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_vsock.h>
15 #include <uapi/linux/vsockmon.h>
18 #include <net/af_vsock.h>
20 #define CREATE_TRACE_POINTS
21 #include <trace/events/vsock_virtio_transport_common.h>
23 /* How long to wait for graceful shutdown of a connection */
24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
26 /* Threshold for detecting small packets to copy */
27 #define GOOD_COPY_LEN 128
29 static const struct virtio_transport *
30 virtio_transport_get_ops(struct vsock_sock *vsk)
32 const struct vsock_transport *t = vsock_core_get_transport(vsk);
37 return container_of(t, struct virtio_transport, transport);
40 /* Returns a new packet on success, otherwise returns NULL.
42 * If NULL is returned, errp is set to a negative errno.
44 static struct sk_buff *
45 virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
52 const size_t skb_len = VIRTIO_VSOCK_SKB_HEADROOM + len;
53 struct virtio_vsock_hdr *hdr;
58 skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
62 hdr = virtio_vsock_hdr(skb);
63 hdr->type = cpu_to_le16(info->type);
64 hdr->op = cpu_to_le16(info->op);
65 hdr->src_cid = cpu_to_le64(src_cid);
66 hdr->dst_cid = cpu_to_le64(dst_cid);
67 hdr->src_port = cpu_to_le32(src_port);
68 hdr->dst_port = cpu_to_le32(dst_port);
69 hdr->flags = cpu_to_le32(info->flags);
70 hdr->len = cpu_to_le32(len);
71 hdr->buf_alloc = cpu_to_le32(0);
72 hdr->fwd_cnt = cpu_to_le32(0);
74 if (info->msg && len > 0) {
75 payload = skb_put(skb, len);
76 err = memcpy_from_msg(payload, info->msg, len);
80 if (msg_data_left(info->msg) == 0 &&
81 info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
82 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
84 if (info->msg->msg_flags & MSG_EOR)
85 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
90 virtio_vsock_skb_set_reply(skb);
92 trace_virtio_transport_alloc_pkt(src_cid, src_port,
99 if (info->vsk && !skb_set_owner_sk_safe(skb, sk_vsock(info->vsk))) {
100 WARN_ONCE(1, "failed to allocate skb on vsock socket with sk_refcnt == 0\n");
112 static struct sk_buff *virtio_transport_build_skb(void *opaque)
114 struct virtio_vsock_hdr *pkt_hdr;
115 struct sk_buff *pkt = opaque;
116 struct af_vsockmon_hdr *hdr;
121 /* A packet could be split to fit the RX buffer, so we can retrieve
122 * the payload length from the header and the buffer pointer taking
123 * care of the offset in the original packet.
125 pkt_hdr = virtio_vsock_hdr(pkt);
126 payload_len = pkt->len;
127 payload_buf = pkt->data;
129 skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
134 hdr = skb_put(skb, sizeof(*hdr));
136 /* pkt->hdr is little-endian so no need to byteswap here */
137 hdr->src_cid = pkt_hdr->src_cid;
138 hdr->src_port = pkt_hdr->src_port;
139 hdr->dst_cid = pkt_hdr->dst_cid;
140 hdr->dst_port = pkt_hdr->dst_port;
142 hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
143 hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
144 memset(hdr->reserved, 0, sizeof(hdr->reserved));
146 switch (le16_to_cpu(pkt_hdr->op)) {
147 case VIRTIO_VSOCK_OP_REQUEST:
148 case VIRTIO_VSOCK_OP_RESPONSE:
149 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
151 case VIRTIO_VSOCK_OP_RST:
152 case VIRTIO_VSOCK_OP_SHUTDOWN:
153 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
155 case VIRTIO_VSOCK_OP_RW:
156 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
158 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
159 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
160 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
163 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
167 skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
170 skb_put_data(skb, payload_buf, payload_len);
176 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
178 if (virtio_vsock_skb_tap_delivered(skb))
181 vsock_deliver_tap(virtio_transport_build_skb, skb);
182 virtio_vsock_skb_set_tap_delivered(skb);
184 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
186 static u16 virtio_transport_get_type(struct sock *sk)
188 if (sk->sk_type == SOCK_STREAM)
189 return VIRTIO_VSOCK_TYPE_STREAM;
191 return VIRTIO_VSOCK_TYPE_SEQPACKET;
194 /* This function can only be used on connecting/connected sockets,
195 * since a socket assigned to a transport is required.
