This patchset implements support of SOCK_SEQPACKET for virtio
transport.
As SOCK_SEQPACKET guarantees to save record boundaries, so to
do it, new bit for field 'flags' was added: SEQ_EOR. This bit is
set to 1 in last RW packet of message.
Now as packets of one socket are not reordered neither on vsock
nor on vhost transport layers, such bit allows to restore original
message on receiver's side. If user's buffer is smaller than message
length, when all out of size data is dropped.
Maximum length of datagram is not limited as in stream socket,
because same credit logic is used. Difference with stream socket is
that user is not woken up until whole record is received or error
occurred. Implementation also supports 'MSG_TRUNC' flags.
Tests also implemented.
Thanks to stsp2@yandex.ru for encouragements and initial design
recommendations.
Arseny Krasnov (18):
af_vsock: update functions for connectible socket
af_vsock: separate wait data loop
af_vsock: separate receive data loop
af_vsock: implement SEQPACKET receive loop
af_vsock: implement send logic for SEQPACKET
af_vsock: rest of SEQPACKET support
af_vsock: update comments for stream sockets
virtio/vsock: set packet's type in virtio_transport_send_pkt_info()
virtio/vsock: simplify credit update function API
virtio/vsock: defines and constants for SEQPACKET
virtio/vsock: dequeue callback for SOCK_SEQPACKET
virtio/vsock: add SEQPACKET receive logic
virtio/vsock: rest of SOCK_SEQPACKET support
virtio/vsock: enable SEQPACKET for transport
vhost/vsock: enable SEQPACKET for transport
vsock/loopback: enable SEQPACKET for transport
vsock_test: add SOCK_SEQPACKET tests
virtio/vsock: update trace event for SEQPACKET
drivers/vhost/vsock.c | 44 +-
include/linux/virtio_vsock.h | 9 +
include/net/af_vsock.h | 7 +
.../events/vsock_virtio_transport_common.h | 5 +-
include/uapi/linux/virtio_vsock.h | 9 +
net/vmw_vsock/af_vsock.c | 465 +++++++++++------
net/vmw_vsock/virtio_transport.c | 25 +
net/vmw_vsock/virtio_transport_common.c | 133 ++++-
net/vmw_vsock/vsock_loopback.c | 11 +
tools/testing/vsock/util.c | 32 +-
tools/testing/vsock/util.h | 3 +
tools/testing/vsock/vsock_test.c | 116 ++++
12 files changed, 672 insertions(+), 187 deletions(-)
v9 -> v10:
General changelog:
- patch for write serialization removed from patchset
- commit messages rephrased
- RFC tag removed
Per patch changelog:
see every patch after '---' line.
v8 -> v9:
General changelog:
- see per patch change log.
Per patch changelog:
see every patch after '---' line.
v7 -> v8:
General changelog:
- whole idea is simplified: channel now considered reliable,
so SEQ_BEGIN, SEQ_END, 'msg_len' and 'msg_id' were removed.
Only thing that is used to mark end of message is bit in
'flags' field of packet header: VIRTIO_VSOCK_SEQ_EOR. Packet
with such bit set to 1 means, that this is last packet of
message.
- POSIX MSG_EOR support is removed, as there is no exact
description how it works.
- all changes to 'include/uapi/linux/virtio_vsock.h' moved
to dedicated patch, as these changes linked with patch to
spec.
- patch 'virtio/vsock: SEQPACKET feature bit support' now merged
to 'virtio/vsock: setup SEQPACKET ops for transport'.
- patch 'vhost/vsock: SEQPACKET feature bit support' now merged
to 'vhost/vsock: setup SEQPACKET ops for transport'.
Per patch changelog:
see every patch after '---' line.
v6 -> v7:
General changelog:
- virtio transport callback for message length now removed
from transport. Length of record is returned by dequeue
callback.
- function which tries to get message length now returns 0
when rx queue is empty. Also length of current message in
progress is set to 0, when message processed or error
happens.
- patches for virtio feature bit moved after patches with
transport ops.
