Commit | Line | Data |
---|---|---|
3a4d5c94 MT |
1 | /* Copyright (C) 2009 Red Hat, Inc. |
2 | * Author: Michael S. Tsirkin <mst@redhat.com> | |
3 | * | |
4 | * This work is licensed under the terms of the GNU GPL, version 2. | |
5 | * | |
6 | * virtio-net server in host kernel. | |
7 | */ | |
8 | ||
9 | #include <linux/compat.h> | |
10 | #include <linux/eventfd.h> | |
11 | #include <linux/vhost.h> | |
12 | #include <linux/virtio_net.h> | |
3a4d5c94 MT |
13 | #include <linux/miscdevice.h> |
14 | #include <linux/module.h> | |
bab632d6 | 15 | #include <linux/moduleparam.h> |
3a4d5c94 MT |
16 | #include <linux/mutex.h> |
17 | #include <linux/workqueue.h> | |
18 | #include <linux/rcupdate.h> | |
19 | #include <linux/file.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
3a4d5c94 MT |
21 | |
22 | #include <linux/net.h> | |
23 | #include <linux/if_packet.h> | |
24 | #include <linux/if_arp.h> | |
25 | #include <linux/if_tun.h> | |
501c774c | 26 | #include <linux/if_macvlan.h> |
c53cff5e | 27 | #include <linux/if_vlan.h> |
3a4d5c94 MT |
28 | |
29 | #include <net/sock.h> | |
30 | ||
31 | #include "vhost.h" | |
32 | ||
f9611c43 | 33 | static int experimental_zcopytx = 1; |
bab632d6 | 34 | module_param(experimental_zcopytx, int, 0444); |
f9611c43 MT |
35 | MODULE_PARM_DESC(experimental_zcopytx, "Enable Zero Copy TX;" |
36 | " 1 -Enable; 0 - Disable"); | |
bab632d6 | 37 | |
3a4d5c94 MT |
38 | /* Max number of bytes transferred before requeueing the job. |
39 | * Using this limit prevents one virtqueue from starving others. */ | |
40 | #define VHOST_NET_WEIGHT 0x80000 | |
41 | ||
bab632d6 MT |
42 | /* MAX number of TX used buffers for outstanding zerocopy */ |
43 | #define VHOST_MAX_PEND 128 | |
44 | #define VHOST_GOODCOPY_LEN 256 | |
45 | ||
eaae8132 MT |
46 | /* |
47 | * For transmit, used buffer len is unused; we override it to track buffer | |
48 | * status internally; used for zerocopy tx only. | |
49 | */ | |
50 | /* Lower device DMA failed */ | |
51 | #define VHOST_DMA_FAILED_LEN 3 | |
52 | /* Lower device DMA done */ | |
53 | #define VHOST_DMA_DONE_LEN 2 | |
54 | /* Lower device DMA in progress */ | |
55 | #define VHOST_DMA_IN_PROGRESS 1 | |
56 | /* Buffer unused */ | |
57 | #define VHOST_DMA_CLEAR_LEN 0 | |
58 | ||
59 | #define VHOST_DMA_IS_DONE(len) ((len) >= VHOST_DMA_DONE_LEN) | |
60 | ||
8570a6e7 AH |
61 | enum { |
62 | VHOST_NET_FEATURES = VHOST_FEATURES | | |
63 | (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) | | |
64 | (1ULL << VIRTIO_NET_F_MRG_RXBUF), | |
65 | }; | |
66 | ||
3a4d5c94 MT |
67 | enum { |
68 | VHOST_NET_VQ_RX = 0, | |
69 | VHOST_NET_VQ_TX = 1, | |
70 | VHOST_NET_VQ_MAX = 2, | |
71 | }; | |
72 | ||
fe729a57 | 73 | struct vhost_net_ubuf_ref { |
2839400f AH |
74 | struct kref kref; |
75 | wait_queue_head_t wait; | |
76 | struct vhost_virtqueue *vq; | |
77 | }; | |
78 | ||
3ab2e420 AH |
79 | struct vhost_net_virtqueue { |
80 | struct vhost_virtqueue vq; | |
81f95a55 MT |
81 | /* hdr is used to store the virtio header. |
82 | * Since each iovec has >= 1 byte length, we never need more than | |
83 | * header length entries to store the header. */ | |
84 | struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)]; | |
85 | size_t vhost_hlen; | |
86 | size_t sock_hlen; | |
2839400f AH |
87 | /* vhost zerocopy support fields below: */ |
88 | /* last used idx for outstanding DMA zerocopy buffers */ | |
89 | int upend_idx; | |
90 | /* first used idx for DMA done zerocopy buffers */ | |
91 | int done_idx; | |
92 | /* an array of userspace buffers info */ | |
93 | struct ubuf_info *ubuf_info; | |
94 | /* Reference counting for outstanding ubufs. | |
95 | * Protected by vq mutex. Writers must also take device mutex. */ | |
fe729a57 | 96 | struct vhost_net_ubuf_ref *ubufs; |
3ab2e420 AH |
97 | }; |
98 | ||
3a4d5c94 MT |
99 | struct vhost_net { |
100 | struct vhost_dev dev; | |
3ab2e420 | 101 | struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX]; |
3a4d5c94 | 102 | struct vhost_poll poll[VHOST_NET_VQ_MAX]; |
eaae8132 MT |
103 | /* Number of TX recently submitted. |
104 | * Protected by tx vq lock. */ | |
105 | unsigned tx_packets; | |
106 | /* Number of times zerocopy TX recently failed. | |
107 | * Protected by tx vq lock. */ | |
108 | unsigned tx_zcopy_err; | |
1280c27f MT |
109 | /* Flush in progress. Protected by tx vq lock. */ |
110 | bool tx_flush; | |
3a4d5c94 MT |
111 | }; |
112 | ||
fe729a57 | 113 | static unsigned vhost_net_zcopy_mask __read_mostly; |
2839400f | 114 | |
fe729a57 | 115 | static void vhost_net_enable_zcopy(int vq) |
2839400f | 116 | { |
fe729a57 | 117 | vhost_net_zcopy_mask |= 0x1 << vq; |
2839400f AH |
118 | } |
119 | ||
fe729a57 | 120 | static void vhost_net_zerocopy_done_signal(struct kref *kref) |
2839400f | 121 | { |
fe729a57 AH |
122 | struct vhost_net_ubuf_ref *ubufs; |
123 | ||
124 | ubufs = container_of(kref, struct vhost_net_ubuf_ref, kref); | |
2839400f AH |
125 | wake_up(&ubufs->wait); |
126 | } | |
127 | ||
fe729a57 AH |
128 | static struct vhost_net_ubuf_ref * |
129 | vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy) | |
2839400f | 130 | { |
fe729a57 | 131 | struct vhost_net_ubuf_ref *ubufs; |
2839400f AH |
132 | /* No zero copy backend? Nothing to count. */ |
133 | if (!zcopy) | |
134 | return NULL; | |
135 | ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL); | |
136 | if (!ubufs) | |
137 | return ERR_PTR(-ENOMEM); | |
138 | kref_init(&ubufs->kref); | |
