmap_fd[i]);
334  }
335  }
336  vui->nregions = 0;
337 }
338 
339 static void
341 {
342  //Let's try to assign one queue to each thread
343  u32 qid = 0;
344  u32 thread_index = 0;
345  vui->use_tx_spinlock = 0;
346  while (1)
347  {
348  for (qid = 0; qid < VHOST_VRING_MAX_N / 2; qid++)
349  {
350  vhost_user_vring_t *rxvq = &vui->vrings[VHOST_VRING_IDX_RX (qid)];
351  if (!rxvq->started || !rxvq->enabled)
352  continue;
353 
354  vui->per_cpu_tx_qid[thread_index] = qid;
355  thread_index++;
356  if (thread_index == vlib_get_thread_main ()->n_vlib_mains)
357  return;
358  }
359  //We need to loop, meaning the spinlock has to be used
360  vui->use_tx_spinlock = 1;
361  if (thread_index == 0)
362  {
363  //Could not find a single valid one
364  for (thread_index = 0;
365  thread_index < vlib_get_thread_main ()->n_vlib_mains;
366  thread_index++)
367  {
368  vui->per_cpu_tx_qid[thread_index] = 0;
369  }
370  return;
371  }
372  }
373 }
374 
375 /**
376  * @brief Unassign existing interface/queue to thread mappings and re-assign
377  * new interface/queue to thread mappings
378  */
379 static void
381 {
383  vhost_user_intf_t *vui;
384  vhost_user_vring_t *txvq;
385  vnet_main_t *vnm = vnet_get_main ();
386  u32 qid;
387  int rv;
388  u16 *queue;
389 
390  // Scrap all existing mappings for all interfaces/queues
391  /* *INDENT-OFF* */
392  pool_foreach (vui, vum->vhost_user_interfaces, {
393  vec_foreach (queue, vui->rx_queues)
394  {
395  rv = vnet_hw_interface_unassign_rx_thread (vnm, vui->hw_if_index,
396  *queue);
397  if (rv)
398  clib_warning ("Warning: unable to unassign interface %d, "
399  "queue %d: rc=%d", vui->hw_if_index, *queue, rv);
400  }
402  });
403  /* *INDENT-ON* */
404 
405  // Create the rx_queues for all interfaces
406  /* *INDENT-OFF* */
407  pool_foreach (vui, vum->vhost_user_interfaces, {
408  for (qid = 0; qid < VHOST_VRING_MAX_N / 2; qid++)
409  {
410  txvq = &vui->vrings[VHOST_VRING_IDX_TX (qid)];
411  if (txvq->started)
412  {
413  if (txvq->mode == VNET_HW_INTERFACE_RX_MODE_UNKNOWN)
414  /* Set polling as the default */
415  txvq->mode = VNET_HW_INTERFACE_RX_MODE_POLLING;
416  vec_add1 (vui->rx_queues, qid);
417  }
418  }
419  });
420  /* *INDENT-ON* */
421 
422  // Assign new mappings for all interfaces/queues
423  /* *INDENT-OFF* */
424  pool_foreach (vui, vum->vhost_user_interfaces, {
425  vnet_hw_interface_set_input_node (vnm, vui->hw_if_index,
426  vhost_user_input_node.index);
427  vec_foreach (queue, vui->rx_queues)
428  {
429  vnet_hw_interface_assign_rx_thread (vnm, vui->hw_if_index, *queue,
430  ~0);
431  txvq = &vui->vrings[VHOST_VRING_IDX_TX (*queue)];
432  rv = vnet_hw_interface_set_rx_mode (vnm, vui->hw_if_index, *queue,
433  txvq->mode);
434  if (rv)
435  clib_warning ("Warning: unable to set rx mode for interface %d, "
436  "queue %d: rc=%d", vui->hw_if_index, *queue, rv);
437  }
438  });
439  /* *INDENT-ON* */
440 }
441 
442 /** @brief Returns whether at least one TX and one RX vring are enabled */
443 int
445 {
446  int i, found[2] = { }; //RX + TX
447 
448  for (i = 0; i < VHOST_VRING_MAX_N; i++)
449  if (vui->vrings[i].started && vui->vrings[i].enabled)
450  found[i & 1] = 1;
451 
452  return found[0] && found[1];
453 }
454 
455 static void
457 {
458  /* if we have pointers to descriptor table, go up */
459  int is_up = vhost_user_intf_ready (vui);
460  if (is_up != vui->is_up)
461  {
462  DBG_SOCK ("interface %d %s", vui->sw_if_index,
463  is_up ? "ready" : "down");
466  0);
467  vui->is_up = is_up;
468  }
471 }
472 
473 static void
475 {
476  u32 qid;
477  vnet_main_t *vnm = vnet_get_main ();
478 
479  qid = ifq & 0xff;
480  if ((qid & 1) == 0)
481  /* Only care about the odd number, or TX, virtqueue */
482  return;
483 
484  if (vhost_user_intf_ready (vui))
485  // qid >> 1 is to convert virtqueue number to vring queue index
487 }
488 
489 static clib_error_t *
491 {
492  __attribute__ ((unused)) int n;
493  u8 buff[8];
494 
495  n = read (uf->file_descriptor, ((char *) &buff), 8);
496 
497  return 0;
498 }
499 
500 static clib_error_t *
502 {
503  __attribute__ ((unused)) int n;
504  u8 buff[8];
505  vhost_user_intf_t *vui =
506  pool_elt_at_index (vhost_user_main.vhost_user_interfaces,
507  uf->private_data >> 8);
508  u32 qid = uf->private_data & 0xff;
509 
510  n = read (uf->file_descriptor, ((char *) &buff), 8);
511  DBG_SOCK ("if %d KICK queue %d", uf->private_data >> 8, qid);
512  if (!vui->vrings[qid].started ||
513  (vhost_user_intf_ready (vui) != vui->is_up))
514  {
516  vui->vrings[qid].started = 1;
519  }
520 
522  return 0;
523 }
524 
525 /**
526  * @brief Try once to lock the vring
527  * @return 0 on success, non-zero on failure.
528  */
529 static inline int
531 {
532  return __sync_lock_test_and_set (vui->vring_locks[qid], 1);
533 }
534 
535 /**
536  * @brief Spin until the vring is successfully locked
537  */
538 static inline void
540 {
541  while (vhost_user_vring_try_lock (vui, qid))
542  ;
543 }
544 
545 /**
546  * @brief Unlock the vring lock
547  */
548 static inline void
550 {
551  *vui->vring_locks[qid] = 0;
552 }
553 
554 static inline void
556 {
557  vhost_user_vring_t *vring = &vui->vrings[qid];
558  memset (vring, 0, sizeof (*vring));
559  vring->kickfd_idx = ~0;
560  vring->callfd_idx = ~0;
561  vring->errfd = -1;
562 
563  /*
564  * We have a bug with some qemu 2.5, and this may be a fix.
565  * Feel like interpretation holy text, but this is from vhost-user.txt.
566  * "
567  * One queue pair is enabled initially. More queues are enabled
568  * dynamically, by sending message VHOST_USER_SET_VRING_ENABLE.
569  * "
570  * Don't know who's right, but this is what DPDK does.
571  */
572  if (qid == 0 || qid == 1)
573  vring->enabled = 1;
574 }
575 
576 static inline void
578 {
579  vhost_user_vring_t *vring = &vui->vrings[qid];
580  if (vring->kickfd_idx != ~0)
581  {
583  vring->kickfd_idx);
584  clib_file_del (&file_main, uf);
585  vring->kickfd_idx = ~0;
586  }
587  if (vring->callfd_idx != ~0)
588  {
590  vring->callfd_idx);
591  clib_file_del (&file_main, uf);
592  vring->callfd_idx = ~0;
593  }
594  if (vring->errfd != -1)
595  {
596  close (vring->errfd);
597  vring->errfd = -1;
598  }
599  vhost_user_vring_init (vui, qid);
600 }
601 
602 static inline void
604 {
605  vnet_main_t *vnm = vnet_get_main ();
606  int q;
607 
609 
610  if (vui->clib_file_index != ~0)
611  {
613  vui->clib_file_index = ~0;
614  }
615 
616  vui->is_up = 0;
617 
618  for (q = 0; q < VHOST_VRING_MAX_N; q++)
619  vhost_user_vring_close (vui, q);
620 
621  unmap_all_mem_regions (vui);
622  DBG_SOCK ("interface ifindex %d disconnected", vui->sw_if_index);
623 }
624 
625 #define VHOST_LOG_PAGE 0x1000
628  u64 addr, u64 len, u8 is_host_address)
629 {
630  if (PREDICT_TRUE (vui->log_base_addr == 0
631  || !(vui->features & (1 << FEAT_VHOST_F_LOG_ALL))))
632  {
633  return;
634  }
635  if (is_host_address)
636  {
637  addr = pointer_to_uword (map_user_mem (vui, (uword) addr));
638  }
639  if (PREDICT_FALSE ((addr + len - 1) / VHOST_LOG_PAGE / 8 >= vui->log_size))
640  {
641  DBG_SOCK ("vhost_user_log_dirty_pages(): out of range\n");
642  return;
643  }
644 
646  u64 page = addr / VHOST_LOG_PAGE;
647  while (page * VHOST_LOG_PAGE < addr + len)
648  {
649  ((u8 *) vui->log_base_addr)[page / 8] |= 1 << page % 8;
650  page++;
651  }
652 }
653 
656 {
657  vhost_user_log_dirty_pages_2 (vui, addr, len, 0);
658 }
659 
660 #define vhost_user_log_dirty_ring(vui, vq, member) \
661  if (PREDICT_FALSE(vq->log_used)) { \
662  vhost_user_log_dirty_pages(vui, vq->log_guest_addr + STRUCT_OFFSET_OF(vring_used_t, member), \
663  sizeof(vq->used->member)); \
664  }
665 
666 static clib_error_t *
668 {
669  int n, i;
670  int fd, number_of_fds = 0;
671  int fds[VHOST_MEMORY_MAX_NREGIONS];
672  vhost_user_msg_t msg;
673  struct msghdr mh;
674  struct iovec iov[1];
676  vhost_user_intf_t *vui;
677  struct cmsghdr *cmsg;
678  u8 q;
679  clib_file_t template = { 0 };
680  vnet_main_t *vnm = vnet_get_main ();
681 
682  vui = pool_elt_at_index (vum->vhost_user_interfaces, uf->private_data);
683 
684  char control[CMSG_SPACE (VHOST_MEMORY_MAX_NREGIONS * sizeof (int))];
685 
686  memset (&mh, 0, sizeof (mh));
687  memset (control, 0, sizeof (control));
688 
689  for (i = 0; i < VHOST_MEMORY_MAX_NREGIONS; i++)
690  fds[i] = -1;
691 
692  /* set the payload */
693  iov[0].iov_base = (void *) &msg;
694  iov[0].iov_len = VHOST_USER_MSG_HDR_SZ;
695 
696  mh.msg_iov = iov;
697  mh.msg_iovlen = 1;
698  mh.msg_control = control;
699  mh.msg_controllen = sizeof (control);
700 
701  n = recvmsg (uf->file_descriptor, &mh, 0);
702 
703  /* Stop workers to avoid end of the world */
705 
706  if (n != VHOST_USER_MSG_HDR_SZ)
707  {
708  if (n == -1)
709  {
710  DBG_SOCK ("recvmsg returned error %d %s", errno, strerror (errno));
711  }
712  else
713  {
714  DBG_SOCK ("n (%d) != VHOST_USER_MSG_HDR_SZ (%d)",
716  }
717  goto close_socket;
718  }
719 
720  if (mh.msg_flags & MSG_CTRUNC)
721  {
722  DBG_SOCK ("MSG_CTRUNC is set");
723  goto close_socket;
724  }
725 
726  cmsg = CMSG_FIRSTHDR (&mh);
727 
728  if (cmsg && (cmsg->cmsg_len > 0) && (cmsg->cmsg_level == SOL_SOCKET) &&
729  (cmsg->cmsg_type == SCM_RIGHTS) &&
730  (cmsg->cmsg_len - CMSG_LEN (0) <=
731  VHOST_MEMORY_MAX_NREGIONS * sizeof (int)))
732  {
733  number_of_fds = (cmsg->cmsg_len - CMSG_LEN (0)) / sizeof (int);
734  clib_memcpy (fds, CMSG_DATA (cmsg), number_of_fds * sizeof (int));
735  }
736 
737  /* version 1, no reply bit set */
738  if ((msg.flags & 7) != 1)
739  {
740  DBG_SOCK ("malformed message received. closing socket");
741  goto close_socket;
742  }
743 
744  {
745  int rv;
746  rv =
747  read (uf->file_descriptor, ((char *) &msg) + VHOST_USER_MSG_HDR_SZ,
748  msg.size);
749  if (rv < 0)
750  {
751  DBG_SOCK ("read failed %s", strerror (errno));
752  goto close_socket;
753  }
754  else if (rv != msg.size)
755  {
756  DBG_SOCK ("message too short (read %dB should be %dB)", rv, msg.size);
757  goto close_socket;
758  }
759  }
760 
761  switch (msg.request)
762  {
764  msg.flags |= 4;
765  msg.u64 = (1ULL << FEAT_VIRTIO_NET_F_MRG_RXBUF) |
766  (1ULL << FEAT_VIRTIO_NET_F_CTRL_VQ) |
767  (1ULL << FEAT_VIRTIO_F_ANY_LAYOUT) |
768  (1ULL << FEAT_VIRTIO_F_INDIRECT_DESC) |
769  (1ULL << FEAT_VHOST_F_LOG_ALL) |
770  (1ULL << FEAT_VIRTIO_NET_F_GUEST_ANNOUNCE) |
771  (1ULL << FEAT_VIRTIO_NET_F_MQ) |
772  (1ULL << FEAT_VHOST_USER_F_PROTOCOL_FEATURES) |
773  (1ULL << FEAT_VIRTIO_F_VERSION_1);
774  msg.u64 &= vui->feature_mask;
775  msg.size = sizeof (msg.u64);
776  DBG_SOCK ("if %d msg VHOST_USER_GET_FEATURES - reply 0x%016llx",
777  vui->hw_if_index, msg.u64);
778  break;
779 
781  DBG_SOCK ("if %d msg VHOST_USER_SET_FEATURES features 0x%016llx",
782  vui->hw_if_index, msg.u64);
783 
784  vui->features = msg.u64;
785 
786  if (vui->features &
787  ((1 << FEAT_VIRTIO_NET_F_MRG_RXBUF) |
788  (1ULL << FEAT_VIRTIO_F_VERSION_1)))
789  vui->virtio_net_hdr_sz = 12;
790  else
791  vui->virtio_net_hdr_sz = 10;
792 
793  vui->is_any_layout =
794  (vui->features & (1 << FEAT_VIRTIO_F_ANY_LAYOUT)) ? 1 : 0;
795 
798  vui->is_up = 0;
799 
800  /*for (q = 0; q < VHOST_VRING_MAX_N; q++)
801  vhost_user_vring_close(&vui->vrings[q]); */
802 
803  break;
804 
806  DBG_SOCK ("if %d msg VHOST_USER_SET_MEM_TABLE nregions %d",
807  vui->hw_if_index, msg.memory.nregions);
808 
809  if ((msg.memory.nregions < 1) ||
810  (msg.memory.nregions > VHOST_MEMORY_MAX_NREGIONS))
811  {
812 
813  DBG_SOCK ("number of mem regions must be between 1 and %i",
814  VHOST_MEMORY_MAX_NREGIONS);
815 
816  goto close_socket;
817  }
818 
819  if (msg.memory.nregions != number_of_fds)
820  {
821  DBG_SOCK ("each memory region must have FD");
822  goto close_socket;
823  }
824  unmap_all_mem_regions (vui);
825  for (i = 0; i < msg.memory.nregions; i++)
826  {
827  clib_memcpy (&(vui->regions[i]), &msg.memory.regions[i],
828  sizeof (vhost_user_memory_region_t));
829 
830  long page_sz = get_huge_page_size (fds[i]);
831 
832  /* align size to 2M page */
833  ssize_t map_sz = (vui->regions[i].memory_size +
834  vui->regions[i].mmap_offset +
835  page_sz - 1) & ~(page_sz - 1);
836 
837  vui->region_mmap_addr[i] = mmap (0, map_sz, PROT_READ | PROT_WRITE,
838  MAP_SHARED, fds[i], 0);
839  vui->region_guest_addr_lo[i] = vui->regions[i].guest_phys_addr;
840  vui->region_guest_addr_hi[i] = vui->regions[i].guest_phys_addr +
841  vui->regions[i].memory_size;
842 
843  DBG_SOCK
844  ("map memory region %d addr 0 len 0x%lx fd %d mapped 0x%lx "
845  "page_sz 0x%x", i, map_sz, fds[i], vui->region_mmap_addr[i],
846  page_sz);
847 
848  if (vui->region_mmap_addr[i] == MAP_FAILED)
849  {
850  clib_warning ("failed to map memory. errno is %d", errno);
851  goto close_socket;
852  }
853  vui->region_mmap_addr[i] += vui->regions[i].mmap_offset;
854  vui->region_mmap_fd[i] = fds[i];
855  }
856  vui->nregions = msg.memory.nregions;
857  break;
858 
860  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_NUM idx %d num %d",
861  vui->hw_if_index, msg.state.index, msg.state.num);
862 
863  if ((msg.state.num > 32768) || /* maximum ring size is 32768 */
864  (msg.state.num == 0) || /* it cannot be zero */
865  ((msg.state.num - 1) & msg.state.num)) /* must be power of 2 */
866  goto close_socket;
867  vui->vrings[msg.state.index].qsz_mask = msg.state.num - 1;
868  break;
869 
871  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_ADDR idx %d",
872  vui->hw_if_index, msg.state.index);
873 
874  if (msg.state.index >= VHOST_VRING_MAX_N)
875  {
876  DBG_SOCK ("invalid vring index VHOST_USER_SET_VRING_ADDR:"
877  " %d >= %d", msg.state.index, VHOST_VRING_MAX_N);
878  goto close_socket;
879  }
880 
881  if (msg.size < sizeof (msg.addr))
882  {
883  DBG_SOCK ("vhost message is too short (%d < %d)",
884  msg.size, sizeof (msg.addr));
885  goto close_socket;
886  }
887 
888  vui->vrings[msg.state.index].desc = (vring_desc_t *)
889  map_user_mem (vui, msg.addr.desc_user_addr);
890  vui->vrings[msg.state.index].used = (vring_used_t *)
891  map_user_mem (vui, msg.addr.used_user_addr);
