1 // SPDX-License-Identifier: GPL-2.0-only
3 * common code for virtio vsock
5 * Copyright (C) 2013-2015 Red Hat, Inc.
6 * Author: Asias He <asias@redhat.com>
7 * Stefan Hajnoczi <stefanha@redhat.com>
9 #include <linux/spinlock.h>
10 #include <linux/module.h>
11 #include <linux/sched/signal.h>
12 #include <linux/ctype.h>
13 #include <linux/list.h>
14 #include <linux/virtio.h>
15 #include <linux/virtio_ids.h>
16 #include <linux/virtio_config.h>
17 #include <linux/virtio_vsock.h>
18 #include <uapi/linux/vsockmon.h>
21 #include <net/af_vsock.h>
23 #define CREATE_TRACE_POINTS
24 #include <trace/events/vsock_virtio_transport_common.h>
26 /* How long to wait for graceful shutdown of a connection */
27 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
29 /* Threshold for detecting small packets to copy */
30 #define GOOD_COPY_LEN 128
32 static const struct virtio_transport *virtio_transport_get_ops(void)
34 const struct vsock_transport *t = vsock_core_get_transport();
36 return container_of(t, struct virtio_transport, transport);
39 static struct virtio_vsock_pkt *
40 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info *info,
47 struct virtio_vsock_pkt *pkt;
50 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
54 pkt->hdr.type = cpu_to_le16(info->type);
55 pkt->hdr.op = cpu_to_le16(info->op);
56 pkt->hdr.src_cid = cpu_to_le64(src_cid);
57 pkt->hdr.dst_cid = cpu_to_le64(dst_cid);
58 pkt->hdr.src_port = cpu_to_le32(src_port);
59 pkt->hdr.dst_port = cpu_to_le32(dst_port);
60 pkt->hdr.flags = cpu_to_le32(info->flags);
62 pkt->hdr.len = cpu_to_le32(len);
63 pkt->reply = info->reply;
66 if (info->msg && len > 0) {
67 pkt->buf = kmalloc(len, GFP_KERNEL);
73 err = memcpy_from_msg(pkt->buf, info->msg, len);
78 trace_virtio_transport_alloc_pkt(src_cid, src_port,
95 static struct sk_buff *virtio_transport_build_skb(void *opaque)
97 struct virtio_vsock_pkt *pkt = opaque;
98 struct af_vsockmon_hdr *hdr;
103 /* A packet could be split to fit the RX buffer, so we can retrieve
104 * the payload length from the header and the buffer pointer taking
105 * care of the offset in the original packet.
107 payload_len = le32_to_cpu(pkt->hdr.len);
108 payload_buf = pkt->buf + pkt->off;
110 skb = alloc_skb(sizeof(*hdr) + sizeof(pkt->hdr) + payload_len,
115 hdr = skb_put(skb, sizeof(*hdr));
117 /* pkt->hdr is little-endian so no need to byteswap here */
118 hdr->src_cid = pkt->hdr.src_cid;
119 hdr->src_port = pkt->hdr.src_port;
120 hdr->dst_cid = pkt->hdr.dst_cid;
121 hdr->dst_port = pkt->hdr.dst_port;
123 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
124 hdr->len = cpu_to_le16(sizeof(pkt->hdr));
125 memset(hdr->reserved, 0, sizeof(hdr->reserved));
127 switch (le16_to_cpu(pkt->hdr.op)) {
128 case VIRTIO_VSOCK_OP_REQUEST:
129 case VIRTIO_VSOCK_OP_RESPONSE:
130 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
132 case VIRTIO_VSOCK_OP_RST:
133 case VIRTIO_VSOCK_OP_SHUTDOWN:
134 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
136 case VIRTIO_VSOCK_OP_RW:
137 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
139 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
140 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
141 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
144 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
148 skb_put_data(skb, &pkt->hdr, sizeof(pkt->hdr));
151 skb_put_data(skb, payload_buf, payload_len);
157 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)
159 vsock_deliver_tap(virtio_transport_build_skb, pkt);
161 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
163 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
164 struct virtio_vsock_pkt_info *info)
