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 static struct virtio_vsock_pkt
*
41 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info
*info
,
48 struct virtio_vsock_pkt
*pkt
;
51 pkt
= kzalloc(sizeof(*pkt
), GFP_KERNEL
);
55 pkt
->hdr
.type
= cpu_to_le16(info
->type
);
56 pkt
->hdr
.op
= cpu_to_le16(info
->op
);
57 pkt
->hdr
.src_cid
= cpu_to_le64(src_cid
);
58 pkt
->hdr
.dst_cid
= cpu_to_le64(dst_cid
);
59 pkt
->hdr
.src_port
= cpu_to_le32(src_port
);
60 pkt
->hdr
.dst_port
= cpu_to_le32(dst_port
);
61 pkt
->hdr
.flags
= cpu_to_le32(info
->flags
);
63 pkt
->hdr
.len
= cpu_to_le32(len
);
64 pkt
->reply
= info
->reply
;
67 if (info
->msg
&& len
> 0) {
68 pkt
->buf
= kmalloc(len
, GFP_KERNEL
);
74 err
= memcpy_from_msg(pkt
->buf
, info
->msg
, len
);
78 if (msg_data_left(info
->msg
) == 0 &&
79 info
->type
== VIRTIO_VSOCK_TYPE_SEQPACKET
)
80 pkt
->hdr
.flags
|= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR
);
83 trace_virtio_transport_alloc_pkt(src_cid
, src_port
,
100 static struct sk_buff
*virtio_transport_build_skb(void *opaque
)
102 struct virtio_vsock_pkt
*pkt
= opaque
;
103 struct af_vsockmon_hdr
*hdr
;
108 /* A packet could be split to fit the RX buffer, so we can retrieve
109 * the payload length from the header and the buffer pointer taking
110 * care of the offset in the original packet.
112 payload_len
= le32_to_cpu(pkt
->hdr
.len
);
113 payload_buf
= pkt
->buf
+ pkt
->off
;
115 skb
= alloc_skb(sizeof(*hdr
) + sizeof(pkt
->hdr
) + payload_len
,
120 hdr
= skb_put(skb
, sizeof(*hdr
));
122 /* pkt->hdr is little-endian so no need to byteswap here */
123 hdr
->src_cid
= pkt
->hdr
.src_cid
;
124 hdr
->src_port
= pkt
->hdr
.src_port
;
125 hdr
->dst_cid
= pkt
->hdr
.dst_cid
;
126 hdr
->dst_port
= pkt
->hdr
.dst_port
;
128 hdr
->transport
= cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO
);
129 hdr
->len
= cpu_to_le16(sizeof(pkt
->hdr
));
130 memset(hdr
->reserved
, 0, sizeof(hdr
->reserved
));
132 switch (le16_to_cpu(pkt
->hdr
.op
)) {
133 case VIRTIO_VSOCK_OP_REQUEST
:
134 case VIRTIO_VSOCK_OP_RESPONSE
:
135 hdr
->op
= cpu_to_le16(AF_VSOCK_OP_CONNECT
);
137 case VIRTIO_VSOCK_OP_RST
:
138 case VIRTIO_VSOCK_OP_SHUTDOWN
:
139 hdr
->op
= cpu_to_le16(AF_VSOCK_OP_DISCONNECT
);
141 case VIRTIO_VSOCK_OP_RW
:
142 hdr
->op
= cpu_to_le16(AF_VSOCK_OP_PAYLOAD
);
144 case VIRTIO_VSOCK_OP_CREDIT_UPDATE
:
145 case VIRTIO_VSOCK_OP_CREDIT_REQUEST
:
146 hdr
->op
= cpu_to_le16(AF_VSOCK_OP_CONTROL
);
149 hdr
->op
= cpu_to_le16(AF_VSOCK_OP_UNKNOWN
);
153 skb_put_data(skb
, &pkt
->hdr
, sizeof(pkt
->hdr
));
156 skb_put_data(skb
, payload_buf
, payload_len
);
162 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt
*pkt
)
164 if (pkt
->tap_delivered
)
167 vsock_deliver_tap(virtio_transport_build_skb
, pkt
);
168 pkt
->tap_delivered
= true;
170 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt
);
172 static u16
virtio_transport_get_type(struct sock
*sk
)
174 if (sk
->sk_type
== SOCK_STREAM
)
175 return VIRTIO_VSOCK_TYPE_STREAM
;
177 return VIRTIO_VSOCK_TYPE_SEQPACKET
;
180 /* This function can only be used on connecting/connected sockets,
181 * since a socket assigned to a transport is required.
183 * Do not use on listener sockets!
