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> | |
3a4d5c94 | 18 | #include <linux/file.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
23cc5a99 | 20 | #include <linux/vmalloc.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 { |
0ad8b480 MT |
74 | /* refcount follows semantics similar to kref: |
75 | * 0: object is released | |
76 | * 1: no outstanding ubufs | |
77 | * >1: outstanding ubufs | |
78 | */ | |
79 | atomic_t refcount; | |
2839400f AH |
80 | wait_queue_head_t wait; |
81 | struct vhost_virtqueue *vq; | |
82 | }; | |
83 | ||
3ab2e420 AH |
84 | struct vhost_net_virtqueue { |
85 | struct vhost_virtqueue vq; | |
81f95a55 MT |
86 | /* hdr is used to store the virtio header. |
87 | * Since each iovec has >= 1 byte length, we never need more than | |
88 | * header length entries to store the header. */ | |
89 | struct iovec hdr[sizeof(struct virtio_net_hdr_mrg_rxbuf)]; | |
90 | size_t vhost_hlen; | |
91 | size_t sock_hlen; | |
2839400f AH |
92 | /* vhost zerocopy support fields below: */ |
93 | /* last used idx for outstanding DMA zerocopy buffers */ | |
94 | int upend_idx; | |
95 | /* first used idx for DMA done zerocopy buffers */ | |
96 | int done_idx; | |
97 | /* an array of userspace buffers info */ | |
98 | struct ubuf_info *ubuf_info; | |
99 | /* Reference counting for outstanding ubufs. | |
100 | * Protected by vq mutex. Writers must also take device mutex. */ | |
fe729a57 | 101 | struct vhost_net_ubuf_ref *ubufs; |
3ab2e420 AH |
102 | }; |
103 | ||
3a4d5c94 MT |
104 | struct vhost_net { |
105 | struct vhost_dev dev; | |
3ab2e420 | 106 | struct vhost_net_virtqueue vqs[VHOST_NET_VQ_MAX]; |
3a4d5c94 | 107 | struct vhost_poll poll[VHOST_NET_VQ_MAX]; |
eaae8132 MT |
108 | /* Number of TX recently submitted. |
109 | * Protected by tx vq lock. */ | |
110 | unsigned tx_packets; | |
111 | /* Number of times zerocopy TX recently failed. | |
112 | * Protected by tx vq lock. */ | |
113 | unsigned tx_zcopy_err; | |
1280c27f MT |
114 | /* Flush in progress. Protected by tx vq lock. */ |
115 | bool tx_flush; | |
3a4d5c94 MT |
116 | }; |
117 | ||
fe729a57 | 118 | static unsigned vhost_net_zcopy_mask __read_mostly; |
2839400f | 119 | |
fe729a57 | 120 | static void vhost_net_enable_zcopy(int vq) |
2839400f | 121 | { |
fe729a57 | 122 | vhost_net_zcopy_mask |= 0x1 << vq; |
2839400f AH |
123 | } |
124 | ||
fe729a57 AH |
125 | static struct vhost_net_ubuf_ref * |
126 | vhost_net_ubuf_alloc(struct vhost_virtqueue *vq, bool zcopy) | |
2839400f | 127 | { |
fe729a57 | 128 | struct vhost_net_ubuf_ref *ubufs; |
2839400f AH |
129 | /* No zero copy backend? Nothing to count. */ |
130 | if (!zcopy) | |
131 | return NULL; | |
132 | ubufs = kmalloc(sizeof(*ubufs), GFP_KERNEL); | |
133 | if (!ubufs) | |
134 | return ERR_PTR(-ENOMEM); | |
0ad8b480 | 135 | atomic_set(&ubufs->refcount, 1); |
2839400f AH |
136 | init_waitqueue_head(&ubufs->wait); |
137 | ubufs->vq = vq; | |
138 | return ubufs; | |
139 | } | |
140 | ||
0ad8b480 | 141 | static int vhost_net_ubuf_put(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 142 | { |
0ad8b480 MT |
143 | int r = atomic_sub_return(1, &ubufs->refcount); |
144 | if (unlikely(!r)) | |
145 | wake_up(&ubufs->wait); | |
146 | return r; | |
2839400f AH |
147 | } |
148 | ||
fe729a57 | 149 | static void vhost_net_ubuf_put_and_wait(struct vhost_net_ubuf_ref *ubufs) |
2839400f | 150 | { |
0ad8b480 MT |
151 | vhost_net_ubuf_put(ubufs); |
152 | wait_event(ubufs->wait, !atomic_read(&ubufs->refcount)); | |
c38e39c3 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 | ||
0a1febf7 | 171 | static int vhost_net_set_ubuf_info(struct vhost_net *n) |
2839400f AH |
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 | ||
0a1febf7 | 192 | static 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 | */ | |
094afe7d JW |
280 | static void 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); | |
c92112ae | 285 | int i, add; |
b211616d MT |
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; | |
b211616d MT |
293 | ++j; |
294 | } else | |
295 | break; | |
296 | } | |
c92112ae JW |
297 | while (j) { |
298 | add = min(UIO_MAXIOV - nvq->done_idx, j); | |
299 | vhost_add_used_and_signal_n(vq->dev, vq, | |
300 | &vq->heads[nvq->done_idx], add); | |
301 | nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV; | |
302 | j -= add; | |
303 | } | |
b211616d MT |
304 | } |
305 | ||
eaae8132 | 306 | static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success) |
b211616d | 307 | { |
fe729a57 | 308 | struct vhost_net_ubuf_ref *ubufs = ubuf->ctx; |