Per patch changelog:
see every patch after '---' line.
v5 -> v6:
General changelog:
- virtio transport specific callbacks which send SEQ_BEGIN or
SEQ_END now hidden inside virtio transport. Only enqueue,
dequeue and record length callbacks are provided by transport.
- virtio feature bit for SEQPACKET socket support introduced:
VIRTIO_VSOCK_F_SEQPACKET.
- 'msg_cnt' field in 'struct virtio_vsock_seq_hdr' renamed to
'msg_id' and used as id.
Per patch changelog:
- 'af_vsock: separate wait data loop':
1) Commit message updated.
2) 'prepare_to_wait()' moved inside while loop(thanks to
Jorgen Hansen).
Marked 'Reviewed-by' with 1), but as 2) I removed R-b.
- 'af_vsock: separate receive data loop': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'af_vsock: implement SEQPACKET receive loop': style fixes.
- 'af_vsock: rest of SEQPACKET support':
1) 'module_put()' added when transport callback check failed.
2) Now only 'seqpacket_allow()' callback called to check
support of SEQPACKET by transport.
- 'af_vsock: update comments for stream sockets': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'virtio/vsock: set packet's type in send':
1) Commit message updated.
2) Parameter 'type' from 'virtio_transport_send_credit_update()'
also removed in this patch instead of in next.
- 'virtio/vsock: dequeue callback for SOCK_SEQPACKET': SEQPACKET
related state wrapped to special struct.
- 'virtio/vsock: update trace event for SEQPACKET': format strings
now not broken by new lines.
v4 -> v5:
- patches reorganized:
1) Setting of packet's type in 'virtio_transport_send_pkt_info()'
is moved to separate patch.
2) Simplifying of 'virtio_transport_send_credit_update()' is
moved to separate patch and before main virtio/vsock patches.
- style problem fixed
- in 'af_vsock: separate receive data loop' extra 'release_sock()'
removed
- added trace event fields for SEQPACKET
- in 'af_vsock: separate wait data loop':
1) 'vsock_wait_data()' removed 'goto out;'
2) Comment for invalid data amount is changed.
- in 'af_vsock: rest of SEQPACKET support', 'new_transport' pointer
check is moved after 'try_module_get()'
- in 'af_vsock: update comments for stream sockets', 'connect-oriented'
replaced with 'connection-oriented'
- in 'loopback/vsock: setup SEQPACKET ops for transport',
'loopback/vsock' replaced with 'vsock/loopback'
v3 -> v4:
- SEQPACKET specific metadata moved from packet header to payload
and called 'virtio_vsock_seq_hdr'
- record integrity check:
1) SEQ_END operation was added, which marks end of record.
2) Both SEQ_BEGIN and SEQ_END carries counter which is incremented
on every marker send.
- af_vsock.c: socket operations for STREAM and SEQPACKET call same
functions instead of having own "gates" differs only by names:
'vsock_seqpacket/stream_getsockopt()' now replaced with
'vsock_connectible_getsockopt()'.
- af_vsock.c: 'seqpacket_dequeue' callback returns error and flag that
record ready. There is no need to return number of copied bytes,
because case when record received successfully is checked at virtio
transport layer, when SEQ_END is processed. Also user doesn't need
number of copied bytes, because 'recv()' from SEQPACKET could return
error, length of users's buffer or length of whole record(both are
known in af_vsock.c).
- af_vsock.c: both wait loops in af_vsock.c(for data and space) moved
to separate functions because now both called from several places.
- af_vsock.c: 'vsock_assign_transport()' checks that 'new_transport'
pointer is not NULL and returns 'ESOCKTNOSUPPORT' instead of 'ENODEV'
if failed to use transport.
- tools/testing/vsock/vsock_test.c: rename tests
v2 -> v3:
- patches reorganized: split for prepare and implementation patches
- local variables are declared in "Reverse Christmas tree" manner
- virtio_transport_common.c: valid leXX_to_cpu() for vsock header
fields access
- af_vsock.c: 'vsock_connectible_*sockopt()' added as shared code
between stream and seqpacket sockets.