166 u32 src_cid, src_port, dst_cid, dst_port;
167 struct virtio_vsock_sock *vvs;
168 struct virtio_vsock_pkt *pkt;
169 u32 pkt_len = info->pkt_len;
171 src_cid = vm_sockets_get_local_cid();
172 src_port = vsk->local_addr.svm_port;
173 if (!info->remote_cid) {
174 dst_cid = vsk->remote_addr.svm_cid;
175 dst_port = vsk->remote_addr.svm_port;
177 dst_cid = info->remote_cid;
178 dst_port = info->remote_port;
183 /* we can send less than pkt_len bytes */
184 if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
185 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
187 /* virtio_transport_get_credit might return less than pkt_len credit */
188 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
190 /* Do not send zero length OP_RW pkt */
191 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
194 pkt = virtio_transport_alloc_pkt(info, pkt_len,
198 virtio_transport_put_credit(vvs, pkt_len);
202 virtio_transport_inc_tx_pkt(vvs, pkt);
204 return virtio_transport_get_ops()->send_pkt(pkt);
207 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
208 struct virtio_vsock_pkt *pkt)
210 if (vvs->rx_bytes + pkt->len > vvs->buf_alloc)
213 vvs->rx_bytes += pkt->len;
217 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
218 struct virtio_vsock_pkt *pkt)
220 vvs->rx_bytes -= pkt->len;
221 vvs->fwd_cnt += pkt->len;
224 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt)
226 spin_lock_bh(&vvs->rx_lock);
227 vvs->last_fwd_cnt = vvs->fwd_cnt;
228 pkt->hdr.fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
229 pkt->hdr.buf_alloc = cpu_to_le32(vvs->buf_alloc);
230 spin_unlock_bh(&vvs->rx_lock);
232 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
234 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
238 spin_lock_bh(&vvs->tx_lock);
239 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
243 spin_unlock_bh(&vvs->tx_lock);
247 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
249 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
251 spin_lock_bh(&vvs->tx_lock);
252 vvs->tx_cnt -= credit;
253 spin_unlock_bh(&vvs->tx_lock);
255 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
257 static int virtio_transport_send_credit_update(struct vsock_sock *vsk,
259 struct virtio_vsock_hdr *hdr)
261 struct virtio_vsock_pkt_info info = {
262 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
267 return virtio_transport_send_pkt_info(vsk, &info);
271 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
275 struct virtio_vsock_sock *vvs = vsk->trans;
276 struct virtio_vsock_pkt *pkt;
277 size_t bytes, total = 0;
281 spin_lock_bh(&vvs->rx_lock);
282 while (total < len && !list_empty(&vvs->rx_queue)) {
283 pkt = list_first_entry(&vvs->rx_queue,
284 struct virtio_vsock_pkt, list);
287 if (bytes > pkt->len - pkt->off)
288 bytes = pkt->len - pkt->off;
290 /* sk_lock is held by caller so no one else can dequeue.
291 * Unlock rx_lock since memcpy_to_msg() may sleep.
293 spin_unlock_bh(&vvs->rx_lock);
295 err = memcpy_to_msg(msg, pkt->buf + pkt->off, bytes);
299 spin_lock_bh(&vvs->rx_lock);
303 if (pkt->off == pkt->len) {
304 virtio_transport_dec_rx_pkt(vvs, pkt);
305 list_del(&pkt->list);
306 virtio_transport_free_pkt(pkt);
310 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
312 spin_unlock_bh(&vvs->rx_lock);
314 /* To reduce the number of credit update messages,
315 * don't update credits as long as lots of space is available.
316 * Note: the limit chosen here is arbitrary. Setting the limit
317 * too high causes extra messages. Too low causes transmitter
318 * stalls. As stalls are in theory more expensive than extra
319 * messages, we set the limit to a high value. TODO: experiment
320 * with different values.