892  vui->vrings[msg.state.index].avail = (vring_avail_t *)
893  map_user_mem (vui, msg.addr.avail_user_addr);
894 
895  if ((vui->vrings[msg.state.index].desc == NULL) ||
896  (vui->vrings[msg.state.index].used == NULL) ||
897  (vui->vrings[msg.state.index].avail == NULL))
898  {
899  DBG_SOCK ("failed to map user memory for hw_if_index %d",
900  vui->hw_if_index);
901  goto close_socket;
902  }
903 
904  vui->vrings[msg.state.index].log_guest_addr = msg.addr.log_guest_addr;
905  vui->vrings[msg.state.index].log_used =
906  (msg.addr.flags & (1 << VHOST_VRING_F_LOG)) ? 1 : 0;
907 
908  /* Spec says: If VHOST_USER_F_PROTOCOL_FEATURES has not been negotiated,
909  the ring is initialized in an enabled state. */
910  if (!(vui->features & (1 << FEAT_VHOST_USER_F_PROTOCOL_FEATURES)))
911  {
912  vui->vrings[msg.state.index].enabled = 1;
913  }
914 
915  vui->vrings[msg.state.index].last_used_idx =
916  vui->vrings[msg.state.index].last_avail_idx =
917  vui->vrings[msg.state.index].used->idx;
918 
919  /* tell driver that we don't want interrupts */
920  vui->vrings[msg.state.index].used->flags = VRING_USED_F_NO_NOTIFY;
921  break;
922 
924  DBG_SOCK ("if %d msg VHOST_USER_SET_OWNER", vui->hw_if_index);
925  break;
926 
928  DBG_SOCK ("if %d msg VHOST_USER_RESET_OWNER", vui->hw_if_index);
929  break;
930 
932  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_CALL %d",
933  vui->hw_if_index, msg.u64);
934 
935  q = (u8) (msg.u64 & 0xFF);
936 
937  /* if there is old fd, delete and close it */
938  if (vui->vrings[q].callfd_idx != ~0)
939  {
941  vui->vrings[q].callfd_idx);
942  clib_file_del (&file_main, uf);
943  vui->vrings[q].callfd_idx = ~0;
944  }
945 
946  if (!(msg.u64 & VHOST_USER_VRING_NOFD_MASK))
947  {
948  if (number_of_fds != 1)
949  {
950  DBG_SOCK ("More than one fd received !");
951  goto close_socket;
952  }
953 
954  template.read_function = vhost_user_callfd_read_ready;
955  template.file_descriptor = fds[0];
956  template.private_data =
957  ((vui - vhost_user_main.vhost_user_interfaces) << 8) + q;
958  vui->vrings[q].callfd_idx = clib_file_add (&file_main, &template);
959  }
960  else
961  vui->vrings[q].callfd_idx = ~0;
962  break;
963 
965  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_KICK %d",
966  vui->hw_if_index, msg.u64);
967 
968  q = (u8) (msg.u64 & 0xFF);
969 
970  if (vui->vrings[q].kickfd_idx != ~0)
971  {
973  vui->vrings[q].kickfd_idx);
974  clib_file_del (&file_main, uf);
975  vui->vrings[q].kickfd_idx = ~0;
976  }
977 
978  if (!(msg.u64 & VHOST_USER_VRING_NOFD_MASK))
979  {
980  if (number_of_fds != 1)
981  {
982  DBG_SOCK ("More than one fd received !");
983  goto close_socket;
984  }
985 
986  template.read_function = vhost_user_kickfd_read_ready;
987  template.file_descriptor = fds[0];
988  template.private_data =
989  (((uword) (vui - vhost_user_main.vhost_user_interfaces)) << 8) +
990  q;
991  vui->vrings[q].kickfd_idx = clib_file_add (&file_main, &template);
992  }
993  else
994  {
995  //When no kickfd is set, the queue is initialized as started
996  vui->vrings[q].kickfd_idx = ~0;
997  vui->vrings[q].started = 1;
998  }
999 
1000  break;
1001 
1003  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_ERR %d",
1004  vui->hw_if_index, msg.u64);
1005 
1006  q = (u8) (msg.u64 & 0xFF);
1007 
1008  if (vui->vrings[q].errfd != -1)
1009  close (vui->vrings[q].errfd);
1010 
1011  if (!(msg.u64 & VHOST_USER_VRING_NOFD_MASK))
1012  {
1013  if (number_of_fds != 1)
1014  goto close_socket;
1015 
1016  vui->vrings[q].errfd = fds[0];
1017  }
1018  else
1019  vui->vrings[q].errfd = -1;
1020 
1021  break;
1022 
1024  DBG_SOCK ("if %d msg VHOST_USER_SET_VRING_BASE idx %d num %d",
1025  vui->hw_if_index, msg.state.index, msg.state.num);
1026 
1027  vui->vrings[msg.state.index].last_avail_idx = msg.state.num;
1028  break;
1029 
1031  if (msg.state.index >= VHOST_VRING_MAX_N)
1032  {
1033  DBG_SOCK ("invalid vring index VHOST_USER_GET_VRING_BASE:"
1034  " %d >= %d", msg.state.index, VHOST_VRING_MAX_N);
1035  goto close_socket;
1036  }
1037 
1038  /*
1039  * Copy last_avail_idx from the vring before closing it because
1040  * closing the vring also initializes the vring last_avail_idx
1041  */
1042  msg.state.num = vui->vrings[msg.state.index].last_avail_idx;
1043  msg.flags |= 4;
1044  msg.size = sizeof (msg.state);
1045 
1046  /* Spec says: Client must [...] stop ring upon receiving VHOST_USER_GET_VRING_BASE. */
1047  vhost_user_vring_close (vui, msg.state.index);
1048  DBG_SOCK ("if %d msg VHOST_USER_GET_VRING_BASE idx %d num %d",
1049  vui->hw_if_index, msg.state.index, msg.state.num);
1050  break;
1051 
1052  case VHOST_USER_NONE:
1053  DBG_SOCK ("if %d msg VHOST_USER_NONE", vui->hw_if_index);
1054 
1055  break;
1056 
1058  {
1059  DBG_SOCK ("if %d msg VHOST_USER_SET_LOG_BASE", vui->hw_if_index);
1060 
1061  if (msg.size != sizeof (msg.log))
1062  {
1063  DBG_SOCK
1064  ("invalid msg size for VHOST_USER_SET_LOG_BASE: %d instead of %d",
1065  msg.size, sizeof (msg.log));
1066  goto close_socket;
1067  }
1068 
1069  if (!
1071  {
1072  DBG_SOCK
1073  ("VHOST_USER_PROTOCOL_F_LOG_SHMFD not set but VHOST_USER_SET_LOG_BASE received");
1074  goto close_socket;
1075  }
1076 
1077  fd = fds[0];
1078  /* align size to 2M page */
1079  long page_sz = get_huge_page_size (fd);
1080  ssize_t map_sz =
1081  (msg.log.size + msg.log.offset + page_sz - 1) & ~(page_sz - 1);
1082 
1083  vui->log_base_addr = mmap (0, map_sz, PROT_READ | PROT_WRITE,
1084  MAP_SHARED, fd, 0);
1085 
1086  DBG_SOCK
1087  ("map log region addr 0 len 0x%lx off 0x%lx fd %d mapped 0x%lx",
1088  map_sz, msg.log.offset, fd, vui->log_base_addr);
1089 
1090  if (vui->log_base_addr == MAP_FAILED)
1091  {
1092  clib_warning ("failed to map memory. errno is %d", errno);
1093  goto close_socket;
1094  }
1095 
1096  vui->log_base_addr += msg.log.offset;
1097  vui->log_size = msg.log.size;
1098 
1099  msg.flags |= 4;
1100  msg.size = sizeof (msg.u64);
1101 
1102  break;
1103  }
1104 
1105  case VHOST_USER_SET_LOG_FD:
1106  DBG_SOCK ("if %d msg VHOST_USER_SET_LOG_FD", vui->hw_if_index);
1107 
1108  break;
1109 
1111  msg.flags |= 4;
1112  msg.u64 = (1 << VHOST_USER_PROTOCOL_F_LOG_SHMFD) |
1113  (1 << VHOST_USER_PROTOCOL_F_MQ);
1114  msg.size = sizeof (msg.u64);
1115  DBG_SOCK
1116  ("if %d msg VHOST_USER_GET_PROTOCOL_FEATURES - reply 0x%016llx",
1117  vui->hw_if_index, msg.u64);
1118  break;
1119 
1121  DBG_SOCK
1122  ("if %d msg VHOST_USER_SET_PROTOCOL_FEATURES features 0x%016llx",
1123  vui->hw_if_index, msg.u64);
1124 
1125  vui->protocol_features = msg.u64;
1126 
1127  break;
1128 
1130  msg.flags |= 4;
1131  msg.u64 = VHOST_VRING_MAX_N;
1132  msg.size = sizeof (msg.u64);
1133  DBG_SOCK ("if %d msg VHOST_USER_GET_QUEUE_NUM - reply %d",
1134  vui->hw_if_index, msg.u64);
1135  break;
1136 
1138  DBG_SOCK ("if %d VHOST_USER_SET_VRING_ENABLE: %s queue %d",
1139  vui->hw_if_index, msg.state.num ? "enable" : "disable",
1140  msg.state.index);
1141  if (msg.state.index >= VHOST_VRING_MAX_N)
1142  {
1143  DBG_SOCK ("invalid vring index VHOST_USER_SET_VRING_ENABLE:"
1144  " %d >= %d", msg.state.index, VHOST_VRING_MAX_N);
1145  goto close_socket;
1146  }
1147 
1148  vui->vrings[msg.state.index].enabled = msg.state.num;
1149  break;
1150 
1151  default:
1152  DBG_SOCK ("unknown vhost-user message %d received. closing socket",
1153  msg.request);
1154  goto close_socket;
1155  }
1156 
1157  /* if we need to reply */
1158  if (msg.flags & 4)
1159  {
1160  n =
1161  send (uf->file_descriptor, &msg, VHOST_USER_MSG_HDR_SZ + msg.size, 0);
1162  if (n != (msg.size + VHOST_USER_MSG_HDR_SZ))
1163  {
1164  DBG_SOCK ("could not send message response");
1165  goto close_socket;
1166  }
1167  }
1168 
1171  return 0;
1172 
1173 close_socket:
1177  return 0;
1178 }
1179 
1180 static clib_error_t *
1182 {
1185  vhost_user_intf_t *vui =
1187 
1188  DBG_SOCK ("socket error on if %d", vui->sw_if_index);
1193  return 0;
1194 }
1195 
1196 static clib_error_t *
1198 {
1199  int client_fd, client_len;
1200  struct sockaddr_un client;
1201  clib_file_t template = { 0 };
1203  vhost_user_intf_t *vui;
1204 
1206 
1207  client_len = sizeof (client);
1208  client_fd = accept (uf->file_descriptor,
1209  (struct sockaddr *) &client,
1210  (socklen_t *) & client_len);
1211 
1212  if (client_fd < 0)
1213  return clib_error_return_unix (0, "accept");
1214 
1215  DBG_SOCK ("New client socket for vhost interface %d", vui->sw_if_index);
1216  template.read_function = vhost_user_socket_read;
1217  template.error_function = vhost_user_socket_error;
1218  template.file_descriptor = client_fd;
1219  template.private_data = vui - vhost_user_main.vhost_user_interfaces;
1220  vui->clib_file_index = clib_file_add (&file_main, &template);
1221  return 0;
1222 }
1223 
1224 static clib_error_t *
1226 {
1227  clib_error_t *error;
1230 
1231  error = vlib_call_init_function (vm, ip4_init);
1232  if (error)
1233  return error;
1234 
1235  vum->coalesce_frames = 32;
1236  vum->coalesce_time = 1e-3;
1237 
1238  vec_validate (vum->cpus, tm->n_vlib_mains - 1);
1239 
1240  vhost_cpu_t *cpu;
1241  vec_foreach (cpu, vum->cpus)
1242  {
1243  /* This is actually not necessary as validate already zeroes it
1244  * Just keeping the loop here for later because I am lazy. */
1245  cpu->rx_buffers_len = 0;
1246  }
1247 
1248  vum->random = random_default_seed ();
1249 
1251 
1252  return 0;
1253 }
1254 
1256 
1257 static u8 *
1258 format_vhost_trace (u8 * s, va_list * va)
1259 {
1260  CLIB_UNUSED (vlib_main_t * vm) = va_arg (*va, vlib_main_t *);
1261  CLIB_UNUSED (vlib_node_t * node) = va_arg (*va, vlib_node_t *);
1262  CLIB_UNUSED (vnet_main_t * vnm) = vnet_get_main ();
1264  vhost_trace_t *t = va_arg (*va, vhost_trace_t *);
1266  t->device_index);
1267 
1269 
1270  uword indent = format_get_indent (s);
1271 
1272  s = format (s, "%U %U queue %d\n", format_white_space, indent,
1273  format_vnet_sw_interface_name, vnm, sw, t->qid);
1274 
1275  s = format (s, "%U virtio flags:\n", format_white_space, indent);
1276 #define _(n,i,st) \
1277  if (t->virtio_ring_flags & (1 << VIRTIO_TRACE_F_##n)) \
1278  s = format (s, "%U %s %s\n", format_white_space, indent, #n, st);
1280 #undef _
1281  s = format (s, "%U virtio_net_hdr first_desc_len %u\n",
1282  format_white_space, indent, t->first_desc_len);
1283 
1284  s = format (s, "%U flags 0x%02x gso_type %u\n",
1285  format_white_space, indent,
1286  t->hdr.hdr.flags, t->hdr.hdr.gso_type);
1287 
1288  if (vui->virtio_net_hdr_sz == 12)
1289  s = format (s, "%U num_buff %u",
1290  format_white_space, indent, t->hdr.num_buffers);
1291 
1292  return s;
1293 }
1294 
1295 void
1297  vhost_user_intf_t * vui, u16 qid,
1298  vlib_buffer_t * b, vhost_user_vring_t * txvq)
1299 {
1301  u32 last_avail_idx = txvq->last_avail_idx;
1302  u32 desc_current = txvq->avail->ring[last_avail_idx & txvq->qsz_mask];
1303  vring_desc_t *hdr_desc = 0;
1304  virtio_net_hdr_mrg_rxbuf_t *hdr;
1305  u32 hint = 0;
1306 
1307  memset (t, 0, sizeof (*t));
1308  t->device_index = vui - vum->vhost_user_interfaces;
1309  t->qid = qid;
1310 
1311  hdr_desc = &txvq->desc[desc_current];
1312  if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
1313  {
1314  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_INDIRECT;
1315  /* Header is the first here */
1316  hdr_desc = map_guest_mem (vui, txvq->desc[desc_current].addr, &hint);
1317  }
1318  if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT)
1319  {
1320  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SIMPLE_CHAINED;
1321  }
1322  if (!(txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT) &&
1323  !(txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT))
1324  {
1325  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SINGLE_DESC;
1326  }
1327 
1328  t->first_desc_len = hdr_desc ? hdr_desc->len : 0;
1329 
1330  if (!hdr_desc || !(hdr = map_guest_mem (vui, hdr_desc->addr, &hint)))
1331  {
1332  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_MAP_ERROR;
1333  }
1334  else
1335  {
1336  u32 len = vui->virtio_net_hdr_sz;
1337  memcpy (&t->hdr, hdr, len > hdr_desc->len ? hdr_desc->len : len);
1338  }
1339 }
1340 
1341 static inline void
1343 {
1345  u64 x = 1;
1346  int fd = UNIX_GET_FD (vq->callfd_idx);
1347  int rv;
1348 
1349  rv = write (fd, &x, sizeof (x));
1350  if (rv <= 0)
1351  {
1353  ("Error: Could not write to unix socket for callfd %d", fd);
1354  return;
1355  }
1356 
1357  vq->n_since_last_int = 0;
1358  vq->int_deadline = vlib_time_now (vm) + vum->coalesce_time;
1359 }
1360 
1363  u16 copy_len, u32 * map_hint)
1364 {
1365  void *src0, *src1, *src2, *src3;
1366  if (PREDICT_TRUE (copy_len >= 4))
1367  {
1368  if (PREDICT_FALSE (!(src2 = map_guest_mem (vui, cpy[0].src, map_hint))))
1369  return 1;
1370  if (PREDICT_FALSE (!(src3 = map_guest_mem (vui, cpy[1].src, map_hint))))
1371  return 1;
1372 
1373  while (PREDICT_TRUE (copy_len >= 4))
1374  {
1375  src0 = src2;
1376  src1 = src3;
1377 
1378  if (PREDICT_FALSE
1379  (!(src2 = map_guest_mem (vui, cpy[2].src, map_hint))))
1380  return 1;
1381  if (PREDICT_FALSE
1382  (!(src3 = map_guest_mem (vui, cpy[3].src, map_hint))))
1383  return 1;
1384 
1385  CLIB_PREFETCH (src2, 64, LOAD);
1386  CLIB_PREFETCH (src3, 64, LOAD);
1387 
1388  clib_memcpy ((void *) cpy[0].dst, src0, cpy[0].len);
1389  clib_memcpy ((void *) cpy[1].dst, src1, cpy[1].len);
1390  copy_len -= 2;
1391  cpy += 2;
1392  }
1393  }
1394  while (copy_len)
1395  {
1396  if (PREDICT_FALSE (!(src0 = map_guest_mem (vui, cpy->src, map_hint))))
1397  return 1;
1398  clib_memcpy ((void *) cpy->dst, src0, cpy->len);
1399  copy_len -= 1;
1400  cpy += 1;
1401  }
1402  return 0;
1403 }
1404 
1405 /**
1406  * Try to discard packets from the tx ring (VPP RX path).
1407  * Returns the number of discarded packets.
1408  */
1409 u32
1411  vhost_user_intf_t * vui,
1412  vhost_user_vring_t * txvq, u32 discard_max)
1413 {
1414  /*
1415  * On the RX side, each packet corresponds to one descriptor
1416  * (it is the same whether it is a shallow descriptor, chained, or indirect).