139 | init_waitqueue_head(&ubufs->wait); | |
140 | ubufs->vq = vq; | |
141 | return ubufs; | |
142 | } | |
143 | ||
fe729a57 | 144 | static void vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 145 | { |
fe729a57 | 146 | kref_put(&ubufs->kref, vhost_net_zerocopy_done_signal); |
2839400f AH |
147 | } |
148 | ||
fe729a57 | 149 | static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 150 | { |
fe729a57 | 151 | kref_put(&ubufs->kref, vhost_net_zerocopy_done_signal); |
2839400f | 152 | wait_event(ubufs->wait, !atomic_read(&ubufs->kref.refcount)); |
f5ce1d25 MT |
153 | } |
154 | ||
155 | static void vhost_net_ubuf_put_wait_and_free(struct vhost_net_ubuf_ref *ubufs) | |
156 | { | |
157 | vhost_net_ubuf_put_and_wait(ubufs); | |
2839400f AH |
158 | kfree(ubufs); |
159 | } | |
160 | ||
b1ad8496 AH |
161 | static void vhost_net_clear_ubuf_info(struct vhost_net *n) |
162 | { | |
b1ad8496 AH |
163 | int i; |
164 | ||
288cfe78 MT |
165 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
166 | kfree(n->vqs[i].ubuf_info); | |
167 | n->vqs[i].ubuf_info = NULL; | |
b1ad8496 AH |
168 | } |
169 | } | |
170 | ||
2839400f AH |
171 | int vhost_net_set_ubuf_info(struct vhost_net *n) |
172 | { | |
173 | bool zcopy; | |
174 | int i; | |
175 | ||
288cfe78 | 176 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
fe729a57 | 177 | zcopy = vhost_net_zcopy_mask & (0x1 << i); |
2839400f AH |
178 | if (!zcopy) |
179 | continue; | |
180 | n->vqs[i].ubuf_info = kmalloc(sizeof(*n->vqs[i].ubuf_info) * | |
181 | UIO_MAXIOV, GFP_KERNEL); | |
182 | if (!n->vqs[i].ubuf_info) | |
183 | goto err; | |
184 | } | |
185 | return 0; | |
186 | ||
187 | err: | |
288cfe78 | 188 | vhost_net_clear_ubuf_info(n); |
2839400f AH |
189 | return -ENOMEM; |
190 | } | |
191 | ||
81f95a55 | 192 | void vhost_net_vq_reset(struct vhost_net *n) |
2839400f AH |
193 | { |
194 | int i; | |
195 | ||
288cfe78 MT |
196 | vhost_net_clear_ubuf_info(n); |
197 | ||
2839400f AH |
198 | for (i = 0; i < VHOST_NET_VQ_MAX; i++) { |
199 | n->vqs[i].done_idx = 0; | |
200 | n->vqs[i].upend_idx = 0; | |
201 | n->vqs[i].ubufs = NULL; | |
81f95a55 MT |
202 | n->vqs[i].vhost_hlen = 0; |
203 | n->vqs[i].sock_hlen = 0; | |
2839400f AH |
204 | } |
205 | ||
206 | } | |
207 | ||
eaae8132 MT |
208 | static void vhost_net_tx_packet(struct vhost_net *net) |
209 | { | |
210 | ++net->tx_packets; | |
211 | if (net->tx_packets < 1024) | |
212 | return; | |
213 | net->tx_packets = 0; | |
214 | net->tx_zcopy_err = 0; | |
215 | } | |
216 | ||
217 | static void vhost_net_tx_err(struct vhost_net *net) | |
218 | { | |
219 | ++net->tx_zcopy_err; | |
220 | } | |
221 | ||
222 | static bool vhost_net_tx_select_zcopy(struct vhost_net *net) | |
223 | { | |
1280c27f MT |
224 | /* TX flush waits for outstanding DMAs to be done. |
225 | * Don't start new DMAs. | |
226 | */ | |
227 | return !net->tx_flush && | |
228 | net->tx_packets / 64 >= net->tx_zcopy_err; | |
eaae8132 MT |
229 | } |
230 | ||
bab632d6 MT |
231 | static bool vhost_sock_zcopy(struct socket *sock) |
232 | { | |
233 | return unlikely(experimental_zcopytx) && | |
234 | sock_flag(sock->sk, SOCK_ZEROCOPY); | |
235 | } | |
236 | ||
3a4d5c94 MT |
237 | /* Pop first len bytes from iovec. Return number of segments used. */ |
238 | static int move_iovec_hdr(struct iovec *from, struct iovec *to, | |
239 | size_t len, int iov_count) | |
240 | { | |
241 | int seg = 0; | |
242 | size_t size; | |
d47effe1 | 243 | |
3a4d5c94 MT |
244 | while (len && seg < iov_count) { |
245 | size = min(from->iov_len, len); | |
246 | to->iov_base = from->iov_base; | |
247 | to->iov_len = size; | |
248 | from->iov_len -= size; | |
249 | from->iov_base += size; | |
250 | len -= size; | |
251 | ++from; | |
252 | ++to; | |
253 | ++seg; | |
254 | } | |
255 | return seg; | |
256 | } | |
8dd014ad DS |
257 | /* Copy iovec entries for len bytes from iovec. */ |
258 | static void copy_iovec_hdr(const struct iovec *from, struct iovec *to, | |
259 | size_t len, int iovcount) | |
260 | { | |
261 | int seg = 0; | |
262 | size_t size; | |
d47effe1 | 263 | |
8dd014ad DS |
264 | while (len && seg < iovcount) { |
265 | size = min(from->iov_len, len); | |
266 | to->iov_base = from->iov_base; | |
267 | to->iov_len = size; | |
268 | len -= size; | |
269 | ++from; | |
270 | ++to; | |
271 | ++seg; | |
272 | } | |
273 | } | |
3a4d5c94 | 274 | |
b211616d MT |
275 | /* In case of DMA done not in order in lower device driver for some reason. |
276 | * upend_idx is used to track end of used idx, done_idx is used to track head | |
277 | * of used idx. Once lower device DMA done contiguously, we will signal KVM | |
278 | * guest used idx. | |
279 | */ | |
eaae8132 MT |
280 | static int vhost_zerocopy_signal_used(struct vhost_net *net, |
281 | struct vhost_virtqueue *vq) | |
b211616d | 282 | { |
2839400f AH |
283 | struct vhost_net_virtqueue *nvq = |
284 | container_of(vq, struct vhost_net_virtqueue, vq); | |
b211616d MT |
285 | int i; |
286 | int j = 0; | |
287 | ||
2839400f | 288 | for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) { |
eaae8132 MT |
289 | if (vq->heads[i].len == VHOST_DMA_FAILED_LEN) |
290 | vhost_net_tx_err(net); | |
b211616d MT |
291 | if (VHOST_DMA_IS_DONE(vq->heads[i].len)) { |
292 | vq->heads[i].len = VHOST_DMA_CLEAR_LEN; | |
293 | vhost_add_used_and_signal(vq->dev, vq, | |
294 | vq->heads[i].id, 0); | |
295 | ++j; | |
296 | } else | |
297 | break; | |
298 | } | |
299 | if (j) | |