b211616d | 309 | struct vhost_virtqueue *vq = ubufs->vq; |
0ad8b480 | 310 | int cnt; |
24eb21a1 | 311 | |
b0c057ca MT |
312 | rcu_read_lock_bh(); |
313 | ||
19c73b3e JW |
314 | /* set len to mark this desc buffers done DMA */ |
315 | vq->heads[ubuf->desc].len = success ? | |
316 | VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN; | |
0ad8b480 | 317 | cnt = vhost_net_ubuf_put(ubufs); |
19c73b3e | 318 | |
24eb21a1 MT |
319 | /* |
320 | * Trigger polling thread if guest stopped submitting new buffers: | |
0ad8b480 | 321 | * in this case, the refcount after decrement will eventually reach 1. |
24eb21a1 MT |
322 | * We also trigger polling periodically after each 16 packets |
323 | * (the value 16 here is more or less arbitrary, it's tuned to trigger | |
324 | * less than 10% of times). | |
325 | */ | |
0ad8b480 | 326 | if (cnt <= 1 || !(cnt % 16)) |
24eb21a1 | 327 | vhost_poll_queue(&vq->poll); |
b0c057ca MT |
328 | |
329 | rcu_read_unlock_bh(); | |
b211616d MT |
330 | } |
331 | ||
3a4d5c94 MT |
332 | /* Expects to be always run from workqueue - which acts as |
333 | * read-size critical section for our kind of RCU. */ | |
334 | static void handle_tx(struct vhost_net *net) | |
335 | { | |
2839400f | 336 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; |
81f95a55 | 337 | struct vhost_virtqueue *vq = &nvq->vq; |
d5675bd2 MT |
338 | unsigned out, in, s; |
339 | int head; | |
3a4d5c94 MT |
340 | struct msghdr msg = { |
341 | .msg_name = NULL, | |
342 | .msg_namelen = 0, | |
343 | .msg_control = NULL, | |
344 | .msg_controllen = 0, | |
3a4d5c94 MT |
345 | .msg_flags = MSG_DONTWAIT, |
346 | }; | |
347 | size_t len, total_len = 0; | |
70181d51 | 348 | int err; |
3a4d5c94 | 349 | size_t hdr_size; |
28457ee6 | 350 | struct socket *sock; |
fe729a57 | 351 | struct vhost_net_ubuf_ref *uninitialized_var(ubufs); |
cedb9bdc | 352 | bool zcopy, zcopy_used; |
28457ee6 | 353 | |
2e26af79 AH |
354 | mutex_lock(&vq->mutex); |
355 | sock = vq->private_data; | |
3a4d5c94 | 356 | if (!sock) |
2e26af79 | 357 | goto out; |
3a4d5c94 | 358 | |
8ea8cf89 | 359 | vhost_disable_notify(&net->dev, vq); |
3a4d5c94 | 360 | |
81f95a55 | 361 | hdr_size = nvq->vhost_hlen; |
2839400f | 362 | zcopy = nvq->ubufs; |
3a4d5c94 MT |
363 | |
364 | for (;;) { | |
bab632d6 MT |
365 | /* Release DMAs done buffers first */ |
366 | if (zcopy) | |
eaae8132 | 367 | vhost_zerocopy_signal_used(net, vq); |
bab632d6 | 368 | |
f7c6be40 JW |
369 | /* If more outstanding DMAs, queue the work. |
370 | * Handle upend_idx wrap around | |
371 | */ | |
372 | if (unlikely((nvq->upend_idx + vq->num - VHOST_MAX_PEND) | |
373 | % UIO_MAXIOV == nvq->done_idx)) | |
374 | break; | |
375 | ||
47283bef | 376 | head = vhost_get_vq_desc(vq, vq->iov, |
3a4d5c94 MT |
377 | ARRAY_SIZE(vq->iov), |
378 | &out, &in, | |
379 | NULL, NULL); | |
d5675bd2 | 380 | /* On error, stop handling until the next kick. */ |
7b3384fc | 381 | if (unlikely(head < 0)) |
d5675bd2 | 382 | break; |
3a4d5c94 MT |
383 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
384 | if (head == vq->num) { | |
8ea8cf89 MT |
385 | if (unlikely(vhost_enable_notify(&net->dev, vq))) { |
386 | vhost_disable_notify(&net->dev, vq); | |
3a4d5c94 MT |
387 | continue; |
388 | } | |
389 | break; | |
390 | } | |
391 | if (in) { | |
392 | vq_err(vq, "Unexpected descriptor format for TX: " | |
393 | "out %d, int %d\n", out, in); | |
394 | break; | |
395 | } | |
396 | /* Skip header. TODO: support TSO. */ | |
81f95a55 | 397 | s = move_iovec_hdr(vq->iov, nvq->hdr, hdr_size, out); |
3a4d5c94 | 398 | len = iov_length(vq->iov, out); |
c0371da6 | 399 | iov_iter_init(&msg.msg_iter, WRITE, vq->iov, out, len); |
3a4d5c94 MT |
400 | /* Sanity check */ |
401 | if (!len) { | |
402 | vq_err(vq, "Unexpected header len for TX: " | |
403 | "%zd expected %zd\n", | |
81f95a55 | 404 | iov_length(nvq->hdr, s), hdr_size); |
3a4d5c94 MT |
405 | break; |
406 | } | |
ce21a029 JW |
407 | |
408 | zcopy_used = zcopy && len >= VHOST_GOODCOPY_LEN | |
409 | && (nvq->upend_idx + 1) % UIO_MAXIOV != | |
410 | nvq->done_idx | |
411 | && vhost_net_tx_select_zcopy(net); | |
cedb9bdc | 412 | |
bab632d6 | 413 | /* use msg_control to pass vhost zerocopy ubuf info to skb */ |
cedb9bdc | 414 | if (zcopy_used) { |
ce21a029 JW |
415 | struct ubuf_info *ubuf; |
416 | ubuf = nvq->ubuf_info + nvq->upend_idx; | |
417 | ||
2839400f | 418 | vq->heads[nvq->upend_idx].id = head; |
ce21a029 JW |
419 | vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS; |
420 | ubuf->callback = vhost_zerocopy_callback; | |