185 static int virtio_transport_send_pkt_info(struct vsock_sock
*vsk
,
186 struct virtio_vsock_pkt_info
*info
)
188 u32 src_cid
, src_port
, dst_cid
, dst_port
;
189 const struct virtio_transport
*t_ops
;
190 struct virtio_vsock_sock
*vvs
;
191 struct virtio_vsock_pkt
*pkt
;
192 u32 pkt_len
= info
->pkt_len
;
194 info
->type
= virtio_transport_get_type(sk_vsock(vsk
));
196 t_ops
= virtio_transport_get_ops(vsk
);
197 if (unlikely(!t_ops
))
200 src_cid
= t_ops
->transport
.get_local_cid();
201 src_port
= vsk
->local_addr
.svm_port
;
202 if (!info
->remote_cid
) {
203 dst_cid
= vsk
->remote_addr
.svm_cid
;
204 dst_port
= vsk
->remote_addr
.svm_port
;
206 dst_cid
= info
->remote_cid
;
207 dst_port
= info
->remote_port
;
212 /* we can send less than pkt_len bytes */
213 if (pkt_len
> VIRTIO_VSOCK_MAX_PKT_BUF_SIZE
)
214 pkt_len
= VIRTIO_VSOCK_MAX_PKT_BUF_SIZE
;
216 /* virtio_transport_get_credit might return less than pkt_len credit */
217 pkt_len
= virtio_transport_get_credit(vvs
, pkt_len
);
219 /* Do not send zero length OP_RW pkt */
220 if (pkt_len
== 0 && info
->op
== VIRTIO_VSOCK_OP_RW
)
223 pkt
= virtio_transport_alloc_pkt(info
, pkt_len
,
227 virtio_transport_put_credit(vvs
, pkt_len
);
231 virtio_transport_inc_tx_pkt(vvs
, pkt
);
233 return t_ops
->send_pkt(pkt
);
236 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock
*vvs
,
237 struct virtio_vsock_pkt
*pkt
)
239 if (vvs
->rx_bytes
+ pkt
->len
> vvs
->buf_alloc
)
242 vvs
->rx_bytes
+= pkt
->len
;
246 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock
*vvs
,
247 struct virtio_vsock_pkt
*pkt
)
249 vvs
->rx_bytes
-= pkt
->len
;
250 vvs
->fwd_cnt
+= pkt
->len
;
253 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock
*vvs
, struct virtio_vsock_pkt
*pkt
)
255 spin_lock_bh(&vvs
->rx_lock
);
256 vvs
->last_fwd_cnt
= vvs
->fwd_cnt
;
257 pkt
->hdr
.fwd_cnt
= cpu_to_le32(vvs
->fwd_cnt
);
258 pkt
->hdr
.buf_alloc
= cpu_to_le32(vvs
->buf_alloc
);
259 spin_unlock_bh(&vvs
->rx_lock
);
261 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt
);
263 u32
virtio_transport_get_credit(struct virtio_vsock_sock
*vvs
, u32 credit
)
267 spin_lock_bh(&vvs
->tx_lock
);
268 ret
= vvs
->peer_buf_alloc
- (vvs
->tx_cnt
- vvs
->peer_fwd_cnt
);
272 spin_unlock_bh(&vvs
->tx_lock
);
276 EXPORT_SYMBOL_GPL(virtio_transport_get_credit
);
278 void virtio_transport_put_credit(struct virtio_vsock_sock
*vvs
, u32 credit
)
280 spin_lock_bh(&vvs
->tx_lock
);
281 vvs
->tx_cnt
-= credit
;
282 spin_unlock_bh(&vvs
->tx_lock
);
284 EXPORT_SYMBOL_GPL(virtio_transport_put_credit
);
286 static int virtio_transport_send_credit_update(struct vsock_sock
*vsk
)
288 struct virtio_vsock_pkt_info info
= {
289 .op
= VIRTIO_VSOCK_OP_CREDIT_UPDATE
,
293 return virtio_transport_send_pkt_info(vsk
, &info
);
297 virtio_transport_stream_do_peek(struct vsock_sock
*vsk
,
301 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
302 struct virtio_vsock_pkt
*pkt
;
303 size_t bytes
, total
= 0, off
;
306 spin_lock_bh(&vvs
->rx_lock
);
308 list_for_each_entry(pkt
, &vvs
->rx_queue
, list
) {
314 while (total
< len
&& off
< pkt
->len
) {
316 if (bytes
> pkt
->len
- off
)
317 bytes
= pkt
->len
- off
;
319 /* sk_lock is held by caller so no one else can dequeue.
320 * Unlock rx_lock since memcpy_to_msg() may sleep.
322 spin_unlock_bh(&vvs
->rx_lock
);
324 err
= memcpy_to_msg(msg
, pkt
->buf
+ off
, bytes
);
328 spin_lock_bh(&vvs
->rx_lock
);
335 spin_unlock_bh(&vvs
->rx_lock
);
346 virtio_transport_stream_do_dequeue(struct vsock_sock
*vsk
,
350 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
351 struct virtio_vsock_pkt
*pkt
;
352 size_t bytes
, total
= 0;
356 spin_lock_bh(&vvs
->rx_lock
);
357 while (total
< len
&& !list_empty(&vvs
->rx_queue
)) {
358 pkt
= list_first_entry(&vvs
->rx_queue
,
359 struct virtio_vsock_pkt
, list
);
362 if (bytes
> pkt
->len
- pkt
->off
)
363 bytes
= pkt
->len
- pkt
->off
;
365 /* sk_lock is held by caller so no one else can dequeue.