197 * Do not use on listener sockets!
199 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
200 struct virtio_vsock_pkt_info *info)
202 u32 src_cid, src_port, dst_cid, dst_port;
203 const struct virtio_transport *t_ops;
204 struct virtio_vsock_sock *vvs;
205 u32 pkt_len = info->pkt_len;
209 info->type = virtio_transport_get_type(sk_vsock(vsk));
211 t_ops = virtio_transport_get_ops(vsk);
212 if (unlikely(!t_ops))
215 src_cid = t_ops->transport.get_local_cid();
216 src_port = vsk->local_addr.svm_port;
217 if (!info->remote_cid) {
218 dst_cid = vsk->remote_addr.svm_cid;
219 dst_port = vsk->remote_addr.svm_port;
221 dst_cid = info->remote_cid;
222 dst_port = info->remote_port;
227 /* virtio_transport_get_credit might return less than pkt_len credit */
228 pkt_len = virtio_transport_get_credit(vvs, pkt_len);
230 /* Do not send zero length OP_RW pkt */
231 if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
240 skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, rest_len);
242 skb = virtio_transport_alloc_skb(info, skb_len,
250 virtio_transport_inc_tx_pkt(vvs, skb);
252 ret = t_ops->send_pkt(skb);
256 /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
257 * but for reliability use 'ret' instead of 'skb_len'.
258 * Also if partial send happens (e.g. 'ret' != 'skb_len')
259 * somehow, we break this loop, but account such returned
260 * value in 'virtio_transport_put_credit()'.
264 if (WARN_ONCE(ret != skb_len,
265 "'send_pkt()' returns %i, but %zu expected\n",
270 virtio_transport_put_credit(vvs, rest_len);
272 /* Return number of bytes, if any data has been sent. */
273 if (rest_len != pkt_len)
274 ret = pkt_len - rest_len;
279 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
282 if (vvs->rx_bytes + len > vvs->buf_alloc)
285 vvs->rx_bytes += len;
289 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
292 vvs->rx_bytes -= len;
296 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
298 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
300 spin_lock_bh(&vvs->rx_lock);
301 vvs->last_fwd_cnt = vvs->fwd_cnt;
302 hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
303 hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
304 spin_unlock_bh(&vvs->rx_lock);
306 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
308 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
315 spin_lock_bh(&vvs->tx_lock);
316 ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
320 spin_unlock_bh(&vvs->tx_lock);
324 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
326 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
331 spin_lock_bh(&vvs->tx_lock);
332 vvs->tx_cnt -= credit;
333 spin_unlock_bh(&vvs->tx_lock);
335 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
337 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
339 struct virtio_vsock_pkt_info info = {
340 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
344 return virtio_transport_send_pkt_info(vsk, &info);
348 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
352 struct virtio_vsock_sock *vvs = vsk->trans;
357 spin_lock_bh(&vvs->rx_lock);
359 skb_queue_walk(&vvs->rx_queue, skb) {
363 if (bytes > skb->len)
366 spin_unlock_bh(&vvs->rx_lock);
368 /* sk_lock is held by caller so no one else can dequeue.
369 * Unlock rx_lock since memcpy_to_msg() may sleep.
371 err = memcpy_to_msg(msg, skb->data, bytes);
377 spin_lock_bh(&vvs->rx_lock);
383 spin_unlock_bh(&vvs->rx_lock);
394 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
398 struct virtio_vsock_sock *vvs = vsk->trans;
399 size_t bytes, total = 0;
404 spin_lock_bh(&vvs->rx_lock);
406 if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
407 "rx_queue is empty, but rx_bytes is non-zero\n")) {
408 spin_unlock_bh(&vvs->rx_lock);
412 while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
413 skb = skb_peek(&vvs->rx_queue);
416 if (bytes > skb->len)
419 /* sk_lock is held by caller so no one else can dequeue.
420 * Unlock rx_lock since memcpy_to_msg() may sleep.
422 spin_unlock_bh(&vvs->rx_lock);
424 err = memcpy_to_msg(msg, skb->data, bytes);
428 spin_lock_bh(&vvs->rx_lock);
431 skb_pull(skb, bytes);
434 u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
436 virtio_transport_dec_rx_pkt(vvs, pkt_len);
437 __skb_unlink(skb, &vvs->rx_queue);
442 free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
444 spin_unlock_bh(&vvs->rx_lock);
446 /* To reduce the number of credit update messages,
447 * don't update credits as long as lots of space is available.