322 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
323 virtio_transport_send_credit_update(vsk,
324 VIRTIO_VSOCK_TYPE_STREAM,
337 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
339 size_t len, int flags)
341 if (flags & MSG_PEEK)
344 return virtio_transport_stream_do_dequeue(vsk, msg, len);
346 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
349 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
351 size_t len, int flags)
355 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
357 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
359 struct virtio_vsock_sock *vvs = vsk->trans;
362 spin_lock_bh(&vvs->rx_lock);
363 bytes = vvs->rx_bytes;
364 spin_unlock_bh(&vvs->rx_lock);
368 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
370 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
372 struct virtio_vsock_sock *vvs = vsk->trans;
375 bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
382 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
384 struct virtio_vsock_sock *vvs = vsk->trans;
387 spin_lock_bh(&vvs->tx_lock);
388 bytes = virtio_transport_has_space(vsk);
389 spin_unlock_bh(&vvs->tx_lock);
393 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
395 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
396 struct vsock_sock *psk)
398 struct virtio_vsock_sock *vvs;
400 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
407 struct virtio_vsock_sock *ptrans = psk->trans;
409 vvs->buf_size = ptrans->buf_size;
410 vvs->buf_size_min = ptrans->buf_size_min;
411 vvs->buf_size_max = ptrans->buf_size_max;
412 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
414 vvs->buf_size = VIRTIO_VSOCK_DEFAULT_BUF_SIZE;
415 vvs->buf_size_min = VIRTIO_VSOCK_DEFAULT_MIN_BUF_SIZE;
416 vvs->buf_size_max = VIRTIO_VSOCK_DEFAULT_MAX_BUF_SIZE;
419 vvs->buf_alloc = vvs->buf_size;
421 spin_lock_init(&vvs->rx_lock);
422 spin_lock_init(&vvs->tx_lock);
423 INIT_LIST_HEAD(&vvs->rx_queue);
427 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
429 u64 virtio_transport_get_buffer_size(struct vsock_sock *vsk)
431 struct virtio_vsock_sock *vvs = vsk->trans;
433 return vvs->buf_size;
435 EXPORT_SYMBOL_GPL(virtio_transport_get_buffer_size);
437 u64 virtio_transport_get_min_buffer_size(struct vsock_sock *vsk)
439 struct virtio_vsock_sock *vvs = vsk->trans;
441 return vvs->buf_size_min;
443 EXPORT_SYMBOL_GPL(virtio_transport_get_min_buffer_size);
445 u64 virtio_transport_get_max_buffer_size(struct vsock_sock *vsk)
447 struct virtio_vsock_sock *vvs = vsk->trans;
449 return vvs->buf_size_max;
451 EXPORT_SYMBOL_GPL(virtio_transport_get_max_buffer_size);
453 void virtio_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
455 struct virtio_vsock_sock *vvs = vsk->trans;
457 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
458 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
459 if (val < vvs->buf_size_min)
460 vvs->buf_size_min = val;
461 if (val > vvs->buf_size_max)
462 vvs->buf_size_max = val;
464 vvs->buf_alloc = val;
466 virtio_transport_send_credit_update(vsk, VIRTIO_VSOCK_TYPE_STREAM,
469 EXPORT_SYMBOL_GPL(virtio_transport_set_buffer_size);
471 void virtio_transport_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
473 struct virtio_vsock_sock *vvs = vsk->trans;
475 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
476 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
477 if (val > vvs->buf_size)
479 vvs->buf_size_min = val;
481 EXPORT_SYMBOL_GPL(virtio_transport_set_min_buffer_size);
483 void virtio_transport_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
485 struct virtio_vsock_sock *vvs = vsk->trans;
487 if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
488 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
489 if (val < vvs->buf_size)
491 vvs->buf_size_max = val;
493 EXPORT_SYMBOL_GPL(virtio_transport_set_max_buffer_size);
496 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
498 bool *data_ready_now)
500 if (vsock_stream_has_data(vsk))
501 *data_ready_now = true;
503 *data_ready_now = false;
507 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
510 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
512 bool *space_avail_now)
516 free_space = vsock_stream_has_space(vsk);
