Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 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> |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 13 | #include <linux/miscdevice.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/mutex.h> |
| 16 | #include <linux/workqueue.h> |
| 17 | #include <linux/rcupdate.h> |
| 18 | #include <linux/file.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 19 | #include <linux/slab.h> |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 20 | |
| 21 | #include <linux/net.h> |
| 22 | #include <linux/if_packet.h> |
| 23 | #include <linux/if_arp.h> |
| 24 | #include <linux/if_tun.h> |
Arnd Bergmann | 501c774 | 2010-02-18 05:46:50 +0000 | [diff] [blame] | 25 | #include <linux/if_macvlan.h> |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 26 | |
| 27 | #include <net/sock.h> |
| 28 | |
| 29 | #include "vhost.h" |
| 30 | |
| 31 | /* Max number of bytes transferred before requeueing the job. |
| 32 | * Using this limit prevents one virtqueue from starving others. */ |
| 33 | #define VHOST_NET_WEIGHT 0x80000 |
| 34 | |
| 35 | enum { |
| 36 | VHOST_NET_VQ_RX = 0, |
| 37 | VHOST_NET_VQ_TX = 1, |
| 38 | VHOST_NET_VQ_MAX = 2, |
| 39 | }; |
| 40 | |
| 41 | enum vhost_net_poll_state { |
| 42 | VHOST_NET_POLL_DISABLED = 0, |
| 43 | VHOST_NET_POLL_STARTED = 1, |
| 44 | VHOST_NET_POLL_STOPPED = 2, |
| 45 | }; |
| 46 | |
| 47 | struct vhost_net { |
| 48 | struct vhost_dev dev; |
| 49 | struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX]; |
| 50 | struct vhost_poll poll[VHOST_NET_VQ_MAX]; |
| 51 | /* Tells us whether we are polling a socket for TX. |
| 52 | * We only do this when socket buffer fills up. |
| 53 | * Protected by tx vq lock. */ |
| 54 | enum vhost_net_poll_state tx_poll_state; |
| 55 | }; |
| 56 | |
| 57 | /* Pop first len bytes from iovec. Return number of segments used. */ |
| 58 | static int move_iovec_hdr(struct iovec *from, struct iovec *to, |
| 59 | size_t len, int iov_count) |
| 60 | { |
| 61 | int seg = 0; |
| 62 | size_t size; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 63 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 64 | while (len && seg < iov_count) { |
| 65 | size = min(from->iov_len, len); |
| 66 | to->iov_base = from->iov_base; |
| 67 | to->iov_len = size; |
| 68 | from->iov_len -= size; |
| 69 | from->iov_base += size; |
| 70 | len -= size; |
| 71 | ++from; |
| 72 | ++to; |
| 73 | ++seg; |
| 74 | } |
| 75 | return seg; |
| 76 | } |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 77 | /* Copy iovec entries for len bytes from iovec. */ |
| 78 | static void copy_iovec_hdr(const struct iovec *from, struct iovec *to, |
| 79 | size_t len, int iovcount) |
| 80 | { |
| 81 | int seg = 0; |
| 82 | size_t size; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 83 | |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 84 | while (len && seg < iovcount) { |
| 85 | size = min(from->iov_len, len); |
| 86 | to->iov_base = from->iov_base; |
| 87 | to->iov_len = size; |
| 88 | len -= size; |
| 89 | ++from; |
| 90 | ++to; |
| 91 | ++seg; |
| 92 | } |
| 93 | } |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 94 | |
| 95 | /* Caller must have TX VQ lock */ |
| 96 | static void tx_poll_stop(struct vhost_net *net) |
| 97 | { |
| 98 | if (likely(net->tx_poll_state != VHOST_NET_POLL_STARTED)) |
| 99 | return; |
| 100 | vhost_poll_stop(net->poll + VHOST_NET_VQ_TX); |
| 101 | net->tx_poll_state = VHOST_NET_POLL_STOPPED; |
| 102 | } |
| 103 | |
| 104 | /* Caller must have TX VQ lock */ |
| 105 | static void tx_poll_start(struct vhost_net *net, struct socket *sock) |
| 106 | { |
| 107 | if (unlikely(net->tx_poll_state != VHOST_NET_POLL_STOPPED)) |
| 108 | return; |
| 109 | vhost_poll_start(net->poll + VHOST_NET_VQ_TX, sock->file); |
| 110 | net->tx_poll_state = VHOST_NET_POLL_STARTED; |
| 111 | } |
| 112 | |
| 113 | /* Expects to be always run from workqueue - which acts as |
| 114 | * read-size critical section for our kind of RCU. */ |
| 115 | static void handle_tx(struct vhost_net *net) |
| 116 | { |
| 117 | struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX]; |
Michael S. Tsirkin | d5675bd | 2010-06-24 16:59:59 +0300 | [diff] [blame] | 118 | unsigned out, in, s; |
| 119 | int head; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 120 | struct msghdr msg = { |