2839400f | 300 | nvq->done_idx = i; |
b211616d MT |
301 | return j; |
302 | } | |
303 | ||
eaae8132 | 304 | static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success) |
b211616d | 305 | { |
fe729a57 | 306 | struct vhost_net_ubuf_ref *ubufs = ubuf->ctx; |
b211616d | 307 | struct vhost_virtqueue *vq = ubufs->vq; |
24eb21a1 MT |
308 | int cnt = atomic_read(&ubufs->kref.refcount); |
309 | ||
bd35c1a7 JW |
310 | /* set len to mark this desc buffers done DMA */ |
311 | vq->heads[ubuf->desc].len = success ? | |
312 | VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN; | |
313 | vhost_net_ubuf_put(ubufs); | |
314 | ||
24eb21a1 MT |
315 | /* |
316 | * Trigger polling thread if guest stopped submitting new buffers: | |
317 | * in this case, the refcount after decrement will eventually reach 1 | |
318 | * so here it is 2. | |
319 | * We also trigger polling periodically after each 16 packets | |
320 | * (the value 16 here is more or less arbitrary, it's tuned to trigger | |
321 | * less than 10% of times). | |
322 | */ | |
323 | if (cnt <= 2 || !(cnt % 16)) | |
324 | vhost_poll_queue(&vq->poll); | |
b211616d MT |
325 | } |
326 | ||
3a4d5c94 MT |
327 | /* Expects to be always run from workqueue - which acts as |
328 | * read-size critical section for our kind of RCU. */ | |
329 | static void handle_tx(struct vhost_net *net) | |
330 | { | |
2839400f | 331 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; |
81f95a55 | 332 | struct vhost_virtqueue *vq = &nvq->vq; |
d5675bd2 MT |
333 | unsigned out, in, s; |
334 | int head; | |
3a4d5c94 MT |
335 | struct msghdr msg = { |
336 | .msg_name = NULL, | |
337 | .msg_namelen = 0, | |
338 | .msg_control = NULL, | |
339 | .msg_controllen = 0, | |
340 | .msg_iov = vq->iov, | |
341 | .msg_flags = MSG_DONTWAIT, | |
342 | }; | |
343 | size_t len, total_len = 0; | |
70181d51 | 344 | int err; |
3a4d5c94 | 345 | size_t hdr_size; |
28457ee6 | 346 | struct socket *sock; |
fe729a57 | 347 | struct vhost_net_ubuf_ref *uninitialized_var(ubufs); |
cedb9bdc | 348 | bool zcopy, zcopy_used; |
28457ee6 | 349 | |
5e18247b | 350 | /* TODO: check that we are running from vhost_worker? */ |
11cd1a8b | 351 | sock = rcu_dereference_check(vq->private_data, 1); |
3a4d5c94 MT |
352 | if (!sock) |
353 | return; | |
354 | ||
3a4d5c94 | 355 | mutex_lock(&vq->mutex); |
8ea8cf89 | 356 | vhost_disable_notify(&net->dev, vq); |
3a4d5c94 | 357 | |
81f95a55 | 358 | hdr_size = nvq->vhost_hlen; |
2839400f | 359 | zcopy = nvq->ubufs; |
3a4d5c94 MT |
360 | |
361 | for (;;) { | |
bab632d6 MT |
362 | /* Release DMAs done buffers first */ |
363 | if (zcopy) | |
eaae8132 | 364 | vhost_zerocopy_signal_used(net, vq); |
bab632d6 | 365 | |
3a4d5c94 MT |
366 | head = vhost_get_vq_desc(&net->dev, vq, vq->iov, |
367 | ARRAY_SIZE(vq->iov), | |
368 | &out, &in, | |
369 | NULL, NULL); | |
d5675bd2 | 370 | /* On error, stop handling until the next kick. */ |
7b3384fc | 371 | if (unlikely(head < 0)) |
d5675bd2 | 372 | break; |
3a4d5c94 MT |
373 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
374 | if (head == vq->num) { | |
9e380825 SM |
375 | int num_pends; |
376 | ||
9e380825 SM |
377 | /* If more outstanding DMAs, queue the work. |
378 | * Handle upend_idx wrap around | |
379 | */ | |
2839400f AH |
380 | num_pends = likely(nvq->upend_idx >= nvq->done_idx) ? |
381 | (nvq->upend_idx - nvq->done_idx) : | |
382 | (nvq->upend_idx + UIO_MAXIOV - | |
383 | nvq->done_idx); | |
70181d51 | 384 | if (unlikely(num_pends > VHOST_MAX_PEND)) |
bab632d6 | 385 | break; |
8ea8cf89 MT |
386 | if (unlikely(vhost_enable_notify(&net->dev, vq))) { |
387 | vhost_disable_notify(&net->dev, vq); | |
3a4d5c94 MT |
388 | continue; |
389 | } | |
390 | break; | |
391 | } | |
392 | if (in) { | |
393 | vq_err(vq, "Unexpected descriptor format for TX: " | |
394 | "out %d, int %d\n", out, in); | |
395 | break; | |
396 | } | |
397 | /* Skip header. TODO: support TSO. */ | |
81f95a55 | 398 | s = move_iovec_hdr(vq->iov, nvq->hdr, hdr_size, out); |
3a4d5c94 MT |
399 | msg.msg_iovlen = out; |
400 | len = iov_length(vq->iov, out); | |
401 | /* Sanity check */ | |
402 | if (!len) { | |
403 | vq_err(vq, "Unexpected header len for TX: " | |
404 | "%zd expected %zd\n", | |
81f95a55 | 405 | iov_length(nvq->hdr, s), hdr_size); |
3a4d5c94 MT |
406 | break; |
407 | } | |
cedb9bdc | 408 | zcopy_used = zcopy && (len >= VHOST_GOODCOPY_LEN || |
2839400f | 409 | nvq->upend_idx != nvq->done_idx); |
cedb9bdc | 410 | |
bab632d6 | 411 | /* use msg_control to pass vhost zerocopy ubuf info to skb */ |
cedb9bdc | 412 | if (zcopy_used) { |
2839400f | 413 | vq->heads[nvq->upend_idx].id = head; |
eaae8132 MT |
414 | if (!vhost_net_tx_select_zcopy(net) || |
415 | len < VHOST_GOODCOPY_LEN) { | |
bab632d6 | 416 | /* copy don't need to wait for DMA done */ |
2839400f | 417 | vq->heads[nvq->upend_idx].len = |
bab632d6 MT |
418 | VHOST_DMA_DONE_LEN; |
419 | msg.msg_control = NULL; | |
420 | msg.msg_controllen = 0; | |
421 | ubufs = NULL; | |
422 | } else { | |
46aa92d1 | 423 | struct ubuf_info *ubuf; |
2839400f | 424 | ubuf = nvq->ubuf_info + nvq->upend_idx; |
bab632d6 | 425 | |
2839400f | 426 | vq->heads[nvq->upend_idx].len = |
70e4cb9a | 427 | VHOST_DMA_IN_PROGRESS; |
bab632d6 | 428 | ubuf->callback = vhost_zerocopy_callback; |
2839400f AH |
429 | ubuf->ctx = nvq->ubufs; |
430 | ubuf->desc = nvq->upend_idx; | |
bab632d6 MT |
431 | msg.msg_control = ubuf; |
432 | msg.msg_controllen = sizeof(ubuf); | |