518 *space_avail_now = true;
519 else if (free_space == 0)
520 *space_avail_now = false;
524 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
526 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
527 size_t target, struct vsock_transport_recv_notify_data *data)
531 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
533 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
534 size_t target, struct vsock_transport_recv_notify_data *data)
538 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
540 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
541 size_t target, struct vsock_transport_recv_notify_data *data)
545 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
547 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
548 size_t target, ssize_t copied, bool data_read,
549 struct vsock_transport_recv_notify_data *data)
553 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
555 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
556 struct vsock_transport_send_notify_data *data)
560 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
562 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
563 struct vsock_transport_send_notify_data *data)
567 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
569 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
570 struct vsock_transport_send_notify_data *data)
574 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
576 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
577 ssize_t written, struct vsock_transport_send_notify_data *data)
581 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
583 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
585 struct virtio_vsock_sock *vvs = vsk->trans;
587 return vvs->buf_size;
589 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
591 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
595 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
597 bool virtio_transport_stream_allow(u32 cid, u32 port)
601 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
603 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
604 struct sockaddr_vm *addr)
608 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
610 bool virtio_transport_dgram_allow(u32 cid, u32 port)
614 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
616 int virtio_transport_connect(struct vsock_sock *vsk)
618 struct virtio_vsock_pkt_info info = {
619 .op = VIRTIO_VSOCK_OP_REQUEST,
620 .type = VIRTIO_VSOCK_TYPE_STREAM,
624 return virtio_transport_send_pkt_info(vsk, &info);
626 EXPORT_SYMBOL_GPL(virtio_transport_connect);
628 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
630 struct virtio_vsock_pkt_info info = {
631 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
632 .type = VIRTIO_VSOCK_TYPE_STREAM,
633 .flags = (mode & RCV_SHUTDOWN ?
634 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
635 (mode & SEND_SHUTDOWN ?
636 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
640 return virtio_transport_send_pkt_info(vsk, &info);
642 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
645 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
646 struct sockaddr_vm *remote_addr,
652 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
655 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
659 struct virtio_vsock_pkt_info info = {
660 .op = VIRTIO_VSOCK_OP_RW,
661 .type = VIRTIO_VSOCK_TYPE_STREAM,
667 return virtio_transport_send_pkt_info(vsk, &info);
669 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
671 void virtio_transport_destruct(struct vsock_sock *vsk)
673 struct virtio_vsock_sock *vvs = vsk->trans;
677 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
679 static int virtio_transport_reset(struct vsock_sock *vsk,
680 struct virtio_vsock_pkt *pkt)
682 struct virtio_vsock_pkt_info info = {
683 .op = VIRTIO_VSOCK_OP_RST,
684 .type = VIRTIO_VSOCK_TYPE_STREAM,
689 /* Send RST only if the original pkt is not a RST pkt */
690 if (pkt && le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
693 return virtio_transport_send_pkt_info(vsk, &info);
696 /* Normally packets are associated with a socket. There may be no socket if an
697 * attempt was made to connect to a socket that does not exist.