| 121 | .msg_name = NULL, |
| 122 | .msg_namelen = 0, |
| 123 | .msg_control = NULL, |
| 124 | .msg_controllen = 0, |
| 125 | .msg_iov = vq->iov, |
| 126 | .msg_flags = MSG_DONTWAIT, |
| 127 | }; |
| 128 | size_t len, total_len = 0; |
| 129 | int err, wmem; |
| 130 | size_t hdr_size; |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 131 | struct socket *sock; |
| 132 | |
Michael S. Tsirkin | 5e18247 | 2011-01-18 13:04:43 +0200 | [diff] [blame] | 133 | /* TODO: check that we are running from vhost_worker? */ |
Michael S. Tsirkin | 11cd1a8 | 2010-11-14 17:31:52 +0200 | [diff] [blame] | 134 | sock = rcu_dereference_check(vq->private_data, 1); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 135 | if (!sock) |
| 136 | return; |
| 137 | |
| 138 | wmem = atomic_read(&sock->sk->sk_wmem_alloc); |
Sridhar Samudrala | 39286fa | 2010-02-28 19:39:16 +0200 | [diff] [blame] | 139 | if (wmem >= sock->sk->sk_sndbuf) { |
| 140 | mutex_lock(&vq->mutex); |
| 141 | tx_poll_start(net, sock); |
| 142 | mutex_unlock(&vq->mutex); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 143 | return; |
Sridhar Samudrala | 39286fa | 2010-02-28 19:39:16 +0200 | [diff] [blame] | 144 | } |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 145 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 146 | mutex_lock(&vq->mutex); |
| 147 | vhost_disable_notify(vq); |
| 148 | |
Michael S. Tsirkin | 0e25557 | 2010-03-08 23:24:22 +0200 | [diff] [blame] | 149 | if (wmem < sock->sk->sk_sndbuf / 2) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 150 | tx_poll_stop(net); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 151 | hdr_size = vq->vhost_hlen; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 152 | |
| 153 | for (;;) { |
| 154 | head = vhost_get_vq_desc(&net->dev, vq, vq->iov, |
| 155 | ARRAY_SIZE(vq->iov), |
| 156 | &out, &in, |
| 157 | NULL, NULL); |
Michael S. Tsirkin | d5675bd | 2010-06-24 16:59:59 +0300 | [diff] [blame] | 158 | /* On error, stop handling until the next kick. */ |
Michael S. Tsirkin | 7b3384f | 2010-07-01 18:40:12 +0300 | [diff] [blame] | 159 | if (unlikely(head < 0)) |
Michael S. Tsirkin | d5675bd | 2010-06-24 16:59:59 +0300 | [diff] [blame] | 160 | break; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 161 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
| 162 | if (head == vq->num) { |
| 163 | wmem = atomic_read(&sock->sk->sk_wmem_alloc); |
| 164 | if (wmem >= sock->sk->sk_sndbuf * 3 / 4) { |
| 165 | tx_poll_start(net, sock); |
| 166 | set_bit(SOCK_ASYNC_NOSPACE, &sock->flags); |
| 167 | break; |
| 168 | } |
| 169 | if (unlikely(vhost_enable_notify(vq))) { |
| 170 | vhost_disable_notify(vq); |
| 171 | continue; |
| 172 | } |
| 173 | break; |
| 174 | } |
| 175 | if (in) { |
| 176 | vq_err(vq, "Unexpected descriptor format for TX: " |
| 177 | "out %d, int %d\n", out, in); |
| 178 | break; |
| 179 | } |
| 180 | /* Skip header. TODO: support TSO. */ |
| 181 | s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, out); |
| 182 | msg.msg_iovlen = out; |
| 183 | len = iov_length(vq->iov, out); |
| 184 | /* Sanity check */ |
| 185 | if (!len) { |
| 186 | vq_err(vq, "Unexpected header len for TX: " |
| 187 | "%zd expected %zd\n", |
| 188 | iov_length(vq->hdr, s), hdr_size); |
| 189 | break; |
| 190 | } |
| 191 | /* TODO: Check specific error and bomb out unless ENOBUFS? */ |
| 192 | err = sock->ops->sendmsg(NULL, sock, &msg, len); |
| 193 | if (unlikely(err < 0)) { |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 194 | vhost_discard_vq_desc(vq, 1); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 195 | tx_poll_start(net, sock); |
| 196 | break; |
| 197 | } |
| 198 | if (err != len) |
Michael S. Tsirkin | 95c0ec6 | 2010-06-24 17:10:25 +0300 | [diff] [blame] | 199 | pr_debug("Truncated TX packet: " |
| 200 | " len %d != %zd\n", err, len); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 201 | vhost_add_used_and_signal(&net->dev, vq, head, 0); |
| 202 | total_len += len; |
| 203 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { |
| 204 | vhost_poll_queue(&vq->poll); |
| 205 | break; |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | mutex_unlock(&vq->mutex); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 210 | } |