1417  * Therefore, discarding a packet is like discarding a descriptor.
1418  */
1419  u32 discarded_packets = 0;
1420  u32 avail_idx = txvq->avail->idx;
1421  while (discarded_packets != discard_max)
1422  {
1423  if (avail_idx == txvq->last_avail_idx)
1424  goto out;
1425 
1426  u16 desc_chain_head =
1427  txvq->avail->ring[txvq->last_avail_idx & txvq->qsz_mask];
1428  txvq->last_avail_idx++;
1429  txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].id =
1430  desc_chain_head;
1431  txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].len = 0;
1432  vhost_user_log_dirty_ring (vui, txvq,
1433  ring[txvq->last_used_idx & txvq->qsz_mask]);
1434  txvq->last_used_idx++;
1435  discarded_packets++;
1436  }
1437 
1438 out:
1440  txvq->used->idx = txvq->last_used_idx;
1441  vhost_user_log_dirty_ring (vui, txvq, idx);
1442  return discarded_packets;
1443 }
1444 
1445 /*
1446  * In case of overflow, we need to rewind the array of allocated buffers.
1447  */
1448 static void
1450  vhost_cpu_t * cpu, vlib_buffer_t * b_head)
1451 {
1452  u32 bi_current = cpu->rx_buffers[cpu->rx_buffers_len];
1453  vlib_buffer_t *b_current = vlib_get_buffer (vm, bi_current);
1454  b_current->current_length = 0;
1455  b_current->flags = 0;
1456  while (b_current != b_head)
1457  {
1458  cpu->rx_buffers_len++;
1459  bi_current = cpu->rx_buffers[cpu->rx_buffers_len];
1460  b_current = vlib_get_buffer (vm, bi_current);
1461  b_current->current_length = 0;
1462  b_current->flags = 0;
1463  }
1464  cpu->rx_buffers_len++;
1465 }
1466 
1467 static u32
1469  vhost_user_main_t * vum,
1470  vhost_user_intf_t * vui,
1471  u16 qid, vlib_node_runtime_t * node,
1473 {
1474  vhost_user_vring_t *txvq = &vui->vrings[VHOST_VRING_IDX_TX (qid)];
1475  u16 n_rx_packets = 0;
1476  u32 n_rx_bytes = 0;
1477  u16 n_left;
1478  u32 n_left_to_next, *to_next;
1480  u32 n_trace = vlib_get_trace_count (vm, node);
1481  u32 map_hint = 0;
1482  u16 thread_index = vlib_get_thread_index ();
1483  u16 copy_len = 0;
1484 
1485  {
1486  /* do we have pending interrupts ? */
1487  vhost_user_vring_t *rxvq = &vui->vrings[VHOST_VRING_IDX_RX (qid)];
1488  f64 now = vlib_time_now (vm);
1489 
1490  if ((txvq->n_since_last_int) && (txvq->int_deadline < now))
1491  vhost_user_send_call (vm, txvq);
1492 
1493  if ((rxvq->n_since_last_int) && (rxvq->int_deadline < now))
1494  vhost_user_send_call (vm, rxvq);
1495  }
1496 
1497  /*
1498  * For adaptive mode, it is optimized to reduce interrupts.
1499  * If the scheduler switches the input node to polling due
1500  * to burst of traffic, we tell the driver no interrupt.
1501  * When the traffic subsides, the scheduler switches the node back to
1502  * interrupt mode. We must tell the driver we want interrupt.
1503  */
1505  {
1506  if ((node->flags &
1508  !(node->flags &
1510  /* Tell driver we want notification */
1511  txvq->used->flags = 0;
1512  else
1513  /* Tell driver we don't want notification */
1514  txvq->used->flags = VRING_USED_F_NO_NOTIFY;
1515  }
1516 
1517  if (PREDICT_FALSE (txvq->avail->flags & 0xFFFE))
1518  return 0;
1519 
1520  n_left = (u16) (txvq->avail->idx - txvq->last_avail_idx);
1521 
1522  /* nothing to do */
1523  if (PREDICT_FALSE (n_left == 0))
1524  return 0;
1525 
1526  if (PREDICT_FALSE (!vui->admin_up || !(txvq->enabled)))
1527  {
1528  /*
1529  * Discard input packet if interface is admin down or vring is not
1530  * enabled.
1531  * "For example, for a networking device, in the disabled state
1532  * client must not supply any new RX packets, but must process
1533  * and discard any TX packets."
1534  */
1535  vhost_user_rx_discard_packet (vm, vui, txvq,
1537  return 0;
1538  }
1539 
1540  if (PREDICT_FALSE (n_left == (txvq->qsz_mask + 1)))
1541  {
1542  /*
1543  * Informational error logging when VPP is not
1544  * receiving packets fast enough.
1545  */
1546  vlib_error_count (vm, node->node_index,
1547  VHOST_USER_INPUT_FUNC_ERROR_FULL_RX_QUEUE, 1);
1548  }
1549 
1550  if (n_left > VLIB_FRAME_SIZE)
1551  n_left = VLIB_FRAME_SIZE;
1552 
1553  /*
1554  * For small packets (<2kB), we will not need more than one vlib buffer
1555  * per packet. In case packets are bigger, we will just yeld at some point
1556  * in the loop and come back later. This is not an issue as for big packet,
1557  * processing cost really comes from the memory copy.
1558  * The assumption is that big packets will fit in 40 buffers.
1559  */
1560  if (PREDICT_FALSE (vum->cpus[thread_index].rx_buffers_len < n_left + 1 ||
1561  vum->cpus[thread_index].rx_buffers_len < 40))
1562  {
1563  u32 curr_len = vum->cpus[thread_index].rx_buffers_len;
1564  vum->cpus[thread_index].rx_buffers_len +=
1566  vum->cpus[thread_index].rx_buffers +
1567  curr_len,
1568  VHOST_USER_RX_BUFFERS_N - curr_len,
1570 
1571  if (PREDICT_FALSE
1572  (vum->cpus[thread_index].rx_buffers_len <
1574  {
1575  /* In case of buffer starvation, discard some packets from the queue
1576  * and log the event.
1577  * We keep doing best effort for the remaining packets. */
1578  u32 flush = (n_left + 1 > vum->cpus[thread_index].rx_buffers_len) ?
1579  n_left + 1 - vum->cpus[thread_index].rx_buffers_len : 1;
1580  flush = vhost_user_rx_discard_packet (vm, vui, txvq, flush);
1581 
1582  n_left -= flush;
1584  interface_main.sw_if_counters +
1587  vui->sw_if_index, flush);
1588 
1590  VHOST_USER_INPUT_FUNC_ERROR_NO_BUFFER, flush);
1591  }
1592  }
1593 
1594  while (n_left > 0)
1595  {
1596  vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
1597 
1598  while (n_left > 0 && n_left_to_next > 0)
1599  {
1600  vlib_buffer_t *b_head, *b_current;
1601  u32 bi_current;
1602  u16 desc_current;
1603  u32 desc_data_offset;
1604  vring_desc_t *desc_table = txvq->desc;
1605 
1606  if (PREDICT_FALSE (vum->cpus[thread_index].rx_buffers_len <= 1))
1607  {
1608  /* Not enough rx_buffers
1609  * Note: We yeld on 1 so we don't need to do an additional
1610  * check for the next buffer prefetch.
1611  */
1612  n_left = 0;
1613  break;
1614  }
1615 
1616  desc_current =
1617  txvq->avail->ring[txvq->last_avail_idx & txvq->qsz_mask];
1618  vum->cpus[thread_index].rx_buffers_len--;
1619  bi_current = (vum->cpus[thread_index].rx_buffers)
1620  [vum->cpus[thread_index].rx_buffers_len];
1621  b_head = b_current = vlib_get_buffer (vm, bi_current);
1622  to_next[0] = bi_current; //We do that now so we can forget about bi_current
1623  to_next++;
1624  n_left_to_next--;
1625 
1627  (vum->
1628  cpus[thread_index].rx_buffers)
1629  [vum->cpus[thread_index].
1630  rx_buffers_len - 1], LOAD);
1631 
1632  /* Just preset the used descriptor id and length for later */
1633  txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].id =
1634  desc_current;
1635  txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].len = 0;
1636  vhost_user_log_dirty_ring (vui, txvq,
1637  ring[txvq->last_used_idx &
1638  txvq->qsz_mask]);
1639 
1640  /* The buffer should already be initialized */
1643 
1644  if (PREDICT_FALSE (n_trace))
1645  {
1646  //TODO: next_index is not exactly known at that point
1647  vlib_trace_buffer (vm, node, next_index, b_head,
1648  /* follow_chain */ 0);
1649  vhost_trace_t *t0 =
1650  vlib_add_trace (vm, node, b_head, sizeof (t0[0]));
1651  vhost_user_rx_trace (t0, vui, qid, b_head, txvq);
1652  n_trace--;
1653  vlib_set_trace_count (vm, node, n_trace);
1654  }
1655 
1656  /* This depends on the setup but is very consistent
1657  * So I think the CPU branch predictor will make a pretty good job
1658  * at optimizing the decision. */
1659  if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
1660  {
1661  desc_table = map_guest_mem (vui, txvq->desc[desc_current].addr,
1662  &map_hint);
1663  desc_current = 0;
1664  if (PREDICT_FALSE (desc_table == 0))
1665  {
1666  vlib_error_count (vm, node->node_index,
1667  VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
1668  goto out;
1669  }
1670  }
1671 
1672  if (PREDICT_TRUE (vui->is_any_layout) ||
1673  (!(desc_table[desc_current].flags & VIRTQ_DESC_F_NEXT)))
1674  {
1675  /* ANYLAYOUT or single buffer */
1676  desc_data_offset = vui->virtio_net_hdr_sz;
1677  }
1678  else
1679  {
1680  /* CSR case without ANYLAYOUT, skip 1st buffer */
1681  desc_data_offset = desc_table[desc_current].len;
1682  }
1683 
1684  while (1)
1685  {
1686  /* Get more input if necessary. Or end of packet. */
1687  if (desc_data_offset == desc_table[desc_current].len)
1688  {
1689  if (PREDICT_FALSE (desc_table[desc_current].flags &
1690  VIRTQ_DESC_F_NEXT))
1691  {
1692  desc_current = desc_table[desc_current].next;
1693  desc_data_offset = 0;
1694  }
1695  else
1696  {
1697  goto out;
1698  }
1699  }
1700 
1701  /* Get more output if necessary. Or end of packet. */
1702  if (PREDICT_FALSE
1703  (b_current->current_length == VLIB_BUFFER_DATA_SIZE))
1704  {
1705  if (PREDICT_FALSE
1706  (vum->cpus[thread_index].rx_buffers_len == 0))
1707  {
1708  /* Cancel speculation */
1709  to_next--;
1710  n_left_to_next++;
1711 
1712  /*
1713  * Checking if there are some left buffers.
1714  * If not, just rewind the used buffers and stop.
1715  * Note: Scheduled copies are not cancelled. This is
1716  * not an issue as they would still be valid. Useless,
1717  * but valid.
1718  */
1720  &vum->cpus
1721  [thread_index],
1722  b_head);
1723  n_left = 0;
1724  goto stop;
1725  }
1726 
1727  /* Get next output */
1728  vum->cpus[thread_index].rx_buffers_len--;
1729  u32 bi_next =
1730  (vum->cpus[thread_index].rx_buffers)[vum->cpus
1731  [thread_index].rx_buffers_len];
1732  b_current->next_buffer = bi_next;
1733  b_current->flags |= VLIB_BUFFER_NEXT_PRESENT;
1734  bi_current = bi_next;
1735  b_current = vlib_get_buffer (vm, bi_current);
1736  }
1737 
1738  /* Prepare a copy order executed later for the data */
1739  vhost_copy_t *cpy = &vum->cpus[thread_index].copy[copy_len];
1740  copy_len++;
1741  u32 desc_data_l =
1742  desc_table[desc_current].len - desc_data_offset;
1743  cpy->len = VLIB_BUFFER_DATA_SIZE - b_current->current_length;
1744  cpy->len = (cpy->len > desc_data_l) ? desc_data_l : cpy->len;
1745  cpy->dst = (uword) (vlib_buffer_get_current (b_current) +
1746  b_current->current_length);
1747  cpy->src = desc_table[desc_current].addr + desc_data_offset;
1748 
1749  desc_data_offset += cpy->len;
1750 
1751  b_current->current_length += cpy->len;
1753  }
1754 
1755  out:
1756  CLIB_PREFETCH (&n_left, sizeof (n_left), LOAD);
1757 
1758  n_rx_bytes += b_head->total_length_not_including_first_buffer;
1759  n_rx_packets++;
1760 
1762  b_head->current_length;
1763 
1764  /* consume the descriptor and return it as used */
1765  txvq->last_avail_idx++;
1766  txvq->last_used_idx++;
1767 
1769 
1770  vnet_buffer (b_head)->sw_if_index[VLIB_RX] = vui->sw_if_index;
1771  vnet_buffer (b_head)->sw_if_index[VLIB_TX] = (u32) ~ 0;
1772  b_head->error = 0;
1773 
1774  {
1776 
1777  /* redirect if feature path enabled */
1779  b_head);
1780 
1781  u32 bi = to_next[-1]; //Cannot use to_next[-1] in the macro
1782  vlib_validate_buffer_enqueue_x1 (vm, node, next_index,
1783  to_next, n_left_to_next,
1784  bi, next0);
1785  }
1786 
1787  n_left--;
1788 
1789  /*
1790  * Although separating memory copies from virtio ring parsing
1791  * is beneficial, we can offer to perform the copies from time
1792  * to time in order to free some space in the ring.