699 static int virtio_transport_reset_no_sock(struct virtio_vsock_pkt *pkt)
701 const struct virtio_transport *t;
702 struct virtio_vsock_pkt *reply;
703 struct virtio_vsock_pkt_info info = {
704 .op = VIRTIO_VSOCK_OP_RST,
705 .type = le16_to_cpu(pkt->hdr.type),
709 /* Send RST only if the original pkt is not a RST pkt */
710 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
713 reply = virtio_transport_alloc_pkt(&info, 0,
714 le64_to_cpu(pkt->hdr.dst_cid),
715 le32_to_cpu(pkt->hdr.dst_port),
716 le64_to_cpu(pkt->hdr.src_cid),
717 le32_to_cpu(pkt->hdr.src_port));
721 t = virtio_transport_get_ops();
723 virtio_transport_free_pkt(reply);
727 return t->send_pkt(reply);
730 static void virtio_transport_wait_close(struct sock *sk, long timeout)
733 DEFINE_WAIT_FUNC(wait, woken_wake_function);
735 add_wait_queue(sk_sleep(sk), &wait);
738 if (sk_wait_event(sk, &timeout,
739 sock_flag(sk, SOCK_DONE), &wait))
741 } while (!signal_pending(current) && timeout);
743 remove_wait_queue(sk_sleep(sk), &wait);
747 static void virtio_transport_do_close(struct vsock_sock *vsk,
750 struct sock *sk = sk_vsock(vsk);
752 sock_set_flag(sk, SOCK_DONE);
753 vsk->peer_shutdown = SHUTDOWN_MASK;
754 if (vsock_stream_has_data(vsk) <= 0)
755 sk->sk_state = TCP_CLOSING;
756 sk->sk_state_change(sk);
758 if (vsk->close_work_scheduled &&
759 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
760 vsk->close_work_scheduled = false;
762 vsock_remove_sock(vsk);
764 /* Release refcnt obtained when we scheduled the timeout */
769 static void virtio_transport_close_timeout(struct work_struct *work)
771 struct vsock_sock *vsk =
772 container_of(work, struct vsock_sock, close_work.work);
773 struct sock *sk = sk_vsock(vsk);
778 if (!sock_flag(sk, SOCK_DONE)) {
779 (void)virtio_transport_reset(vsk, NULL);
781 virtio_transport_do_close(vsk, false);
784 vsk->close_work_scheduled = false;
790 /* User context, vsk->sk is locked */
791 static bool virtio_transport_close(struct vsock_sock *vsk)
793 struct sock *sk = &vsk->sk;
795 if (!(sk->sk_state == TCP_ESTABLISHED ||
796 sk->sk_state == TCP_CLOSING))
799 /* Already received SHUTDOWN from peer, reply with RST */
800 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
801 (void)virtio_transport_reset(vsk, NULL);
805 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
806 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
808 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
809 virtio_transport_wait_close(sk, sk->sk_lingertime);
811 if (sock_flag(sk, SOCK_DONE)) {
816 INIT_DELAYED_WORK(&vsk->close_work,
817 virtio_transport_close_timeout);
818 vsk->close_work_scheduled = true;
819 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
823 void virtio_transport_release(struct vsock_sock *vsk)
825 struct virtio_vsock_sock *vvs = vsk->trans;
826 struct virtio_vsock_pkt *pkt, *tmp;
827 struct sock *sk = &vsk->sk;
828 bool remove_sock = true;
830 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
831 if (sk->sk_type == SOCK_STREAM)
832 remove_sock = virtio_transport_close(vsk);
834 list_for_each_entry_safe(pkt, tmp, &vvs->rx_queue, list) {
835 list_del(&pkt->list);
836 virtio_transport_free_pkt(pkt);
841 vsock_remove_sock(vsk);
843 EXPORT_SYMBOL_GPL(virtio_transport_release);
846 virtio_transport_recv_connecting(struct sock *sk,
847 struct virtio_vsock_pkt *pkt)
849 struct vsock_sock *vsk = vsock_sk(sk);
853 switch (le16_to_cpu(pkt->hdr.op)) {
854 case VIRTIO_VSOCK_OP_RESPONSE:
855 sk->sk_state = TCP_ESTABLISHED;
856 sk->sk_socket->state = SS_CONNECTED;
857 vsock_insert_connected(vsk);
858 sk->sk_state_change(sk);
860 case VIRTIO_VSOCK_OP_INVALID:
862 case VIRTIO_VSOCK_OP_RST:
874 virtio_transport_reset(vsk, pkt);
875 sk->sk_state = TCP_CLOSE;
877 sk->sk_error_report(sk);
882 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
883 struct virtio_vsock_pkt *pkt)
885 struct virtio_vsock_sock *vvs = vsk->trans;
886 bool can_enqueue, free_pkt = false;
888 pkt->len = le32_to_cpu(pkt->hdr.len);
891 spin_lock_bh(&vvs->rx_lock);
893 can_enqueue = virtio_transport_inc_rx_pkt(vvs, pkt);
899 /* Try to copy small packets into the buffer of last packet queued,
900 * to avoid wasting memory queueing the entire buffer with a small
903 if (pkt->len <= GOOD_COPY_LEN && !list_empty(&vvs->rx_queue)) {
904 struct virtio_vsock_pkt *last_pkt;
906 last_pkt = list_last_entry(&vvs->rx_queue,
907 struct virtio_vsock_pkt, list);
909 /* If there is space in the last packet queued, we copy the
910 * new packet in its buffer.