- af_vsock.c: loops in '__vsock_*_recvmsg()' refactored.
- af_vsock.c: 'vsock_wait_data()' refactored.
v1 -> v2:
- patches reordered: af_vsock.c related changes now before virtio vsock
- patches reorganized: more small patches, where +/- are not mixed
- tests for SOCK_SEQPACKET added
- all commit messages updated
- af_vsock.c: 'vsock_pre_recv_check()' inlined to
'vsock_connectible_recvmsg()'
- af_vsock.c: 'vsock_assign_transport()' returns ENODEV if transport
was not found
- virtio_transport_common.c: transport callback for seqpacket dequeue
- virtio_transport_common.c: simplified
'virtio_transport_recv_connected()'
- virtio_transport_common.c: send reset on socket and packet type
mismatch.
Signed-off-by: Arseny Krasnov <redacted>
--
2.25.1
Prepare af_vsock.c for SEQPACKET support: rename some functions such
as setsockopt(), getsockopt(), connect(), recvmsg(), sendmsg() in general
manner, because they are shared with stream sockets.
Signed-off-by: Arseny Krasnov <redacted>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
---
net/vmw_vsock/af_vsock.c | 64 +++++++++++++++++++++-------------------
1 file changed, 34 insertions(+), 30 deletions(-)
Wait loop for data could be shared between SEQPACKET and STREAM
sockets, so move it to dedicated function. While moving the code
around, let's update an old comment.
Signed-off-by: Arseny Krasnov <redacted>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
---
net/vmw_vsock/af_vsock.c | 156 +++++++++++++++++++++------------------
1 file changed, 84 insertions(+), 72 deletions(-)
@@ -1833,6 +1833,69 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,returnerr;}+staticintvsock_wait_data(structsock*sk,structwait_queue_entry*wait,+longtimeout,+structvsock_transport_recv_notify_data*recv_data,+size_ttarget)+{+conststructvsock_transport*transport;+structvsock_sock*vsk;+s64data;+interr;++vsk=vsock_sk(sk);+err=0;+transport=vsk->transport;++while((data=vsock_stream_has_data(vsk))==0){+prepare_to_wait(sk_sleep(sk),wait,TASK_INTERRUPTIBLE);++if(sk->sk_err!=0||+(sk->sk_shutdown&RCV_SHUTDOWN)||+(vsk->peer_shutdown&SEND_SHUTDOWN)){+break;+}++/* Don't wait for non-blocking sockets. */+if(timeout==0){+err=-EAGAIN;+break;+}++if(recv_data){+err=transport->notify_recv_pre_block(vsk,target,recv_data);+if(err<0)+break;+}++release_sock(sk);+timeout=schedule_timeout(timeout);+lock_sock(sk);++if(signal_pending(current)){+err=sock_intr_errno(timeout);+break;+}elseif(timeout==0){+err=-EAGAIN;+break;+}+}++finish_wait(sk_sleep(sk),wait);++if(err)+returnerr;++/* Internal transport error when checking for available+*data.XXXThisshouldbechangedtoaconnection+*resetinalaterchange.+*/+if(data<0)+return-ENOMEM;++returndata;+}+staticintvsock_connectible_recvmsg(structsocket*sock,structmsghdr*msg,size_tlen,intflags)
Some code in receive data loop could be shared between SEQPACKET
and STREAM sockets, while another part is type specific, so move STREAM
specific data receive logic to '__vsock_stream_recvmsg()' dedicated
function, while checks, that will be same for both STREAM and SEQPACKET
sockets, stays in 'vsock_connectible_recvmsg()'.