421 | ubuf->ctx = nvq->ubufs; | |
422 | ubuf->desc = nvq->upend_idx; | |
423 | msg.msg_control = ubuf; | |
424 | msg.msg_controllen = sizeof(ubuf); | |
425 | ubufs = nvq->ubufs; | |
0ad8b480 | 426 | atomic_inc(&ubufs->refcount); |
2839400f | 427 | nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV; |
ce21a029 | 428 | } else { |
4364d5f9 | 429 | msg.msg_control = NULL; |
ce21a029 JW |
430 | ubufs = NULL; |
431 | } | |
3a4d5c94 MT |
432 | /* TODO: Check specific error and bomb out unless ENOBUFS? */ |
433 | err = sock->ops->sendmsg(NULL, sock, &msg, len); | |
434 | if (unlikely(err < 0)) { | |
cedb9bdc | 435 | if (zcopy_used) { |
ce21a029 | 436 | vhost_net_ubuf_put(ubufs); |
2839400f AH |
437 | nvq->upend_idx = ((unsigned)nvq->upend_idx - 1) |
438 | % UIO_MAXIOV; | |
bab632d6 | 439 | } |
8dd014ad | 440 | vhost_discard_vq_desc(vq, 1); |
3a4d5c94 MT |
441 | break; |
442 | } | |
443 | if (err != len) | |
95c0ec6a MT |
444 | pr_debug("Truncated TX packet: " |
445 | " len %d != %zd\n", err, len); | |
cedb9bdc | 446 | if (!zcopy_used) |
bab632d6 | 447 | vhost_add_used_and_signal(&net->dev, vq, head, 0); |
c8fb217a | 448 | else |
eaae8132 | 449 | vhost_zerocopy_signal_used(net, vq); |
3a4d5c94 | 450 | total_len += len; |
eaae8132 | 451 | vhost_net_tx_packet(net); |
3a4d5c94 MT |
452 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { |
453 | vhost_poll_queue(&vq->poll); | |
454 | break; | |
455 | } | |
456 | } | |
2e26af79 | 457 | out: |
3a4d5c94 | 458 | mutex_unlock(&vq->mutex); |
3a4d5c94 MT |
459 | } |
460 | ||
8dd014ad DS |
461 | static int peek_head_len(struct sock *sk) |
462 | { | |
463 | struct sk_buff *head; | |
464 | int len = 0; | |
783e3988 | 465 | unsigned long flags; |
8dd014ad | 466 | |
783e3988 | 467 | spin_lock_irqsave(&sk->sk_receive_queue.lock, flags); |
8dd014ad | 468 | head = skb_peek(&sk->sk_receive_queue); |
c53cff5e | 469 | if (likely(head)) { |
8dd014ad | 470 | len = head->len; |
c53cff5e BG |
471 | if (vlan_tx_tag_present(head)) |
472 | len += VLAN_HLEN; | |
473 | } | |
474 | ||
783e3988 | 475 | spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags); |
8dd014ad DS |
476 | return len; |
477 | } | |
478 | ||
479 | /* This is a multi-buffer version of vhost_get_desc, that works if | |
480 | * vq has read descriptors only. | |
481 | * @vq - the relevant virtqueue | |
482 | * @datalen - data length we'll be reading | |
483 | * @iovcount - returned count of io vectors we fill | |
484 | * @log - vhost log | |
485 | * @log_num - log offset | |
94249369 | 486 | * @quota - headcount quota, 1 for big buffer |
8dd014ad DS |
487 | * returns number of buffer heads allocated, negative on error |
488 | */ | |
489 | static int get_rx_bufs(struct vhost_virtqueue *vq, | |
490 | struct vring_used_elem *heads, | |
491 | int datalen, | |
492 | unsigned *iovcount, | |
493 | struct vhost_log *log, | |
94249369 JW |
494 | unsigned *log_num, |
495 | unsigned int quota) | |
8dd014ad DS |
496 | { |
497 | unsigned int out, in; | |
498 | int seg = 0; | |
499 | int headcount = 0; | |
500 | unsigned d; | |
501 | int r, nlogs = 0; | |
502 | ||
94249369 | 503 | while (datalen > 0 && headcount < quota) { |
e0e9b406 | 504 | if (unlikely(seg >= UIO_MAXIOV)) { |
8dd014ad DS |
505 | r = -ENOBUFS; |
506 | goto err; | |
507 | } | |
47283bef | 508 | r = vhost_get_vq_desc(vq, vq->iov + seg, |
8dd014ad DS |
509 | ARRAY_SIZE(vq->iov) - seg, &out, |
510 | &in, log, log_num); | |
a39ee449 MT |
511 | if (unlikely(r < 0)) |
512 | goto err; | |
513 | ||
514 | d = r; | |
8dd014ad DS |
515 | if (d == vq->num) { |
516 | r = 0; | |
517 | goto err; | |
518 | } | |
519 | if (unlikely(out || in <= 0)) { | |
520 | vq_err(vq, "unexpected descriptor format for RX: " | |
521 | "out %d, in %d\n", out, in); | |
522 | r = -EINVAL; | |
523 | goto err; | |
524 | } | |
525 | if (unlikely(log)) { | |
526 | nlogs += *log_num; | |
527 | log += *log_num; | |
528 | } | |
529 | heads[headcount].id = d; | |
530 | heads[headcount].len = iov_length(vq->iov + seg, in); | |
531 | datalen -= heads[headcount].len; | |
532 | ++headcount; | |
533 | seg += in; | |
534 | } | |
535 | heads[headcount - 1].len += datalen; | |
536 | *iovcount = seg; | |
537 | if (unlikely(log)) | |
538 | *log_num = nlogs; | |
d8316f39 MT |
539 | |
540 | /* Detect overrun */ | |
541 | if (unlikely(datalen > 0)) { | |
542 | r = UIO_MAXIOV + 1; | |
543 | goto err; | |
544 | } | |
8dd014ad DS |
545 | return headcount; |
546 | err: | |
547 | vhost_discard_vq_desc(vq, headcount); | |
548 | return r; | |
549 | } | |
550 | ||