912 if (pkt->len <= last_pkt->buf_len - last_pkt->len) {
913 memcpy(last_pkt->buf + last_pkt->len, pkt->buf,
915 last_pkt->len += pkt->len;
921 list_add_tail(&pkt->list, &vvs->rx_queue);
924 spin_unlock_bh(&vvs->rx_lock);
926 virtio_transport_free_pkt(pkt);
930 virtio_transport_recv_connected(struct sock *sk,
931 struct virtio_vsock_pkt *pkt)
933 struct vsock_sock *vsk = vsock_sk(sk);
936 switch (le16_to_cpu(pkt->hdr.op)) {
937 case VIRTIO_VSOCK_OP_RW:
938 virtio_transport_recv_enqueue(vsk, pkt);
939 sk->sk_data_ready(sk);
941 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
942 sk->sk_write_space(sk);
944 case VIRTIO_VSOCK_OP_SHUTDOWN:
945 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
946 vsk->peer_shutdown |= RCV_SHUTDOWN;
947 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
948 vsk->peer_shutdown |= SEND_SHUTDOWN;
949 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
950 vsock_stream_has_data(vsk) <= 0 &&
951 !sock_flag(sk, SOCK_DONE)) {
952 (void)virtio_transport_reset(vsk, NULL);
954 virtio_transport_do_close(vsk, true);
956 if (le32_to_cpu(pkt->hdr.flags))
957 sk->sk_state_change(sk);
959 case VIRTIO_VSOCK_OP_RST:
960 virtio_transport_do_close(vsk, true);
967 virtio_transport_free_pkt(pkt);
972 virtio_transport_recv_disconnecting(struct sock *sk,
973 struct virtio_vsock_pkt *pkt)
975 struct vsock_sock *vsk = vsock_sk(sk);
977 if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
978 virtio_transport_do_close(vsk, true);
982 virtio_transport_send_response(struct vsock_sock *vsk,
983 struct virtio_vsock_pkt *pkt)
985 struct virtio_vsock_pkt_info info = {
986 .op = VIRTIO_VSOCK_OP_RESPONSE,
987 .type = VIRTIO_VSOCK_TYPE_STREAM,
988 .remote_cid = le64_to_cpu(pkt->hdr.src_cid),
989 .remote_port = le32_to_cpu(pkt->hdr.src_port),
994 return virtio_transport_send_pkt_info(vsk, &info);
997 /* Handle server socket */
999 virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt)
1001 struct vsock_sock *vsk = vsock_sk(sk);
1002 struct vsock_sock *vchild;
1005 if (le16_to_cpu(pkt->hdr.op) != VIRTIO_VSOCK_OP_REQUEST) {
1006 virtio_transport_reset(vsk, pkt);
1010 if (sk_acceptq_is_full(sk)) {
1011 virtio_transport_reset(vsk, pkt);
1015 child = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1018 virtio_transport_reset(vsk, pkt);
1022 sk->sk_ack_backlog++;
1024 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1026 child->sk_state = TCP_ESTABLISHED;
1028 vchild = vsock_sk(child);
1029 vsock_addr_init(&vchild->local_addr, le64_to_cpu(pkt->hdr.dst_cid),
1030 le32_to_cpu(pkt->hdr.dst_port));
1031 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(pkt->hdr.src_cid),
1032 le32_to_cpu(pkt->hdr.src_port));
1034 vsock_insert_connected(vchild);
1035 vsock_enqueue_accept(sk, child);