448 * Note: the limit chosen here is arbitrary. Setting the limit
449 * too high causes extra messages. Too low causes transmitter
450 * stalls. As stalls are in theory more expensive than extra
451 * messages, we set the limit to a high value. TODO: experiment
452 * with different values.
454 if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
455 virtio_transport_send_credit_update(vsk);
466 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
469 struct virtio_vsock_sock *vvs = vsk->trans;
473 spin_lock_bh(&vvs->rx_lock);
475 if (!vvs->msg_count) {
476 spin_unlock_bh(&vvs->rx_lock);
481 len = msg_data_left(msg);
483 skb_queue_walk(&vvs->rx_queue, skb) {
484 struct virtio_vsock_hdr *hdr;
491 if (bytes > skb->len)
494 spin_unlock_bh(&vvs->rx_lock);
496 /* sk_lock is held by caller so no one else can dequeue.
497 * Unlock rx_lock since memcpy_to_msg() may sleep.
499 err = memcpy_to_msg(msg, skb->data, bytes);
503 spin_lock_bh(&vvs->rx_lock);
507 hdr = virtio_vsock_hdr(skb);
509 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
510 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
511 msg->msg_flags |= MSG_EOR;
517 spin_unlock_bh(&vvs->rx_lock);
522 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
526 struct virtio_vsock_sock *vvs = vsk->trans;
527 int dequeued_len = 0;
528 size_t user_buf_len = msg_data_left(msg);
529 bool msg_ready = false;
532 spin_lock_bh(&vvs->rx_lock);
534 if (vvs->msg_count == 0) {
535 spin_unlock_bh(&vvs->rx_lock);
540 struct virtio_vsock_hdr *hdr;
543 skb = __skb_dequeue(&vvs->rx_queue);
546 hdr = virtio_vsock_hdr(skb);
547 pkt_len = (size_t)le32_to_cpu(hdr->len);
549 if (dequeued_len >= 0) {
550 size_t bytes_to_copy;
552 bytes_to_copy = min(user_buf_len, pkt_len);
557 /* sk_lock is held by caller so no one else can dequeue.
558 * Unlock rx_lock since memcpy_to_msg() may sleep.
560 spin_unlock_bh(&vvs->rx_lock);
562 err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
564 /* Copy of message failed. Rest of
565 * fragments will be freed without copy.
569 user_buf_len -= bytes_to_copy;
572 spin_lock_bh(&vvs->rx_lock);
575 if (dequeued_len >= 0)
576 dequeued_len += pkt_len;
579 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
583 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
584 msg->msg_flags |= MSG_EOR;
587 virtio_transport_dec_rx_pkt(vvs, pkt_len);
591 spin_unlock_bh(&vvs->rx_lock);
593 virtio_transport_send_credit_update(vsk);
599 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
601 size_t len, int flags)
603 if (flags & MSG_PEEK)
604 return virtio_transport_stream_do_peek(vsk, msg, len);
606 return virtio_transport_stream_do_dequeue(vsk, msg, len);
608 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
611 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
615 if (flags & MSG_PEEK)
616 return virtio_transport_seqpacket_do_peek(vsk, msg);
618 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
620 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
623 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
627 struct virtio_vsock_sock *vvs = vsk->trans;
629 spin_lock_bh(&vvs->tx_lock);
631 if (len > vvs->peer_buf_alloc) {
632 spin_unlock_bh(&vvs->tx_lock);
636 spin_unlock_bh(&vvs->tx_lock);
638 return virtio_transport_stream_enqueue(vsk, msg, len);
640 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
643 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
645 size_t len, int flags)
649 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
651 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
653 struct virtio_vsock_sock *vvs = vsk->trans;
656 spin_lock_bh(&vvs->rx_lock);
657 bytes = vvs->rx_bytes;
658 spin_unlock_bh(&vvs->rx_lock);
662 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
664 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
666 struct virtio_vsock_sock *vvs = vsk->trans;
669 spin_lock_bh(&vvs->rx_lock);
670 msg_count = vvs->msg_count;
671 spin_unlock_bh(&vvs->rx_lock);
675 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
677 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
679 struct virtio_vsock_sock *vvs = vsk->trans;
682 bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
689 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
691 struct virtio_vsock_sock *vvs = vsk->trans;
694 spin_lock_bh(&vvs->tx_lock);
695 bytes = virtio_transport_has_space(vsk);
696 spin_unlock_bh(&vvs->tx_lock);
700 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
702 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
703 struct vsock_sock *psk)
705 struct virtio_vsock_sock *vvs;
707 vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
713 if (psk && psk->trans) {
714 struct virtio_vsock_sock *ptrans = psk->trans;
716 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
719 if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
720 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