366 * Unlock rx_lock since memcpy_to_msg() may sleep.
368 spin_unlock_bh(&vvs
->rx_lock
);
370 err
= memcpy_to_msg(msg
, pkt
->buf
+ pkt
->off
, bytes
);
374 spin_lock_bh(&vvs
->rx_lock
);
378 if (pkt
->off
== pkt
->len
) {
379 virtio_transport_dec_rx_pkt(vvs
, pkt
);
380 list_del(&pkt
->list
);
381 virtio_transport_free_pkt(pkt
);
385 free_space
= vvs
->buf_alloc
- (vvs
->fwd_cnt
- vvs
->last_fwd_cnt
);
387 spin_unlock_bh(&vvs
->rx_lock
);
389 /* To reduce the number of credit update messages,
390 * don't update credits as long as lots of space is available.
391 * Note: the limit chosen here is arbitrary. Setting the limit
392 * too high causes extra messages. Too low causes transmitter
393 * stalls. As stalls are in theory more expensive than extra
394 * messages, we set the limit to a high value. TODO: experiment
395 * with different values.
397 if (free_space
< VIRTIO_VSOCK_MAX_PKT_BUF_SIZE
)
398 virtio_transport_send_credit_update(vsk
);
408 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock
*vsk
,
412 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
413 struct virtio_vsock_pkt
*pkt
;
414 int dequeued_len
= 0;
415 size_t user_buf_len
= msg_data_left(msg
);
416 bool msg_ready
= false;
418 spin_lock_bh(&vvs
->rx_lock
);
420 if (vvs
->msg_count
== 0) {
421 spin_unlock_bh(&vvs
->rx_lock
);
426 pkt
= list_first_entry(&vvs
->rx_queue
, struct virtio_vsock_pkt
, list
);
428 if (dequeued_len
>= 0) {
430 size_t bytes_to_copy
;
432 pkt_len
= (size_t)le32_to_cpu(pkt
->hdr
.len
);
433 bytes_to_copy
= min(user_buf_len
, pkt_len
);
438 /* sk_lock is held by caller so no one else can dequeue.
439 * Unlock rx_lock since memcpy_to_msg() may sleep.
441 spin_unlock_bh(&vvs
->rx_lock
);
443 err
= memcpy_to_msg(msg
, pkt
->buf
, bytes_to_copy
);
445 /* Copy of message failed. Rest of
446 * fragments will be freed without copy.
450 user_buf_len
-= bytes_to_copy
;
453 spin_lock_bh(&vvs
->rx_lock
);
456 if (dequeued_len
>= 0)
457 dequeued_len
+= pkt_len
;
460 if (le32_to_cpu(pkt
->hdr
.flags
) & VIRTIO_VSOCK_SEQ_EOR
) {
465 virtio_transport_dec_rx_pkt(vvs
, pkt
);
466 list_del(&pkt
->list
);
467 virtio_transport_free_pkt(pkt
);
470 spin_unlock_bh(&vvs
->rx_lock
);
472 virtio_transport_send_credit_update(vsk
);
478 virtio_transport_stream_dequeue(struct vsock_sock
*vsk
,
480 size_t len
, int flags
)
482 if (flags
& MSG_PEEK
)
483 return virtio_transport_stream_do_peek(vsk
, msg
, len
);
485 return virtio_transport_stream_do_dequeue(vsk
, msg
, len
);
487 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue
);
490 virtio_transport_seqpacket_dequeue(struct vsock_sock
*vsk
,
494 if (flags
& MSG_PEEK
)
497 return virtio_transport_seqpacket_do_dequeue(vsk
, msg
, flags
);
499 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue
);
502 virtio_transport_seqpacket_enqueue(struct vsock_sock
*vsk
,
506 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
508 spin_lock_bh(&vvs
->tx_lock
);
510 if (len
> vvs
->peer_buf_alloc
) {
511 spin_unlock_bh(&vvs
->tx_lock
);
515 spin_unlock_bh(&vvs
->tx_lock
);
517 return virtio_transport_stream_enqueue(vsk
, msg
, len
);
519 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue
);
522 virtio_transport_dgram_dequeue(struct vsock_sock
*vsk
,
524 size_t len
, int flags
)
528 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue
);
530 s64
virtio_transport_stream_has_data(struct vsock_sock
*vsk
)
532 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
535 spin_lock_bh(&vvs
->rx_lock
);
536 bytes
= vvs
->rx_bytes
;
537 spin_unlock_bh(&vvs
->rx_lock
);
541 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data
);
543 u32
virtio_transport_seqpacket_has_data(struct vsock_sock
*vsk
)
545 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
548 spin_lock_bh(&vvs
->rx_lock
);
549 msg_count
= vvs
->msg_count
;
550 spin_unlock_bh(&vvs
->rx_lock
);
554 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data
);
556 static s64
virtio_transport_has_space(struct vsock_sock
*vsk
)
558 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
561 bytes
= vvs
->peer_buf_alloc
- (vvs
->tx_cnt
- vvs
->peer_fwd_cnt
);
568 s64
virtio_transport_stream_has_space(struct vsock_sock
*vsk
)
570 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
573 spin_lock_bh(&vvs
->tx_lock
);
574 bytes
= virtio_transport_has_space(vsk
);
575 spin_unlock_bh(&vvs
->tx_lock
);