Signed-off-by: Arseny Krasnov <redacted>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
---
net/vmw_vsock/af_vsock.c | 116 ++++++++++++++++++++++-----------------
1 file changed, 67 insertions(+), 49 deletions(-)
@@ -1896,65 +1896,22 @@ static int vsock_wait_data(struct sock *sk, struct wait_queue_entry *wait,returndata;}-staticint-vsock_connectible_recvmsg(structsocket*sock,structmsghdr*msg,size_tlen,-intflags)+staticint__vsock_stream_recvmsg(structsock*sk,structmsghdr*msg,+size_tlen,intflags){-structsock*sk;-structvsock_sock*vsk;+structvsock_transport_recv_notify_datarecv_data;conststructvsock_transport*transport;-interr;-size_ttarget;+structvsock_sock*vsk;ssize_tcopied;+size_ttarget;longtimeout;-structvsock_transport_recv_notify_datarecv_data;+interr;DEFINE_WAIT(wait);-sk=sock->sk;vsk=vsock_sk(sk);-err=0;--lock_sock(sk);-transport=vsk->transport;-if(!transport||sk->sk_state!=TCP_ESTABLISHED){-/* Recvmsg is supposed to return 0 if a peer performs an-*orderlyshutdown.Differentiatebetweenthatcaseandwhena-*peerhasnotconnectedoralocalshutdownoccurredwiththe-*SOCK_DONEflag.-*/-if(sock_flag(sk,SOCK_DONE))-err=0;-else-err=-ENOTCONN;--gotoout;-}--if(flags&MSG_OOB){-err=-EOPNOTSUPP;-gotoout;-}--/* We don't check peer_shutdown flag here since peer may actually shut-*down,buttherecanbedatainthequeuethatalocalsocketcan-*receive.-*/-if(sk->sk_shutdown&RCV_SHUTDOWN){-err=0;-gotoout;-}--/* It is valid on Linux to pass in a zero-length receive buffer. This-*isnotanerror.Wemayaswellbailoutnow.-*/-if(!len){-err=0;-gotoout;-}-/* We must not copy less than target bytes into the user's buffer*beforereturningsuccessfully,sowewaitfortheconsumequeueto*havethatmuchdatatoconsumebeforedequeueing.Notethatthis
@@ -2013,6 +1970,67 @@ vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,if(copied>0)err=copied;+out:+returnerr;+}++staticint+vsock_connectible_recvmsg(structsocket*sock,structmsghdr*msg,size_tlen,+intflags)+{+structsock*sk;+structvsock_sock*vsk;+conststructvsock_transport*transport;+interr;++DEFINE_WAIT(wait);++sk=sock->sk;+vsk=vsock_sk(sk);+err=0;++lock_sock(sk);++transport=vsk->transport;++if(!transport||sk->sk_state!=TCP_ESTABLISHED){+/* Recvmsg is supposed to return 0 if a peer performs an+*orderlyshutdown.Differentiatebetweenthatcaseandwhena+*peerhasnotconnectedoralocalshutdownoccurredwiththe+*SOCK_DONEflag.+*/+if(sock_flag(sk,SOCK_DONE))+err=0;+else+err=-ENOTCONN;++gotoout;+}++if(flags&MSG_OOB){+err=-EOPNOTSUPP;+gotoout;+}++/* We don't check peer_shutdown flag here since peer may actually shut+*down,buttherecanbedatainthequeuethatalocalsocketcan+*receive.+*/+if(sk->sk_shutdown&RCV_SHUTDOWN){+err=0;+gotoout;+}++/* It is valid on Linux to pass in a zero-length receive buffer. This+*isnotanerror.Wemayaswellbailoutnow.+*/+if(!len){+err=0;+gotoout;+}++err=__vsock_stream_recvmsg(sk,msg,len,flags);+out:release_sock(sk);returnerr;
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