722 vvs->buf_alloc = vsk->buffer_size;
724 spin_lock_init(&vvs->rx_lock);
725 spin_lock_init(&vvs->tx_lock);
726 skb_queue_head_init(&vvs->rx_queue);
730 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
732 /* sk_lock held by the caller */
733 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
735 struct virtio_vsock_sock *vvs = vsk->trans;
737 if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
738 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
740 vvs->buf_alloc = *val;
742 virtio_transport_send_credit_update(vsk);
744 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
747 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
749 bool *data_ready_now)
751 *data_ready_now = vsock_stream_has_data(vsk) >= target;
755 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
758 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
760 bool *space_avail_now)
764 free_space = vsock_stream_has_space(vsk);
766 *space_avail_now = true;
767 else if (free_space == 0)
768 *space_avail_now = false;
772 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
774 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
775 size_t target, struct vsock_transport_recv_notify_data *data)
779 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
781 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
782 size_t target, struct vsock_transport_recv_notify_data *data)
786 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
788 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
789 size_t target, struct vsock_transport_recv_notify_data *data)
793 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
795 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
796 size_t target, ssize_t copied, bool data_read,
797 struct vsock_transport_recv_notify_data *data)
801 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
803 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
804 struct vsock_transport_send_notify_data *data)
808 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
810 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
811 struct vsock_transport_send_notify_data *data)
815 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
817 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
818 struct vsock_transport_send_notify_data *data)
822 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
824 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
825 ssize_t written, struct vsock_transport_send_notify_data *data)
829 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
831 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
833 return vsk->buffer_size;
835 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
837 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
841 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
843 bool virtio_transport_stream_allow(u32 cid, u32 port)
847 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
849 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
850 struct sockaddr_vm *addr)
854 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
856 bool virtio_transport_dgram_allow(u32 cid, u32 port)
860 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
862 int virtio_transport_connect(struct vsock_sock *vsk)
864 struct virtio_vsock_pkt_info info = {
865 .op = VIRTIO_VSOCK_OP_REQUEST,
869 return virtio_transport_send_pkt_info(vsk, &info);
871 EXPORT_SYMBOL_GPL(virtio_transport_connect);
873 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
875 struct virtio_vsock_pkt_info info = {
876 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
877 .flags = (mode & RCV_SHUTDOWN ?
878 VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
879 (mode & SEND_SHUTDOWN ?
880 VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
884 return virtio_transport_send_pkt_info(vsk, &info);
886 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
889 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
890 struct sockaddr_vm *remote_addr,
896 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
899 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
903 struct virtio_vsock_pkt_info info = {
904 .op = VIRTIO_VSOCK_OP_RW,
910 return virtio_transport_send_pkt_info(vsk, &info);
912 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
914 void virtio_transport_destruct(struct vsock_sock *vsk)
916 struct virtio_vsock_sock *vvs = vsk->trans;
920 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
922 static int virtio_transport_reset(struct vsock_sock *vsk,
925 struct virtio_vsock_pkt_info info = {
926 .op = VIRTIO_VSOCK_OP_RST,
931 /* Send RST only if the original pkt is not a RST pkt */
932 if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
935 return virtio_transport_send_pkt_info(vsk, &info);
938 /* Normally packets are associated with a socket. There may be no socket if an
939 * attempt was made to connect to a socket that does not exist.