2839400f | 433 | ubufs = nvq->ubufs; |
bab632d6 MT |
434 | kref_get(&ubufs->kref); |
435 | } | |
2839400f | 436 | nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV; |
4364d5f9 JW |
437 | } else |
438 | msg.msg_control = NULL; | |
3a4d5c94 MT |
439 | /* TODO: Check specific error and bomb out unless ENOBUFS? */ |
440 | err = sock->ops->sendmsg(NULL, sock, &msg, len); | |
441 | if (unlikely(err < 0)) { | |
cedb9bdc | 442 | if (zcopy_used) { |
bab632d6 | 443 | if (ubufs) |
fe729a57 | 444 | vhost_net_ubuf_put(ubufs); |
2839400f AH |
445 | nvq->upend_idx = ((unsigned)nvq->upend_idx - 1) |
446 | % UIO_MAXIOV; | |
bab632d6 | 447 | } |
8dd014ad | 448 | vhost_discard_vq_desc(vq, 1); |
3a4d5c94 MT |
449 | break; |
450 | } | |
451 | if (err != len) | |
95c0ec6a MT |
452 | pr_debug("Truncated TX packet: " |
453 | " len %d != %zd\n", err, len); | |
cedb9bdc | 454 | if (!zcopy_used) |
bab632d6 | 455 | vhost_add_used_and_signal(&net->dev, vq, head, 0); |
c8fb217a | 456 | else |
eaae8132 | 457 | vhost_zerocopy_signal_used(net, vq); |
3a4d5c94 | 458 | total_len += len; |
eaae8132 | 459 | vhost_net_tx_packet(net); |
3a4d5c94 MT |
460 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { |
461 | vhost_poll_queue(&vq->poll); | |
462 | break; | |
463 | } | |
464 | } | |
465 | ||
466 | mutex_unlock(&vq->mutex); | |
3a4d5c94 MT |
467 | } |
468 | ||
8dd014ad DS |
469 | static int peek_head_len(struct sock *sk) |
470 | { | |
471 | struct sk_buff *head; | |
472 | int len = 0; | |
783e3988 | 473 | unsigned long flags; |
8dd014ad | 474 | |
783e3988 | 475 | spin_lock_irqsave(&sk->sk_receive_queue.lock, flags); |
8dd014ad | 476 | head = skb_peek(&sk->sk_receive_queue); |
c53cff5e | 477 | if (likely(head)) { |
8dd014ad | 478 | len = head->len; |
c53cff5e BG |
479 | if (vlan_tx_tag_present(head)) |
480 | len += VLAN_HLEN; | |
481 | } | |
482 | ||
783e3988 | 483 | spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags); |
8dd014ad DS |
484 | return len; |
485 | } | |
486 | ||
487 | /* This is a multi-buffer version of vhost_get_desc, that works if | |
488 | * vq has read descriptors only. | |
489 | * @vq - the relevant virtqueue | |
490 | * @datalen - data length we'll be reading | |
491 | * @iovcount - returned count of io vectors we fill | |
492 | * @log - vhost log | |
493 | * @log_num - log offset | |
94249369 | 494 | * @quota - headcount quota, 1 for big buffer |
8dd014ad DS |
495 | * returns number of buffer heads allocated, negative on error |
496 | */ | |
497 | static int get_rx_bufs(struct vhost_virtqueue *vq, | |
498 | struct vring_used_elem *heads, | |
499 | int datalen, | |
500 | unsigned *iovcount, | |
501 | struct vhost_log *log, | |
94249369 JW |
502 | unsigned *log_num, |
503 | unsigned int quota) | |
8dd014ad DS |
504 | { |
505 | unsigned int out, in; | |
506 | int seg = 0; | |
507 | int headcount = 0; | |
508 | unsigned d; | |
509 | int r, nlogs = 0; | |
510 | ||
94249369 | 511 | while (datalen > 0 && headcount < quota) { |
e0e9b406 | 512 | if (unlikely(seg >= UIO_MAXIOV)) { |
8dd014ad DS |
513 | r = -ENOBUFS; |
514 | goto err; | |
515 | } | |
57962c47 | 516 | r = vhost_get_vq_desc(vq->dev, vq, vq->iov + seg, |
8dd014ad DS |
517 | ARRAY_SIZE(vq->iov) - seg, &out, |
518 | &in, log, log_num); | |
57962c47 MT |
519 | if (unlikely(r < 0)) |
520 | goto err; | |
521 | ||
522 | d = r; | |
8dd014ad DS |
523 | if (d == vq->num) { |
524 | r = 0; | |
525 | goto err; | |
526 | } | |
527 | if (unlikely(out || in <= 0)) { | |
528 | vq_err(vq, "unexpected descriptor format for RX: " | |
529 | "out %d, in %d\n", out, in); | |
530 | r = -EINVAL; | |
531 | goto err; | |
532 | } | |
533 | if (unlikely(log)) { | |
534 | nlogs += *log_num; | |
535 | log += *log_num; | |
536 | } | |
537 | heads[headcount].id = d; | |
538 | heads[headcount].len = iov_length(vq->iov + seg, in); | |
539 | datalen -= heads[headcount].len; | |
540 | ++headcount; | |
541 | seg += in; | |
542 | } | |
543 | heads[headcount - 1].len += datalen; | |
544 | *iovcount = seg; | |
545 | if (unlikely(log)) | |
546 | *log_num = nlogs; | |
f78f1512 MT |
547 | |
548 | /* Detect overrun */ | |
549 | if (unlikely(datalen > 0)) { | |
550 | r = UIO_MAXIOV + 1; | |
551 | goto err; | |
552 | } | |
8dd014ad DS |
553 | return headcount; |
554 | err: | |
555 | vhost_discard_vq_desc(vq, headcount); | |
556 | return r; | |
557 | } | |
558 | ||
3a4d5c94 MT |
559 | /* Expects to be always run from workqueue - which acts as |
560 | * read-size critical section for our kind of RCU. */ | |
94249369 | 561 | static void handle_rx(struct vhost_net *net) |
3a4d5c94 | 562 | { |
81f95a55 MT |
563 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX]; |
564 | struct vhost_virtqueue *vq = &nvq->vq; | |
8dd014ad DS |
565 | unsigned uninitialized_var(in), log; |
566 | struct vhost_log *vq_log; | |
567 | struct msghdr msg = { | |
568 | .msg_name = NULL, | |
569 | .msg_namelen = 0, | |
570 | .msg_control = NULL, /* FIXME: get and handle RX aux data. */ | |
571 | .msg_controllen = 0, | |
572 | .msg_iov = vq->iov, | |
573 | .msg_flags = MSG_DONTWAIT, | |
574 | }; | |
8dd014ad DS |
575 | struct virtio_net_hdr_mrg_rxbuf hdr = { |
576 | .hdr.flags = 0, | |
577 | .hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE | |
578 | }; | |
8dd014ad | 579 | size_t total_len = 0; |
910a578f MT |
580 | int err, mergeable; |
581 | s16 headcount; | |
8dd014ad DS |
582 | size_t vhost_hlen, sock_hlen; |
583 | size_t vhost_len, sock_len; | |