@@ -1974,6 +1974,73 @@ static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,returnerr;}+staticint__vsock_seqpacket_recvmsg(structsock*sk,structmsghdr*msg,+size_tlen,intflags)+{+conststructvsock_transport*transport;+boolmsg_ready;+structvsock_sock*vsk;+ssize_trecord_len;+longtimeout;+interr=0;+DEFINE_WAIT(wait);++vsk=vsock_sk(sk);+transport=vsk->transport;++timeout=sock_rcvtimeo(sk,flags&MSG_DONTWAIT);+msg_ready=false;+record_len=0;++while(1){+ssize_tfragment_len;++if(vsock_wait_data(sk,&wait,timeout,NULL,0)<=0){+/* In case of any loop break(timeout, signal+*interruptorshutdown),wereportuserthat+*nothingwascopied.+*/+err=0;+break;+}++fragment_len=transport->seqpacket_dequeue(vsk,msg,flags,&msg_ready);++if(fragment_len<0){+err=-ENOMEM;+break;+}++record_len+=fragment_len;++if(msg_ready)+break;+}++if(sk->sk_err)+err=-sk->sk_err;+elseif(sk->sk_shutdown&RCV_SHUTDOWN)+err=0;++if(msg_ready&&err==0){+/* User sets MSG_TRUNC, so return real length of+*packet.+*/+if(flags&MSG_TRUNC)+err=record_len;+else+err=len-msg_data_left(msg);++/* Always set MSG_TRUNC if real length of packet is+*biggerthanuser'sbuffer.+*/+if(record_len>len)+msg->msg_flags|=MSG_TRUNC;+}++returnerr;+}+staticintvsock_connectible_recvmsg(structsocket*sock,structmsghdr*msg,size_tlen,intflags)
@@ -1826,8 +1830,14 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,}out_err:-if(total_written>0)-err=total_written;+if(total_written>0){+/* Return number of written bytes only if:+*1)SOCK_STREAMsocket.+*2)SOCK_SEQPACKETsocketwhenwholebufferissent.+*/+if(sk->sk_type==SOCK_STREAM||total_written==len)+err=total_written;+}out:release_sock(sk);returnerr;
To make SEQPACKET socket functional, socket ops was added
for SEQPACKET type and such type of socket was allowed
to create.
Signed-off-by: Arseny Krasnov <redacted>
---
include/net/af_vsock.h | 1 +
net/vmw_vsock/af_vsock.c | 36 +++++++++++++++++++++++++++++++++++-
2 files changed, 36 insertions(+), 1 deletion(-)
@@ -415,8 +415,8 @@ static void vsock_deassign_transport(struct vsock_sock *vsk)/* Assign a transport to a socket and call the .init transport callback.*-*Note:forstreamsocketthismustbecalledwhenvsk->remote_addrisset-*(e.g.duringtheconnect()orwhenaconnectionrequestonalistener+*Note:forconnectionorientedsocketthismustbecalledwhenvsk->remote_addr+*isset(e.g.duringtheconnect()orwhenaconnectionrequestonalistener*socketisreceived).*Thevsk->remote_addrisusedtodecidewhichtransporttouse:*-remoteCID==VMADDR_CID_LOCALorg2h->local_cidorVMADDR_CID_HOSTif
@@ -470,10 +470,10 @@ int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)return0;/* transport->release() must be called with sock lock acquired.-*Thispathcanonlybetakenduringvsock_stream_connect(),-*wherewehavealreadyheldthesocklock.-*Intheothercases,thisfunctioniscalledonanewsocket-*whichisnotassignedtoanytransport.+*Thispathcanonlybetakenduringvsock_connect(),wherewe+*havealreadyheldthesocklock.Intheothercases,this+*functioniscalledonanewsocketwhichisnotassignedto+*anytransport.*/vsk->transport->release(vsk);vsock_deassign_transport(vsk);
@@ -658,9 +658,10 @@ static int __vsock_bind_connectible(struct vsock_sock *vsk,vsock_addr_init(&vsk->local_addr,new_addr.svm_cid,new_addr.svm_port);-/* Remove stream sockets from the unbound list and add them to the hash-*tableforeasylookupbyitsaddress.Theunboundlistissimplyan-*extraentryattheendofthehashtable,atrickusedbyAF_UNIX.+/* Remove connection oriented sockets from the unbound list and add them+*tothehashtableforeasylookupbyitsaddress.Theunboundlist+*issimplyanextraentryattheendofthehashtable,atrickused+*byAF_UNIX.*/__vsock_remove_bound(vsk);__vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr),vsk);