941 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
944 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
945 struct virtio_vsock_pkt_info info = {
946 .op = VIRTIO_VSOCK_OP_RST,
947 .type = le16_to_cpu(hdr->type),
950 struct sk_buff *reply;
952 /* Send RST only if the original pkt is not a RST pkt */
953 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
959 reply = virtio_transport_alloc_skb(&info, 0,
960 le64_to_cpu(hdr->dst_cid),
961 le32_to_cpu(hdr->dst_port),
962 le64_to_cpu(hdr->src_cid),
963 le32_to_cpu(hdr->src_port));
967 return t->send_pkt(reply);
970 /* This function should be called with sk_lock held and SOCK_DONE set */
971 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
973 struct virtio_vsock_sock *vvs = vsk->trans;
975 /* We don't need to take rx_lock, as the socket is closing and we are
978 __skb_queue_purge(&vvs->rx_queue);
979 vsock_remove_sock(vsk);
982 static void virtio_transport_wait_close(struct sock *sk, long timeout)
985 DEFINE_WAIT_FUNC(wait, woken_wake_function);
987 add_wait_queue(sk_sleep(sk), &wait);
990 if (sk_wait_event(sk, &timeout,
991 sock_flag(sk, SOCK_DONE), &wait))
993 } while (!signal_pending(current) && timeout);
995 remove_wait_queue(sk_sleep(sk), &wait);
999 static void virtio_transport_do_close(struct vsock_sock *vsk,
1000 bool cancel_timeout)
1002 struct sock *sk = sk_vsock(vsk);
1004 sock_set_flag(sk, SOCK_DONE);
1005 vsk->peer_shutdown = SHUTDOWN_MASK;
1006 if (vsock_stream_has_data(vsk) <= 0)
1007 sk->sk_state = TCP_CLOSING;
1008 sk->sk_state_change(sk);
1010 if (vsk->close_work_scheduled &&
1011 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1012 vsk->close_work_scheduled = false;
1014 virtio_transport_remove_sock(vsk);
1016 /* Release refcnt obtained when we scheduled the timeout */
1021 static void virtio_transport_close_timeout(struct work_struct *work)
1023 struct vsock_sock *vsk =
1024 container_of(work, struct vsock_sock, close_work.work);
1025 struct sock *sk = sk_vsock(vsk);
1030 if (!sock_flag(sk, SOCK_DONE)) {
1031 (void)virtio_transport_reset(vsk, NULL);
1033 virtio_transport_do_close(vsk, false);
1036 vsk->close_work_scheduled = false;
1042 /* User context, vsk->sk is locked */
1043 static bool virtio_transport_close(struct vsock_sock *vsk)
1045 struct sock *sk = &vsk->sk;
1047 if (!(sk->sk_state == TCP_ESTABLISHED ||
1048 sk->sk_state == TCP_CLOSING))
1051 /* Already received SHUTDOWN from peer, reply with RST */
1052 if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1053 (void)virtio_transport_reset(vsk, NULL);
1057 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1058 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1060 if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1061 virtio_transport_wait_close(sk, sk->sk_lingertime);
1063 if (sock_flag(sk, SOCK_DONE)) {
1068 INIT_DELAYED_WORK(&vsk->close_work,
1069 virtio_transport_close_timeout);
1070 vsk->close_work_scheduled = true;
1071 schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1075 void virtio_transport_release(struct vsock_sock *vsk)
1077 struct sock *sk = &vsk->sk;
1078 bool remove_sock = true;
1080 if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1081 remove_sock = virtio_transport_close(vsk);
1084 sock_set_flag(sk, SOCK_DONE);
1085 virtio_transport_remove_sock(vsk);
1088 EXPORT_SYMBOL_GPL(virtio_transport_release);
1091 virtio_transport_recv_connecting(struct sock *sk,
1092 struct sk_buff *skb)
1094 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1095 struct vsock_sock *vsk = vsock_sk(sk);
1099 switch (le16_to_cpu(hdr->op)) {
1100 case VIRTIO_VSOCK_OP_RESPONSE:
1101 sk->sk_state = TCP_ESTABLISHED;
1102 sk->sk_socket->state = SS_CONNECTED;
1103 vsock_insert_connected(vsk);
1104 sk->sk_state_change(sk);
1106 case VIRTIO_VSOCK_OP_INVALID:
1108 case VIRTIO_VSOCK_OP_RST:
1120 virtio_transport_reset(vsk, skb);
1121 sk->sk_state = TCP_CLOSE;
1123 sk_error_report(sk);
1128 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1129 struct sk_buff *skb)
1131 struct virtio_vsock_sock *vvs = vsk->trans;
1132 bool can_enqueue, free_pkt = false;
1133 struct virtio_vsock_hdr *hdr;
1136 hdr = virtio_vsock_hdr(skb);
1137 len = le32_to_cpu(hdr->len);
1139 spin_lock_bh(&vvs->rx_lock);
1141 can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1147 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1150 /* Try to copy small packets into the buffer of last packet queued,
1151 * to avoid wasting memory queueing the entire buffer with a small
1154 if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1155 struct virtio_vsock_hdr *last_hdr;
1156 struct sk_buff *last_skb;