3a4d5c94 MT |
551 | /* Expects to be always run from workqueue - which acts as |
552 | * read-size critical section for our kind of RCU. */ | |
94249369 | 553 | static void handle_rx(struct vhost_net *net) |
3a4d5c94 | 554 | { |
81f95a55 MT |
555 | struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX]; |
556 | struct vhost_virtqueue *vq = &nvq->vq; | |
8dd014ad DS |
557 | unsigned uninitialized_var(in), log; |
558 | struct vhost_log *vq_log; | |
559 | struct msghdr msg = { | |
560 | .msg_name = NULL, | |
561 | .msg_namelen = 0, | |
562 | .msg_control = NULL, /* FIXME: get and handle RX aux data. */ | |
563 | .msg_controllen = 0, | |
8dd014ad DS |
564 | .msg_flags = MSG_DONTWAIT, |
565 | }; | |
8dd014ad DS |
566 | struct virtio_net_hdr_mrg_rxbuf hdr = { |
567 | .hdr.flags = 0, | |
568 | .hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE | |
569 | }; | |
8dd014ad | 570 | size_t total_len = 0; |
910a578f MT |
571 | int err, mergeable; |
572 | s16 headcount; | |
8dd014ad DS |
573 | size_t vhost_hlen, sock_hlen; |
574 | size_t vhost_len, sock_len; | |
2e26af79 | 575 | struct socket *sock; |
8dd014ad | 576 | |
8dd014ad | 577 | mutex_lock(&vq->mutex); |
2e26af79 AH |
578 | sock = vq->private_data; |
579 | if (!sock) | |
580 | goto out; | |
8ea8cf89 | 581 | vhost_disable_notify(&net->dev, vq); |
2e26af79 | 582 | |
81f95a55 MT |
583 | vhost_hlen = nvq->vhost_hlen; |
584 | sock_hlen = nvq->sock_hlen; | |
8dd014ad | 585 | |
ea16c514 | 586 | vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ? |
8dd014ad | 587 | vq->log : NULL; |
ea16c514 | 588 | mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF); |
8dd014ad DS |
589 | |
590 | while ((sock_len = peek_head_len(sock->sk))) { | |
591 | sock_len += sock_hlen; | |
592 | vhost_len = sock_len + vhost_hlen; | |
593 | headcount = get_rx_bufs(vq, vq->heads, vhost_len, | |
94249369 JW |
594 | &in, vq_log, &log, |
595 | likely(mergeable) ? UIO_MAXIOV : 1); | |
8dd014ad DS |
596 | /* On error, stop handling until the next kick. */ |
597 | if (unlikely(headcount < 0)) | |
598 | break; | |
d8316f39 MT |
599 | /* On overrun, truncate and discard */ |
600 | if (unlikely(headcount > UIO_MAXIOV)) { | |
c0371da6 | 601 | iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1); |
d8316f39 MT |
602 | err = sock->ops->recvmsg(NULL, sock, &msg, |
603 | 1, MSG_DONTWAIT | MSG_TRUNC); | |
604 | pr_debug("Discarded rx packet: len %zd\n", sock_len); | |
605 | continue; | |
606 | } | |
8dd014ad DS |
607 | /* OK, now we need to know about added descriptors. */ |
608 | if (!headcount) { | |
8ea8cf89 | 609 | if (unlikely(vhost_enable_notify(&net->dev, vq))) { |
8dd014ad DS |
610 | /* They have slipped one in as we were |
611 | * doing that: check again. */ | |
8ea8cf89 | 612 | vhost_disable_notify(&net->dev, vq); |
8dd014ad DS |
613 | continue; |
614 | } | |
615 | /* Nothing new? Wait for eventfd to tell us | |
616 | * they refilled. */ | |
617 | break; | |
618 | } | |
619 | /* We don't need to be notified again. */ | |
620 | if (unlikely((vhost_hlen))) | |
621 | /* Skip header. TODO: support TSO. */ | |
81f95a55 | 622 | move_iovec_hdr(vq->iov, nvq->hdr, vhost_hlen, in); |
8dd014ad DS |
623 | else |
624 | /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF: | |
a290aec8 | 625 | * needed because recvmsg can modify msg_iov. */ |
81f95a55 | 626 | copy_iovec_hdr(vq->iov, nvq->hdr, sock_hlen, in); |
c0371da6 | 627 | iov_iter_init(&msg.msg_iter, READ, vq->iov, in, sock_len); |
8dd014ad DS |
628 | err = sock->ops->recvmsg(NULL, sock, &msg, |
629 | sock_len, MSG_DONTWAIT | MSG_TRUNC); | |
630 | /* Userspace might have consumed the packet meanwhile: | |
631 | * it's not supposed to do this usually, but might be hard | |
632 | * to prevent. Discard data we got (if any) and keep going. */ | |
633 | if (unlikely(err != sock_len)) { | |
634 | pr_debug("Discarded rx packet: " | |
635 | " len %d, expected %zd\n", err, sock_len); | |
636 | vhost_discard_vq_desc(vq, headcount); | |
637 | continue; | |
638 | } | |
639 | if (unlikely(vhost_hlen) && | |
81f95a55 | 640 | memcpy_toiovecend(nvq->hdr, (unsigned char *)&hdr, 0, |
8dd014ad DS |
641 | vhost_hlen)) { |
642 | vq_err(vq, "Unable to write vnet_hdr at addr %p\n", | |
643 | vq->iov->iov_base); | |
644 | break; | |
645 | } | |
646 | /* TODO: Should check and handle checksum. */ | |
cfbdab95 | 647 | if (likely(mergeable) && |
81f95a55 | 648 | memcpy_toiovecend(nvq->hdr, (unsigned char *)&headcount, |
8dd014ad DS |
649 | offsetof(typeof(hdr), num_buffers), |
650 | sizeof hdr.num_buffers)) { | |