5e18247b MT |
584 | /* TODO: check that we are running from vhost_worker? */ |
585 | struct socket *sock = rcu_dereference_check(vq->private_data, 1); | |
d47effe1 | 586 | |
de4d768a | 587 | if (!sock) |
8dd014ad DS |
588 | return; |
589 | ||
8dd014ad | 590 | mutex_lock(&vq->mutex); |
8ea8cf89 | 591 | vhost_disable_notify(&net->dev, vq); |
81f95a55 MT |
592 | vhost_hlen = nvq->vhost_hlen; |
593 | sock_hlen = nvq->sock_hlen; | |
8dd014ad DS |
594 | |
595 | vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ? | |
596 | vq->log : NULL; | |
cfbdab95 | 597 | mergeable = vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF); |
8dd014ad DS |
598 | |
599 | while ((sock_len = peek_head_len(sock->sk))) { | |
600 | sock_len += sock_hlen; | |
601 | vhost_len = sock_len + vhost_hlen; | |
602 | headcount = get_rx_bufs(vq, vq->heads, vhost_len, | |
94249369 JW |
603 | &in, vq_log, &log, |
604 | likely(mergeable) ? UIO_MAXIOV : 1); | |
8dd014ad DS |
605 | /* On error, stop handling until the next kick. */ |
606 | if (unlikely(headcount < 0)) | |
607 | break; | |
f78f1512 MT |
608 | /* On overrun, truncate and discard */ |
609 | if (unlikely(headcount > UIO_MAXIOV)) { | |
610 | msg.msg_iovlen = 1; | |
611 | err = sock->ops->recvmsg(NULL, sock, &msg, | |
612 | 1, MSG_DONTWAIT | MSG_TRUNC); | |
613 | pr_debug("Discarded rx packet: len %zd\n", sock_len); | |
614 | continue; | |
615 | } | |
8dd014ad DS |
616 | /* OK, now we need to know about added descriptors. */ |
617 | if (!headcount) { | |
8ea8cf89 | 618 | if (unlikely(vhost_enable_notify(&net->dev, vq))) { |
8dd014ad DS |
619 | /* They have slipped one in as we were |
620 | * doing that: check again. */ | |
8ea8cf89 | 621 | vhost_disable_notify(&net->dev, vq); |
8dd014ad DS |
622 | continue; |
623 | } | |
624 | /* Nothing new? Wait for eventfd to tell us | |
625 | * they refilled. */ | |
626 | break; | |
627 | } | |
628 | /* We don't need to be notified again. */ | |
629 | if (unlikely((vhost_hlen))) | |
630 | /* Skip header. TODO: support TSO. */ | |
81f95a55 | 631 | move_iovec_hdr(vq->iov, nvq->hdr, vhost_hlen, in); |
8dd014ad DS |
632 | else |
633 | /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF: | |
a290aec8 | 634 | * needed because recvmsg can modify msg_iov. */ |
81f95a55 | 635 | copy_iovec_hdr(vq->iov, nvq->hdr, sock_hlen, in); |
8dd014ad DS |
636 | msg.msg_iovlen = in; |
637 | err = sock->ops->recvmsg(NULL, sock, &msg, | |
638 | sock_len, MSG_DONTWAIT | MSG_TRUNC); | |
639 | /* Userspace might have consumed the packet meanwhile: | |
640 | * it's not supposed to do this usually, but might be hard | |
641 | * to prevent. Discard data we got (if any) and keep going. */ | |
642 | if (unlikely(err != sock_len)) { | |
643 | pr_debug("Discarded rx packet: " | |
644 | " len %d, expected %zd\n", err, sock_len); | |
645 | vhost_discard_vq_desc(vq, headcount); | |
646 | continue; | |
647 | } | |
648 | if (unlikely(vhost_hlen) && | |
81f95a55 | 649 | memcpy_toiovecend(nvq->hdr, (unsigned char *)&hdr, 0, |
8dd014ad DS |
650 | vhost_hlen)) { |
651 | vq_err(vq, "Unable to write vnet_hdr at addr %p\n", | |
652 | vq->iov->iov_base); | |
653 | break; | |
654 | } | |
655 | /* TODO: Should check and handle checksum. */ | |
cfbdab95 | 656 | if (likely(mergeable) && |
81f95a55 | 657 | memcpy_toiovecend(nvq->hdr, (unsigned char *)&headcount, |
8dd014ad DS |
658 | offsetof(typeof(hdr), num_buffers), |
659 | sizeof hdr.num_buffers)) { | |
660 | vq_err(vq, "Failed num_buffers write"); | |
661 | vhost_discard_vq_desc(vq, headcount); | |
662 | break; | |
663 | } | |
664 | vhost_add_used_and_signal_n(&net->dev, vq, vq->heads, | |
665 | headcount); | |
666 | if (unlikely(vq_log)) | |
667 | vhost_log_write(vq, vq_log, log, vhost_len); | |
668 | total_len += vhost_len; | |
669 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { | |
670 | vhost_poll_queue(&vq->poll); | |
671 | break; | |
672 | } | |
673 | } | |
674 | ||
675 | mutex_unlock(&vq->mutex); | |
8dd014ad DS |
676 | } |
677 | ||
c23f3445 | 678 | static void handle_tx_kick(struct vhost_work *work) |
3a4d5c94 | 679 | { |
c23f3445 TH |
680 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
681 | poll.work); | |
682 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); | |
683 | ||
3a4d5c94 MT |
684 | handle_tx(net); |
685 | } | |
686 | ||
c23f3445 | 687 | static void handle_rx_kick(struct vhost_work *work) |
3a4d5c94 | 688 | { |
c23f3445 TH |
689 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
690 | poll.work); | |
691 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); | |
692 | ||
3a4d5c94 MT |
693 | handle_rx(net); |
694 | } | |
695 | ||
c23f3445 | 696 | static void handle_tx_net(struct vhost_work *work) |
3a4d5c94 | 697 | { |
c23f3445 TH |
698 | struct vhost_net *net = container_of(work, struct vhost_net, |
699 | poll[VHOST_NET_VQ_TX].work); | |
3a4d5c94 MT |
700 | handle_tx(net); |
701 | } | |
702 | ||
c23f3445 | 703 | static void handle_rx_net(struct vhost_work *work) |
3a4d5c94 | 704 | { |
c23f3445 TH |
705 | struct vhost_net *net = container_of(work, struct vhost_net, |
706 | poll[VHOST_NET_VQ_RX].work); | |
3a4d5c94 MT |
707 | handle_rx(net); |
708 | } | |
709 | ||
710 | static int vhost_net_open(struct inode *inode, struct file *f) | |
711 | { | |
712 | struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL); | |
c23f3445 | 713 | struct vhost_dev *dev; |
3ab2e420 | 714 | struct vhost_virtqueue **vqs; |
2839400f | 715 | int r, i; |