@@ -952,10 +953,10 @@ static int vsock_shutdown(struct socket *sock, int mode)if((mode&~SHUTDOWN_MASK)||!mode)return-EINVAL;-/* If this is a STREAM socket and it is not connected then bail out-*immediately.IfitisaDGRAMsocketthenwemustfirstkickthe-*socketsothatitwakesupfromanysleepingcalls,forexample-*recv(),andthenafterwardsreturntheerror.+/* If this is a connection oriented socket and it is not connected then+*bailoutimmediately.IfitisaDGRAMsocketthenwemustfirst+*kickthesocketsothatitwakesupfromanysleepingcalls,for+*examplerecv(),andthenafterwardsreturntheerror.*/sk=sock->sk;
@@ -1727,7 +1728,9 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,transport=vsk->transport;-/* Callers should not provide a destination with stream sockets. */+/* Callers should not provide a destination with connection oriented+*sockets.+*/if(msg->msg_namelen){err=sk->sk_state==TCP_ESTABLISHED?-EISCONN:-EOPNOTSUPP;gotoout;
There is no need to set type of packet which differs from type
of socket, so move passing type of packet from 'info' structure
to 'virtio_transport_send_pkt_info()' function. Since at current
time only stream type is supported, set it directly in 'virtio_
transport_send_pkt_info()', so callers don't need to set it.
Signed-off-by: Arseny Krasnov <redacted>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
---
net/vmw_vsock/virtio_transport_common.c | 19 +++++--------------
1 file changed, 5 insertions(+), 14 deletions(-)
@@ -624,7 +620,6 @@ int virtio_transport_connect(struct vsock_sock *vsk){structvirtio_vsock_pkt_infoinfo={.op=VIRTIO_VSOCK_OP_REQUEST,-.type=VIRTIO_VSOCK_TYPE_STREAM,.vsk=vsk,};
@@ -636,7 +631,6 @@ int virtio_transport_shutdown(struct vsock_sock *vsk, int mode){structvirtio_vsock_pkt_infoinfo={.op=VIRTIO_VSOCK_OP_SHUTDOWN,-.type=VIRTIO_VSOCK_TYPE_STREAM,.flags=(mode&RCV_SHUTDOWN?VIRTIO_VSOCK_SHUTDOWN_RCV:0)|(mode&SEND_SHUTDOWN?
Add set of defines and constants for SOCK_SEQPACKET support
in vsock.
Signed-off-by: Arseny Krasnov <redacted>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
---
include/uapi/linux/virtio_vsock.h | 9 +++++++++
1 file changed, 9 insertions(+)
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
@@ -393,6 +393,59 @@ virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,returnerr;}+staticintvirtio_transport_seqpacket_do_dequeue(structvsock_sock*vsk,+structmsghdr*msg,+intflags,+bool*msg_ready)+{+structvirtio_vsock_sock*vvs=vsk->trans;+structvirtio_vsock_pkt*pkt;+intdequeued_len=0;+size_tuser_buf_len=msg_data_left(msg);++*msg_ready=false;+spin_lock_bh(&vvs->rx_lock);++while(!*msg_ready&&!list_empty(&vvs->rx_queue)&&dequeued_len>=0){+size_tbytes_to_copy;+size_tpkt_len;++pkt=list_first_entry(&vvs->rx_queue,structvirtio_vsock_pkt,list);+pkt_len=(size_t)le32_to_cpu(pkt->hdr.len);+bytes_to_copy=min(user_buf_len,pkt_len);++if(bytes_to_copy){+/* sk_lock is held by caller so no one else can dequeue.+*Unlockrx_locksincememcpy_to_msg()maysleep.+*/+spin_unlock_bh(&vvs->rx_lock);++if(memcpy_to_msg(msg,pkt->buf,bytes_to_copy))+dequeued_len=-EINVAL;+else+user_buf_len-=bytes_to_copy;++spin_lock_bh(&vvs->rx_lock);+}++if(dequeued_len>=0)+dequeued_len+=pkt_len;++if(le32_to_cpu(pkt->hdr.flags)&VIRTIO_VSOCK_SEQ_EOR)+*msg_ready=true;++virtio_transport_dec_rx_pkt(vvs,pkt);+list_del(&pkt->list);+virtio_transport_free_pkt(pkt);+}++spin_unlock_bh(&vvs->rx_lock);++virtio_transport_send_credit_update(vsk);++returndequeued_len;+}+ssize_tvirtio_transport_stream_dequeue(structvsock_sock*vsk,structmsghdr*msg,