651 | vq_err(vq, "Failed num_buffers write"); | |
652 | vhost_discard_vq_desc(vq, headcount); | |
653 | break; | |
654 | } | |
655 | vhost_add_used_and_signal_n(&net->dev, vq, vq->heads, | |
656 | headcount); | |
657 | if (unlikely(vq_log)) | |
658 | vhost_log_write(vq, vq_log, log, vhost_len); | |
659 | total_len += vhost_len; | |
660 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { | |
661 | vhost_poll_queue(&vq->poll); | |
662 | break; | |
663 | } | |
664 | } | |
2e26af79 | 665 | out: |
8dd014ad | 666 | mutex_unlock(&vq->mutex); |
8dd014ad DS |
667 | } |
668 | ||
c23f3445 | 669 | static void handle_tx_kick(struct vhost_work *work) |
3a4d5c94 | 670 | { |
c23f3445 TH |
671 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
672 | poll.work); | |
673 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); | |
674 | ||
3a4d5c94 MT |
675 | handle_tx(net); |
676 | } | |
677 | ||
c23f3445 | 678 | static void handle_rx_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_rx(net); |
685 | } | |
686 | ||
c23f3445 | 687 | static void handle_tx_net(struct vhost_work *work) |
3a4d5c94 | 688 | { |
c23f3445 TH |
689 | struct vhost_net *net = container_of(work, struct vhost_net, |
690 | poll[VHOST_NET_VQ_TX].work); | |
3a4d5c94 MT |
691 | handle_tx(net); |
692 | } | |
693 | ||
c23f3445 | 694 | static void handle_rx_net(struct vhost_work *work) |
3a4d5c94 | 695 | { |
c23f3445 TH |
696 | struct vhost_net *net = container_of(work, struct vhost_net, |
697 | poll[VHOST_NET_VQ_RX].work); | |
3a4d5c94 MT |
698 | handle_rx(net); |
699 | } | |
700 | ||
701 | static int vhost_net_open(struct inode *inode, struct file *f) | |
702 | { | |
23cc5a99 | 703 | struct vhost_net *n; |
c23f3445 | 704 | struct vhost_dev *dev; |
3ab2e420 | 705 | struct vhost_virtqueue **vqs; |
59566b6e | 706 | int i; |
c23f3445 | 707 | |
23cc5a99 MT |
708 | n = kmalloc(sizeof *n, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT); |
709 | if (!n) { | |
710 | n = vmalloc(sizeof *n); | |
711 | if (!n) | |
712 | return -ENOMEM; | |
713 | } | |
3ab2e420 AH |
714 | vqs = kmalloc(VHOST_NET_VQ_MAX * sizeof(*vqs), GFP_KERNEL); |
715 | if (!vqs) { | |
d04257b0 | 716 | kvfree(n); |
3ab2e420 AH |
717 | return -ENOMEM; |
718 | } | |
c23f3445 TH |
719 | |
720 | dev = &n->dev; | |
3ab2e420 AH |
721 | vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq; |
722 | vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq; | |
723 | n->vqs[VHOST_NET_VQ_TX].vq.handle_kick = handle_tx_kick; | |
724 | n->vqs[VHOST_NET_VQ_RX].vq.handle_kick = handle_rx_kick; | |
2839400f AH |
725 | for (i = 0; i < VHOST_NET_VQ_MAX; i++) { |
726 | n->vqs[i].ubufs = NULL; | |
727 | n->vqs[i].ubuf_info = NULL; | |
728 | n->vqs[i].upend_idx = 0; | |
729 | n->vqs[i].done_idx = 0; | |
81f95a55 MT |
730 | n->vqs[i].vhost_hlen = 0; |
731 | n->vqs[i].sock_hlen = 0; | |
2839400f | 732 | } |
59566b6e | 733 | vhost_dev_init(dev, vqs, VHOST_NET_VQ_MAX); |
3a4d5c94 | 734 | |
c23f3445 TH |
735 | vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev); |
736 | vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev); | |
3a4d5c94 MT |
737 | |
738 | f->private_data = n; | |
739 | ||
740 | return 0; | |
741 | } | |
742 | ||
743 | static void vhost_net_disable_vq(struct vhost_net *n, | |
744 | struct vhost_virtqueue *vq) | |
745 | { | |
3ab2e420 AH |
746 | struct vhost_net_virtqueue *nvq = |
747 | container_of(vq, struct vhost_net_virtqueue, vq); | |
748 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
3a4d5c94 MT |
749 | if (!vq->private_data) |
750 | return; | |
70181d51 | 751 | vhost_poll_stop(poll); |
3a4d5c94 MT |
752 | } |
753 | ||
2b8b328b | 754 | static int vhost_net_enable_vq(struct vhost_net *n, |
3a4d5c94 MT |
755 | struct vhost_virtqueue *vq) |
756 | { | |
3ab2e420 AH |
757 | struct vhost_net_virtqueue *nvq = |
758 | container_of(vq, struct vhost_net_virtqueue, vq); | |
759 | struct vhost_poll *poll = n->poll + (nvq - n->vqs); | |
28457ee6 AB |
760 | struct socket *sock; |
761 | ||
22fa90c7 | 762 | sock = vq->private_data; |
3a4d5c94 | 763 | if (!sock) |
2b8b328b | 764 | return 0; |
2b8b328b | 765 | |
70181d51 | 766 | return vhost_poll_start(poll, sock->file); |
3a4d5c94 MT |
767 | } |
768 | ||
769 | static struct socket *vhost_net_stop_vq(struct vhost_net *n, | |
770 | struct vhost_virtqueue *vq) | |
771 | { | |
772 | struct socket *sock; | |
773 | ||
774 | mutex_lock(&vq->mutex); | |
22fa90c7 | 775 | sock = vq->private_data; |
3a4d5c94 | 776 | vhost_net_disable_vq(n, vq); |
22fa90c7 | 777 | vq->private_data = NULL; |