c23f3445 | 716 | |
3a4d5c94 MT |
717 | if (!n) |
718 | return -ENOMEM; | |
3ab2e420 AH |
719 | vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL); |
720 | if (!vqs) { | |
721 | kfree(n); | |
722 | return -ENOMEM; | |
723 | } | |
c23f3445 TH |
724 | |
725 | dev = &n->dev; | |
3ab2e420 AH |
726 | vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq; |
727 | vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq; | |
728 | n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick; | |
729 | n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick; | |
2839400f AH |
730 | for (i = 0; i < VHOST_NET_VQ_MAX; i++) { |
731 | n->vqs[i].ubufs = NULL; | |
732 | n->vqs[i].ubuf_info = NULL; | |
733 | n->vqs[i].upend_idx = 0; | |
734 | n->vqs[i].done_idx = 0; | |
81f95a55 MT |
735 | n->vqs[i].vhost_hlen = 0; |
736 | n->vqs[i].sock_hlen = 0; | |
2839400f | 737 | } |
3ab2e420 | 738 | r = vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX); |
3a4d5c94 MT |
739 | if (r < 0) { |
740 | kfree(n); | |
3ab2e420 | 741 | kfree(vqs); |
3a4d5c94 MT |
742 | return r; |
743 | } | |
744 | ||
c23f3445 TH |
745 | vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev); |
746 | vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev); | |
3a4d5c94 MT |
747 | |
748 | f->private_data = n; | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | static void vhost_net_disable_vq(struct vhost_net *n, | |
754 | struct vhost_virtqueue *vq) | |
755 | { | |
3ab2e420 AH |
756 | struct vhost_net_virtqueue *nvq = |
757 | container_of(vq, struct vhost_net_virtqueue, vq); | |
758 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
3a4d5c94 MT |
759 | if (!vq->private_data) |
760 | return; | |
70181d51 | 761 | vhost_poll_stop(poll); |
3a4d5c94 MT |
762 | } |
763 | ||
2b8b328b | 764 | static int vhost_net_enable_vq(struct vhost_net *n, |
3a4d5c94 MT |
765 | struct vhost_virtqueue *vq) |
766 | { | |
3ab2e420 AH |
767 | struct vhost_net_virtqueue *nvq = |
768 | container_of(vq, struct vhost_net_virtqueue, vq); | |
769 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
28457ee6 AB |
770 | struct socket *sock; |
771 | ||
772 | sock = rcu_dereference_protected(vq->private_data, | |
773 | lockdep_is_held(&vq->mutex)); | |
3a4d5c94 | 774 | if (!sock) |
2b8b328b | 775 | return 0; |
2b8b328b | 776 | |
70181d51 | 777 | return vhost_poll_start(poll, sock->file); |
3a4d5c94 MT |
778 | } |
779 | ||
780 | static struct socket *vhost_net_stop_vq(struct vhost_net *n, | |
781 | struct vhost_virtqueue *vq) | |
782 | { | |
783 | struct socket *sock; | |
784 | ||
785 | mutex_lock(&vq->mutex); | |
28457ee6 AB |
786 | sock = rcu_dereference_protected(vq->private_data, |
787 | lockdep_is_held(&vq->mutex)); | |
3a4d5c94 MT |
788 | vhost_net_disable_vq(n, vq); |
789 | rcu_assign_pointer(vq->private_data, NULL); | |
790 | mutex_unlock(&vq->mutex); | |
791 | return sock; | |
792 | } | |
793 | ||
794 | static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock, | |
795 | struct socket **rx_sock) | |
796 | { | |
3ab2e420 AH |
797 | *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq); |
798 | *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq); | |
3a4d5c94 MT |
799 | } |
800 | ||
801 | static void vhost_net_flush_vq(struct vhost_net *n, int index) | |
802 | { | |
803 | vhost_poll_flush(n->poll + index); | |
3ab2e420 | 804 | vhost_poll_flush(&n->vqs[index].vq.poll); |
3a4d5c94 MT |
805 | } |
806 | ||
807 | static void vhost_net_flush(struct vhost_net *n) | |
808 | { | |
809 | vhost_net_flush_vq(n, VHOST_NET_VQ_TX); | |
810 | vhost_net_flush_vq(n, VHOST_NET_VQ_RX); | |
2839400f | 811 | if (n->vqs[VHOST_NET_VQ_TX].ubufs) { |
3ab2e420 | 812 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 813 | n->tx_flush = true; |
3ab2e420 | 814 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 815 | /* Wait for all lower device DMAs done. */ |
fe729a57 | 816 | vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs); |
3ab2e420 | 817 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 818 | n->tx_flush = false; |
2839400f | 819 | kref_init(&n->vqs[VHOST_NET_VQ_TX].ubufs->kref); |
3ab2e420 | 820 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 821 | } |
3a4d5c94 MT |
822 | } |
823 | ||
824 | static int vhost_net_release(struct inode *inode, struct file *f) | |
825 | { | |
826 | struct vhost_net *n = f->private_data; | |
827 | struct socket *tx_sock; | |
828 | struct socket *rx_sock; | |
829 | ||
830 | vhost_net_stop(n, &tx_sock, &rx_sock); | |
831 | vhost_net_flush(n); | |
b211616d | 832 | vhost_dev_stop(&n->dev); |
ea5d4046 | 833 | vhost_dev_cleanup(&n->dev, false); |
81f95a55 | 834 | vhost_net_vq_reset(n); |
3a4d5c94 MT |
835 | if (tx_sock) |
836 | fput(tx_sock->file); | |
837 | if (rx_sock) | |
838 | fput(rx_sock->file); | |
839 | /* We do an extra flush before freeing memory, | |
840 | * since jobs can re-queue themselves. */ | |
841 | vhost_net_flush(n); | |
3ab2e420 | 842 | kfree(n->dev.vqs); |
3a4d5c94 MT |
843 | kfree(n); |
844 | return 0; | |
845 | } | |
846 | ||
847 | static struct socket *get_raw_socket(int fd) | |
848 | { | |
849 | struct { | |
850 | struct sockaddr_ll sa; | |
851 | char buf[MAX_ADDR_LEN]; | |
852 | } uaddr; | |
853 | int uaddr_len = sizeof uaddr, r; | |
854 | struct socket *sock = sockfd_lookup(fd, &r); | |
d47effe1 | 855 | |
3a4d5c94 MT |
856 | if (!sock) |
857 | return ERR_PTR(-ENOTSOCK); | |
858 | ||
859 | /* Parameter checking */ | |