579 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space
);
581 int virtio_transport_do_socket_init(struct vsock_sock
*vsk
,
582 struct vsock_sock
*psk
)
584 struct virtio_vsock_sock
*vvs
;
586 vvs
= kzalloc(sizeof(*vvs
), GFP_KERNEL
);
592 if (psk
&& psk
->trans
) {
593 struct virtio_vsock_sock
*ptrans
= psk
->trans
;
595 vvs
->peer_buf_alloc
= ptrans
->peer_buf_alloc
;
598 if (vsk
->buffer_size
> VIRTIO_VSOCK_MAX_BUF_SIZE
)
599 vsk
->buffer_size
= VIRTIO_VSOCK_MAX_BUF_SIZE
;
601 vvs
->buf_alloc
= vsk
->buffer_size
;
603 spin_lock_init(&vvs
->rx_lock
);
604 spin_lock_init(&vvs
->tx_lock
);
605 INIT_LIST_HEAD(&vvs
->rx_queue
);
609 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init
);
611 /* sk_lock held by the caller */
612 void virtio_transport_notify_buffer_size(struct vsock_sock
*vsk
, u64
*val
)
614 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
616 if (*val
> VIRTIO_VSOCK_MAX_BUF_SIZE
)
617 *val
= VIRTIO_VSOCK_MAX_BUF_SIZE
;
619 vvs
->buf_alloc
= *val
;
621 virtio_transport_send_credit_update(vsk
);
623 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size
);
626 virtio_transport_notify_poll_in(struct vsock_sock
*vsk
,
628 bool *data_ready_now
)
630 if (vsock_stream_has_data(vsk
))
631 *data_ready_now
= true;
633 *data_ready_now
= false;
637 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in
);
640 virtio_transport_notify_poll_out(struct vsock_sock
*vsk
,
642 bool *space_avail_now
)
646 free_space
= vsock_stream_has_space(vsk
);
648 *space_avail_now
= true;
649 else if (free_space
== 0)
650 *space_avail_now
= false;
654 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out
);
656 int virtio_transport_notify_recv_init(struct vsock_sock
*vsk
,
657 size_t target
, struct vsock_transport_recv_notify_data
*data
)
661 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init
);
663 int virtio_transport_notify_recv_pre_block(struct vsock_sock
*vsk
,
664 size_t target
, struct vsock_transport_recv_notify_data
*data
)
668 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block
);
670 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock
*vsk
,
671 size_t target
, struct vsock_transport_recv_notify_data
*data
)
675 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue
);
677 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock
*vsk
,
678 size_t target
, ssize_t copied
, bool data_read
,
679 struct vsock_transport_recv_notify_data
*data
)
683 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue
);
685 int virtio_transport_notify_send_init(struct vsock_sock
*vsk
,
686 struct vsock_transport_send_notify_data
*data
)
690 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init
);
692 int virtio_transport_notify_send_pre_block(struct vsock_sock
*vsk
,
693 struct vsock_transport_send_notify_data
*data
)
697 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block
);
699 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock
*vsk
,
700 struct vsock_transport_send_notify_data
*data
)
704 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue
);
706 int virtio_transport_notify_send_post_enqueue(struct vsock_sock
*vsk
,
707 ssize_t written
, struct vsock_transport_send_notify_data
*data
)
711 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue
);
713 u64
virtio_transport_stream_rcvhiwat(struct vsock_sock
*vsk
)
715 return vsk
->buffer_size
;
717 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat
);
719 bool virtio_transport_stream_is_active(struct vsock_sock
*vsk
)
723 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active
);
725 bool virtio_transport_stream_allow(u32 cid
, u32 port
)
729 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow
);
731 int virtio_transport_dgram_bind(struct vsock_sock
*vsk
,
732 struct sockaddr_vm
*addr
)
736 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind
);
738 bool virtio_transport_dgram_allow(u32 cid
, u32 port
)
742 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow
);
744 int virtio_transport_connect(struct vsock_sock
*vsk
)
746 struct virtio_vsock_pkt_info info
= {
747 .op
= VIRTIO_VSOCK_OP_REQUEST
,
751 return virtio_transport_send_pkt_info(vsk
, &info
);
753 EXPORT_SYMBOL_GPL(virtio_transport_connect
);
755 int virtio_transport_shutdown(struct vsock_sock
*vsk
, int mode
)
757 struct virtio_vsock_pkt_info info
= {
758 .op
= VIRTIO_VSOCK_OP_SHUTDOWN
,
759 .flags
= (mode
& RCV_SHUTDOWN
?