3a4d5c94 MT |
778 | mutex_unlock(&vq->mutex); |
779 | return sock; | |
780 | } | |
781 | ||
782 | static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock, | |
783 | struct socket **rx_sock) | |
784 | { | |
3ab2e420 AH |
785 | *tx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_TX].vq); |
786 | *rx_sock = vhost_net_stop_vq(n, &n->vqs[VHOST_NET_VQ_RX].vq); | |
3a4d5c94 MT |
787 | } |
788 | ||
789 | static void vhost_net_flush_vq(struct vhost_net *n, int index) | |
790 | { | |
791 | vhost_poll_flush(n->poll + index); | |
3ab2e420 | 792 | vhost_poll_flush(&n->vqs[index].vq.poll); |
3a4d5c94 MT |
793 | } |
794 | ||
795 | static void vhost_net_flush(struct vhost_net *n) | |
796 | { | |
797 | vhost_net_flush_vq(n, VHOST_NET_VQ_TX); | |
798 | vhost_net_flush_vq(n, VHOST_NET_VQ_RX); | |
2839400f | 799 | if (n->vqs[VHOST_NET_VQ_TX].ubufs) { |
3ab2e420 | 800 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 801 | n->tx_flush = true; |
3ab2e420 | 802 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 803 | /* Wait for all lower device DMAs done. */ |
fe729a57 | 804 | vhost_net_ubuf_put_and_wait(n->vqs[VHOST_NET_VQ_TX].ubufs); |
3ab2e420 | 805 | mutex_lock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 806 | n->tx_flush = false; |
0ad8b480 | 807 | atomic_set(&n->vqs[VHOST_NET_VQ_TX].ubufs->refcount, 1); |
3ab2e420 | 808 | mutex_unlock(&n->vqs[VHOST_NET_VQ_TX].vq.mutex); |
1280c27f | 809 | } |
3a4d5c94 MT |
810 | } |
811 | ||
812 | static int vhost_net_release(struct inode *inode, struct file *f) | |
813 | { | |
814 | struct vhost_net *n = f->private_data; | |
815 | struct socket *tx_sock; | |
816 | struct socket *rx_sock; | |
817 | ||
818 | vhost_net_stop(n, &tx_sock, &rx_sock); | |
819 | vhost_net_flush(n); | |
b211616d | 820 | vhost_dev_stop(&n->dev); |
ea5d4046 | 821 | vhost_dev_cleanup(&n->dev, false); |
81f95a55 | 822 | vhost_net_vq_reset(n); |
3a4d5c94 | 823 | if (tx_sock) |
09aaacf0 | 824 | sockfd_put(tx_sock); |
3a4d5c94 | 825 | if (rx_sock) |
09aaacf0 | 826 | sockfd_put(rx_sock); |
b0c057ca MT |
827 | /* Make sure no callbacks are outstanding */ |
828 | synchronize_rcu_bh(); | |
3a4d5c94 MT |
829 | /* We do an extra flush before freeing memory, |
830 | * since jobs can re-queue themselves. */ | |
831 | vhost_net_flush(n); | |
3ab2e420 | 832 | kfree(n->dev.vqs); |
d04257b0 | 833 | kvfree(n); |
3a4d5c94 MT |
834 | return 0; |
835 | } | |
836 | ||
837 | static struct socket *get_raw_socket(int fd) | |
838 | { | |
839 | struct { | |
840 | struct sockaddr_ll sa; | |
841 | char buf[MAX_ADDR_LEN]; | |
842 | } uaddr; | |
843 | int uaddr_len = sizeof uaddr, r; | |
844 | struct socket *sock = sockfd_lookup(fd, &r); | |
d47effe1 | 845 | |
3a4d5c94 MT |
846 | if (!sock) |
847 | return ERR_PTR(-ENOTSOCK); | |
848 | ||
849 | /* Parameter checking */ | |
850 | if (sock->sk->sk_type != SOCK_RAW) { | |
851 | r = -ESOCKTNOSUPPORT; | |
852 | goto err; | |
853 | } | |
854 | ||
855 | r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa, | |
856 | &uaddr_len, 0); | |
857 | if (r) | |
858 | goto err; | |
859 | ||
860 | if (uaddr.sa.sll_family != AF_PACKET) { | |
861 | r = -EPFNOSUPPORT; | |
862 | goto err; | |
863 | } | |
864 | return sock; | |
865 | err: | |
09aaacf0 | 866 | sockfd_put(sock); |
3a4d5c94 MT |
867 | return ERR_PTR(r); |
868 | } | |
869 | ||
501c774c | 870 | static struct socket *get_tap_socket(int fd) |
3a4d5c94 MT |
871 | { |
872 | struct file *file = fget(fd); | |
873 | struct socket *sock; | |
d47effe1 | 874 | |
3a4d5c94 MT |
875 | if (!file) |
876 | return ERR_PTR(-EBADF); | |
877 | sock = tun_get_socket(file); | |
501c774c AB |
878 | if (!IS_ERR(sock)) |
879 | return sock; | |
880 | sock = macvtap_get_socket(file); | |
3a4d5c94 MT |
881 | if (IS_ERR(sock)) |
882 | fput(file); | |
883 | return sock; | |
884 | } | |
885 | ||
886 | static struct socket *get_socket(int fd) | |
887 | { | |
888 | struct socket *sock; | |
d47effe1 | 889 | |
3a4d5c94 MT |
890 | /* special case to disable backend */ |
891 | if (fd == -1) | |
892 | return NULL; | |
893 | sock = get_raw_socket(fd); | |
894 | if (!IS_ERR(sock)) | |
895 | return sock; | |
501c774c | 896 | sock = get_tap_socket(fd); |
3a4d5c94 MT |
897 | if (!IS_ERR(sock)) |
898 | return sock; | |
899 | return ERR_PTR(-ENOTSOCK); | |
900 | } | |
901 | ||
902 | static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) | |
903 | { | |
904 | struct socket *sock, *oldsock; | |
905 | struct vhost_virtqueue *vq; | |
2839400f | 906 | struct vhost_net_virtqueue *nvq; |