860 | if (sock->sk->sk_type != SOCK_RAW) { | |
861 | r = -ESOCKTNOSUPPORT; | |
862 | goto err; | |
863 | } | |
864 | ||
865 | r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa, | |
866 | &uaddr_len, 0); | |
867 | if (r) | |
868 | goto err; | |
869 | ||
870 | if (uaddr.sa.sll_family != AF_PACKET) { | |
871 | r = -EPFNOSUPPORT; | |
872 | goto err; | |
873 | } | |
874 | return sock; | |
875 | err: | |
876 | fput(sock->file); | |
877 | return ERR_PTR(r); | |
878 | } | |
879 | ||
501c774c | 880 | static struct socket *get_tap_socket(int fd) |
3a4d5c94 MT |
881 | { |
882 | struct file *file = fget(fd); | |
883 | struct socket *sock; | |
d47effe1 | 884 | |
3a4d5c94 MT |
885 | if (!file) |
886 | return ERR_PTR(-EBADF); | |
887 | sock = tun_get_socket(file); | |
501c774c AB |
888 | if (!IS_ERR(sock)) |
889 | return sock; | |
890 | sock = macvtap_get_socket(file); | |
3a4d5c94 MT |
891 | if (IS_ERR(sock)) |
892 | fput(file); | |
893 | return sock; | |
894 | } | |
895 | ||
896 | static struct socket *get_socket(int fd) | |
897 | { | |
898 | struct socket *sock; | |
d47effe1 | 899 | |
3a4d5c94 MT |
900 | /* special case to disable backend */ |
901 | if (fd == -1) | |
902 | return NULL; | |
903 | sock = get_raw_socket(fd); | |
904 | if (!IS_ERR(sock)) | |
905 | return sock; | |
501c774c | 906 | sock = get_tap_socket(fd); |
3a4d5c94 MT |
907 | if (!IS_ERR(sock)) |
908 | return sock; | |
909 | return ERR_PTR(-ENOTSOCK); | |
910 | } | |
911 | ||
912 | static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) | |
913 | { | |
914 | struct socket *sock, *oldsock; | |
915 | struct vhost_virtqueue *vq; | |
2839400f | 916 | struct vhost_net_virtqueue *nvq; |
fe729a57 | 917 | struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL; |
3a4d5c94 MT |
918 | int r; |
919 | ||
920 | mutex_lock(&n->dev.mutex); | |
921 | r = vhost_dev_check_owner(&n->dev); | |
922 | if (r) | |
923 | goto err; | |
924 | ||
925 | if (index >= VHOST_NET_VQ_MAX) { | |
926 | r = -ENOBUFS; | |
927 | goto err; | |
928 | } | |
3ab2e420 | 929 | vq = &n->vqs[index].vq; |
2839400f | 930 | nvq = &n->vqs[index]; |
3a4d5c94 MT |
931 | mutex_lock(&vq->mutex); |
932 | ||
933 | /* Verify that ring has been setup correctly. */ | |
934 | if (!vhost_vq_access_ok(vq)) { | |
935 | r = -EFAULT; | |
1dace8c8 | 936 | goto err_vq; |
3a4d5c94 MT |
937 | } |
938 | sock = get_socket(fd); | |
939 | if (IS_ERR(sock)) { | |
940 | r = PTR_ERR(sock); | |
1dace8c8 | 941 | goto err_vq; |
3a4d5c94 MT |
942 | } |
943 | ||
944 | /* start polling new socket */ | |
28457ee6 AB |
945 | oldsock = rcu_dereference_protected(vq->private_data, |
946 | lockdep_is_held(&vq->mutex)); | |
11fe8839 | 947 | if (sock != oldsock) { |
fe729a57 AH |
948 | ubufs = vhost_net_ubuf_alloc(vq, |
949 | sock && vhost_sock_zcopy(sock)); | |
bab632d6 MT |
950 | if (IS_ERR(ubufs)) { |
951 | r = PTR_ERR(ubufs); | |
952 | goto err_ubufs; | |
953 | } | |
692a998b | 954 | |
d47effe1 KK |
955 | vhost_net_disable_vq(n, vq); |
956 | rcu_assign_pointer(vq->private_data, sock); | |
f59281da JW |
957 | r = vhost_init_used(vq); |
958 | if (r) | |
692a998b | 959 | goto err_used; |
2b8b328b JW |
960 | r = vhost_net_enable_vq(n, vq); |
961 | if (r) | |
962 | goto err_used; | |
692a998b | 963 | |
2839400f AH |
964 | oldubufs = nvq->ubufs; |
965 | nvq->ubufs = ubufs; | |
64e9a9b8 MT |
966 | |
967 | n->tx_packets = 0; | |
968 | n->tx_zcopy_err = 0; | |
1280c27f | 969 | n->tx_flush = false; |
dd1f4078 | 970 | } |
3a4d5c94 | 971 | |
1680e906 MT |
972 | mutex_unlock(&vq->mutex); |
973 | ||
c047e5f3 | 974 | if (oldubufs) { |
f5ce1d25 | 975 | vhost_net_ubuf_put_wait_and_free(oldubufs); |
c047e5f3 | 976 | mutex_lock(&vq->mutex); |
eaae8132 | 977 | vhost_zerocopy_signal_used(n, vq); |
c047e5f3 MT |
978 | mutex_unlock(&vq->mutex); |
979 | } | |
bab632d6 | 980 | |
3a4d5c94 MT |
981 | if (oldsock) { |
982 | vhost_net_flush_vq(n, index); | |
983 | fput(oldsock->file); | |
984 | } | |
1dace8c8 | 985 | |
1680e906 MT |
986 | mutex_unlock(&n->dev.mutex); |
987 | return 0; | |
988 | ||
692a998b JW |
989 | err_used: |
990 | rcu_assign_pointer(vq->private_data, oldsock); | |
991 | vhost_net_enable_vq(n, vq); | |
992 | if (ubufs) | |
f5ce1d25 | 993 | vhost_net_ubuf_put_wait_and_free(ubufs); |
bab632d6 MT |
994 | err_ubufs: |
995 | fput(sock->file); | |
1dace8c8 JD |
996 | err_vq: |
997 | mutex_unlock(&vq->mutex); | |
3a4d5c94 MT |
998 | err: |
999 | mutex_unlock(&n->dev.mutex); | |
1000 | return r; | |
1001 | } | |
1002 | ||
1003 | static long vhost_net_reset_owner(struct vhost_net *n) | |
1004 | { | |
1005 | struct socket *tx_sock = NULL; | |
1006 | struct socket *rx_sock = NULL; | |
1007 | long err; | |
150b9e51 | 1008 | struct vhost_memory *memory; |
d47effe1 | 1009 | |
3a4d5c94 MT |
1010 | mutex_lock(&n->dev.mutex); |
1011 | err = vhost_dev_check_owner(&n->dev); | |
1012 | if (err) | |
1013 | goto done; | |
150b9e51 MT |
1014 | memory = vhost_dev_reset_owner_prepare(); |
1015 | if (!memory) { | |
1016 | err = -ENOMEM; | |
1017 | goto done; | |
1018 | } | |
3a4d5c94 MT |
1019 | vhost_net_stop(n, &tx_sock, &rx_sock); |
1020 | vhost_net_flush(n); | |
150b9e51 | 1021 | vhost_dev_reset_owner(&n->dev, memory); |
81f95a55 | 1022 | vhost_net_vq_reset(n); |
3a4d5c94 MT |
1023 | done: |
1024 | mutex_unlock(&n->dev.mutex); | |
1025 | if (tx_sock) | |
1026 | fput(tx_sock->file); | |
1027 | if (rx_sock) | |
1028 | fput(rx_sock->file); | |
1029 | return err; | |
1030 | } | |
1031 | ||