760 VIRTIO_VSOCK_SHUTDOWN_RCV
: 0) |
761 (mode
& SEND_SHUTDOWN
?
762 VIRTIO_VSOCK_SHUTDOWN_SEND
: 0),
766 return virtio_transport_send_pkt_info(vsk
, &info
);
768 EXPORT_SYMBOL_GPL(virtio_transport_shutdown
);
771 virtio_transport_dgram_enqueue(struct vsock_sock
*vsk
,
772 struct sockaddr_vm
*remote_addr
,
778 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue
);
781 virtio_transport_stream_enqueue(struct vsock_sock
*vsk
,
785 struct virtio_vsock_pkt_info info
= {
786 .op
= VIRTIO_VSOCK_OP_RW
,
792 return virtio_transport_send_pkt_info(vsk
, &info
);
794 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue
);
796 void virtio_transport_destruct(struct vsock_sock
*vsk
)
798 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
802 EXPORT_SYMBOL_GPL(virtio_transport_destruct
);
804 static int virtio_transport_reset(struct vsock_sock
*vsk
,
805 struct virtio_vsock_pkt
*pkt
)
807 struct virtio_vsock_pkt_info info
= {
808 .op
= VIRTIO_VSOCK_OP_RST
,
813 /* Send RST only if the original pkt is not a RST pkt */
814 if (pkt
&& le16_to_cpu(pkt
->hdr
.op
) == VIRTIO_VSOCK_OP_RST
)
817 return virtio_transport_send_pkt_info(vsk
, &info
);
820 /* Normally packets are associated with a socket. There may be no socket if an
821 * attempt was made to connect to a socket that does not exist.
823 static int virtio_transport_reset_no_sock(const struct virtio_transport
*t
,
824 struct virtio_vsock_pkt
*pkt
)
826 struct virtio_vsock_pkt
*reply
;
827 struct virtio_vsock_pkt_info info
= {
828 .op
= VIRTIO_VSOCK_OP_RST
,
829 .type
= le16_to_cpu(pkt
->hdr
.type
),
833 /* Send RST only if the original pkt is not a RST pkt */
834 if (le16_to_cpu(pkt
->hdr
.op
) == VIRTIO_VSOCK_OP_RST
)
837 reply
= virtio_transport_alloc_pkt(&info
, 0,
838 le64_to_cpu(pkt
->hdr
.dst_cid
),
839 le32_to_cpu(pkt
->hdr
.dst_port
),
840 le64_to_cpu(pkt
->hdr
.src_cid
),
841 le32_to_cpu(pkt
->hdr
.src_port
));
846 virtio_transport_free_pkt(reply
);
850 return t
->send_pkt(reply
);
853 /* This function should be called with sk_lock held and SOCK_DONE set */
854 static void virtio_transport_remove_sock(struct vsock_sock
*vsk
)
856 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
857 struct virtio_vsock_pkt
*pkt
, *tmp
;
859 /* We don't need to take rx_lock, as the socket is closing and we are
862 list_for_each_entry_safe(pkt
, tmp
, &vvs
->rx_queue
, list
) {
863 list_del(&pkt
->list
);
864 virtio_transport_free_pkt(pkt
);
867 vsock_remove_sock(vsk
);
870 static void virtio_transport_wait_close(struct sock
*sk
, long timeout
)
873 DEFINE_WAIT_FUNC(wait
, woken_wake_function
);
875 add_wait_queue(sk_sleep(sk
), &wait
);
878 if (sk_wait_event(sk
, &timeout
,
879 sock_flag(sk
, SOCK_DONE
), &wait
))
881 } while (!signal_pending(current
) && timeout
);
883 remove_wait_queue(sk_sleep(sk
), &wait
);
887 static void virtio_transport_do_close(struct vsock_sock
*vsk
,
890 struct sock
*sk
= sk_vsock(vsk
);
892 sock_set_flag(sk
, SOCK_DONE
);
893 vsk
->peer_shutdown
= SHUTDOWN_MASK
;
894 if (vsock_stream_has_data(vsk
) <= 0)
895 sk
->sk_state
= TCP_CLOSING
;
896 sk
->sk_state_change(sk
);
898 if (vsk
->close_work_scheduled
&&
899 (!cancel_timeout
|| cancel_delayed_work(&vsk
->close_work
))) {
900 vsk
->close_work_scheduled
= false;
902 virtio_transport_remove_sock(vsk
);
904 /* Release refcnt obtained when we scheduled the timeout */
909 static void virtio_transport_close_timeout(struct work_struct
*work
)
911 struct vsock_sock
*vsk
=
912 container_of(work
, struct vsock_sock
, close_work
.work
);
913 struct sock