1036 virtio_transport_send_response(vchild, pkt);
1038 release_sock(child);
1040 sk->sk_data_ready(sk);
1044 static bool virtio_transport_space_update(struct sock *sk,
1045 struct virtio_vsock_pkt *pkt)
1047 struct vsock_sock *vsk = vsock_sk(sk);
1048 struct virtio_vsock_sock *vvs = vsk->trans;
1049 bool space_available;
1051 /* buf_alloc and fwd_cnt is always included in the hdr */
1052 spin_lock_bh(&vvs->tx_lock);
1053 vvs->peer_buf_alloc = le32_to_cpu(pkt->hdr.buf_alloc);
1054 vvs->peer_fwd_cnt = le32_to_cpu(pkt->hdr.fwd_cnt);
1055 space_available = virtio_transport_has_space(vsk);
1056 spin_unlock_bh(&vvs->tx_lock);
1057 return space_available;
1060 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1063 void virtio_transport_recv_pkt(struct virtio_vsock_pkt *pkt)
1065 struct sockaddr_vm src, dst;
1066 struct vsock_sock *vsk;
1068 bool space_available;
1070 vsock_addr_init(&src, le64_to_cpu(pkt->hdr.src_cid),
1071 le32_to_cpu(pkt->hdr.src_port));
1072 vsock_addr_init(&dst, le64_to_cpu(pkt->hdr.dst_cid),
1073 le32_to_cpu(pkt->hdr.dst_port));
1075 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1076 dst.svm_cid, dst.svm_port,
1077 le32_to_cpu(pkt->hdr.len),
1078 le16_to_cpu(pkt->hdr.type),
1079 le16_to_cpu(pkt->hdr.op),
1080 le32_to_cpu(pkt->hdr.flags),
1081 le32_to_cpu(pkt->hdr.buf_alloc),
1082 le32_to_cpu(pkt->hdr.fwd_cnt));
1084 if (le16_to_cpu(pkt->hdr.type) != VIRTIO_VSOCK_TYPE_STREAM) {
1085 (void)virtio_transport_reset_no_sock(pkt);
1089 /* The socket must be in connected or bound table
1090 * otherwise send reset back
1092 sk = vsock_find_connected_socket(&src, &dst);
1094 sk = vsock_find_bound_socket(&dst);
1096 (void)virtio_transport_reset_no_sock(pkt);
1103 space_available = virtio_transport_space_update(sk, pkt);
1107 /* Update CID in case it has changed after a transport reset event */
1108 vsk->local_addr.svm_cid = dst.svm_cid;
1110 if (space_available)
1111 sk->sk_write_space(sk);
1113 switch (sk->sk_state) {
1115 virtio_transport_recv_listen(sk, pkt);
1116 virtio_transport_free_pkt(pkt);
1119 virtio_transport_recv_connecting(sk, pkt);
1120 virtio_transport_free_pkt(pkt);
1122 case TCP_ESTABLISHED:
1123 virtio_transport_recv_connected(sk, pkt);
1126 virtio_transport_recv_disconnecting(sk, pkt);
1127 virtio_transport_free_pkt(pkt);
1130 virtio_transport_free_pkt(pkt);
1135 /* Release refcnt obtained when we fetched this socket out of the
1136 * bound or connected list.
1142 virtio_transport_free_pkt(pkt);
1144 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1146 void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt)
1151 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt);
1153 MODULE_LICENSE("GPL v2");
1154 MODULE_AUTHOR("Asias He");
1155 MODULE_DESCRIPTION("common code for virtio vsock");