fe729a57 | 907 | struct vhost_net_ubuf_ref *ubufs, *oldubufs = NULL; |
3a4d5c94 MT |
908 | int r; |
909 | ||
910 | mutex_lock(&n->dev.mutex); | |
911 | r = vhost_dev_check_owner(&n->dev); | |
912 | if (r) | |
913 | goto err; | |
914 | ||
915 | if (index >= VHOST_NET_VQ_MAX) { | |
916 | r = -ENOBUFS; | |
917 | goto err; | |
918 | } | |
3ab2e420 | 919 | vq = &n->vqs[index].vq; |
2839400f | 920 | nvq = &n->vqs[index]; |
3a4d5c94 MT |
921 | mutex_lock(&vq->mutex); |
922 | ||
923 | /* Verify that ring has been setup correctly. */ | |
924 | if (!vhost_vq_access_ok(vq)) { | |
925 | r = -EFAULT; | |
1dace8c8 | 926 | goto err_vq; |
3a4d5c94 MT |
927 | } |
928 | sock = get_socket(fd); | |
929 | if (IS_ERR(sock)) { | |
930 | r = PTR_ERR(sock); | |
1dace8c8 | 931 | goto err_vq; |
3a4d5c94 MT |
932 | } |
933 | ||
934 | /* start polling new socket */ | |
22fa90c7 | 935 | oldsock = vq->private_data; |
11fe8839 | 936 | if (sock != oldsock) { |
fe729a57 AH |
937 | ubufs = vhost_net_ubuf_alloc(vq, |
938 | sock && vhost_sock_zcopy(sock)); | |
bab632d6 MT |
939 | if (IS_ERR(ubufs)) { |
940 | r = PTR_ERR(ubufs); | |
941 | goto err_ubufs; | |
942 | } | |
692a998b | 943 | |
d47effe1 | 944 | vhost_net_disable_vq(n, vq); |
22fa90c7 | 945 | vq->private_data = sock; |
f59281da JW |
946 | r = vhost_init_used(vq); |
947 | if (r) | |
692a998b | 948 | goto err_used; |
2b8b328b JW |
949 | r = vhost_net_enable_vq(n, vq); |
950 | if (r) | |
951 | goto err_used; | |
692a998b | 952 | |
2839400f AH |
953 | oldubufs = nvq->ubufs; |
954 | nvq->ubufs = ubufs; | |
64e9a9b8 MT |
955 | |
956 | n->tx_packets = 0; | |
957 | n->tx_zcopy_err = 0; | |
1280c27f | 958 | n->tx_flush = false; |
dd1f4078 | 959 | } |
3a4d5c94 | 960 | |
1680e906 MT |
961 | mutex_unlock(&vq->mutex); |
962 | ||
c047e5f3 | 963 | if (oldubufs) { |
c38e39c3 | 964 | vhost_net_ubuf_put_wait_and_free(oldubufs); |
c047e5f3 | 965 | mutex_lock(&vq->mutex); |
eaae8132 | 966 | vhost_zerocopy_signal_used(n, vq); |
c047e5f3 MT |
967 | mutex_unlock(&vq->mutex); |
968 | } | |
bab632d6 | 969 | |
3a4d5c94 MT |
970 | if (oldsock) { |
971 | vhost_net_flush_vq(n, index); | |
09aaacf0 | 972 | sockfd_put(oldsock); |
3a4d5c94 | 973 | } |
1dace8c8 | 974 | |
1680e906 MT |
975 | mutex_unlock(&n->dev.mutex); |
976 | return 0; | |
977 | ||
692a998b | 978 | err_used: |
22fa90c7 | 979 | vq->private_data = oldsock; |
692a998b JW |
980 | vhost_net_enable_vq(n, vq); |
981 | if (ubufs) | |
c38e39c3 | 982 | vhost_net_ubuf_put_wait_and_free(ubufs); |
bab632d6 | 983 | err_ubufs: |
09aaacf0 | 984 | sockfd_put(sock); |
1dace8c8 JD |
985 | err_vq: |
986 | mutex_unlock(&vq->mutex); | |
3a4d5c94 MT |
987 | err: |
988 | mutex_unlock(&n->dev.mutex); | |
989 | return r; | |
990 | } | |
991 | ||
992 | static long vhost_net_reset_owner(struct vhost_net *n) | |
993 | { | |
994 | struct socket *tx_sock = NULL; | |
995 | struct socket *rx_sock = NULL; | |
996 | long err; | |
150b9e51 | 997 | struct vhost_memory *memory; |
d47effe1 | 998 | |
3a4d5c94 MT |
999 | mutex_lock(&n->dev.mutex); |
1000 | err = vhost_dev_check_owner(&n->dev); | |
1001 | if (err) | |
1002 | goto done; | |
150b9e51 MT |
1003 | memory = vhost_dev_reset_owner_prepare(); |
1004 | if (!memory) { | |
1005 | err = -ENOMEM; | |
1006 | goto done; | |
1007 | } | |
3a4d5c94 MT |
1008 | vhost_net_stop(n, &tx_sock, &rx_sock); |
1009 | vhost_net_flush(n); | |
150b9e51 | 1010 | vhost_dev_reset_owner(&n->dev, memory); |
81f95a55 | 1011 | vhost_net_vq_reset(n); |
3a4d5c94 MT |
1012 | done: |
1013 | mutex_unlock(&n->dev.mutex); | |
1014 | if (tx_sock) | |
09aaacf0 | 1015 | sockfd_put(tx_sock); |
3a4d5c94 | 1016 | if (rx_sock) |
09aaacf0 | 1017 | sockfd_put(rx_sock); |
3a4d5c94 MT |
1018 | return err; |
1019 | } | |
1020 | ||
1021 | static int vhost_net_set_features(struct vhost_net *n, u64 features) | |
1022 | { | |
8dd014ad | 1023 | size_t vhost_hlen, sock_hlen, hdr_len; |
3a4d5c94 | 1024 | int i; |
8dd014ad DS |
1025 | |
1026 | hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? | |
1027 | sizeof(struct virtio_net_hdr_mrg_rxbuf) : | |
1028 | sizeof(struct virtio_net_hdr); | |
1029 | if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) { | |
1030 | /* vhost provides vnet_hdr */ | |
1031 | vhost_hlen = hdr_len; | |
1032 | sock_hlen = 0; | |
1033 | } else { | |
1034 | /* socket provides vnet_hdr */ | |
1035 | vhost_hlen = 0; | |
1036 | sock_hlen = hdr_len; | |
1037 | } | |
3a4d5c94 MT |
1038 | mutex_lock(&n->dev.mutex); |
1039 | if ((features & (1 << VHOST_F_LOG_ALL)) && | |
1040 | !vhost_log_access_ok(&n->dev)) { | |