| 211 | |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 212 | static int peek_head_len(struct sock *sk) |
| 213 | { |
| 214 | struct sk_buff *head; |
| 215 | int len = 0; |
Jason Wang | 783e398 | 2011-01-17 16:11:17 +0800 | [diff] [blame] | 216 | unsigned long flags; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 217 | |
Jason Wang | 783e398 | 2011-01-17 16:11:17 +0800 | [diff] [blame] | 218 | spin_lock_irqsave(&sk->sk_receive_queue.lock, flags); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 219 | head = skb_peek(&sk->sk_receive_queue); |
Jason Wang | 783e398 | 2011-01-17 16:11:17 +0800 | [diff] [blame] | 220 | if (likely(head)) |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 221 | len = head->len; |
Jason Wang | 783e398 | 2011-01-17 16:11:17 +0800 | [diff] [blame] | 222 | spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 223 | return len; |
| 224 | } |
| 225 | |
| 226 | /* This is a multi-buffer version of vhost_get_desc, that works if |
| 227 | * vq has read descriptors only. |
| 228 | * @vq - the relevant virtqueue |
| 229 | * @datalen - data length we'll be reading |
| 230 | * @iovcount - returned count of io vectors we fill |
| 231 | * @log - vhost log |
| 232 | * @log_num - log offset |
Jason Wang | 9424936 | 2011-01-17 16:11:08 +0800 | [diff] [blame] | 233 | * @quota - headcount quota, 1 for big buffer |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 234 | * returns number of buffer heads allocated, negative on error |
| 235 | */ |
| 236 | static int get_rx_bufs(struct vhost_virtqueue *vq, |
| 237 | struct vring_used_elem *heads, |
| 238 | int datalen, |
| 239 | unsigned *iovcount, |
| 240 | struct vhost_log *log, |
Jason Wang | 9424936 | 2011-01-17 16:11:08 +0800 | [diff] [blame] | 241 | unsigned *log_num, |
| 242 | unsigned int quota) |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 243 | { |
| 244 | unsigned int out, in; |
| 245 | int seg = 0; |
| 246 | int headcount = 0; |
| 247 | unsigned d; |
| 248 | int r, nlogs = 0; |
| 249 | |
Jason Wang | 9424936 | 2011-01-17 16:11:08 +0800 | [diff] [blame] | 250 | while (datalen > 0 && headcount < quota) { |
Jason Wang | e0e9b40 | 2010-09-14 23:53:05 +0800 | [diff] [blame] | 251 | if (unlikely(seg >= UIO_MAXIOV)) { |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 252 | r = -ENOBUFS; |
| 253 | goto err; |
| 254 | } |
| 255 | d = vhost_get_vq_desc(vq->dev, vq, vq->iov + seg, |
| 256 | ARRAY_SIZE(vq->iov) - seg, &out, |
| 257 | &in, log, log_num); |
| 258 | if (d == vq->num) { |
| 259 | r = 0; |
| 260 | goto err; |
| 261 | } |
| 262 | if (unlikely(out || in <= 0)) { |
| 263 | vq_err(vq, "unexpected descriptor format for RX: " |
| 264 | "out %d, in %d\n", out, in); |
| 265 | r = -EINVAL; |
| 266 | goto err; |
| 267 | } |
| 268 | if (unlikely(log)) { |
| 269 | nlogs += *log_num; |
| 270 | log += *log_num; |
| 271 | } |
| 272 | heads[headcount].id = d; |
| 273 | heads[headcount].len = iov_length(vq->iov + seg, in); |
| 274 | datalen -= heads[headcount].len; |
| 275 | ++headcount; |
| 276 | seg += in; |
| 277 | } |
| 278 | heads[headcount - 1].len += datalen; |
| 279 | *iovcount = seg; |
| 280 | if (unlikely(log)) |
| 281 | *log_num = nlogs; |
| 282 | return headcount; |
| 283 | err: |
| 284 | vhost_discard_vq_desc(vq, headcount); |
| 285 | return r; |
| 286 | } |
| 287 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 288 | /* Expects to be always run from workqueue - which acts as |
| 289 | * read-size critical section for our kind of RCU. */ |
Jason Wang | 9424936 | 2011-01-17 16:11:08 +0800 | [diff] [blame] | 290 | static void handle_rx(struct vhost_net *net) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 291 | { |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 292 | struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX]; |
| 293 | unsigned uninitialized_var(in), log; |
| 294 | struct vhost_log *vq_log; |
| 295 | struct msghdr msg = { |
| 296 | .msg_name = NULL, |
| 297 | .msg_namelen = 0, |
| 298 | .msg_control = NULL, /* FIXME: get and handle RX aux data. */ |
| 299 | .msg_controllen = 0, |
| 300 | .msg_iov = vq->iov, |
| 301 | .msg_flags = MSG_DONTWAIT, |
| 302 | }; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 303 | struct virtio_net_hdr_mrg_rxbuf hdr = { |
| 304 | .hdr.flags = 0, |