1158 last_skb = skb_peek_tail(&vvs->rx_queue);
1159 last_hdr = virtio_vsock_hdr(last_skb);
1161 /* If there is space in the last packet queued, we copy the
1162 * new packet in its buffer. We avoid this if the last packet
1163 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1164 * delimiter of SEQPACKET message, so 'pkt' is the first packet
1167 if (skb->len < skb_tailroom(last_skb) &&
1168 !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1169 memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1171 last_hdr->flags |= hdr->flags;
1172 le32_add_cpu(&last_hdr->len, len);
1177 __skb_queue_tail(&vvs->rx_queue, skb);
1180 spin_unlock_bh(&vvs->rx_lock);
1186 virtio_transport_recv_connected(struct sock *sk,
1187 struct sk_buff *skb)
1189 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1190 struct vsock_sock *vsk = vsock_sk(sk);
1193 switch (le16_to_cpu(hdr->op)) {
1194 case VIRTIO_VSOCK_OP_RW:
1195 virtio_transport_recv_enqueue(vsk, skb);
1196 vsock_data_ready(sk);
1198 case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1199 virtio_transport_send_credit_update(vsk);
1201 case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1202 sk->sk_write_space(sk);
1204 case VIRTIO_VSOCK_OP_SHUTDOWN:
1205 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1206 vsk->peer_shutdown |= RCV_SHUTDOWN;
1207 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1208 vsk->peer_shutdown |= SEND_SHUTDOWN;
1209 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1210 if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1211 (void)virtio_transport_reset(vsk, NULL);
1212 virtio_transport_do_close(vsk, true);
1214 /* Remove this socket anyway because the remote peer sent
1215 * the shutdown. This way a new connection will succeed
1216 * if the remote peer uses the same source port,
1217 * even if the old socket is still unreleased, but now disconnected.
1219 vsock_remove_sock(vsk);
1221 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1222 sk->sk_state_change(sk);
1224 case VIRTIO_VSOCK_OP_RST:
1225 virtio_transport_do_close(vsk, true);
1237 virtio_transport_recv_disconnecting(struct sock *sk,
1238 struct sk_buff *skb)
1240 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1241 struct vsock_sock *vsk = vsock_sk(sk);
1243 if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1244 virtio_transport_do_close(vsk, true);
1248 virtio_transport_send_response(struct vsock_sock *vsk,
1249 struct sk_buff *skb)
1251 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1252 struct virtio_vsock_pkt_info info = {
1253 .op = VIRTIO_VSOCK_OP_RESPONSE,
1254 .remote_cid = le64_to_cpu(hdr->src_cid),
1255 .remote_port = le32_to_cpu(hdr->src_port),
1260 return virtio_transport_send_pkt_info(vsk, &info);
1263 static bool virtio_transport_space_update(struct sock *sk,
1264 struct sk_buff *skb)
1266 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1267 struct vsock_sock *vsk = vsock_sk(sk);
1268 struct virtio_vsock_sock *vvs = vsk->trans;
1269 bool space_available;
1271 /* Listener sockets are not associated with any transport, so we are
1272 * not able to take the state to see if there is space available in the
1273 * remote peer, but since they are only used to receive requests, we
1274 * can assume that there is always space available in the other peer.
1279 /* buf_alloc and fwd_cnt is always included in the hdr */
1280 spin_lock_bh(&vvs->tx_lock);
1281 vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1282 vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1283 space_available = virtio_transport_has_space(vsk);
1284 spin_unlock_bh(&vvs->tx_lock);
1285 return space_available;
1288 /* Handle server socket */
1290 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1291 struct virtio_transport *t)
1293 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1294 struct vsock_sock *vsk = vsock_sk(sk);
1295 struct vsock_sock *vchild;
1299 if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1300 virtio_transport_reset_no_sock(t, skb);
1304 if (sk_acceptq_is_full(sk)) {
1305 virtio_transport_reset_no_sock(t, skb);
1309 child = vsock_create_connected(sk);
1311 virtio_transport_reset_no_sock(t, skb);
1315 sk_acceptq_added(sk);
1317 lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1319 child->sk_state = TCP_ESTABLISHED;
1321 vchild = vsock_sk(child);
1322 vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1323 le32_to_cpu(hdr->dst_port));
1324 vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1325 le32_to_cpu(hdr->src_port));
1327 ret = vsock_assign_transport(vchild, vsk);
1328 /* Transport assigned (looking at remote_addr) must be the same
1329 * where we received the request.