*sk
= sk_vsock(vsk
);
918 if (!sock_flag(sk
, SOCK_DONE
)) {
919 (void)virtio_transport_reset(vsk
, NULL
);
921 virtio_transport_do_close(vsk
, false);
924 vsk
->close_work_scheduled
= false;
930 /* User context, vsk->sk is locked */
931 static bool virtio_transport_close(struct vsock_sock
*vsk
)
933 struct sock
*sk
= &vsk
->sk
;
935 if (!(sk
->sk_state
== TCP_ESTABLISHED
||
936 sk
->sk_state
== TCP_CLOSING
))
939 /* Already received SHUTDOWN from peer, reply with RST */
940 if ((vsk
->peer_shutdown
& SHUTDOWN_MASK
) == SHUTDOWN_MASK
) {
941 (void)virtio_transport_reset(vsk
, NULL
);
945 if ((sk
->sk_shutdown
& SHUTDOWN_MASK
) != SHUTDOWN_MASK
)
946 (void)virtio_transport_shutdown(vsk
, SHUTDOWN_MASK
);
948 if (sock_flag(sk
, SOCK_LINGER
) && !(current
->flags
& PF_EXITING
))
949 virtio_transport_wait_close(sk
, sk
->sk_lingertime
);
951 if (sock_flag(sk
, SOCK_DONE
)) {
956 INIT_DELAYED_WORK(&vsk
->close_work
,
957 virtio_transport_close_timeout
);
958 vsk
->close_work_scheduled
= true;
959 schedule_delayed_work(&vsk
->close_work
, VSOCK_CLOSE_TIMEOUT
);
963 void virtio_transport_release(struct vsock_sock
*vsk
)
965 struct sock
*sk
= &vsk
->sk
;
966 bool remove_sock
= true;
968 if (sk
->sk_type
== SOCK_STREAM
|| sk
->sk_type
== SOCK_SEQPACKET
)
969 remove_sock
= virtio_transport_close(vsk
);
972 sock_set_flag(sk
, SOCK_DONE
);
973 virtio_transport_remove_sock(vsk
);
976 EXPORT_SYMBOL_GPL(virtio_transport_release
);
979 virtio_transport_recv_connecting(struct sock
*sk
,
980 struct virtio_vsock_pkt
*pkt
)
982 struct vsock_sock
*vsk
= vsock_sk(sk
);
986 switch (le16_to_cpu(pkt
->hdr
.op
)) {
987 case VIRTIO_VSOCK_OP_RESPONSE
:
988 sk
->sk_state
= TCP_ESTABLISHED
;
989 sk
->sk_socket
->state
= SS_CONNECTED
;
990 vsock_insert_connected(vsk
);
991 sk
->sk_state_change(sk
);
993 case VIRTIO_VSOCK_OP_INVALID
:
995 case VIRTIO_VSOCK_OP_RST
:
1007 virtio_transport_reset(vsk
, pkt
);
1008 sk
->sk_state
= TCP_CLOSE
;
1010 sk_error_report(sk
);
1015 virtio_transport_recv_enqueue(struct vsock_sock
*vsk
,
1016 struct virtio_vsock_pkt
*pkt
)
1018 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
1019 bool can_enqueue
, free_pkt
= false;
1021 pkt
->len
= le32_to_cpu(pkt
->hdr
.len
);
1024 spin_lock_bh(&vvs
->rx_lock
);
1026 can_enqueue
= virtio_transport_inc_rx_pkt(vvs
, pkt
);
1032 if (le32_to_cpu(pkt
->hdr
.flags
) & VIRTIO_VSOCK_SEQ_EOR
)
1035 /* Try to copy small packets into the buffer of last packet queued,
1036 * to avoid wasting memory queueing the entire buffer with a small
1039 if (pkt
->len
<= GOOD_COPY_LEN
&& !list_empty(&vvs
->rx_queue
)) {
1040 struct virtio_vsock_pkt
*last_pkt
;
1042 last_pkt
= list_last_entry(&vvs
->rx_queue
,
1043 struct virtio_vsock_pkt
, list
);
1045 /* If there is space in the last packet queued, we copy the
1046 * new packet in its buffer. We avoid this if the last packet
1047 * queued has VIRTIO_VSOCK_SEQ_EOR set, because this is
1048 * delimiter of SEQPACKET record, so 'pkt' is the first packet
1051 if ((pkt
->len
<= last_pkt
->buf_len
- last_pkt
->len
) &&
1052 !(le32_to_cpu(last_pkt
->hdr
.flags
) & VIRTIO_VSOCK_SEQ_EOR
)) {
1053 memcpy(last_pkt
->buf
+ last_pkt
->len
, pkt
->buf
,
1055 last_pkt
->len
+= pkt
->len
;
1057 last_pkt
->hdr
.flags
|= pkt
->hdr
.flags
;
1062 list_add_tail(&pkt
->list
, &vvs
->rx_queue
);
1065 spin_unlock_bh(&vvs
->rx_lock
);
1067 virtio_transport_free_pkt(pkt
);
1071 virtio_transport_recv_connected(struct sock
*sk
,
1072 struct virtio_vsock_pkt
*pkt
)
1074 struct vsock_sock
*vsk
= vsock_sk(sk