| 305 | .hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE |
| 306 | }; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 307 | size_t total_len = 0; |
Jason Wang | cfbdab9 | 2011-01-17 16:10:59 +0800 | [diff] [blame] | 308 | int err, headcount, mergeable; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 309 | size_t vhost_hlen, sock_hlen; |
| 310 | size_t vhost_len, sock_len; |
Michael S. Tsirkin | 5e18247 | 2011-01-18 13:04:43 +0200 | [diff] [blame] | 311 | /* TODO: check that we are running from vhost_worker? */ |
| 312 | struct socket *sock = rcu_dereference_check(vq->private_data, 1); |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 313 | |
Michael S. Tsirkin | de4d768 | 2011-03-13 23:00:52 +0200 | [diff] [blame] | 314 | if (!sock) |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 315 | return; |
| 316 | |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 317 | mutex_lock(&vq->mutex); |
| 318 | vhost_disable_notify(vq); |
| 319 | vhost_hlen = vq->vhost_hlen; |
| 320 | sock_hlen = vq->sock_hlen; |
| 321 | |
| 322 | vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ? |
| 323 | vq->log : NULL; |
Jason Wang | cfbdab9 | 2011-01-17 16:10:59 +0800 | [diff] [blame] | 324 | mergeable = vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 325 | |
| 326 | while ((sock_len = peek_head_len(sock->sk))) { |
| 327 | sock_len += sock_hlen; |
| 328 | vhost_len = sock_len + vhost_hlen; |
| 329 | headcount = get_rx_bufs(vq, vq->heads, vhost_len, |
Jason Wang | 9424936 | 2011-01-17 16:11:08 +0800 | [diff] [blame] | 330 | &in, vq_log, &log, |
| 331 | likely(mergeable) ? UIO_MAXIOV : 1); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 332 | /* On error, stop handling until the next kick. */ |
| 333 | if (unlikely(headcount < 0)) |
| 334 | break; |
| 335 | /* OK, now we need to know about added descriptors. */ |
| 336 | if (!headcount) { |
| 337 | if (unlikely(vhost_enable_notify(vq))) { |
| 338 | /* They have slipped one in as we were |
| 339 | * doing that: check again. */ |
| 340 | vhost_disable_notify(vq); |
| 341 | continue; |
| 342 | } |
| 343 | /* Nothing new? Wait for eventfd to tell us |
| 344 | * they refilled. */ |
| 345 | break; |
| 346 | } |
| 347 | /* We don't need to be notified again. */ |
| 348 | if (unlikely((vhost_hlen))) |
| 349 | /* Skip header. TODO: support TSO. */ |
| 350 | move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in); |
| 351 | else |
| 352 | /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF: |
Jason Wang | a290aec | 2010-11-29 13:48:40 +0800 | [diff] [blame] | 353 | * needed because recvmsg can modify msg_iov. */ |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 354 | copy_iovec_hdr(vq->iov, vq->hdr, sock_hlen, in); |
| 355 | msg.msg_iovlen = in; |
| 356 | err = sock->ops->recvmsg(NULL, sock, &msg, |
| 357 | sock_len, MSG_DONTWAIT | MSG_TRUNC); |
| 358 | /* Userspace might have consumed the packet meanwhile: |
| 359 | * it's not supposed to do this usually, but might be hard |
| 360 | * to prevent. Discard data we got (if any) and keep going. */ |
| 361 | if (unlikely(err != sock_len)) { |
| 362 | pr_debug("Discarded rx packet: " |
| 363 | " len %d, expected %zd\n", err, sock_len); |
| 364 | vhost_discard_vq_desc(vq, headcount); |
| 365 | continue; |
| 366 | } |
| 367 | if (unlikely(vhost_hlen) && |
| 368 | memcpy_toiovecend(vq->hdr, (unsigned char *)&hdr, 0, |
| 369 | vhost_hlen)) { |
| 370 | vq_err(vq, "Unable to write vnet_hdr at addr %p\n", |
| 371 | vq->iov->iov_base); |
| 372 | break; |
| 373 | } |
| 374 | /* TODO: Should check and handle checksum. */ |
Jason Wang | cfbdab9 | 2011-01-17 16:10:59 +0800 | [diff] [blame] | 375 | if (likely(mergeable) && |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 376 | memcpy_toiovecend(vq->hdr, (unsigned char *)&headcount, |
| 377 | offsetof(typeof(hdr), num_buffers), |
| 378 | sizeof hdr.num_buffers)) { |
| 379 | vq_err(vq, "Failed num_buffers write"); |
| 380 | vhost_discard_vq_desc(vq, headcount); |
| 381 | break; |
| 382 | } |
| 383 | vhost_add_used_and_signal_n(&net->dev, vq, vq->heads, |
| 384 | headcount); |
| 385 | if (unlikely(vq_log)) |
| 386 | vhost_log_write(vq, vq_log, log, vhost_len); |
| 387 | total_len += vhost_len; |
| 388 | if (unlikely(total_len >= VHOST_NET_WEIGHT)) { |
| 389 | vhost_poll_queue(&vq->poll); |