1793  */
1794  if (PREDICT_FALSE (copy_len >= VHOST_USER_RX_COPY_THRESHOLD))
1795  {
1796  if (PREDICT_FALSE
1797  (vhost_user_input_copy (vui, vum->cpus[thread_index].copy,
1798  copy_len, &map_hint)))
1799  {
1800  vlib_error_count (vm, node->node_index,
1801  VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
1802  }
1803  copy_len = 0;
1804 
1805  /* give buffers back to driver */
1807  txvq->used->idx = txvq->last_used_idx;
1808  vhost_user_log_dirty_ring (vui, txvq, idx);
1809  }
1810  }
1811  stop:
1812  vlib_put_next_frame (vm, node, next_index, n_left_to_next);
1813  }
1814 
1815  /* Do the memory copies */
1816  if (PREDICT_FALSE
1817  (vhost_user_input_copy (vui, vum->cpus[thread_index].copy,
1818  copy_len, &map_hint)))
1819  {
1820  vlib_error_count (vm, node->node_index,
1821  VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
1822  }
1823 
1824  /* give buffers back to driver */
1826  txvq->used->idx = txvq->last_used_idx;
1827  vhost_user_log_dirty_ring (vui, txvq, idx);
1828 
1829  /* interrupt (call) handling */
1830  if ((txvq->callfd_idx != ~0) &&
1831  !(txvq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
1832  {
1833  txvq->n_since_last_int += n_rx_packets;
1834 
1835  if (txvq->n_since_last_int > vum->coalesce_frames)
1836  vhost_user_send_call (vm, txvq);
1837  }
1838 
1839  /* increase rx counters */
1843  vlib_get_thread_index (), vui->sw_if_index, n_rx_packets, n_rx_bytes);
1844 
1845  vnet_device_increment_rx_packets (thread_index, n_rx_packets);
1846 
1847  return n_rx_packets;
1848 }
1849 
1850 static uword
1852  vlib_node_runtime_t * node, vlib_frame_t * f)
1853 {
1855  uword n_rx_packets = 0;
1856  vhost_user_intf_t *vui;
1860 
1862  {
1863  if (clib_smp_swap (&dq->interrupt_pending, 0) ||
1864  (node->state == VLIB_NODE_STATE_POLLING))
1865  {
1866  vui =
1867  pool_elt_at_index (vum->vhost_user_interfaces, dq->dev_instance);
1868  n_rx_packets = vhost_user_if_input (vm, vum, vui, dq->queue_id, node,
1869  dq->mode);
1870  }
1871  }
1872 
1873  return n_rx_packets;
1874 }
1875 
1876 /* *INDENT-OFF* */
1878  .function = vhost_user_input,
1879  .type = VLIB_NODE_TYPE_INPUT,
1880  .name = "vhost-user-input",
1881  .sibling_of = "device-input",
1882 
1883  /* Will be enabled if/when hardware is detected. */
1884  .state = VLIB_NODE_STATE_DISABLED,
1885 
1886  .format_buffer = format_ethernet_header_with_length,
1887  .format_trace = format_vhost_trace,
1888 
1889  .n_errors = VHOST_USER_INPUT_FUNC_N_ERROR,
1890  .error_strings = vhost_user_input_func_error_strings,
1891 };
1892 
1894 /* *INDENT-ON* */
1895 
1896 
1897 void
1899  vhost_user_intf_t * vui, u16 qid,
1900  vlib_buffer_t * b, vhost_user_vring_t * rxvq)
1901 {
1903  u32 last_avail_idx = rxvq->last_avail_idx;
1904  u32 desc_current = rxvq->avail->ring[last_avail_idx & rxvq->qsz_mask];
1905  vring_desc_t *hdr_desc = 0;
1906  u32 hint = 0;
1907 
1908  memset (t, 0, sizeof (*t));
1909  t->device_index = vui - vum->vhost_user_interfaces;
1910  t->qid = qid;
1911 
1912  hdr_desc = &rxvq->desc[desc_current];
1913  if (rxvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
1914  {
1915  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_INDIRECT;
1916  /* Header is the first here */
1917  hdr_desc = map_guest_mem (vui, rxvq->desc[desc_current].addr, &hint);
1918  }
1919  if (rxvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT)
1920  {
1921  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SIMPLE_CHAINED;
1922  }
1923  if (!(rxvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT) &&
1924  !(rxvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT))
1925  {
1926  t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SINGLE_DESC;
1927  }
1928 
1929  t->first_desc_len = hdr_desc ? hdr_desc->len : 0;
1930 }
1931 
1934  u16 copy_len, u32 * map_hint)
1935 {
1936  void *dst0, *dst1, *dst2, *dst3;
1937  if (PREDICT_TRUE (copy_len >= 4))
1938  {
1939  if (PREDICT_FALSE (!(dst2 = map_guest_mem (vui, cpy[0].dst, map_hint))))
1940  return 1;
1941  if (PREDICT_FALSE (!(dst3 = map_guest_mem (vui, cpy[1].dst, map_hint))))
1942  return 1;
1943  while (PREDICT_TRUE (copy_len >= 4))
1944  {
1945  dst0 = dst2;
1946  dst1 = dst3;
1947 
1948  if (PREDICT_FALSE
1949  (!(dst2 = map_guest_mem (vui, cpy[2].dst, map_hint))))
1950  return 1;
1951  if (PREDICT_FALSE
1952  (!(dst3 = map_guest_mem (vui, cpy[3].dst, map_hint))))
1953  return 1;
1954 
1955  CLIB_PREFETCH ((void *) cpy[2].src, 64, LOAD);
1956  CLIB_PREFETCH ((void *) cpy[3].src, 64, LOAD);
1957 
1958  clib_memcpy (dst0, (void *) cpy[0].src, cpy[0].len);
1959  clib_memcpy (dst1, (void *) cpy[1].src, cpy[1].len);
1960 
1961  vhost_user_log_dirty_pages_2 (vui, cpy[0].dst, cpy[0].len, 1);
1962  vhost_user_log_dirty_pages_2 (vui, cpy[1].dst, cpy[1].len, 1);
1963  copy_len -= 2;
1964  cpy += 2;
1965  }
1966  }
1967  while (copy_len)
1968  {
1969  if (PREDICT_FALSE (!(dst0 = map_guest_mem (vui, cpy->dst, map_hint))))
1970  return 1;
1971  clib_memcpy (dst0, (void *) cpy->src, cpy->len);
1972  vhost_user_log_dirty_pages_2 (vui, cpy->dst, cpy->len, 1);
1973  copy_len -= 1;
1974  cpy += 1;
1975  }
1976  return 0;
1977 }
1978 
1979 
1980 static uword
1982  vlib_node_runtime_t * node, vlib_frame_t * frame)
1983 {
1984  u32 *buffers = vlib_frame_args (frame);
1985  u32 n_left = frame->n_vectors;
1987  vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
1988  vhost_user_intf_t *vui =
1990  u32 qid = ~0;
1991  vhost_user_vring_t *rxvq;
1992  u8 error;
1993  u32 thread_index = vlib_get_thread_index ();
1994  u32 map_hint = 0;
1995  u8 retry = 8;
1996  u16 copy_len;
1997  u16 tx_headers_len;
1998 
1999  if (PREDICT_FALSE (!vui->admin_up))
2000  {
2001  error = VHOST_USER_TX_FUNC_ERROR_DOWN;
2002  goto done3;
2003  }
2004 
2005  if (PREDICT_FALSE (!vui->is_up))
2006  {
2007  error = VHOST_USER_TX_FUNC_ERROR_NOT_READY;
2008  goto done3;
2009  }
2010 
2011  qid =
2013  (vui->per_cpu_tx_qid, thread_index));
2014  rxvq = &vui->vrings[qid];
2015  if (PREDICT_FALSE (vui->use_tx_spinlock))
2016  vhost_user_vring_lock (vui, qid);
2017 
2018 retry:
2019  error = VHOST_USER_TX_FUNC_ERROR_NONE;
2020  tx_headers_len = 0;
2021  copy_len = 0;
2022  while (n_left > 0)
2023  {
2024  vlib_buffer_t *b0, *current_b0;
2025  u16 desc_head, desc_index, desc_len;
2026  vring_desc_t *desc_table;
2027  uword buffer_map_addr;
2028  u32 buffer_len;
2029  u16 bytes_left;
2030 
2031  if (PREDICT_TRUE (n_left > 1))
2032  vlib_prefetch_buffer_with_index (vm, buffers[1], LOAD);
2033 
2034  b0 = vlib_get_buffer (vm, buffers[0]);
2035 
2037  {
2038  vum->cpus[thread_index].current_trace =
2039  vlib_add_trace (vm, node, b0,
2040  sizeof (*vum->cpus[thread_index].current_trace));
2041  vhost_user_tx_trace (vum->cpus[thread_index].current_trace,
2042  vui, qid / 2, b0, rxvq);
2043  }
2044 
2045  if (PREDICT_FALSE (rxvq->last_avail_idx == rxvq->avail->idx))
2046  {
2047  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
2048  goto done;
2049  }
2050 
2051  desc_table = rxvq->desc;
2052  desc_head = desc_index =
2053  rxvq->avail->ring[rxvq->last_avail_idx & rxvq->qsz_mask];
2054 
2055  /* Go deeper in case of indirect descriptor
2056  * I don't know of any driver providing indirect for RX. */
2057  if (PREDICT_FALSE (rxvq->desc[desc_head].flags & VIRTQ_DESC_F_INDIRECT))
2058  {
2059  if (PREDICT_FALSE
2060  (rxvq->desc[desc_head].len < sizeof (vring_desc_t)))
2061  {
2062  error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
2063  goto done;
2064  }
2065  if (PREDICT_FALSE
2066  (!(desc_table =
2067  map_guest_mem (vui, rxvq->desc[desc_index].addr,
2068  &map_hint))))
2069  {
2070  error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
2071  goto done;
2072  }
2073  desc_index = 0;
2074  }
2075 
2076  desc_len = vui->virtio_net_hdr_sz;
2077  buffer_map_addr = desc_table[desc_index].addr;
2078  buffer_len = desc_table[desc_index].len;
2079 
2080  {
2081  // Get a header from the header array
2082  virtio_net_hdr_mrg_rxbuf_t *hdr =
2083  &vum->cpus[thread_index].tx_headers[tx_headers_len];
2084  tx_headers_len++;
2085  hdr->hdr.flags = 0;
2086  hdr->hdr.gso_type = 0;
2087  hdr->num_buffers = 1; //This is local, no need to check
2088 
2089  // Prepare a copy order executed later for the header
2090  vhost_copy_t *cpy = &vum->cpus[thread_index].copy[copy_len];
2091  copy_len++;
2092  cpy->len = vui->virtio_net_hdr_sz;
2093  cpy->dst = buffer_map_addr;
2094  cpy->src = (uword) hdr;
2095  }
2096 
2097  buffer_map_addr += vui->virtio_net_hdr_sz;
2098  buffer_len -= vui->virtio_net_hdr_sz;
2099  bytes_left = b0->current_length;
2100  current_b0 = b0;
2101  while (1)
2102  {
2103  if (buffer_len == 0)
2104  { //Get new output
2105  if (desc_table[desc_index].flags & VIRTQ_DESC_F_NEXT)
2106  {
2107  //Next one is chained
2108  desc_index = desc_table[desc_index].next;
2109  buffer_map_addr = desc_table[desc_index].addr;
2110  buffer_len = desc_table[desc_index].len;
2111  }
2112  else if (vui->virtio_net_hdr_sz == 12) //MRG is available
2113  {
2114  virtio_net_hdr_mrg_rxbuf_t *hdr =
2115  &vum->cpus[thread_index].tx_headers[tx_headers_len - 1];
2116 
2117  //Move from available to used buffer
2118  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].id =
2119  desc_head;
2120  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].len =
2121  desc_len;
2122  vhost_user_log_dirty_ring (vui, rxvq,
2123  ring[rxvq->last_used_idx &
2124  rxvq->qsz_mask]);
2125 
2126  rxvq->last_avail_idx++;
2127  rxvq->last_used_idx++;
2128  hdr->num_buffers++;
2129  desc_len = 0;
2130 
2131  if (PREDICT_FALSE
2132  (rxvq->last_avail_idx == rxvq->avail->idx))
2133  {
2134  //Dequeue queued descriptors for this packet
2135  rxvq->last_used_idx -= hdr->num_buffers - 1;
2136  rxvq->last_avail_idx -= hdr->num_buffers - 1;
2137  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
2138  goto done;
2139  }
2140 
2141  desc_table = rxvq->desc;
2142  desc_head = desc_index =
2143  rxvq->avail->ring[rxvq->last_avail_idx & rxvq->qsz_mask];
2144  if (PREDICT_FALSE
2145  (rxvq->desc[desc_head].flags & VIRTQ_DESC_F_INDIRECT))
2146  {
2147  //It is seriously unlikely that a driver will put indirect descriptor
2148  //after non-indirect descriptor.
2149  if (PREDICT_FALSE
2150  (rxvq->desc[desc_head].len < sizeof (vring_desc_t)))
2151  {
2152  error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
2153  goto done;
2154  }
2155  if (PREDICT_FALSE
2156  (!(desc_table =
2157  map_guest_mem (vui,
2158  rxvq->desc[desc_index].addr,
2159  &map_hint))))
2160  {
2161  error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
2162  goto done;
2163  }
2164  desc_index = 0;
2165  }
2166  buffer_map_addr = desc_table[desc_index].addr;
2167  buffer_len = desc_table[desc_index].len;
2168  }
2169  else
2170  {
2171  error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOMRG;
2172  goto done;
2173  }
2174  }
2175 
2176  {
2177  vhost_copy_t *cpy = &vum->cpus[thread_index].copy[copy_len];
2178  copy_len++;
2179  cpy->len = bytes_left;
2180  cpy->len = (cpy->len > buffer_len) ? buffer_len : cpy->len;
2181  cpy->dst = buffer_map_addr;
2182  cpy->src = (uword) vlib_buffer_get_current (current_b0) +
2183  current_b0->current_length - bytes_left;
2184 
2185  bytes_left -= cpy->len;
2186  buffer_len -= cpy->len;
2187  buffer_map_addr += cpy->len;
2188  desc_len += cpy->len;
2189 
2190  CLIB_PREFETCH (&rxvq->desc, CLIB_CACHE_LINE_BYTES, LOAD);
2191  }
2192 
2193  // Check if vlib buffer has more data. If not, get more or break.
2194  if (PREDICT_TRUE (!bytes_left))
2195  {
2196  if (PREDICT_FALSE
2197  (current_b0->flags & VLIB_BUFFER_NEXT_PRESENT))
2198  {
2199  current_b0 = vlib_get_buffer (vm, current_b0->next_buffer);
2200  bytes_left = current_b0->current_length;
2201  }
2202  else
2203  {
2204  //End of packet
2205  break;
2206  }
2207  }
2208  }
2209 
2210  //Move from available to used ring
2211  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].id = desc_head;
2212  rxvq->used->ring[rxvq->last_used_idx & rxvq->qsz_mask].len = desc_len;
2213  vhost_user_log_dirty_ring (vui, rxvq,
2214  ring[rxvq->last_used_idx & rxvq->qsz_mask]);
2215  rxvq->last_avail_idx++;
2216  rxvq->last_used_idx++;
2217 
2219  {
2220  vum->cpus[thread_index].current_trace->hdr =
2221  vum->cpus[thread_index].tx_headers[tx_headers_len - 1];
2222  }
2223 
2224  n_left--; //At the end for error counting when 'goto done' is invoked
2225 
2226  /*
2227  * Do the copy periodically to prevent
2228  * vum->cpus[thread_index].copy array overflow and corrupt memory
2229  */
2230  if (PREDICT_FALSE (copy_len >= VHOST_USER_TX_COPY_THRESHOLD))
2231  {
2232  if (PREDICT_FALSE
2233  (vhost_user_tx_copy (vui, vum->cpus[thread_index].copy,
2234  copy_len, &map_hint)))
2235  {
2236  vlib_error_count (vm, node->node_index,
2237  VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);
2238  }
2239  copy_len = 0;
2240 
2241  /* give buffers back to driver */
2243  rxvq->used->idx = rxvq->last_used_idx;
2244  vhost_user_log_dirty_ring (vui, rxvq, idx);
2245  }
2246  buffers++;
2247  }
2248 
2249 done:
2250  //Do the memory copies
2251  if (PREDICT_FALSE
2252  (vhost_user_tx_copy (vui, vum->cpus[thread_index].copy,
2253  copy_len, &map_hint)))
2254  {
2255  vlib_error_count (vm, node->node_index,
2256  VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL, 1);
2257  }
2258 
2260  rxvq->used->idx = rxvq->last_used_idx;
2261  vhost_user_log_dirty_ring (vui, rxvq, idx);
2262 
2263  /*
2264  * When n_left is set, error is always set to something too.
2265  * In case error is due to lack of remaining buffers, we go back up and
2266  * retry.
2267  * The idea is that it is better to waste some time on packets
2268  * that have been processed already than dropping them and get
2269  * more fresh packets with a good likelyhood that they will be dropped too.
2270  * This technique also gives more time to VM driver to pick-up packets.
2271  * In case the traffic flows from physical to virtual interfaces, this
2272  * technique will end-up leveraging the physical NIC buffer in order to
2273  * absorb the VM's CPU jitter.
2274  */
2275  if (n_left && (error == VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF) && retry)
2276  {
2277  retry--;
2278  goto retry;
2279  }
2280 
2281  /* interrupt (call) handling */
2282  if ((rxvq->callfd_idx != ~0) &&
2283  !(rxvq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
2284  {
2285  rxvq->n_since_last_int += frame->n_vectors - n_left;
2286 
2287  if (rxvq->n_since_last_int > vum->coalesce_frames)
2288  vhost_user_send_call (vm, rxvq);
2289  }
2290 
2291  vhost_user_vring_unlock (vui, qid);
2292 
2293 done3:
2294  if (PREDICT_FALSE (n_left && error != VHOST_USER_TX_FUNC_ERROR_NONE))
2295  {
2296  vlib_error_count (vm, node->node_index, error, n_left);
2300  thread_index, vui->sw_if_index, n_left);
2301  }
2302 
2303  vlib_buffer_free (vm, vlib_frame_args (frame), frame->n_vectors);
2304  return frame->n_vectors;
2305 }
2306 
2307 static uword
2310 {
2311  vhost_user_intf_t *vui;
2312  f64 timeout = 3153600000.0 /* 100 years */ ;
2313  uword event_type, *event_data = 0;
2315  u16 *queue;
2316  f64 now, poll_time_remaining;
2317  f64 next_timeout;
2318  u8 stop_timer = 0;
2319 
2320  while (1)
2321  {
2322  poll_time_remaining =
2324  event_type = vlib_process_get_events (vm, &event_data);
2325  vec_reset_length (event_data);
2326 
2327  /*
2328  * Use the remaining timeout if it is less than coalesce time to avoid
2329  * resetting the existing timer in the middle of expiration
2330  */
2331  timeout = poll_time_remaining;
2332  if (vlib_process_suspend_time_is_zero (timeout) ||
2333  (timeout > vum->coalesce_time))
2334  timeout = vum->coalesce_time;
2335 
2336  now = vlib_time_now (vm);
2337  switch (event_type)
2338  {
2340  stop_timer = 1;
2341  break;
2342 
2344  stop_timer = 0;
2345  if (!vlib_process_suspend_time_is_zero (poll_time_remaining))
2346  break;
2347  /* fall through */
2348 
2349  case ~0:
2350  /* *INDENT-OFF* */
2351  pool_foreach (vui, vum->vhost_user_interfaces, {
2352  next_timeout = timeout;
2353  vec_foreach (queue, vui->rx_queues)
2354  {
2355  vhost_user_vring_t *rxvq =
2356  &vui->vrings[VHOST_VRING_IDX_RX (*queue)];
2357  vhost_user_vring_t *txvq =
2358  &vui->vrings[VHOST_VRING_IDX_TX (*queue)];
2359 
2360  if (txvq->n_since_last_int)
2361  {
2362  if (now >= txvq->int_deadline)
2363  vhost_user_send_call (vm, txvq);
2364  else
2365  next_timeout = txvq->int_deadline - now;
2366  }
2367 
2368  if (rxvq->n_since_last_int)
2369  {
2370  if (now >= rxvq->int_deadline)
2371  vhost_user_send_call (vm, rxvq);
2372  else
2373  next_timeout = rxvq->int_deadline - now;
2374  }
2375 
2376  if ((next_timeout < timeout) && (next_timeout > 0.0))
2377  timeout = next_timeout;
2378  }
2379  });
2380  /* *INDENT-ON* */
2381  break;
2382 
2383  default:
2384  clib_warning ("BUG: unhandled event type %d", event_type);
2385  break;
2386  }
2387  /* No less than 1 millisecond */
2388  if (timeout < 1e-3)
2389  timeout = 1e-3;
2390  if (stop_timer)
2391  timeout = 3153600000.0;
2392  }
2393  return 0;
2394 }
2395 
2396 /* *INDENT-OFF* */
2399  .type = VLIB_NODE_TYPE_PROCESS,
2400  .name = "vhost-user-send-interrupt-process",
2401 };
2402 /* *INDENT-ON* */
2403 
2404 static clib_error_t *
2406  u32 qid, vnet_hw_interface_rx_mode mode)
2407 {
2408  vlib_main_t *vm = vnm->vlib_main;
2409  vnet_hw_interface_t *hif = vnet_get_hw_interface (vnm, hw_if_index);
2411  vhost_user_intf_t *vui =
2413  vhost_user_vring_t *txvq = &vui->vrings[VHOST_VRING_IDX_TX (qid)];
2414 
2415  if ((mode == VNET_HW_INTERFACE_RX_MODE_INTERRUPT) ||
2417  {
2418  if (txvq->kickfd_idx == ~0)
2419  {
2420  // We cannot support interrupt mode if the driver opts out
2421  return clib_error_return (0, "Driver does not support interrupt");
2422  }
2424  {
2425  vum->ifq_count++;
2426  // Start the timer if this is the first encounter on interrupt
2427  // interface/queue
2428  if ((vum->ifq_count == 1) &&
2429  (vum->coalesce_time > 0.0) && (vum->coalesce_frames > 0))
2433  }
2434  }
2435  else if (mode == VNET_HW_INTERFACE_RX_MODE_POLLING)
2436  {
2437  if (((txvq->mode == VNET_HW_INTERFACE_RX_MODE_INTERRUPT) ||
2439  vum->ifq_count)
2440  {
2441  vum->ifq_count--;
2442  // Stop the timer if there is no more interrupt interface/queue
2443  if ((vum->ifq_count == 0) &&
2444  (vum->coalesce_time > 0.0) && (vum->coalesce_frames > 0))
2448  }
2449  }
2450 
2451  txvq->mode = mode;
2453  txvq->used->flags = VRING_USED_F_NO_NOTIFY;
2454  else if ((mode == VNET_HW_INTERFACE_RX_MODE_ADAPTIVE) ||
2456  txvq->used->flags = 0;
2457  else
2458  {
2459  clib_warning ("BUG: unhandled mode %d changed for if %d queue %d", mode,
2460  hw_if_index, qid);
2461  return clib_error_return (0, "unsupported");
2462  }
2463 
2464  return 0;
2465 }
2466 
2467 static clib_error_t *
2469  u32 flags)
2470 {
2471  vnet_hw_interface_t *hif = vnet_get_hw_interface (vnm, hw_if_index);
2472  uword is_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0;
2474  vhost_user_intf_t *vui =
2476 
2477  vui->admin_up = is_up;
2478 
2479  if (is_up && vui->is_up)
2482 
2483  return /* no error */ 0;
2484 }
2485 
2486 /* *INDENT-OFF* */
2487 VNET_DEVICE_CLASS (vhost_user_dev_class,static) = {
2488  .name = "vhost-user",
2489  .tx_function = vhost_user_tx,
2490  .tx_function_n_errors = VHOST_USER_TX_FUNC_N_ERROR,
2491  .tx_function_error_strings = vhost_user_tx_func_error_strings,
2492  .format_device_name = format_vhost_user_interface_name,
2493  .name_renumber = vhost_user_name_renumber,
2494  .admin_up_down_function = vhost_user_interface_admin_up_down,
2495  .rx_mode_change_function = vhost_user_interface_rx_mode_change,
2496  .format_tx_trace = format_vhost_trace,
2497 };
2498 
2500  vhost_user_tx)
2501 /* *INDENT-ON* */
2502 
2503 static uword
2504 vhost_user_process (vlib_main_t * vm,
2506 {
2508  vhost_user_intf_t *vui;
2509  struct sockaddr_un sun;
2510  int sockfd;
2511  clib_file_t template = { 0 };
2512  f64 timeout = 3153600000.0 /* 100 years */ ;
2513  uword *event_data = 0;
2514 
2515  sockfd = -1;
2516  sun.sun_family = AF_UNIX;
2518  template.error_function = vhost_user_socket_error;
2519 
2520  while (1)
2521  {
2523  vlib_process_get_events (vm, &event_data);
2524  vec_reset_length (event_data);
2525 
2526  timeout = 3.0;
2527 
2528  /* *INDENT-OFF* */
2529  pool_foreach (vui, vum->vhost_user_interfaces, {
2530 
2531  if (vui->unix_server_index == ~0) { //Nothing to do for server sockets
2532  if (vui->clib_file_index == ~0)
2533  {
2534  if ((sockfd < 0) &&
2535  ((sockfd = socket (AF_UNIX, SOCK_STREAM, 0)) < 0))
2536  {
2537  /*
2538  * 1st time error or new error for this interface,
2539  * spit out the message and record the error
2540  */
2541  if (!vui->sock_errno || (vui->sock_errno != errno))
2542  {
2543  clib_unix_warning
2544  ("Error: Could not open unix socket for %s",
2545  vui->sock_filename);
2546  vui->sock_errno = errno;
2547  }
2548  continue;
2549  }
2550 
2551  /* try to connect */
2552  strncpy (sun.sun_path, (char *) vui->sock_filename,
2553  sizeof (sun.sun_path) - 1);
2554 
2555  /* Avoid hanging VPP if the other end does not accept */
2556  if (fcntl(sockfd, F_SETFL, O_NONBLOCK) < 0)
2557  clib_unix_warning ("fcntl");
2558 
2559  if (connect (sockfd, (struct sockaddr *) &sun,
2560  sizeof (struct sockaddr_un)) == 0)
2561  {
2562  /* Set the socket to blocking as it was before */
2563  if (fcntl(sockfd, F_SETFL, 0) < 0)
2564  clib_unix_warning ("fcntl2");
2565 
2566  vui->sock_errno = 0;
2567  template.file_descriptor = sockfd;
2568  template.private_data =
2569  vui - vhost_user_main.vhost_user_interfaces;
2570  vui->clib_file_index = clib_file_add (&file_main, &template);
2571 
2572  /* This sockfd is considered consumed */
2573  sockfd = -1;
2574  }
2575  else
2576  {
2577  vui->sock_errno = errno;
2578  }
2579  }
2580  else
2581  {
2582  /* check if socket is alive */
2583  int error = 0;
2584  socklen_t len = sizeof (error);
2585  int fd = UNIX_GET_FD(vui->clib_file_index);
2586  int retval =
2587  getsockopt (fd, SOL_SOCKET, SO_ERROR, &error, &len);
2588 
2589  if (retval)
2590  {
2591  DBG_SOCK ("getsockopt returned %d", retval);
2592  vhost_user_if_disconnect (vui);
2593  }
2594  }
2595  }
2596  });
2597  /* *INDENT-ON* */
2598  }
2599  return 0;
2600 }
2601 
2602 /* *INDENT-OFF* */
2604  .function = vhost_user_process,
2605  .type = VLIB_NODE_TYPE_PROCESS,
2606  .name = "vhost-user-process",
2607 };
2608 /* *INDENT-ON* */
2609 
2610 /**
2611  * Disables and reset interface structure.