);
1077 switch (le16_to_cpu(pkt
->hdr
.op
)) {
1078 case VIRTIO_VSOCK_OP_RW
:
1079 virtio_transport_recv_enqueue(vsk
, pkt
);
1080 sk
->sk_data_ready(sk
);
1082 case VIRTIO_VSOCK_OP_CREDIT_REQUEST
:
1083 virtio_transport_send_credit_update(vsk
);
1085 case VIRTIO_VSOCK_OP_CREDIT_UPDATE
:
1086 sk
->sk_write_space(sk
);
1088 case VIRTIO_VSOCK_OP_SHUTDOWN
:
1089 if (le32_to_cpu(pkt
->hdr
.flags
) & VIRTIO_VSOCK_SHUTDOWN_RCV
)
1090 vsk
->peer_shutdown
|= RCV_SHUTDOWN
;
1091 if (le32_to_cpu(pkt
->hdr
.flags
) & VIRTIO_VSOCK_SHUTDOWN_SEND
)
1092 vsk
->peer_shutdown
|= SEND_SHUTDOWN
;
1093 if (vsk
->peer_shutdown
== SHUTDOWN_MASK
&&
1094 vsock_stream_has_data(vsk
) <= 0 &&
1095 !sock_flag(sk
, SOCK_DONE
)) {
1096 (void)virtio_transport_reset(vsk
, NULL
);
1098 virtio_transport_do_close(vsk
, true);
1100 if (le32_to_cpu(pkt
->hdr
.flags
))
1101 sk
->sk_state_change(sk
);
1103 case VIRTIO_VSOCK_OP_RST
:
1104 virtio_transport_do_close(vsk
, true);
1111 virtio_transport_free_pkt(pkt
);
1116 virtio_transport_recv_disconnecting(struct sock
*sk
,
1117 struct virtio_vsock_pkt
*pkt
)
1119 struct vsock_sock
*vsk
= vsock_sk(sk
);
1121 if (le16_to_cpu(pkt
->hdr
.op
) == VIRTIO_VSOCK_OP_RST
)
1122 virtio_transport_do_close(vsk
, true);
1126 virtio_transport_send_response(struct vsock_sock
*vsk
,
1127 struct virtio_vsock_pkt
*pkt
)
1129 struct virtio_vsock_pkt_info info
= {
1130 .op
= VIRTIO_VSOCK_OP_RESPONSE
,
1131 .remote_cid
= le64_to_cpu(pkt
->hdr
.src_cid
),
1132 .remote_port
= le32_to_cpu(pkt
->hdr
.src_port
),
1137 return virtio_transport_send_pkt_info(vsk
, &info
);
1140 static bool virtio_transport_space_update(struct sock
*sk
,
1141 struct virtio_vsock_pkt
*pkt
)
1143 struct vsock_sock
*vsk
= vsock_sk(sk
);
1144 struct virtio_vsock_sock
*vvs
= vsk
->trans
;
1145 bool space_available
;
1147 /* Listener sockets are not associated with any transport, so we are
1148 * not able to take the state to see if there is space available in the
1149 * remote peer, but since they are only used to receive requests, we
1150 * can assume that there is always space available in the other peer.
1155 /* buf_alloc and fwd_cnt is always included in the hdr */
1156 spin_lock_bh(&vvs
->tx_lock
);
1157 vvs
->peer_buf_alloc
= le32_to_cpu(pkt
->hdr
.buf_alloc
);
1158 vvs
->peer_fwd_cnt
= le32_to_cpu(pkt
->hdr
.fwd_cnt
);
1159 space_available
= virtio_transport_has_space(vsk
);
1160 spin_unlock_bh(&vvs
->tx_lock
);
1161 return space_available
;
1164 /* Handle server socket */
1166 virtio_transport_recv_listen(struct sock
*sk
, struct virtio_vsock_pkt
*pkt
,
1167 struct virtio_transport
*t
)
1169 struct vsock_sock
*vsk
= vsock_sk(sk
);
1170 struct vsock_sock
*vchild
;
1174 if (le16_to_cpu(pkt
->hdr
.op
) != VIRTIO_VSOCK_OP_REQUEST
) {
1175 virtio_transport_reset_no_sock(t
, pkt
);
1179 if (sk_acceptq_is_full(sk
)) {
1180 virtio_transport_reset_no_sock(t
, pkt
);
1184 child
= vsock_create_connected(sk
);
1186 virtio_transport_reset_no_sock(t
, pkt
);
1190 sk_acceptq_added(sk
);
1192 lock_sock_nested(child
, SINGLE_DEPTH_NESTING
);
1194 child
->sk_state
= TCP_ESTABLISHED
;
1196 vchild
= vsock_sk(child
);
1197 vsock_addr_init(&vchild
->local_addr
, le64_to_cpu(pkt
->hdr
.dst_cid
),
1198 le32_to_cpu(pkt
->hdr
.dst_port
));
1199 vsock_addr_init(&vchild
->remote_addr
, le64_to_cpu(pkt
->hdr
.src_cid
),
1200 le32_to_cpu(pkt
->hdr
.src_port
));
1202 ret
= vsock_assign_transport(vchild
, vsk
);
1203 /* Transport assigned (looking at remote_addr) must be the same
1204 * where we received the request.