| 390 | break; |
| 391 | } |
| 392 | } |
| 393 | |
| 394 | mutex_unlock(&vq->mutex); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 395 | } |
| 396 | |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 397 | static void handle_tx_kick(struct vhost_work *work) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 398 | { |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 399 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 400 | poll.work); |
| 401 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); |
| 402 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 403 | handle_tx(net); |
| 404 | } |
| 405 | |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 406 | static void handle_rx_kick(struct vhost_work *work) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 407 | { |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 408 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, |
| 409 | poll.work); |
| 410 | struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev); |
| 411 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 412 | handle_rx(net); |
| 413 | } |
| 414 | |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 415 | static void handle_tx_net(struct vhost_work *work) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 416 | { |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 417 | struct vhost_net *net = container_of(work, struct vhost_net, |
| 418 | poll[VHOST_NET_VQ_TX].work); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 419 | handle_tx(net); |
| 420 | } |
| 421 | |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 422 | static void handle_rx_net(struct vhost_work *work) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 423 | { |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 424 | struct vhost_net *net = container_of(work, struct vhost_net, |
| 425 | poll[VHOST_NET_VQ_RX].work); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 426 | handle_rx(net); |
| 427 | } |
| 428 | |
| 429 | static int vhost_net_open(struct inode *inode, struct file *f) |
| 430 | { |
| 431 | struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL); |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 432 | struct vhost_dev *dev; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 433 | int r; |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 434 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 435 | if (!n) |
| 436 | return -ENOMEM; |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 437 | |
| 438 | dev = &n->dev; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 439 | n->vqs[VHOST_NET_VQ_TX].handle_kick = handle_tx_kick; |
| 440 | n->vqs[VHOST_NET_VQ_RX].handle_kick = handle_rx_kick; |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 441 | r = vhost_dev_init(dev, n->vqs, VHOST_NET_VQ_MAX); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 442 | if (r < 0) { |
| 443 | kfree(n); |
| 444 | return r; |
| 445 | } |
| 446 | |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 447 | vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev); |
| 448 | vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 449 | n->tx_poll_state = VHOST_NET_POLL_DISABLED; |
| 450 | |
| 451 | f->private_data = n; |
| 452 | |
| 453 | return 0; |
| 454 | } |
| 455 | |
| 456 | static void vhost_net_disable_vq(struct vhost_net *n, |
| 457 | struct vhost_virtqueue *vq) |
| 458 | { |
| 459 | if (!vq->private_data) |
| 460 | return; |
| 461 | if (vq == n->vqs + VHOST_NET_VQ_TX) { |
| 462 | tx_poll_stop(n); |
| 463 | n->tx_poll_state = VHOST_NET_POLL_DISABLED; |
| 464 | } else |
| 465 | vhost_poll_stop(n->poll + VHOST_NET_VQ_RX); |
| 466 | } |
| 467 | |
| 468 | static void vhost_net_enable_vq(struct vhost_net *n, |
| 469 | struct vhost_virtqueue *vq) |
| 470 | { |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 471 | struct socket *sock; |
| 472 | |
| 473 | sock = rcu_dereference_protected(vq->private_data, |
| 474 | lockdep_is_held(&vq->mutex)); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 475 | if (!sock) |
| 476 | return; |
| 477 | if (vq == n->vqs + VHOST_NET_VQ_TX) { |
| 478 | n->tx_poll_state = VHOST_NET_POLL_STOPPED; |
| 479 | tx_poll_start(n, sock); |
| 480 | } else |
| 481 | vhost_poll_start(n->poll + VHOST_NET_VQ_RX, sock->file); |