2612  * It can then be either init again, or removed from used interfaces.
2613  */
2614 static void
2616 {
2617  int q;
2619 
2620  // disconnect interface sockets
2623 
2624  for (q = 0; q < VHOST_VRING_MAX_N; q++)
2625  {
2626  clib_mem_free ((void *) vui->vring_locks[q]);
2627  }
2628 
2629  if (vui->unix_server_index != ~0)
2630  {
2631  //Close server socket
2633  vui->unix_server_index);
2634  clib_file_del (&file_main, uf);
2635  vui->unix_server_index = ~0;
2636  unlink (vui->sock_filename);
2637  }
2638 
2640  &vui->if_index);
2641 }
2642 
2643 int
2645 {
2647  vhost_user_intf_t *vui;
2648  int rv = 0;
2649  vnet_hw_interface_t *hwif;
2650  u16 *queue;
2651 
2652  if (!(hwif = vnet_get_sup_hw_interface (vnm, sw_if_index)) ||
2653  hwif->dev_class_index != vhost_user_dev_class.index)
2654  return VNET_API_ERROR_INVALID_SW_IF_INDEX;
2655 
2656  DBG_SOCK ("Deleting vhost-user interface %s (instance %d)",
2657  hwif->name, hwif->dev_instance);
2658 
2660 
2661  vec_foreach (queue, vui->rx_queues)
2662  {
2663  vhost_user_vring_t *txvq;
2664 
2665  txvq = &vui->vrings[VHOST_VRING_IDX_TX (*queue)];
2666  if ((vum->ifq_count > 0) &&
2669  {
2670  vum->ifq_count--;
2671  // Stop the timer if there is no more interrupt interface/queue
2672  if ((vum->ifq_count == 0) &&
2673  (vum->coalesce_time > 0.0) && (vum->coalesce_frames > 0))
2674  {
2678  break;
2679  }
2680  }
2681  }
2682 
2683  // Disable and reset interface
2684  vhost_user_term_if (vui);
2685 
2686  // Reset renumbered iface
2687  if (hwif->dev_instance <
2690 
2691  // Delete ethernet interface
2693 
2694  // Back to pool
2695  pool_put (vum->vhost_user_interfaces, vui);
2696 
2697  return rv;
2698 }
2699 
2700 static clib_error_t *
2702 {
2703  vnet_main_t *vnm = vnet_get_main ();
2705  vhost_user_intf_t *vui;
2706 
2708  /* *INDENT-OFF* */
2709  pool_foreach (vui, vum->vhost_user_interfaces, {
2710  vhost_user_delete_if (vnm, vm, vui->sw_if_index);
2711  });
2712  /* *INDENT-ON* */
2714  return 0;
2715 }
2716 
2718 
2719 /**
2720  * Open server unix socket on specified sock_filename.
2721  */
2722 static int
2723 vhost_user_init_server_sock (const char *sock_filename, int *sock_fd)
2724 {
2725  int rv = 0;
2726  struct sockaddr_un un = { };
2727  int fd;
2728  /* create listening socket */
2729  if ((fd = socket (AF_UNIX, SOCK_STREAM, 0)) < 0)
2730  return VNET_API_ERROR_SYSCALL_ERROR_1;
2731 
2732  un.sun_family = AF_UNIX;
2733  strncpy ((char *) un.sun_path, (char *) sock_filename,
2734  sizeof (un.sun_path) - 1);
2735 
2736  /* remove if exists */
2737  unlink ((char *) sock_filename);
2738 
2739  if (bind (fd, (struct sockaddr *) &un, sizeof (un)) == -1)
2740  {
2741  rv = VNET_API_ERROR_SYSCALL_ERROR_2;
2742  goto error;
2743  }
2744 
2745  if (listen (fd, 1) == -1)
2746  {
2747  rv = VNET_API_ERROR_SYSCALL_ERROR_3;
2748  goto error;
2749  }
2750 
2751  *sock_fd = fd;
2752  return 0;
2753 
2754 error:
2755  close (fd);
2756  return rv;
2757 }
2758 
2759 /**
2760  * Create ethernet interface for vhost user interface.
2761  */
2762 static void
2764  vhost_user_intf_t * vui, u8 * hwaddress)
2765 {
2767  u8 hwaddr[6];
2768  clib_error_t *error;
2769 
2770  /* create hw and sw interface */
2771  if (hwaddress)
2772  {
2773  clib_memcpy (hwaddr, hwaddress, 6);
2774  }
2775  else
2776  {
2777  random_u32 (&vum->random);
2778  clib_memcpy (hwaddr + 2, &vum->random, sizeof (vum->random));
2779  hwaddr[0] = 2;
2780  hwaddr[1] = 0xfe;
2781  }
2782 
2784  (vnm,
2785  vhost_user_dev_class.index,
2786  vui - vum->vhost_user_interfaces /* device instance */ ,
2787  hwaddr /* ethernet address */ ,
2788  &vui->hw_if_index, 0 /* flag change */ );
2789 
2790  if (error)
2791  clib_error_report (error);
2792 
2795 }
2796 
2797 /*
2798  * Initialize vui with specified attributes
2799  */
2800 static void
2802  vhost_user_intf_t * vui,
2803  int server_sock_fd,
2804  const char *sock_filename,
2805  u64 feature_mask, u32 * sw_if_index)
2806 {
2807  vnet_sw_interface_t *sw;
2808  int q;
2810  vnet_hw_interface_t *hw;
2811 
2812  hw = vnet_get_hw_interface (vnm, vui->hw_if_index);
2813  sw = vnet_get_hw_sw_interface (vnm, vui->hw_if_index);
2814  if (server_sock_fd != -1)
2815  {
2816  clib_file_t template = { 0 };
2818  template.file_descriptor = server_sock_fd;
2819  template.private_data = vui - vum->vhost_user_interfaces; //hw index
2820  vui->unix_server_index = clib_file_add (&file_main, &template);
2821  }
2822  else
2823  {
2824  vui->unix_server_index = ~0;
2825  }
2826 
2827  vui->sw_if_index = sw->sw_if_index;
2828  strncpy (vui->sock_filename, sock_filename,
2829  ARRAY_LEN (vui->sock_filename) - 1);
2830  vui->sock_errno = 0;
2831  vui->is_up = 0;
2832  vui->feature_mask = feature_mask;
2833  vui->clib_file_index = ~0;
2834  vui->log_base_addr = 0;
2835  vui->if_index = vui - vum->vhost_user_interfaces;
2837  &vui->if_index, 0);
2838 
2839  for (q = 0; q < VHOST_VRING_MAX_N; q++)
2840  vhost_user_vring_init (vui, q);
2841 
2844 
2845  if (sw_if_index)
2846  *sw_if_index = vui->sw_if_index;
2847 
2848  for (q = 0; q < VHOST_VRING_MAX_N; q++)
2849  {
2852  memset ((void *) vui->vring_locks[q], 0, CLIB_CACHE_LINE_BYTES);
2853  }
2854 
2856  vlib_get_thread_main ()->n_vlib_mains - 1);
2858 }
2859 
2860 int
2862  const char *sock_filename,
2863  u8 is_server,
2864  u32 * sw_if_index,
2865  u64 feature_mask,
2866  u8 renumber, u32 custom_dev_instance, u8 * hwaddr)
2867 {
2868  vhost_user_intf_t *vui = NULL;
2869  u32 sw_if_idx = ~0;
2870  int rv = 0;
2871  int server_sock_fd = -1;
2873  uword *if_index;
2874 
2875  if (sock_filename == NULL || !(strlen (sock_filename) > 0))
2876  {
2877  return VNET_API_ERROR_INVALID_ARGUMENT;
2878  }
2879 
2880  if_index = mhash_get (&vum->if_index_by_sock_name, (void *) sock_filename);
2881  if (if_index)
2882  {
2883  if (sw_if_index)
2884  {
2885  vui = &vum->vhost_user_interfaces[*if_index];
2886  *sw_if_index = vui->sw_if_index;
2887  }
2888  return VNET_API_ERROR_IF_ALREADY_EXISTS;
2889  }
2890 
2891  if (is_server)
2892  {
2893  if ((rv =
2894  vhost_user_init_server_sock (sock_filename, &server_sock_fd)) != 0)
2895  {
2896  return rv;
2897  }
2898  }
2899 
2900  pool_get (vhost_user_main.vhost_user_interfaces, vui);
2901 
2902  vhost_user_create_ethernet (vnm, vm, vui, hwaddr);
2903  vhost_user_vui_init (vnm, vui, server_sock_fd, sock_filename,
2904  feature_mask, &sw_if_idx);
2905 
2906  if (renumber)
2907  vnet_interface_name_renumber (sw_if_idx, custom_dev_instance);
2908 
2909  if (sw_if_index)
2910  *sw_if_index = sw_if_idx;
2911 
2912  // Process node must connect
2914 
2915  return rv;
2916 }
2917 
2918 int
2920  const char *sock_filename,
2921  u8 is_server,
2922  u32 sw_if_index,
2923  u64 feature_mask, u8 renumber, u32 custom_dev_instance)
2924 {
2926  vhost_user_intf_t *vui = NULL;
2927  u32 sw_if_idx = ~0;
2928  int server_sock_fd = -1;
2929  int rv = 0;
2930  vnet_hw_interface_t *hwif;
2931  uword *if_index;
2932 
2933  if (!(hwif = vnet_get_sup_hw_interface (vnm, sw_if_index)) ||
2934  hwif->dev_class_index != vhost_user_dev_class.index)
2935  return VNET_API_ERROR_INVALID_SW_IF_INDEX;
2936 
2937  if (sock_filename == NULL || !(strlen (sock_filename) > 0))
2938  return VNET_API_ERROR_INVALID_ARGUMENT;
2939 
2941 
2942  /*
2943  * Disallow changing the interface to have the same path name
2944  * as other interface
2945  */
2946  if_index = mhash_get (&vum->if_index_by_sock_name, (void *) sock_filename);
2947  if (if_index && (*if_index != vui->if_index))
2948  return VNET_API_ERROR_IF_ALREADY_EXISTS;
2949 
2950  // First try to open server socket
2951  if (is_server)
2952  if ((rv = vhost_user_init_server_sock (sock_filename,
2953  &server_sock_fd)) != 0)
2954  return rv;
2955 
2956  vhost_user_term_if (vui);
2957  vhost_user_vui_init (vnm, vui, server_sock_fd,
2958  sock_filename, feature_mask, &sw_if_idx);
2959 
2960  if (renumber)
2961  vnet_interface_name_renumber (sw_if_idx, custom_dev_instance);
2962 
2963  // Process node must connect
2965 
2966  return rv;
2967 }
2968 
2969 clib_error_t *
2971  unformat_input_t * input,
2972  vlib_cli_command_t * cmd)
2973 {
2974  unformat_input_t _line_input, *line_input = &_line_input;
2975  u8 *sock_filename = NULL;
2976  u32 sw_if_index;
2977  u8 is_server = 0;
2978  u64 feature_mask = (u64) ~ (0ULL);
2979  u8 renumber = 0;
2980  u32 custom_dev_instance = ~0;
2981  u8 hwaddr[6];
2982  u8 *hw = NULL;
2983  clib_error_t *error = NULL;
2984 
2985  /* Get a line of input. */
2986  if (!unformat_user (input, unformat_line_input, line_input))
2987  return 0;
2988 
2989  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
2990  {
2991  if (unformat (line_input, "socket %s", &sock_filename))
2992  ;
2993  else if (unformat (line_input, "server"))
2994  is_server = 1;
2995  else if (unformat (line_input, "feature-mask 0x%llx", &feature_mask))
2996  ;
2997  else
2998  if (unformat
2999  (line_input, "hwaddr %U", unformat_ethernet_address, hwaddr))
3000  hw = hwaddr;
3001  else if (unformat (line_input, "renumber %d", &custom_dev_instance))
3002  {
3003  renumber = 1;
3004  }
3005  else
3006  {
3007  error = clib_error_return (0, "unknown input `%U'",
3008  format_unformat_error, line_input);
3009  goto done;
3010  }
3011  }
3012 
3013  vnet_main_t *vnm = vnet_get_main ();
3014 
3015  int rv;
3016  if ((rv = vhost_user_create_if (vnm, vm, (char *) sock_filename,
3017  is_server, &sw_if_index, feature_mask,
3018  renumber, custom_dev_instance, hw)))
3019  {
3020  error = clib_error_return (0, "vhost_user_create_if returned %d", rv);
3021  goto done;
3022  }
3023 
3025  sw_if_index);
3026 
3027 done:
3028  vec_free (sock_filename);
3029  unformat_free (line_input);
3030 
3031  return error;
3032 }
3033 
3034 clib_error_t *
3036  unformat_input_t * input,
3037  vlib_cli_command_t * cmd)
3038 {
3039  unformat_input_t _line_input, *line_input = &_line_input;
3040  u32 sw_if_index = ~0;
3041  vnet_main_t *vnm = vnet_get_main ();
3042  clib_error_t *error = NULL;
3043 
3044  /* Get a line of input. */
3045  if (!unformat_user (input, unformat_line_input, line_input))
3046  return 0;
3047 
3048  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
3049  {
3050  if (unformat (line_input, "sw_if_index %d", &sw_if_index))
3051  ;
3052  else if (unformat
3053  (line_input, "%U", unformat_vnet_sw_interface, vnm,
3054  &sw_if_index))
3055  {
3056  vnet_hw_interface_t *hwif =
3057  vnet_get_sup_hw_interface (vnm, sw_if_index);
3058  if (hwif == NULL ||
3059  vhost_user_dev_class.index != hwif->dev_class_index)
3060  {
3061  error = clib_error_return (0, "Not a vhost interface");
3062  goto done;
3063  }
3064  }
3065  else
3066  {
3067  error = clib_error_return (0, "unknown input `%U'",
3068  format_unformat_error, line_input);
3069  goto done;
3070  }
3071  }
3072 
3073  vhost_user_delete_if (vnm, vm, sw_if_index);
3074 
3075 done:
3076  unformat_free (line_input);
3077 
3078  return error;
3079 }
3080 
3081 int
3083  vhost_user_intf_details_t ** out_vuids)
3084 {
3085  int rv = 0;
3087  vhost_user_intf_t *vui;
3088  vhost_user_intf_details_t *r_vuids = NULL;
3090  u32 *hw_if_indices = 0;
3092  u8 *s = NULL;
3093  int i;
3094 
3095  if (!out_vuids)
3096  return -1;
3097 
3099  vec_add1 (hw_if_indices, vui->hw_if_index);
3100  );
3101 
3102  for (i = 0; i < vec_len (hw_if_indices); i++)
3103  {
3104  hi = vnet_get_hw_interface (vnm, hw_if_indices[i]);
3106 
3107  vec_add2 (r_vuids, vuid, 1);
3108  vuid->sw_if_index = vui->sw_if_index;
3109  vuid->virtio_net_hdr_sz = vui->virtio_net_hdr_sz;
3110  vuid->features = vui->features;
3111  vuid->num_regions = vui->nregions;
3112  vuid->is_server = vui->unix_server_index != ~0;
3113  vuid->sock_errno = vui->sock_errno;
3114  strncpy ((char *) vuid->sock_filename, (char *) vui->sock_filename,
3115  ARRAY_LEN (vuid->sock_filename) - 1);
3116 
3117  s = format (s, "%v%c", hi->name, 0);
3118 
3119  strncpy ((char *) vuid->if_name, (char *) s,
3120  ARRAY_LEN (vuid->if_name) - 1);
3121  _vec_len (s) = 0;
3122  }
3123 
3124  vec_free (s);
3125  vec_free (hw_if_indices);
3126 
3127  *out_vuids = r_vuids;
3128 
3129  return rv;
3130 }
3131 
3132 clib_error_t *
3134  unformat_input_t * input,
3135  vlib_cli_command_t * cmd)
3136 {
3137  clib_error_t *error = 0;
3138  vnet_main_t *vnm = vnet_get_main ();
3140  vhost_user_intf_t *vui;
3141  u32 hw_if_index, *hw_if_indices = 0;
3143  u16 *queue;
3144  u32 ci;
3145  int i, j, q;
3146  int show_descr = 0;
3147  struct feat_struct
3148  {
3149  u8 bit;
3150  char *str;
3151  };
3152  struct feat_struct *feat_entry;
3153 
3154  static struct feat_struct feat_array[] = {
3155 #define _(s,b) { .str = #s, .bit = b, },
3157 #undef _
3158  {.str = NULL}
3159  };
3160 
3161 #define foreach_protocol_feature \