1206 if (ret
|| vchild
->transport
!= &t
->transport
) {
1207 release_sock(child
);
1208 virtio_transport_reset_no_sock(t
, pkt
);
1213 if (virtio_transport_space_update(child
, pkt
))
1214 child
->sk_write_space(child
);
1216 vsock_insert_connected(vchild
);
1217 vsock_enqueue_accept(sk
, child
);
1218 virtio_transport_send_response(vchild
, pkt
);
1220 release_sock(child
);
1222 sk
->sk_data_ready(sk
);
1226 static bool virtio_transport_valid_type(u16 type
)
1228 return (type
== VIRTIO_VSOCK_TYPE_STREAM
) ||
1229 (type
== VIRTIO_VSOCK_TYPE_SEQPACKET
);
1232 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1235 void virtio_transport_recv_pkt(struct virtio_transport
*t
,
1236 struct virtio_vsock_pkt
*pkt
)
1238 struct sockaddr_vm src
, dst
;
1239 struct vsock_sock
*vsk
;
1241 bool space_available
;
1243 vsock_addr_init(&src
, le64_to_cpu(pkt
->hdr
.src_cid
),
1244 le32_to_cpu(pkt
->hdr
.src_port
));
1245 vsock_addr_init(&dst
, le64_to_cpu(pkt
->hdr
.dst_cid
),
1246 le32_to_cpu(pkt
->hdr
.dst_port
));
1248 trace_virtio_transport_recv_pkt(src
.svm_cid
, src
.svm_port
,
1249 dst
.svm_cid
, dst
.svm_port
,
1250 le32_to_cpu(pkt
->hdr
.len
),
1251 le16_to_cpu(pkt
->hdr
.type
),
1252 le16_to_cpu(pkt
->hdr
.op
),
1253 le32_to_cpu(pkt
->hdr
.flags
),
1254 le32_to_cpu(pkt
->hdr
.buf_alloc
),
1255 le32_to_cpu(pkt
->hdr
.fwd_cnt
));
1257 if (!virtio_transport_valid_type(le16_to_cpu(pkt
->hdr
.type
))) {
1258 (void)virtio_transport_reset_no_sock(t
, pkt
);
1262 /* The socket must be in connected or bound table
1263 * otherwise send reset back
1265 sk
= vsock_find_connected_socket(&src
, &dst
);
1267 sk
= vsock_find_bound_socket(&dst
);
1269 (void)virtio_transport_reset_no_sock(t
, pkt
);
1274 if (virtio_transport_get_type(sk
) != le16_to_cpu(pkt
->hdr
.type
)) {
1275 (void)virtio_transport_reset_no_sock(t
, pkt
);
1284 /* Check if sk has been closed before lock_sock */
1285 if (sock_flag(sk
, SOCK_DONE
)) {
1286 (void)virtio_transport_reset_no_sock(t
, pkt
);
1292 space_available
= virtio_transport_space_update(sk
, pkt
);
1294 /* Update CID in case it has changed after a transport reset event */
1295 vsk
->local_addr
.svm_cid
= dst
.svm_cid
;
1297 if (space_available
)
1298 sk
->sk_write_space(sk
);
1300 switch (sk
->sk_state
) {
1302 virtio_transport_recv_listen(sk
, pkt
, t
);
1303 virtio_transport_free_pkt(pkt
);
1306 virtio_transport_recv_connecting(sk
, pkt
);
1307 virtio_transport_free_pkt(pkt
);
1309 case TCP_ESTABLISHED
:
1310 virtio_transport_recv_connected(sk
, pkt
);
1313 virtio_transport_recv_disconnecting(sk
, pkt
);
1314 virtio_transport_free_pkt(pkt
);
1317 (void)virtio_transport_reset_no_sock(t
, pkt
);
1318 virtio_transport_free_pkt(pkt
);
1324 /* Release refcnt obtained when we fetched this socket out of the
1325 * bound or connected list.
1331 virtio_transport_free_pkt(pkt
);
1333 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt
);
1335 void virtio_transport_free_pkt(struct virtio_vsock_pkt
*pkt
)
1340 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt
);
1342 MODULE_LICENSE("GPL v2");
1343 MODULE_AUTHOR("Asias He");
1344 MODULE_DESCRIPTION("common code for virtio vsock");