| 482 | } |
| 483 | |
| 484 | static struct socket *vhost_net_stop_vq(struct vhost_net *n, |
| 485 | struct vhost_virtqueue *vq) |
| 486 | { |
| 487 | struct socket *sock; |
| 488 | |
| 489 | mutex_lock(&vq->mutex); |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 490 | sock = rcu_dereference_protected(vq->private_data, |
| 491 | lockdep_is_held(&vq->mutex)); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 492 | vhost_net_disable_vq(n, vq); |
| 493 | rcu_assign_pointer(vq->private_data, NULL); |
| 494 | mutex_unlock(&vq->mutex); |
| 495 | return sock; |
| 496 | } |
| 497 | |
| 498 | static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock, |
| 499 | struct socket **rx_sock) |
| 500 | { |
| 501 | *tx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_TX); |
| 502 | *rx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_RX); |
| 503 | } |
| 504 | |
| 505 | static void vhost_net_flush_vq(struct vhost_net *n, int index) |
| 506 | { |
| 507 | vhost_poll_flush(n->poll + index); |
| 508 | vhost_poll_flush(&n->dev.vqs[index].poll); |
| 509 | } |
| 510 | |
| 511 | static void vhost_net_flush(struct vhost_net *n) |
| 512 | { |
| 513 | vhost_net_flush_vq(n, VHOST_NET_VQ_TX); |
| 514 | vhost_net_flush_vq(n, VHOST_NET_VQ_RX); |
| 515 | } |
| 516 | |
| 517 | static int vhost_net_release(struct inode *inode, struct file *f) |
| 518 | { |
| 519 | struct vhost_net *n = f->private_data; |
| 520 | struct socket *tx_sock; |
| 521 | struct socket *rx_sock; |
| 522 | |
| 523 | vhost_net_stop(n, &tx_sock, &rx_sock); |
| 524 | vhost_net_flush(n); |
| 525 | vhost_dev_cleanup(&n->dev); |
| 526 | if (tx_sock) |
| 527 | fput(tx_sock->file); |
| 528 | if (rx_sock) |
| 529 | fput(rx_sock->file); |
| 530 | /* We do an extra flush before freeing memory, |
| 531 | * since jobs can re-queue themselves. */ |
| 532 | vhost_net_flush(n); |
| 533 | kfree(n); |
| 534 | return 0; |
| 535 | } |
| 536 | |
| 537 | static struct socket *get_raw_socket(int fd) |
| 538 | { |
| 539 | struct { |
| 540 | struct sockaddr_ll sa; |
| 541 | char buf[MAX_ADDR_LEN]; |
| 542 | } uaddr; |
| 543 | int uaddr_len = sizeof uaddr, r; |
| 544 | struct socket *sock = sockfd_lookup(fd, &r); |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 545 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 546 | if (!sock) |
| 547 | return ERR_PTR(-ENOTSOCK); |
| 548 | |
| 549 | /* Parameter checking */ |
| 550 | if (sock->sk->sk_type != SOCK_RAW) { |
| 551 | r = -ESOCKTNOSUPPORT; |
| 552 | goto err; |
| 553 | } |
| 554 | |
| 555 | r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa, |
| 556 | &uaddr_len, 0); |
| 557 | if (r) |
| 558 | goto err; |
| 559 | |
| 560 | if (uaddr.sa.sll_family != AF_PACKET) { |
| 561 | r = -EPFNOSUPPORT; |
| 562 | goto err; |
| 563 | } |
| 564 | return sock; |
| 565 | err: |
| 566 | fput(sock->file); |
| 567 | return ERR_PTR(r); |
| 568 | } |
| 569 | |
Arnd Bergmann | 501c774 | 2010-02-18 05:46:50 +0000 | [diff] [blame] | 570 | static struct socket *get_tap_socket(int fd) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 571 | { |
| 572 | struct file *file = fget(fd); |
| 573 | struct socket *sock; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 574 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 575 | if (!file) |
| 576 | return ERR_PTR(-EBADF); |
| 577 | sock = tun_get_socket(file); |
Arnd Bergmann | 501c774 | 2010-02-18 05:46:50 +0000 | [diff] [blame] | 578 | if (!IS_ERR(sock)) |
| 579 | return sock; |
| 580 | sock = macvtap_get_socket(file); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 581 | if (IS_ERR(sock)) |
| 582 | fput(file); |
| 583 | return sock; |
| 584 | } |
| 585 | |
| 586 | static struct socket *get_socket(int fd) |
| 587 | { |
| 588 | struct socket *sock; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 589 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 590 | /* special case to disable backend */ |
| 591 | if (fd == -1) |
| 592 | return NULL; |
| 593 | sock = get_raw_socket(fd); |
| 594 | if (!IS_ERR(sock)) |
| 595 | return sock; |
Arnd Bergmann | 501c774 | 2010-02-18 05:46:50 +0000 | [diff] [blame] | 596 | sock = get_tap_socket(fd); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 597 | if (!IS_ERR(sock)) |
| 598 | return sock; |
| 599 | return ERR_PTR(-ENOTSOCK); |
| 600 | } |
| 601 | |