1041 | mutex_unlock(&n->dev.mutex); | |
1042 | return -EFAULT; | |
1043 | } | |
3a4d5c94 | 1044 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
3ab2e420 | 1045 | mutex_lock(&n->vqs[i].vq.mutex); |
ea16c514 | 1046 | n->vqs[i].vq.acked_features = features; |
81f95a55 MT |
1047 | n->vqs[i].vhost_hlen = vhost_hlen; |
1048 | n->vqs[i].sock_hlen = sock_hlen; | |
3ab2e420 | 1049 | mutex_unlock(&n->vqs[i].vq.mutex); |
3a4d5c94 | 1050 | } |
3a4d5c94 MT |
1051 | mutex_unlock(&n->dev.mutex); |
1052 | return 0; | |
1053 | } | |
1054 | ||
b1ad8496 AH |
1055 | static long vhost_net_set_owner(struct vhost_net *n) |
1056 | { | |
1057 | int r; | |
1058 | ||
1059 | mutex_lock(&n->dev.mutex); | |
05c05351 MT |
1060 | if (vhost_dev_has_owner(&n->dev)) { |
1061 | r = -EBUSY; | |
1062 | goto out; | |
1063 | } | |
b1ad8496 AH |
1064 | r = vhost_net_set_ubuf_info(n); |
1065 | if (r) | |
1066 | goto out; | |
1067 | r = vhost_dev_set_owner(&n->dev); | |
1068 | if (r) | |
1069 | vhost_net_clear_ubuf_info(n); | |
1070 | vhost_net_flush(n); | |
1071 | out: | |
1072 | mutex_unlock(&n->dev.mutex); | |
1073 | return r; | |
1074 | } | |
1075 | ||
3a4d5c94 MT |
1076 | static long vhost_net_ioctl(struct file *f, unsigned int ioctl, |
1077 | unsigned long arg) | |
1078 | { | |
1079 | struct vhost_net *n = f->private_data; | |
1080 | void __user *argp = (void __user *)arg; | |
1081 | u64 __user *featurep = argp; | |
1082 | struct vhost_vring_file backend; | |
1083 | u64 features; | |
1084 | int r; | |
d47effe1 | 1085 | |
3a4d5c94 MT |
1086 | switch (ioctl) { |
1087 | case VHOST_NET_SET_BACKEND: | |
d3553a52 TY |
1088 | if (copy_from_user(&backend, argp, sizeof backend)) |
1089 | return -EFAULT; | |
3a4d5c94 MT |
1090 | return vhost_net_set_backend(n, backend.index, backend.fd); |
1091 | case VHOST_GET_FEATURES: | |
0dd05a3b | 1092 | features = VHOST_NET_FEATURES; |
d3553a52 TY |
1093 | if (copy_to_user(featurep, &features, sizeof features)) |
1094 | return -EFAULT; | |
1095 | return 0; | |
3a4d5c94 | 1096 | case VHOST_SET_FEATURES: |
d3553a52 TY |
1097 | if (copy_from_user(&features, featurep, sizeof features)) |
1098 | return -EFAULT; | |
0dd05a3b | 1099 | if (features & ~VHOST_NET_FEATURES) |
3a4d5c94 MT |
1100 | return -EOPNOTSUPP; |
1101 | return vhost_net_set_features(n, features); | |
1102 | case VHOST_RESET_OWNER: | |
1103 | return vhost_net_reset_owner(n); | |
b1ad8496 AH |
1104 | case VHOST_SET_OWNER: |
1105 | return vhost_net_set_owner(n); | |
3a4d5c94 MT |
1106 | default: |
1107 | mutex_lock(&n->dev.mutex); | |
935cdee7 MT |
1108 | r = vhost_dev_ioctl(&n->dev, ioctl, argp); |
1109 | if (r == -ENOIOCTLCMD) | |
1110 | r = vhost_vring_ioctl(&n->dev, ioctl, argp); | |
1111 | else | |
1112 | vhost_net_flush(n); | |
3a4d5c94 MT |
1113 | mutex_unlock(&n->dev.mutex); |
1114 | return r; | |
1115 | } | |
1116 | } | |
1117 | ||
1118 | #ifdef CONFIG_COMPAT | |
1119 | static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl, | |
1120 | unsigned long arg) | |
1121 | { | |
1122 | return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); | |
1123 | } | |
1124 | #endif | |
1125 | ||
373a83a6 | 1126 | static const struct file_operations vhost_net_fops = { |
3a4d5c94 MT |
1127 | .owner = THIS_MODULE, |
1128 | .release = vhost_net_release, | |
1129 | .unlocked_ioctl = vhost_net_ioctl, | |
1130 | #ifdef CONFIG_COMPAT | |
1131 | .compat_ioctl = vhost_net_compat_ioctl, | |
1132 | #endif | |
1133 | .open = vhost_net_open, | |
6038f373 | 1134 | .llseek = noop_llseek, |
3a4d5c94 MT |
1135 | }; |
1136 | ||
1137 | static struct miscdevice vhost_net_misc = { | |
7c7c7f01 | 1138 | .minor = VHOST_NET_MINOR, |
1139 | .name = "vhost-net", | |
1140 | .fops = &vhost_net_fops, | |
3a4d5c94 MT |
1141 | }; |
1142 | ||
a8d3782f | 1143 | static int vhost_net_init(void) |
3a4d5c94 | 1144 | { |
bab632d6 | 1145 | if (experimental_zcopytx) |
fe729a57 | 1146 | vhost_net_enable_zcopy(VHOST_NET_VQ_TX); |
c23f3445 | 1147 | return misc_register(&vhost_net_misc); |
3a4d5c94 MT |
1148 | } |
1149 | module_init(vhost_net_init); | |
1150 | ||
a8d3782f | 1151 | static void vhost_net_exit(void) |
3a4d5c94 MT |
1152 | { |
1153 | misc_deregister(&vhost_net_misc); | |
3a4d5c94 MT |
1154 | } |
1155 | module_exit(vhost_net_exit); | |
1156 | ||
1157 | MODULE_VERSION("0.0.1"); | |
1158 | MODULE_LICENSE("GPL v2"); | |
1159 | MODULE_AUTHOR("Michael S. Tsirkin"); | |
1160 | MODULE_DESCRIPTION("Host kernel accelerator for virtio net"); | |
7c7c7f01 | 1161 | MODULE_ALIAS_MISCDEV(VHOST_NET_MINOR); |
1162 | MODULE_ALIAS("devname:vhost-net"); |