1331 if (ret || vchild->transport != &t->transport) {
1332 release_sock(child);
1333 virtio_transport_reset_no_sock(t, skb);
1338 if (virtio_transport_space_update(child, skb))
1339 child->sk_write_space(child);
1341 vsock_insert_connected(vchild);
1342 vsock_enqueue_accept(sk, child);
1343 virtio_transport_send_response(vchild, skb);
1345 release_sock(child);
1347 sk->sk_data_ready(sk);
1351 static bool virtio_transport_valid_type(u16 type)
1353 return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1354 (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1357 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1360 void virtio_transport_recv_pkt(struct virtio_transport *t,
1361 struct sk_buff *skb)
1363 struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1364 struct sockaddr_vm src, dst;
1365 struct vsock_sock *vsk;
1367 bool space_available;
1369 vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1370 le32_to_cpu(hdr->src_port));
1371 vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1372 le32_to_cpu(hdr->dst_port));
1374 trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1375 dst.svm_cid, dst.svm_port,
1376 le32_to_cpu(hdr->len),
1377 le16_to_cpu(hdr->type),
1378 le16_to_cpu(hdr->op),
1379 le32_to_cpu(hdr->flags),
1380 le32_to_cpu(hdr->buf_alloc),
1381 le32_to_cpu(hdr->fwd_cnt));
1383 if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1384 (void)virtio_transport_reset_no_sock(t, skb);
1388 /* The socket must be in connected or bound table
1389 * otherwise send reset back
1391 sk = vsock_find_connected_socket(&src, &dst);
1393 sk = vsock_find_bound_socket(&dst);
1395 (void)virtio_transport_reset_no_sock(t, skb);
1400 if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1401 (void)virtio_transport_reset_no_sock(t, skb);
1406 if (!skb_set_owner_sk_safe(skb, sk)) {
1407 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1415 /* Check if sk has been closed before lock_sock */
1416 if (sock_flag(sk, SOCK_DONE)) {
1417 (void)virtio_transport_reset_no_sock(t, skb);
1423 space_available = virtio_transport_space_update(sk, skb);
1425 /* Update CID in case it has changed after a transport reset event */
1426 if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1427 vsk->local_addr.svm_cid = dst.svm_cid;
1429 if (space_available)
1430 sk->sk_write_space(sk);
1432 switch (sk->sk_state) {
1434 virtio_transport_recv_listen(sk, skb, t);
1438 virtio_transport_recv_connecting(sk, skb);
1441 case TCP_ESTABLISHED:
1442 virtio_transport_recv_connected(sk, skb);
1445 virtio_transport_recv_disconnecting(sk, skb);
1449 (void)virtio_transport_reset_no_sock(t, skb);
1456 /* Release refcnt obtained when we fetched this socket out of the
1457 * bound or connected list.
1465 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1467 /* Remove skbs found in a queue that have a vsk that matches.
1469 * Each skb is freed.
1471 * Returns the count of skbs that were reply packets.
1473 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1475 struct sk_buff_head freeme;
1476 struct sk_buff *skb, *tmp;
1479 skb_queue_head_init(&freeme);
1481 spin_lock_bh(&queue->lock);
1482 skb_queue_walk_safe(queue, skb, tmp) {
1483 if (vsock_sk(skb->sk) != vsk)
1486 __skb_unlink(skb, queue);
1487 __skb_queue_tail(&freeme, skb);
1489 if (virtio_vsock_skb_reply(skb))
1492 spin_unlock_bh(&queue->lock);
1494 __skb_queue_purge(&freeme);
1498 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1500 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1502 struct virtio_vsock_sock *vvs = vsk->trans;
1503 struct sock *sk = sk_vsock(vsk);
1504 struct sk_buff *skb;
1508 spin_lock_bh(&vvs->rx_lock);
1509 /* Use __skb_recv_datagram() for race-free handling of the receive. It
1510 * works for types other than dgrams.
1512 skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1513 spin_unlock_bh(&vvs->rx_lock);
1518 return recv_actor(sk, skb);
1520 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1522 MODULE_LICENSE("GPL v2");
1523 MODULE_AUTHOR("Asias He");
1524 MODULE_DESCRIPTION("common code for virtio vsock");