| 602 | static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) |
| 603 | { |
| 604 | struct socket *sock, *oldsock; |
| 605 | struct vhost_virtqueue *vq; |
| 606 | int r; |
| 607 | |
| 608 | mutex_lock(&n->dev.mutex); |
| 609 | r = vhost_dev_check_owner(&n->dev); |
| 610 | if (r) |
| 611 | goto err; |
| 612 | |
| 613 | if (index >= VHOST_NET_VQ_MAX) { |
| 614 | r = -ENOBUFS; |
| 615 | goto err; |
| 616 | } |
| 617 | vq = n->vqs + index; |
| 618 | mutex_lock(&vq->mutex); |
| 619 | |
| 620 | /* Verify that ring has been setup correctly. */ |
| 621 | if (!vhost_vq_access_ok(vq)) { |
| 622 | r = -EFAULT; |
Jeff Dike | 1dace8c | 2010-03-04 16:10:14 -0500 | [diff] [blame] | 623 | goto err_vq; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 624 | } |
| 625 | sock = get_socket(fd); |
| 626 | if (IS_ERR(sock)) { |
| 627 | r = PTR_ERR(sock); |
Jeff Dike | 1dace8c | 2010-03-04 16:10:14 -0500 | [diff] [blame] | 628 | goto err_vq; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 629 | } |
| 630 | |
| 631 | /* start polling new socket */ |
Arnd Bergmann | 28457ee | 2010-03-09 19:24:45 +0100 | [diff] [blame] | 632 | oldsock = rcu_dereference_protected(vq->private_data, |
| 633 | lockdep_is_held(&vq->mutex)); |
David S. Miller | 11fe883 | 2010-07-20 18:25:24 -0700 | [diff] [blame] | 634 | if (sock != oldsock) { |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 635 | vhost_net_disable_vq(n, vq); |
| 636 | rcu_assign_pointer(vq->private_data, sock); |
| 637 | vhost_net_enable_vq(n, vq); |
Jeff Dike | dd1f407 | 2010-03-04 16:10:14 -0500 | [diff] [blame] | 638 | } |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 639 | |
Michael S. Tsirkin | 1680e90 | 2010-07-15 15:19:12 +0300 | [diff] [blame] | 640 | mutex_unlock(&vq->mutex); |
| 641 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 642 | if (oldsock) { |
| 643 | vhost_net_flush_vq(n, index); |
| 644 | fput(oldsock->file); |
| 645 | } |
Jeff Dike | 1dace8c | 2010-03-04 16:10:14 -0500 | [diff] [blame] | 646 | |
Michael S. Tsirkin | 1680e90 | 2010-07-15 15:19:12 +0300 | [diff] [blame] | 647 | mutex_unlock(&n->dev.mutex); |
| 648 | return 0; |
| 649 | |
Jeff Dike | 1dace8c | 2010-03-04 16:10:14 -0500 | [diff] [blame] | 650 | err_vq: |
| 651 | mutex_unlock(&vq->mutex); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 652 | err: |
| 653 | mutex_unlock(&n->dev.mutex); |
| 654 | return r; |
| 655 | } |
| 656 | |
| 657 | static long vhost_net_reset_owner(struct vhost_net *n) |
| 658 | { |
| 659 | struct socket *tx_sock = NULL; |
| 660 | struct socket *rx_sock = NULL; |
| 661 | long err; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 662 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 663 | mutex_lock(&n->dev.mutex); |
| 664 | err = vhost_dev_check_owner(&n->dev); |
| 665 | if (err) |
| 666 | goto done; |
| 667 | vhost_net_stop(n, &tx_sock, &rx_sock); |
| 668 | vhost_net_flush(n); |
| 669 | err = vhost_dev_reset_owner(&n->dev); |
| 670 | done: |
| 671 | mutex_unlock(&n->dev.mutex); |
| 672 | if (tx_sock) |
| 673 | fput(tx_sock->file); |
| 674 | if (rx_sock) |
| 675 | fput(rx_sock->file); |
| 676 | return err; |
| 677 | } |
| 678 | |
| 679 | static int vhost_net_set_features(struct vhost_net *n, u64 features) |
| 680 | { |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 681 | size_t vhost_hlen, sock_hlen, hdr_len; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 682 | int i; |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 683 | |
| 684 | hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? |
| 685 | sizeof(struct virtio_net_hdr_mrg_rxbuf) : |
| 686 | sizeof(struct virtio_net_hdr); |
| 687 | if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) { |
| 688 | /* vhost provides vnet_hdr */ |
| 689 | vhost_hlen = hdr_len; |
| 690 | sock_hlen = 0; |
| 691 | } else { |
| 692 | /* socket provides vnet_hdr */ |
| 693 | vhost_hlen = 0; |
| 694 | sock_hlen = hdr_len; |
| 695 | } |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 696 | mutex_lock(&n->dev.mutex); |
| 697 | if ((features & (1 << VHOST_F_LOG_ALL)) && |
| 698 | !vhost_log_access_ok(&n->dev)) { |
| 699 | mutex_unlock(&n->dev.mutex); |
| 700 | return -EFAULT; |
| 701 | } |
| 702 | n->dev.acked_features = features; |
| 703 | smp_wmb(); |
| 704 | for (i = 0; i < VHOST_NET_VQ_MAX; ++i) { |