3162  _(VHOST_USER_PROTOCOL_F_MQ) \
3163  _(VHOST_USER_PROTOCOL_F_LOG_SHMFD)
3164 
3165  static struct feat_struct proto_feat_array[] = {
3166 #define _(s) { .str = #s, .bit = s},
3168 #undef _
3169  {.str = NULL}
3170  };
3171 
3172  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
3173  {
3174  if (unformat
3175  (input, "%U", unformat_vnet_hw_interface, vnm, &hw_if_index))
3176  {
3177  vec_add1 (hw_if_indices, hw_if_index);
3178  }
3179  else if (unformat (input, "descriptors") || unformat (input, "desc"))
3180  show_descr = 1;
3181  else
3182  {
3183  error = clib_error_return (0, "unknown input `%U'",
3184  format_unformat_error, input);
3185  goto done;
3186  }
3187  }
3188  if (vec_len (hw_if_indices) == 0)
3189  {
3191  vec_add1 (hw_if_indices, vui->hw_if_index);
3192  );
3193  }
3194  vlib_cli_output (vm, "Virtio vhost-user interfaces");
3195  vlib_cli_output (vm, "Global:\n coalesce frames %d time %e",
3196  vum->coalesce_frames, vum->coalesce_time);
3197  vlib_cli_output (vm, " number of rx virtqueues in interrupt mode: %d",
3198  vum->ifq_count);
3199 
3200  for (i = 0; i < vec_len (hw_if_indices); i++)
3201  {
3202  hi = vnet_get_hw_interface (vnm, hw_if_indices[i]);
3204  vlib_cli_output (vm, "Interface: %s (ifindex %d)",
3205  hi->name, hw_if_indices[i]);
3206 
3207  vlib_cli_output (vm, "virtio_net_hdr_sz %d\n"
3208  " features mask (0x%llx): \n"
3209  " features (0x%llx): \n",
3210  vui->virtio_net_hdr_sz, vui->feature_mask,
3211  vui->features);
3212 
3213  feat_entry = (struct feat_struct *) &feat_array;
3214  while (feat_entry->str)
3215  {
3216  if (vui->features & (1ULL << feat_entry->bit))
3217  vlib_cli_output (vm, " %s (%d)", feat_entry->str,
3218  feat_entry->bit);
3219  feat_entry++;
3220  }
3221 
3222  vlib_cli_output (vm, " protocol features (0x%llx)",
3223  vui->protocol_features);
3224  feat_entry = (struct feat_struct *) &proto_feat_array;
3225  while (feat_entry->str)
3226  {
3227  if (vui->protocol_features & (1ULL << feat_entry->bit))
3228  vlib_cli_output (vm, " %s (%d)", feat_entry->str,
3229  feat_entry->bit);
3230  feat_entry++;
3231  }
3232 
3233  vlib_cli_output (vm, "\n");
3234 
3235  vlib_cli_output (vm, " socket filename %s type %s errno \"%s\"\n\n",
3236  vui->sock_filename,
3237  (vui->unix_server_index != ~0) ? "server" : "client",
3238  strerror (vui->sock_errno));
3239 
3240  vlib_cli_output (vm, " rx placement: ");
3241 
3242  vec_foreach (queue, vui->rx_queues)
3243  {
3244  vnet_main_t *vnm = vnet_get_main ();
3245  uword thread_index;
3247 
3248  thread_index = vnet_get_device_input_thread_index (vnm,
3249  vui->hw_if_index,
3250  *queue);
3251  vnet_hw_interface_get_rx_mode (vnm, vui->hw_if_index, *queue, &mode);
3252  vlib_cli_output (vm, " thread %d on vring %d, %U\n",
3253  thread_index, VHOST_VRING_IDX_TX (*queue),
3255  }
3256 
3257  vlib_cli_output (vm, " tx placement: %s\n",
3258  vui->use_tx_spinlock ? "spin-lock" : "lock-free");
3259 
3261  {
3262  vlib_cli_output (vm, " thread %d on vring %d\n", ci,
3263  VHOST_VRING_IDX_RX (vui->per_cpu_tx_qid[ci]));
3264  }
3265 
3266  vlib_cli_output (vm, "\n");
3267 
3268  vlib_cli_output (vm, " Memory regions (total %d)\n", vui->nregions);
3269 
3270  if (vui->nregions)
3271  {
3272  vlib_cli_output (vm,
3273  " region fd guest_phys_addr memory_size userspace_addr mmap_offset mmap_addr\n");
3274  vlib_cli_output (vm,
3275  " ====== ===== ================== ================== ================== ================== ==================\n");
3276  }
3277  for (j = 0; j < vui->nregions; j++)
3278  {
3279  vlib_cli_output (vm,
3280  " %d %-5d 0x%016lx 0x%016lx 0x%016lx 0x%016lx 0x%016lx\n",
3281  j, vui->region_mmap_fd[j],
3282  vui->regions[j].guest_phys_addr,
3283  vui->regions[j].memory_size,
3284  vui->regions[j].userspace_addr,
3285  vui->regions[j].mmap_offset,
3287  }
3288  for (q = 0; q < VHOST_VRING_MAX_N; q++)
3289  {
3290  if (!vui->vrings[q].started)
3291  continue;
3292 
3293  vlib_cli_output (vm, "\n Virtqueue %d (%s%s)\n", q,
3294  (q & 1) ? "RX" : "TX",
3295  vui->vrings[q].enabled ? "" : " disabled");
3296 
3297  vlib_cli_output (vm,
3298  " qsz %d last_avail_idx %d last_used_idx %d\n",
3299  vui->vrings[q].qsz_mask + 1,
3300  vui->vrings[q].last_avail_idx,
3301  vui->vrings[q].last_used_idx);
3302 
3303  if (vui->vrings[q].avail && vui->vrings[q].used)
3304  vlib_cli_output (vm,
3305  " avail.flags %x avail.idx %d used.flags %x used.idx %d\n",
3306  vui->vrings[q].avail->flags,
3307  vui->vrings[q].avail->idx,
3308  vui->vrings[q].used->flags,
3309  vui->vrings[q].used->idx);
3310 
3311  int kickfd = UNIX_GET_FD (vui->vrings[q].kickfd_idx);
3312  int callfd = UNIX_GET_FD (vui->vrings[q].callfd_idx);
3313  vlib_cli_output (vm, " kickfd %d callfd %d errfd %d\n",
3314  kickfd, callfd, vui->vrings[q].errfd);
3315 
3316  if (show_descr)
3317  {
3318  vlib_cli_output (vm, "\n descriptor table:\n");
3319  vlib_cli_output (vm,
3320  " id addr len flags next user_addr\n");
3321  vlib_cli_output (vm,
3322  " ===== ================== ===== ====== ===== ==================\n");
3323  for (j = 0; j < vui->vrings[q].qsz_mask + 1; j++)
3324  {
3325  u32 mem_hint = 0;
3326  vlib_cli_output (vm,
3327  " %-5d 0x%016lx %-5d 0x%04x %-5d 0x%016lx\n",
3328  j, vui->vrings[q].desc[j].addr,
3329  vui->vrings[q].desc[j].len,
3330  vui->vrings[q].desc[j].flags,
3331  vui->vrings[q].desc[j].next,
3333  (vui,
3334  vui->vrings[q].desc[j].
3335  addr, &mem_hint)));
3336  }
3337  }
3338  }
3339  vlib_cli_output (vm, "\n");
3340  }
3341 done:
3342  vec_free (hw_if_indices);
3343  return error;
3344 }
3345 
3346 /*
3347  * CLI functions
3348  */
3349 
3350 /*?
3351  * Create a vHost User interface. Once created, a new virtual interface
3352  * will exist with the name '<em>VirtualEthernet0/0/x</em>', where '<em>x</em>'
3353  * is the next free index.
3354  *
3355  * There are several parameters associated with a vHost interface:
3356  *
3357  * - <b>socket <socket-filename></b> - Name of the linux socket used by QEMU/VM and
3358  * VPP to manage the vHost interface. If socket does not already exist, VPP will
3359  * create the socket.
3360  *
3361  * - <b>server</b> - Optional flag to indicate that VPP should be the server for the
3362  * linux socket. If not provided, VPP will be the client.
3363  *
3364  * - <b>feature-mask <hex></b> - Optional virtio/vhost feature set negotiated at
3365  * startup. By default, all supported features will be advertised. Otherwise,
3366  * provide the set of features desired.
3367  * - 0x000008000 (15) - VIRTIO_NET_F_MRG_RXBUF
3368  * - 0x000020000 (17) - VIRTIO_NET_F_CTRL_VQ
3369  * - 0x000200000 (21) - VIRTIO_NET_F_GUEST_ANNOUNCE
3370  * - 0x000400000 (22) - VIRTIO_NET_F_MQ
3371  * - 0x004000000 (26) - VHOST_F_LOG_ALL
3372  * - 0x008000000 (27) - VIRTIO_F_ANY_LAYOUT
3373  * - 0x010000000 (28) - VIRTIO_F_INDIRECT_DESC
3374  * - 0x040000000 (30) - VHOST_USER_F_PROTOCOL_FEATURES
3375  * - 0x100000000 (32) - VIRTIO_F_VERSION_1
3376  *
3377  * - <b>hwaddr <mac-addr></b> - Optional ethernet address, can be in either
3378  * X:X:X:X:X:X unix or X.X.X cisco format.
3379  *
3380  * - <b>renumber <dev_instance></b> - Optional parameter which allows the instance
3381  * in the name to be specified. If instance already exists, name will be used
3382  * anyway and multiple instances will have the same name. Use with caution.
3383  *
3384  * - <b>mode [interrupt | polling]</b> - Optional parameter specifying
3385  * the input thread polling policy.
3386  *
3387  * @cliexpar
3388  * Example of how to create a vhost interface with VPP as the client and all features enabled:
3389  * @cliexstart{create vhost-user socket /tmp/vhost1.sock}
3390  * VirtualEthernet0/0/0
3391  * @cliexend
3392  * Example of how to create a vhost interface with VPP as the server and with just
3393  * multiple queues enabled:
3394  * @cliexstart{create vhost-user socket /tmp/vhost2.sock server feature-mask 0x40400000}
3395  * VirtualEthernet0/0/1
3396  * @cliexend
3397  * Once the vHost interface is created, enable the interface using:
3398  * @cliexcmd{set interface state VirtualEthernet0/0/0 up}
3399 ?*/
3400 /* *INDENT-OFF* */
3401 VLIB_CLI_COMMAND (vhost_user_connect_command, static) = {
3402  .path = "create vhost-user",
3403  .short_help = "create vhost-user socket <socket-filename> [server] "
3404  "[feature-mask <hex>] [hwaddr <mac-addr>] [renumber <dev_instance>] ",
3405  .function = vhost_user_connect_command_fn,
3406 };
3407 /* *INDENT-ON* */
3408 
3409 /*?
3410  * Delete a vHost User interface using the interface name or the
3411  * software interface index. Use the '<em>show interface</em>'
3412  * command to determine the software interface index. On deletion,
3413  * the linux socket will not be deleted.
3414  *
3415  * @cliexpar
3416  * Example of how to delete a vhost interface by name:
3417  * @cliexcmd{delete vhost-user VirtualEthernet0/0/1}
3418  * Example of how to delete a vhost interface by software interface index:
3419  * @cliexcmd{delete vhost-user sw_if_index 1}
3420 ?*/
3421 /* *INDENT-OFF* */
3422 VLIB_CLI_COMMAND (vhost_user_delete_command, static) = {
3423  .path = "delete vhost-user",
3424  .short_help = "delete vhost-user {<interface> | sw_if_index <sw_idx>}",
3425  .function = vhost_user_delete_command_fn,
3426 };
3427 
3428 /*?
3429  * Display the attributes of a single vHost User interface (provide interface
3430  * name), multiple vHost User interfaces (provide a list of interface names seperated
3431  * by spaces) or all Vhost User interfaces (omit an interface name to display all
3432  * vHost interfaces).
3433  *
3434  * @cliexpar
3435  * @parblock
3436  * Example of how to display a vhost interface:
3437  * @cliexstart{show vhost-user VirtualEthernet0/0/0}
3438  * Virtio vhost-user interfaces
3439  * Global:
3440  * coalesce frames 32 time 1e-3
3441  * Interface: VirtualEthernet0/0/0 (ifindex 1)
3442  * virtio_net_hdr_sz 12
3443  * features mask (0xffffffffffffffff):
3444  * features (0x50408000):
3445  * VIRTIO_NET_F_MRG_RXBUF (15)
3446  * VIRTIO_NET_F_MQ (22)
3447  * VIRTIO_F_INDIRECT_DESC (28)
3448  * VHOST_USER_F_PROTOCOL_FEATURES (30)
3449  * protocol features (0x3)
3450  * VHOST_USER_PROTOCOL_F_MQ (0)
3451  * VHOST_USER_PROTOCOL_F_LOG_SHMFD (1)
3452  *
3453  * socket filename /tmp/vhost1.sock type client errno "Success"
3454  *
3455  * rx placement:
3456  * thread 1 on vring 1
3457  * thread 1 on vring 5
3458  * thread 2 on vring 3
3459  * thread 2 on vring 7
3460  * tx placement: spin-lock
3461  * thread 0 on vring 0
3462  * thread 1 on vring 2
3463  * thread 2 on vring 0
3464  *
3465  * Memory regions (total 2)
3466  * region fd guest_phys_addr memory_size userspace_addr mmap_offset mmap_addr
3467  * ====== ===== ================== ================== ================== ================== ==================
3468  * 0 60 0x0000000000000000 0x00000000000a0000 0x00002aaaaac00000 0x0000000000000000 0x00002aab2b400000
3469  * 1 61 0x00000000000c0000 0x000000003ff40000 0x00002aaaaacc0000 0x00000000000c0000 0x00002aababcc0000
3470  *
3471  * Virtqueue 0 (TX)
3472  * qsz 256 last_avail_idx 0 last_used_idx 0
3473  * avail.flags 1 avail.idx 128 used.flags 1 used.idx 0
3474  * kickfd 62 callfd 64 errfd -1
3475  *
3476  * Virtqueue 1 (RX)
3477  * qsz 256 last_avail_idx 0 last_used_idx 0
3478  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3479  * kickfd 65 callfd 66 errfd -1
3480  *
3481  * Virtqueue 2 (TX)
3482  * qsz 256 last_avail_idx 0 last_used_idx 0
3483  * avail.flags 1 avail.idx 128 used.flags 1 used.idx 0
3484  * kickfd 63 callfd 70 errfd -1
3485  *
3486  * Virtqueue 3 (RX)
3487  * qsz 256 last_avail_idx 0 last_used_idx 0
3488  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3489  * kickfd 72 callfd 74 errfd -1
3490  *
3491  * Virtqueue 4 (TX disabled)
3492  * qsz 256 last_avail_idx 0 last_used_idx 0
3493  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3494  * kickfd 76 callfd 78 errfd -1
3495  *
3496  * Virtqueue 5 (RX disabled)
3497  * qsz 256 last_avail_idx 0 last_used_idx 0
3498  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3499  * kickfd 80 callfd 82 errfd -1
3500  *
3501  * Virtqueue 6 (TX disabled)
3502  * qsz 256 last_avail_idx 0 last_used_idx 0
3503  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3504  * kickfd 84 callfd 86 errfd -1
3505  *
3506  * Virtqueue 7 (RX disabled)
3507  * qsz 256 last_avail_idx 0 last_used_idx 0
3508  * avail.flags 1 avail.idx 0 used.flags 1 used.idx 0
3509  * kickfd 88 callfd 90 errfd -1
3510  *
3511  * @cliexend
3512  *
3513  * The optional '<em>descriptors</em>' parameter will display the same output as
3514  * the previous example but will include the descriptor table for each queue.