1032 | static int vhost_net_set_features(struct vhost_net *n, u64 features) | |
1033 | { | |
8dd014ad | 1034 | size_t vhost_hlen, sock_hlen, hdr_len; |
3a4d5c94 | 1035 | int i; |
8dd014ad DS |
1036 | |
1037 | hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? | |
1038 | sizeof(struct virtio_net_hdr_mrg_rxbuf) : | |
1039 | sizeof(struct virtio_net_hdr); | |
1040 | if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) { | |
1041 | /* vhost provides vnet_hdr */ | |
1042 | vhost_hlen = hdr_len; | |
1043 | sock_hlen = 0; | |
1044 | } else { | |
1045 | /* socket provides vnet_hdr */ | |
1046 | vhost_hlen = 0; | |
1047 | sock_hlen = hdr_len; | |
1048 | } | |
3a4d5c94 MT |
1049 | mutex_lock(&n->dev.mutex); |
1050 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
1051 | !vhost_log_access_ok(&n->dev)) { | |
1052 | mutex_unlock(&n->dev.mutex); | |
1053 | return -EFAULT; | |
1054 | } | |
1055 | n->dev.acked_features = features; | |
1056 | smp_wmb(); | |
1057 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { | |
3ab2e420 | 1058 | mutex_lock(&n->vqs[i].vq.mutex); |
81f95a55 MT |
1059 | n->vqs[i].vhost_hlen = vhost_hlen; |
1060 | n->vqs[i].sock_hlen = sock_hlen; | |
3ab2e420 | 1061 | mutex_unlock(&n->vqs[i].vq.mutex); |
3a4d5c94 MT |
1062 | } |
1063 | vhost_net_flush(n); | |
1064 | mutex_unlock(&n->dev.mutex); | |
1065 | return 0; | |
1066 | } | |
1067 | ||
b1ad8496 AH |
1068 | static long vhost_net_set_owner(struct vhost_net *n) |
1069 | { | |
1070 | int r; | |
1071 | ||
1072 | mutex_lock(&n->dev.mutex); | |
05c05351 MT |
1073 | if (vhost_dev_has_owner(&n->dev)) { |
1074 | r = -EBUSY; | |
1075 | goto out; | |
1076 | } | |
b1ad8496 AH |
1077 | r = vhost_net_set_ubuf_info(n); |
1078 | if (r) | |
1079 | goto out; | |
1080 | r = vhost_dev_set_owner(&n->dev); | |
1081 | if (r) | |
1082 | vhost_net_clear_ubuf_info(n); | |
1083 | vhost_net_flush(n); | |
1084 | out: | |
1085 | mutex_unlock(&n->dev.mutex); | |
1086 | return r; | |
1087 | } | |
1088 | ||
3a4d5c94 MT |
1089 | static long vhost_net_ioctl(struct file *f, unsigned int ioctl, |
1090 | unsigned long arg) | |
1091 | { | |
1092 | struct vhost_net *n = f->private_data; | |
1093 | void __user *argp = (void __user *)arg; | |
1094 | u64 __user *featurep = argp; | |
1095 | struct vhost_vring_file backend; | |
1096 | u64 features; | |
1097 | int r; | |
d47effe1 | 1098 | |
3a4d5c94 MT |
1099 | switch (ioctl) { |
1100 | case VHOST_NET_SET_BACKEND: | |
d3553a52 TY |
1101 | if (copy_from_user(&backend, argp, sizeof backend)) |
1102 | return -EFAULT; | |
3a4d5c94 MT |
1103 | return vhost_net_set_backend(n, backend.index, backend.fd); |
1104 | case VHOST_GET_FEATURES: | |
0dd05a3b | 1105 | features = VHOST_NET_FEATURES; |
d3553a52 TY |
1106 | if (copy_to_user(featurep, &features, sizeof features)) |
1107 | return -EFAULT; | |
1108 | return 0; | |
3a4d5c94 | 1109 | case VHOST_SET_FEATURES: |
d3553a52 TY |
1110 | if (copy_from_user(&features, featurep, sizeof features)) |
1111 | return -EFAULT; | |
0dd05a3b | 1112 | if (features & ~VHOST_NET_FEATURES) |
3a4d5c94 MT |
1113 | return -EOPNOTSUPP; |
1114 | return vhost_net_set_features(n, features); | |
1115 | case VHOST_RESET_OWNER: | |
1116 | return vhost_net_reset_owner(n); | |
b1ad8496 AH |
1117 | case VHOST_SET_OWNER: |
1118 | return vhost_net_set_owner(n); | |
3a4d5c94 MT |
1119 | default: |
1120 | mutex_lock(&n->dev.mutex); | |
935cdee7 MT |
1121 | r = vhost_dev_ioctl(&n->dev, ioctl, argp); |
1122 | if (r == -ENOIOCTLCMD) | |
1123 | r = vhost_vring_ioctl(&n->dev, ioctl, argp); | |
1124 | else | |
1125 | vhost_net_flush(n); | |
3a4d5c94 MT |
1126 | mutex_unlock(&n->dev.mutex); |
1127 | return r; | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | #ifdef CONFIG_COMPAT | |
1132 | static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl, | |
1133 | unsigned long arg) | |
1134 | { | |
1135 | return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); | |
1136 | } | |
1137 | #endif | |
1138 | ||
373a83a6 | 1139 | static const struct file_operations vhost_net_fops = { |
3a4d5c94 MT |
1140 | .owner = THIS_MODULE, |
1141 | .release = vhost_net_release, | |
1142 | .unlocked_ioctl = vhost_net_ioctl, | |
1143 | #ifdef CONFIG_COMPAT | |
1144 | .compat_ioctl = vhost_net_compat_ioctl, | |
1145 | #endif | |
1146 | .open = vhost_net_open, | |
6038f373 | 1147 | .llseek = noop_llseek, |
3a4d5c94 MT |
1148 | }; |
1149 | ||
1150 | static struct miscdevice vhost_net_misc = { | |
7c7c7f01 | 1151 | .minor = VHOST_NET_MINOR, |
1152 | .name = "vhost-net", | |
1153 | .fops = &vhost_net_fops, | |
3a4d5c94 MT |
1154 | }; |
1155 | ||
a8d3782f | 1156 | static int vhost_net_init(void) |
3a4d5c94 | 1157 | { |
bab632d6 | 1158 | if (experimental_zcopytx) |
fe729a57 | 1159 | vhost_net_enable_zcopy(VHOST_NET_VQ_TX); |
c23f3445 | 1160 | return misc_register(&vhost_net_misc); |
3a4d5c94 MT |
1161 | } |
1162 | module_init(vhost_net_init); | |
1163 | ||
a8d3782f | 1164 | static void vhost_net_exit(void) |
3a4d5c94 MT |
1165 | { |
1166 | misc_deregister(&vhost_net_misc); | |
3a4d5c94 MT |
1167 | } |
1168 | module_exit(vhost_net_exit); | |
1169 | ||
1170 | MODULE_VERSION("0.0.1"); | |
1171 | MODULE_LICENSE("GPL v2"); | |
1172 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
1173 | MODULE_DESCRIPTION("Host kernel accelerator for virtio net"); | |
7c7c7f01 | 1174 | MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR); |
1175 | MODULE_ALIAS("devname:vhost-net"); |