| 705 | mutex_lock(&n->vqs[i].mutex); |
David Stevens | 8dd014a | 2010-07-27 18:52:21 +0300 | [diff] [blame] | 706 | n->vqs[i].vhost_hlen = vhost_hlen; |
| 707 | n->vqs[i].sock_hlen = sock_hlen; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 708 | mutex_unlock(&n->vqs[i].mutex); |
| 709 | } |
| 710 | vhost_net_flush(n); |
| 711 | mutex_unlock(&n->dev.mutex); |
| 712 | return 0; |
| 713 | } |
| 714 | |
| 715 | static long vhost_net_ioctl(struct file *f, unsigned int ioctl, |
| 716 | unsigned long arg) |
| 717 | { |
| 718 | struct vhost_net *n = f->private_data; |
| 719 | void __user *argp = (void __user *)arg; |
| 720 | u64 __user *featurep = argp; |
| 721 | struct vhost_vring_file backend; |
| 722 | u64 features; |
| 723 | int r; |
Krishna Kumar | d47effe | 2011-03-01 17:06:37 +0530 | [diff] [blame] | 724 | |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 725 | switch (ioctl) { |
| 726 | case VHOST_NET_SET_BACKEND: |
Takuya Yoshikawa | d3553a5 | 2010-05-27 19:01:58 +0900 | [diff] [blame] | 727 | if (copy_from_user(&backend, argp, sizeof backend)) |
| 728 | return -EFAULT; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 729 | return vhost_net_set_backend(n, backend.index, backend.fd); |
| 730 | case VHOST_GET_FEATURES: |
| 731 | features = VHOST_FEATURES; |
Takuya Yoshikawa | d3553a5 | 2010-05-27 19:01:58 +0900 | [diff] [blame] | 732 | if (copy_to_user(featurep, &features, sizeof features)) |
| 733 | return -EFAULT; |
| 734 | return 0; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 735 | case VHOST_SET_FEATURES: |
Takuya Yoshikawa | d3553a5 | 2010-05-27 19:01:58 +0900 | [diff] [blame] | 736 | if (copy_from_user(&features, featurep, sizeof features)) |
| 737 | return -EFAULT; |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 738 | if (features & ~VHOST_FEATURES) |
| 739 | return -EOPNOTSUPP; |
| 740 | return vhost_net_set_features(n, features); |
| 741 | case VHOST_RESET_OWNER: |
| 742 | return vhost_net_reset_owner(n); |
| 743 | default: |
| 744 | mutex_lock(&n->dev.mutex); |
| 745 | r = vhost_dev_ioctl(&n->dev, ioctl, arg); |
| 746 | vhost_net_flush(n); |
| 747 | mutex_unlock(&n->dev.mutex); |
| 748 | return r; |
| 749 | } |
| 750 | } |
| 751 | |
| 752 | #ifdef CONFIG_COMPAT |
| 753 | static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl, |
| 754 | unsigned long arg) |
| 755 | { |
| 756 | return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg)); |
| 757 | } |
| 758 | #endif |
| 759 | |
Tobias Klauser | 373a83a | 2010-05-17 15:12:49 +0200 | [diff] [blame] | 760 | static const struct file_operations vhost_net_fops = { |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 761 | .owner = THIS_MODULE, |
| 762 | .release = vhost_net_release, |
| 763 | .unlocked_ioctl = vhost_net_ioctl, |
| 764 | #ifdef CONFIG_COMPAT |
| 765 | .compat_ioctl = vhost_net_compat_ioctl, |
| 766 | #endif |
| 767 | .open = vhost_net_open, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 768 | .llseek = noop_llseek, |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 769 | }; |
| 770 | |
| 771 | static struct miscdevice vhost_net_misc = { |
Alan Cox | 79907d8 | 2010-06-09 09:39:49 +0100 | [diff] [blame] | 772 | MISC_DYNAMIC_MINOR, |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 773 | "vhost-net", |
| 774 | &vhost_net_fops, |
| 775 | }; |
| 776 | |
Christoph Hellwig | a8d3782 | 2010-04-13 14:11:25 -0400 | [diff] [blame] | 777 | static int vhost_net_init(void) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 778 | { |
Tejun Heo | c23f3445 | 2010-06-02 20:40:00 +0200 | [diff] [blame] | 779 | return misc_register(&vhost_net_misc); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 780 | } |
| 781 | module_init(vhost_net_init); |
| 782 | |
Christoph Hellwig | a8d3782 | 2010-04-13 14:11:25 -0400 | [diff] [blame] | 783 | static void vhost_net_exit(void) |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 784 | { |
| 785 | misc_deregister(&vhost_net_misc); |
Michael S. Tsirkin | 3a4d5c94e | 2010-01-14 06:17:27 +0000 | [diff] [blame] | 786 | } |
| 787 | module_exit(vhost_net_exit); |
| 788 | |
| 789 | MODULE_VERSION("0.0.1"); |
| 790 | MODULE_LICENSE("GPL v2"); |
| 791 | MODULE_AUTHOR("Michael S. Tsirkin"); |
| 792 | MODULE_DESCRIPTION("Host kernel accelerator for virtio net"); |