3515  * The output is truncated below:
3516  * @cliexstart{show vhost-user VirtualEthernet0/0/0 descriptors}
3517  * Virtio vhost-user interfaces
3518  * Global:
3519  * coalesce frames 32 time 1e-3
3520  * Interface: VirtualEthernet0/0/0 (ifindex 1)
3521  * virtio_net_hdr_sz 12
3522  * features mask (0xffffffffffffffff):
3523  * features (0x50408000):
3524  * VIRTIO_NET_F_MRG_RXBUF (15)
3525  * VIRTIO_NET_F_MQ (22)
3526  * :
3527  * Virtqueue 0 (TX)
3528  * qsz 256 last_avail_idx 0 last_used_idx 0
3529  * avail.flags 1 avail.idx 128 used.flags 1 used.idx 0
3530  * kickfd 62 callfd 64 errfd -1
3531  *
3532  * descriptor table:
3533  * id addr len flags next user_addr
3534  * ===== ================== ===== ====== ===== ==================
3535  * 0 0x0000000010b6e974 2060 0x0002 1 0x00002aabbc76e974
3536  * 1 0x0000000010b6e034 2060 0x0002 2 0x00002aabbc76e034
3537  * 2 0x0000000010b6d6f4 2060 0x0002 3 0x00002aabbc76d6f4
3538  * 3 0x0000000010b6cdb4 2060 0x0002 4 0x00002aabbc76cdb4
3539  * 4 0x0000000010b6c474 2060 0x0002 5 0x00002aabbc76c474
3540  * 5 0x0000000010b6bb34 2060 0x0002 6 0x00002aabbc76bb34
3541  * 6 0x0000000010b6b1f4 2060 0x0002 7 0x00002aabbc76b1f4
3542  * 7 0x0000000010b6a8b4 2060 0x0002 8 0x00002aabbc76a8b4
3543  * 8 0x0000000010b69f74 2060 0x0002 9 0x00002aabbc769f74
3544  * 9 0x0000000010b69634 2060 0x0002 10 0x00002aabbc769634
3545  * 10 0x0000000010b68cf4 2060 0x0002 11 0x00002aabbc768cf4
3546  * :
3547  * 249 0x0000000000000000 0 0x0000 250 0x00002aab2b400000
3548  * 250 0x0000000000000000 0 0x0000 251 0x00002aab2b400000
3549  * 251 0x0000000000000000 0 0x0000 252 0x00002aab2b400000
3550  * 252 0x0000000000000000 0 0x0000 253 0x00002aab2b400000
3551  * 253 0x0000000000000000 0 0x0000 254 0x00002aab2b400000
3552  * 254 0x0000000000000000 0 0x0000 255 0x00002aab2b400000
3553  * 255 0x0000000000000000 0 0x0000 32768 0x00002aab2b400000
3554  *
3555  * Virtqueue 1 (RX)
3556  * qsz 256 last_avail_idx 0 last_used_idx 0
3557  * :
3558  * @cliexend
3559  * @endparblock
3560 ?*/
3561 /* *INDENT-OFF* */
3562 VLIB_CLI_COMMAND (show_vhost_user_command, static) = {
3563  .path = "show vhost-user",
3564  .short_help = "show vhost-user [<interface> [<interface> [..]]] [descriptors]",
3565  .function = show_vhost_user_command_fn,
3566 };
3567 /* *INDENT-ON* */
3568 
3569 clib_error_t *
3571  unformat_input_t * input,
3572  vlib_cli_command_t * cmd)
3573 {
3574  unformat_input_t _line_input, *line_input = &_line_input;
3575  clib_error_t *error = NULL;
3577  u8 onoff = 0;
3578  u8 input_found = 0;
3579 
3580  /* Get a line of input. */
3581  if (!unformat_user (input, unformat_line_input, line_input))
3582  return clib_error_return (0, "missing argument");
3583 
3584  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
3585  {
3586  if (input_found)
3587  {
3588  error = clib_error_return (0, "unknown input `%U'",
3589  format_unformat_error, line_input);
3590  goto done;
3591  }
3592 
3593  if (unformat (line_input, "on"))
3594  {
3595  input_found = 1;
3596  onoff = 1;
3597  }
3598  else if (unformat (line_input, "off"))
3599  {
3600  input_found = 1;
3601  onoff = 0;
3602  }
3603  else
3604  {
3605  error = clib_error_return (0, "unknown input `%U'",
3606  format_unformat_error, line_input);
3607  goto done;
3608  }
3609  }
3610 
3611  vum->debug = onoff;
3612 
3613 done:
3614  unformat_free (line_input);
3615 
3616  return error;
3617 }
3618 
3619 /* *INDENT-OFF* */
3620 VLIB_CLI_COMMAND (debug_vhost_user_command, static) = {
3621  .path = "debug vhost-user",
3622  .short_help = "debug vhost-user <on | off>",
3623  .function = debug_vhost_user_command_fn,
3624 };
3625 /* *INDENT-ON* */
3626 
3627 static clib_error_t *
3629 {
3631 
3632  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
3633  {
3634  if (unformat (input, "coalesce-frames %d", &vum->coalesce_frames))
3635  ;
3636  else if (unformat (input, "coalesce-time %f", &vum->coalesce_time))
3637  ;
3638  else if (unformat (input, "dont-dump-memory"))
3639  vum->dont_dump_vhost_user_memory = 1;
3640  else
3641  return clib_error_return (0, "unknown input `%U'",
3642  format_unformat_error, input);
3643  }
3644 
3645  return 0;
3646 }
3647 
3648 /* vhost-user { ... } configuration. */
3649 VLIB_CONFIG_FUNCTION (vhost_user_config, "vhost-user");
3650 
3651 void
3653 {
3655  vhost_user_intf_t *vui;
3656 
3657  if (vum->dont_dump_vhost_user_memory)
3658  {
3660  unmap_all_mem_regions (vui);
3661  );
3662  }
3663 }
3664 
3665 /*
3666  * fd.io coding-style-patch-verification: ON
3667  *
3668  * Local Variables:
3669  * eval: (c-set-style "gnu")
3670  * End:
3671  */
unformat_function_t unformat_vnet_hw_interface
#define vec_validate(V, I)
Make sure vector is long enough for given index (no header, unspecified alignment) ...
Definition: vec.h:432
static clib_error_t * vhost_user_init(vlib_main_t *vm)
Definition: vhost-user.c:1225
static void vhost_user_vring_close(vhost_user_intf_t *vui, u32 qid)
Definition: vhost-user.c:577
static void vnet_device_increment_rx_packets(u32 thread_index, u64 count)
Definition: devices.h:109
vmrglw vmrglh hi
static void vhost_user_if_disconnect(vhost_user_intf_t *vui)
Definition: vhost-user.c:603
#define vec_foreach_index(var, v)
Iterate over vector indices.
static void clib_file_del(clib_file_main_t *um, clib_file_t *f)
Definition: file.h:94
vnet_device_and_queue_t * devices_and_queues
Definition: devices.h:69
sll srl srl sll sra u16x4 i
Definition: vector_sse2.h:337
vring_desc_t * desc
Definition: vhost-user.h:199
#define VRING_AVAIL_F_NO_INTERRUPT
Definition: vhost-user.h:46
#define CLIB_UNUSED(x)
Definition: clib.h:79
u32 virtio_ring_flags
The device index.
Definition: vhost-user.h:274
virtio_net_hdr_mrg_rxbuf_t hdr
Length of the first data descriptor.
Definition: vhost-user.h:276
static uword random_default_seed(void)
Default random seed (unix/linux user-mode)
Definition: random.h:91
clib_error_t * vnet_hw_interface_set_flags(vnet_main_t *vnm, u32 hw_if_index, u32 flags)
Definition: interface.c:538
static u32 vlib_get_trace_count(vlib_main_t *vm, vlib_node_runtime_t *rt)
Definition: trace_funcs.h:143
static f64 vlib_process_wait_for_event_or_clock(vlib_main_t *vm, f64 dt)
Suspend a cooperative multi-tasking thread Waits for an event, or for the indicated number of seconds...
Definition: node_funcs.h:699
#define clib_smp_swap(addr, new)
Definition: smp.h:45
vhost_cpu_t * cpus
Per-CPU data for vhost-user.
Definition: vhost-user.h:307
static void vhost_user_create_ethernet(vnet_main_t *vnm, vlib_main_t *vm, vhost_user_intf_t *vui, u8 *hwaddress)
Create ethernet interface for vhost user interface.
Definition: vhost-user.c:2763
#define VHOST_USER_DOWN_DISCARD_COUNT
Definition: vhost-user.c:72
static void vlib_increment_combined_counter(vlib_combined_counter_main_t *cm, u32 thread_index, u32 index, u64 n_packets, u64 n_bytes)
Increment a combined counter.
Definition: counter.h:211
static void vlib_buffer_free(vlib_main_t *vm, u32 *buffers, u32 n_buffers)
Free buffers Frees the entire buffer chain for each buffer.
Definition: buffer_funcs.h:317
u8 runtime_data[0]
Function dependent node-runtime data.
Definition: node.h:464
void ethernet_delete_interface(vnet_main_t *vnm, u32 hw_if_index)
Definition: interface.c:322
#define VHOST_VRING_IDX_TX(qid)
Definition: vhost-user.h:24
vnet_main_t * vnet_get_main(void)
Definition: misc.c:46
static vnet_hw_interface_t * vnet_get_sup_hw_interface(vnet_main_t *vnm, u32 sw_if_index)
void vhost_user_rx_trace(vhost_trace_t *t, vhost_user_intf_t *vui, u16 qid, vlib_buffer_t *b, vhost_user_vring_t *txvq)
Definition: vhost-user.c:1296
u64 region_guest_addr_hi[VHOST_MEMORY_MAX_NREGIONS]
Definition: vhost-user.h:242
vnet_interface_main_t interface_main
Definition: vnet.h:56
#define PREDICT_TRUE(x)
Definition: clib.h:98
static void vlib_error_count(vlib_main_t *vm, uword node_index, uword counter, uword increment)
Definition: error_funcs.h:57
#define NULL
Definition: clib.h:55
uword mhash_unset(mhash_t *h, void *key, uword *old_value)
Definition: mhash.c:353
static f64 vlib_time_now(vlib_main_t *vm)
Definition: main.h:221
#define foreach_virtio_trace_flags
Definition: vhost-user.c:105
vhost_copy_t copy[VHOST_USER_COPY_ARRAY_N]
Definition: vhost-user.h:289
static void vhost_user_term_if(vhost_user_intf_t *vui)
Disables and reset interface structure.
Definition: vhost-user.c:2615
#define VLIB_BUFFER_PRE_DATA_SIZE
Definition: buffer.h:52
vring_avail_t * avail
Definition: vhost-user.h:200
static uword vhost_user_input(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *f)
Definition: vhost-user.c:1851
static vnet_hw_interface_t * vnet_get_hw_interface(vnet_main_t *vnm, u32 hw_if_index)
#define VHOST_USER_EVENT_START_TIMER
Definition: vhost-user.h:217
u32 file_descriptor
Definition: file.h:53
#define vec_add1(V, E)
Add 1 element to end of vector (unspecified alignment).
Definition: vec.h:518
int vnet_interface_name_renumber(u32 sw_if_index, u32 new_show_dev_instance)
Definition: interface.c:1268
struct _vlib_node_registration vlib_node_registration_t
static_always_inline u32 vhost_user_input_copy(vhost_user_intf_t *vui, vhost_copy_t *cpy, u16 copy_len, u32 *map_hint)
Definition: vhost-user.c:1362
#define VHOST_USER_MSG_HDR_SZ
Definition: vhost-user.h:20
static clib_error_t * vhost_user_interface_admin_up_down(vnet_main_t *vnm, u32 hw_if_index, u32 flags)
Definition: vhost-user.c:2468
#define vec_add2(V, P, N)
Add N elements to end of vector V, return pointer to new elements in P.
Definition: vec.h:557
uword unformat_user(unformat_input_t *input, unformat_function_t *func,...)
Definition: unformat.c:983
static vnet_sw_interface_t * vnet_get_sw_interface(vnet_main_t *vnm, u32 sw_if_index)
u8 * format(u8 *s, const char *fmt,...)
Definition: format.c:419
clib_error_t * show_vhost_user_command_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd)
Definition: vhost-user.c:3133
static void vlib_increment_simple_counter(vlib_simple_counter_main_t *cm, u32 thread_index, u32 index, u64 increment)
Increment a simple counter.
Definition: counter.h:78
unformat_function_t unformat_vnet_sw_interface
#define VNET_HW_INTERFACE_FLAG_LINK_UP
Definition: interface.h:394
static char * vhost_user_input_func_error_strings[]
Definition: vhost-user.c:159
static char * vhost_user_tx_func_error_strings[]
Definition: vhost-user.c:137
#define pool_get(P, E)
Allocate an object E from a pool P (unspecified alignment).
Definition: pool.h:225
vring_used_t * used
Definition: vhost-user.h:201
format_function_t format_vnet_sw_if_index_name
vhost_trace_t * current_trace
Definition: vhost-user.h:293
static uword vlib_process_suspend_time_is_zero(f64 dt)
Returns TRUE if a process suspend time is less than 10us.
Definition: node_funcs.h:436
#define vec_reset_length(v)
Reset vector length to zero NULL-pointer tolerant.
static int vhost_user_name_renumber(vnet_hw_interface_t *hi, u32 new_dev_instance)
Definition: vhost-user.c:193
clib_file_function_t * read_function
Definition: file.h:63
#define vlib_worker_thread_barrier_sync(X)
Definition: threads.h:212
static void vhost_user_vui_init(vnet_main_t *vnm, vhost_user_intf_t *vui, int server_sock_fd, const char *sock_filename, u64 feature_mask, u32 *sw_if_index)
Definition: vhost-user.c:2801
static vnet_sw_interface_t * vnet_get_hw_sw_interface(vnet_main_t *vnm, u32 hw_if_index)
static void vlib_trace_buffer(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, vlib_buffer_t *b, int follow_chain)
Definition: trace_funcs.h:104
static clib_error_t * vhost_user_socket_error(clib_file_t *uf)
Definition: vhost-user.c:1181
clib_file_t * file_pool
Definition: file.h:76
#define VHOST_VRING_F_LOG
Definition: vhost-user.h:33
vnet_hw_interface_rx_mode
Definition: interface.h:51
VNET_DEVICE_CLASS(vhost_user_dev_class, static)
#define VLIB_BUFFER_NEXT_PRESENT
Definition: buffer.h:95
static clib_error_t * vhost_user_socket_read(clib_file_t *uf)
Definition: vhost-user.c:667
static u8 * format_vhost_user_interface_name(u8 *s, va_list *args)
Definition: vhost-user.c:176
#define static_always_inline
Definition: clib.h:85
#define pool_foreach(VAR, POOL, BODY)
Iterate through pool.
Definition: pool.h:437
#define vlib_prefetch_buffer_with_index(vm, bi, type)
Prefetch buffer metadata by buffer index The first 64 bytes of buffer contains most header informatio...
Definition: buffer_funcs.h:180
#define VLIB_INIT_FUNCTION(x)
Definition: init.h:111
static uword vlib_process_get_events(vlib_main_t *vm, uword **data_vector)
Return the first event type which has occurred and a vector of per-event data of that type...
Definition: node_funcs.h:542
static clib_error_t * ip4_init(vlib_main_t *vm)
Definition: ip4_input.c:464
static uword format_get_indent(u8 *s)
Definition: format.h:72
vlib_combined_counter_main_t * combined_sw_if_counters
Definition: interface.h:668
u8 * format_white_space(u8 *s, va_list *va)
Definition: std-formats.c:113
void * log_base_addr
Definition: vhost-user.h:252
static_always_inline void vnet_device_input_set_interrupt_pending(vnet_main_t *vnm, u32 hw_if_index, u16 queue_id)
Definition: devices.h:135
#define VLIB_BUFFER_TOTAL_LENGTH_VALID
Definition: buffer.h:97
#define foreach_protocol_feature
#define vec_elt_at_index(v, i)
Get vector value at index i checking that i is in bounds.
#define clib_error_return(e, args...)
Definition: error.h:99
clib_file_main_t file_main
Definition: main.c:63
vhost_user_tx_func_error_t
Definition: vhost-user.c:129
unsigned long u64
Definition: types.h:89
static void unmap_all_mem_regions(vhost_user_intf_t *vui)
Definition: vhost-user.c:304
static void vhost_user_set_interrupt_pending(vhost_user_intf_t *vui, u32 ifq)
Definition: vhost-user.c:474
vhost_user_input_func_error_t
Definition: vhost-user.c:151
#define vlib_call_init_function(vm, x)
Definition: init.h:162
#define VHOST_USER_TX_COPY_THRESHOLD
Definition: vhost-user.c:98
static uword pointer_to_uword(const void *p)
Definition: types.h:131
#define UNIX_GET_FD(unixfd_idx)
Definition: vhost-user.c:100
unformat_function_t unformat_line_input
Definition: format.h:281
static int vhost_user_init_server_sock(const char *sock_filename, int *sock_fd)
Open server unix socket on specified sock_filename.
Definition: vhost-user.c:2723
static uword vhost_user_send_interrupt_process(vlib_main_t *vm, vlib_node_runtime_t *rt, vlib_frame_t *f)
Definition: vhost-user.c:2308
VLIB_DEVICE_TX_FUNCTION_MULTIARCH(vhost_user_dev_class, vhost_user_tx)
Definition: vhost-user.c:2499
#define VHOST_USER_EVENT_STOP_TIMER
Definition: vhost-user.h:218
static void vhost_user_vring_unlock(vhost_user_intf_t *vui, u32 qid)
Unlock the vring lock.
Definition: vhost-user.c:549
format_function_t format_vnet_sw_interface_name
#define pool_elt_at_index(p, i)
Returns pointer to element at given index.
Definition: pool.h:458
static_always_inline uword vnet_get_device_input_thread_index(vnet_main_t *vnm, u32 hw_if_index, u16 queue_id)
Definition: devices.h:126
u16 state
Input node state.
Definition: node.h:452
vlib_main_t * vlib_main
Definition: vnet.h:78
u16 current_length
Nbytes between current data and the end of this buffer.
Definition: buffer.h:72
static void vlib_process_signal_event(vlib_main_t *vm, uword node_index, uword type_opaque, uword data)
Definition: node_funcs.h:950
int vhost_user_delete_if(vnet_main_t *vnm, vlib_main_t *vm, u32 sw_if_index)
Definition: vhost-user.c:2644
static void * map_user_mem(vhost_user_intf_t *vui, uword addr)
Definition: vhost-user.c:279
u32 random
Pseudo random iterator.
Definition: vhost-user.h:310
static_always_inline void vhost_user_log_dirty_pages(vhost_user_intf_t *vui, u64 addr, u64 len)
Definition: vhost-user.c:655
uword mhash_set_mem(mhash_t *h, void *key, uword *new_value, uword *old_value)
Definition: mhash.c:271
struct _unformat_input_t unformat_input_t
#define VIRTQ_DESC_F_INDIRECT
Definition: vhost-user.h:28
#define clib_error_return_unix(e, args...)
Definition: error.h:102
static void * vlib_buffer_get_current(vlib_buffer_t *b)
Get pointer to current data to process.
Definition: buffer.h:193
int vnet_hw_interface_get_rx_mode(vnet_main_t *vnm, u32 hw_if_index, u16 queue_id, vnet_hw_interface_rx_mode *mode)
Definition: devices.c:302
#define VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX
Definition: buffer.h:422
#define pool_put(P, E)
Free an object E in pool P.
Definition: pool.h:270
format_function_t format_vnet_hw_interface_rx_mode
void vhost_user_tx_trace(vhost_trace_t *t, vhost_user_intf_t *vui, u16 qid, vlib_buffer_t *b, vhost_user_vring_t *rxvq)
Definition: vhost-user.c:1898
#define PREDICT_FALSE(x)
Definition: clib.h:97
#define VLIB_CONFIG_FUNCTION(x, n,...)
Definition: init.h:119
#define vhost_user_log_dirty_ring(vui, vq, member)
Definition: vhost-user.c:660
static vlib_node_registration_t vhost_user_process_node
(constructor) VLIB_REGISTER_NODE (vhost_user_process_node)
Definition: vhost-user.c:2603
void vhost_user_unmap_all(void)
Definition: vhost-user.c:3652
char sock_filename[256]
Definition: vhost-user.h:227
vnet_main_t vnet_main
Definition: misc.c:43
#define VLIB_FRAME_SIZE
Definition: node.h:328
vlib_simple_counter_main_t * sw_if_counters
Definition: interface.h:667
u32 region_mmap_fd[VHOST_MEMORY_MAX_NREGIONS]
Definition: vhost-user.h:243
static void vhost_user_send_call(vlib_main_t *vm, vhost_user_vring_t *vq)
Definition: vhost-user.c:1342
#define VNET_HW_INTERFACE_FLAG_SUPPORTS_INT_MODE
Definition: interface.h:420
u32 node_index
Node index.
Definition: node.h:437
vhost_user_memory_region_t regions[VHOST_MEMORY_MAX_NREGIONS]
Definition: vhost-user.h:239
#define vlib_validate_buffer_enqueue_x1(vm, node, next_index, to_next, n_left_to_next, bi0, next0)
Finish enqueueing one buffer forward in the graph.
Definition: buffer_node.h:216
#define vlib_get_next_frame(vm, node, next_index, vectors, n_vectors_left)
Get pointer to next frame vector data by (vlib_node_runtime_t, next_index).
Definition: node_funcs.h:364
int vhost_user_dump_ifs(vnet_main_t *vnm, vlib_main_t *vm, vhost_user_intf_details_t **out_vuids)
Definition: vhost-user.c:3082
vlib_error_t error
Error code for buffers to be enqueued to error handler.
Definition: buffer.h:113
static clib_error_t * vhost_user_exit(vlib_main_t *vm)
Definition: vhost-user.c:2701
static void vhost_user_tx_thread_placement(vhost_user_intf_t *vui)
Definition: vhost-user.c:340
static void vhost_user_vring_init(vhost_user_intf_t *vui, u32 qid)
Definition: vhost-user.c:555
u8 * format_ethernet_header_with_length(u8 *s, va_list *args)
Definition: format.c:91
static vlib_node_registration_t vhost_user_send_interrupt_node
(constructor) VLIB_REGISTER_NODE (vhost_user_send_interrupt_node)
Definition: vhost-user.c:2397
u32 * show_dev_instance_by_real_dev_instance
Definition: vhost-user.h:301
int vhost_user_create_if(vnet_main_t *vnm, vlib_main_t *vm, const char *sock_filename, u8 is_server, u32 *sw_if_index, u64 feature_mask, u8 renumber, u32 custom_dev_instance, u8 *hwaddr)
Definition: vhost-user.c:2861
u16 device_index
The interface queue index (Not the virtio vring idx)
Definition: vhost-user.h:273
vhost_user_intf_t * vhost_user_interfaces
Definition: vhost-user.h:300
static void mhash_init_c_string(mhash_t *h, uword n_value_bytes)
Definition: mhash.h:78
#define UNFORMAT_END_OF_INPUT
Definition: format.h:143
u16 n_vectors
Definition: node.h:344
static_always_inline uword vlib_get_thread_index(void)
Definition: threads.h:221
#define CLIB_PREFETCH(addr, size, type)
Definition: cache.h:82
vlib_main_t * vm
Definition: buffer.c:283
static_always_inline void vhost_user_log_dirty_pages_2(vhost_user_intf_t *vui, u64 addr, u64 len, u8 is_host_address)
Definition: vhost-user.c:627
#define vec_free(V)
Free vector&#39;s memory (no header).
Definition: vec.h:336
static int vhost_user_vring_try_lock(vhost_user_intf_t *vui, u32 qid)
Try once to lock the vring.
Definition: vhost-user.c:530
#define VLIB_MAIN_LOOP_EXIT_FUNCTION(x)
Definition: init.h:116
int vhost_user_modify_if(vnet_main_t *vnm, vlib_main_t *vm, const char *sock_filename, u8 is_server, u32 sw_if_index, u64 feature_mask, u8 renumber, u32 custom_dev_instance)
Definition: vhost-user.c:2919
#define clib_warning(format, args...)
Definition: error.h:59
#define VLIB_BUFFER_IS_TRACED
Definition: buffer.h:93
#define clib_memcpy(a, b, c)
Definition: string.h:69
#define VHOST_MEMORY_MAX_NREGIONS
Definition: vhost-user.h:19
static_always_inline void * map_guest_mem(vhost_user_intf_t *vui, uword addr, u32 *hint)
Definition: vhost-user.c:210
clib_error_t * debug_vhost_user_command_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd)
Definition: vhost-user.c:3570
static u32 vlib_buffer_alloc_from_free_list(vlib_main_t *vm, u32 *buffers, u32 n_buffers, u32 free_list_index)
Allocate buffers from specific freelist into supplied array.
Definition: buffer_funcs.h:296
u32 nregions
Definition: vhost-user.h:76
#define ARRAY_LEN(x)
Definition: clib.h:59
void vlib_put_next_frame(vlib_main_t *vm, vlib_node_runtime_t *r, u32 next_index, u32 n_vectors_left)
Release pointer to next frame vector data.
Definition: main.c:454
static uword vhost_user_tx(vlib_main_t *vm, vlib_node_runtime_t *node, vlib_frame_t *frame)
Definition: vhost-user.c:1981
u16 first_desc_len
Runtime queue flags.
Definition: vhost-user.h:275
#define VHOST_USER_PROTOCOL_F_LOG_SHMFD
Definition: vhost-user.h:32
#define VLIB_BUFFER_DATA_SIZE
Definition: buffer.h:51
static void vhost_user_input_rewind_buffers(vlib_main_t *vm, vhost_cpu_t *cpu, vlib_buffer_t *b_head)
Definition: vhost-user.c:1449
#define VLIB_CLI_COMMAND(x,...)
Definition: cli.h:154
#define VHOST_USER_VRING_NOFD_MASK
Definition: vhost-user.h:26
#define VNET_SW_INTERFACE_FLAG_ADMIN_UP
Definition: interface.h:572
u32 max_l3_packet_bytes[VLIB_N_RX_TX]
Definition: interface.h:468
u32 rx_buffers[VHOST_USER_RX_BUFFERS_N]
Definition: vhost-user.h:286
uword unformat_ethernet_address(unformat_input_t *input, va_list *args)
Definition: format.c:227
#define ASSERT(truth)
#define VHOST_USER_RX_BUFFER_STARVATION
Definition: vhost-user.c:78
unsigned int u32
Definition: types.h:88
static uword * mhash_get(mhash_t *h, const void *key)
Definition: mhash.h:110
static uword clib_file_add(clib_file_main_t *um, clib_file_t *template)
Definition: file.h:84
static long get_huge_page_size(int fd)
Definition: vhost-user.c:296
u32 next_buffer
Next buffer for this linked-list of buffers.
Definition: buffer.h:109
clib_error_t * vhost_user_delete_command_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd)
Definition: vhost-user.c:3035
static void clib_mem_free(void *p)
Definition: mem.h:179
#define VIRTQ_DESC_F_NEXT
Definition: vhost-user.h:27
volatile u32 * vring_locks[VHOST_VRING_MAX_N]
Definition: vhost-user.h:247
clib_error_t * ethernet_register_interface(vnet_main_t *vnm, u32 dev_class_index, u32 dev_instance, u8 *address, u32 *hw_if_index_return, ethernet_flag_change_function_t flag_change)
Definition: interface.c:273
mhash_t if_index_by_sock_name
Definition: vhost-user.h:298
virtio_trace_flag_t
Definition: vhost-user.c:111
#define clib_error_report(e)
Definition: error.h:113
static void * vlib_frame_args(vlib_frame_t *f)
Get pointer to frame scalar data.
Definition: node_funcs.h:284
static clib_error_t * vhost_user_kickfd_read_ready(clib_file_t *uf)
Definition: vhost-user.c:501
static u8 * format_vhost_trace(u8 *s, va_list *va)
Definition: vhost-user.c:1258
#define VHOST_USER_RX_COPY_THRESHOLD
Definition: vhost-user.c:88
static void vhost_user_vring_lock(vhost_user_intf_t *vui, u32 qid)
Spin until the vring is successfully locked.
Definition: vhost-user.c:539
static vlib_main_t * vlib_get_main(void)
Definition: global_funcs.h:23
static void vhost_user_rx_thread_placement()
Unassign existing interface/queue to thread mappings and re-assign new interface/queue to thread mapp...
Definition: vhost-user.c:380
u64 uword
Definition: types.h:112
static void * vlib_add_trace(vlib_main_t *vm, vlib_node_runtime_t *r, vlib_buffer_t *b, u32 n_data_bytes)
Definition: trace_funcs.h:55
static void vhost_user_update_iface_state(vhost_user_intf_t *vui)
Definition: vhost-user.c:456
u32 total_length_not_including_first_buffer
Only valid for first buffer in chain.
Definition: buffer.h:141
#define foreach_vhost_user_tx_func_error
Definition: vhost-user.c:120
void * region_mmap_addr[VHOST_MEMORY_MAX_NREGIONS]
Definition: vhost-user.h:240
Definition: defs.h:47
unsigned short u16
Definition: types.h:57
static u32 vhost_user_if_input(vlib_main_t *vm, vhost_user_main_t *vum, vhost_user_intf_t *vui, u16 qid, vlib_node_runtime_t *node, vnet_hw_interface_rx_mode mode)
Definition: vhost-user.c:1468
static clib_error_t * vhost_user_config(vlib_main_t *vm, unformat_input_t *input)
Definition: vhost-user.c:3628
#define vec_len(v)
Number of elements in vector (rvalue-only, NULL tolerant)
double f64
Definition: types.h:142
#define VRING_USED_F_NO_NOTIFY
Definition: vhost-user.h:45
#define VHOST_USER_RX_BUFFERS_N
Definition: vhost-user.h:280
unsigned char u8
Definition: types.h:56
static clib_error_t * vhost_user_socksvr_accept_ready(clib_file_t *uf)
Definition: vhost-user.c:1197
int vhost_user_intf_ready(vhost_user_intf_t *vui)
Returns whether at least one TX and one RX vring are enabled.
Definition: vhost-user.c:444
vhost_user_vring_t vrings[VHOST_VRING_MAX_N]
Definition: vhost-user.h:246
#define VLIB_BUFFER_TRACE_TRAJECTORY_INIT(b)
Definition: buffer.h:517
#define VHOST_VRING_MAX_N
Definition: vhost-user.h:22
static void unformat_free(unformat_input_t *i)
Definition: format.h:161
#define VLIB_NODE_FLAG_SWITCH_FROM_POLLING_TO_INTERRUPT_MODE
Definition: node.h:262
#define clib_unix_warning(format, args...)
Definition: error.h:68
vlib_node_registration_t vhost_user_input_node
(constructor) VLIB_REGISTER_NODE (vhost_user_input_node)
Definition: vhost-user.c:118
u32 rx_buffers_len
Definition: vhost-user.h:285
#define DBG_SOCK(args...)
Definition: vhost-user.c:54
u32 vhost_user_rx_discard_packet(vlib_main_t *vm, vhost_user_intf_t *vui, vhost_user_vring_t *txvq, u32 discard_max)
Try to discard packets from the tx ring (VPP RX path).
Definition: vhost-user.c:1410
static vhost_user_main_t vhost_user_main
Definition: vhost-user.c:166
static void * clib_mem_alloc_aligned(uword size, uword align)
Definition: mem.h:120
#define vnet_buffer(b)
Definition: buffer.h:306
#define DBG_VQ(args...)
Definition: vhost-user.c:64
#define VLIB_NODE_FUNCTION_MULTIARCH(node, fn)
Definition: node.h:158
static u32 random_u32(u32 *seed)
32-bit random number generator
Definition: random.h:69
u8 * format_unformat_error(u8 *s, va_list *va)
Definition: unformat.c:91
void vlib_worker_thread_barrier_release(vlib_main_t *vm)
Definition: threads.c:1488
static_always_inline void vnet_feature_start_device_input_x1(u32 sw_if_index, u32 *next0, vlib_buffer_t *b0)
Definition: feature.h:227
#define VLIB_REGISTER_NODE(x,...)
Definition: node.h:143
static vlib_thread_main_t * vlib_get_thread_main()
Definition: global_funcs.h:32
#define VLIB_NODE_FLAG_SWITCH_FROM_INTERRUPT_TO_POLLING_MODE
Definition: node.h:261
u64 region_guest_addr_lo[VHOST_MEMORY_MAX_NREGIONS]
Definition: vhost-user.h:241
#define vec_foreach(var, vec)
Vector iterator.
uword private_data
Definition: file.h:60
#define foreach_vhost_user_input_func_error
Definition: vhost-user.c:143
u16 flags
Copy of main node flags.
Definition: node.h:450
#define CLIB_MEMORY_BARRIER()
Definition: clib.h:101
Definition: file.h:50
vhost_vring_addr_t addr
Definition: vhost-user.h:83
virtio_net_hdr_mrg_rxbuf_t tx_headers[VLIB_FRAME_SIZE]
Definition: vhost-user.h:288
static void vlib_set_trace_count(vlib_main_t *vm, vlib_node_runtime_t *rt, u32 count)
Definition: trace_funcs.h:159
u32 flags
Definition: vhost-user.h:77
#define vec_validate_init_empty(V, I, INIT)
Make sure vector is long enough for given index and initialize empty space (no header, unspecified alignment)
Definition: vec.h:481
#define CLIB_CACHE_LINE_BYTES
Definition: cache.h:67
u32 flags
buffer flags: VLIB_BUFFER_FREE_LIST_INDEX_MASK: bits used to store free list index, VLIB_BUFFER_IS_TRACED: trace this buffer.
Definition: buffer.h:75
static_always_inline u32 vhost_user_tx_copy(vhost_user_intf_t *vui, vhost_copy_t *cpy, u16 copy_len, u32 *map_hint)
Definition: vhost-user.c:1933
void vlib_cli_output(vlib_main_t *vm, char *fmt,...)
Definition: cli.c:680
static clib_error_t * vhost_user_interface_rx_mode_change(vnet_main_t *vnm, u32 hw_if_index, u32 qid, vnet_hw_interface_rx_mode mode)
Definition: vhost-user.c:2405
#define VHOST_USER_PROTOCOL_F_MQ
Definition: vhost-user.h:31
#define VHOST_LOG_PAGE
Definition: vhost-user.c:625
static vlib_buffer_t * vlib_get_buffer(vlib_main_t *vm, u32 buffer_index)
Translate buffer index into buffer pointer.
Definition: buffer_funcs.h:57
clib_error_t * vhost_user_connect_command_fn(vlib_main_t *vm, unformat_input_t *input, vlib_cli_command_t *cmd)
Definition: vhost-user.c:2970
VNET_HW_INTERFACE_CLASS(vhost_interface_class, static)
uword unformat(unformat_input_t *i, const char *fmt,...)
Definition: unformat.c:972
Definition: defs.h:46
static clib_error_t * vhost_user_callfd_read_ready(clib_file_t *uf)
Definition: vhost-user.c:490
static uword unformat_check_input(unformat_input_t *i)
Definition: format.h:169
int dont_dump_vhost_user_memory
Definition: vhost-user.h:304
#define VHOST_VRING_IDX_RX(qid)
Definition: vhost-user.h:23