@@ -165,6 +165,14 @@ void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)}EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);+staticu16virtio_transport_get_type(structsock*sk)+{+if(sk->sk_type==SOCK_STREAM)+returnVIRTIO_VSOCK_TYPE_STREAM;+else+returnVIRTIO_VSOCK_TYPE_SEQPACKET;+}+/* This function can only be used on connecting/connected sockets,*sinceasocketassignedtoatransportisrequired.*
@@ -979,13 +987,17 @@ virtio_transport_recv_enqueue(struct vsock_sock *vsk,structvirtio_vsock_pkt,list);/* If there is space in the last packet queued, we copy the-*newpacketinitsbuffer.+*newpacketinitsbuffer(exceptSEQPACKETcase,whenwe+*alsocheckthatlastpacketisnotlastpacketofprevious+*record).*/-if(pkt->len<=last_pkt->buf_len-last_pkt->len){+if((pkt->len<=last_pkt->buf_len-last_pkt->len)&&+!(le32_to_cpu(last_pkt->hdr.flags)&VIRTIO_VSOCK_SEQ_EOR)){memcpy(last_pkt->buf+last_pkt->len,pkt->buf,pkt->len);last_pkt->len+=pkt->len;free_pkt=true;+last_pkt->hdr.flags|=pkt->hdr.flags;gotoout;}}
@@ -1151,6 +1163,12 @@ virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt,return0;}+staticboolvirtio_transport_valid_type(u16type)+{+return(type==VIRTIO_VSOCK_TYPE_STREAM)||+(type==VIRTIO_VSOCK_TYPE_SEQPACKET);+}+/* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex*lock.*/
Small updates to make SOCK_SEQPACKET work:
1) Send SHUTDOWN on socket close for SEQPACKET type.
2) Set SEQPACKET packet type during send.
3) Set 'VIRTIO_VSOCK_SEQ_EOR' bit in flags for last
packet of message.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
2) Commit message updated.
3) Add check for socket type when setting SEQ_EOR bit.
include/linux/virtio_vsock.h | 4 ++++
net/vmw_vsock/virtio_transport_common.c | 18 ++++++++++++++++--
2 files changed, 20 insertions(+), 2 deletions(-)
To make transport work with SOCK_SEQPACKET two updates were
added:
1) SOCK_SEQPACKET ops for virtio transport and 'seqpacket_allow()'
callback.
2) Handling of SEQPACKET bit: guest tries to negotiate it with vhost.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'virtio_has_feature()' to check feature bit.
2) Move assignment to 'seqpacket_allow' before 'rcu_assign_pointer()'.
net/vmw_vsock/virtio_transport.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
As vhost places data in buffers of guest's rx queue, keep SEQ_EOR
bit set only when last piece of data is copied. Otherwise we get
sequence packets for one socket in guest's rx queue with SEQ_EOR bit
set. Also remove ignore of non-stream type of packets, handle SEQPACKET
feature bit.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Move 'restore_flag' handling to 'payload_len' calculation
block.
drivers/vhost/vsock.c | 44 +++++++++++++++++++++++++++++++++++++++----
1 file changed, 40 insertions(+), 4 deletions(-)
@@ -168,9 +171,15 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock,/* If the packet is greater than the space available in the*buffer,wesplititusingmultiplebuffers.*/-if(payload_len>iov_len-sizeof(pkt->hdr))+if(payload_len>iov_len-sizeof(pkt->hdr)){payload_len=iov_len-sizeof(pkt->hdr);+if(le32_to_cpu(pkt->hdr.flags)&VIRTIO_VSOCK_SEQ_EOR){+pkt->hdr.flags&=~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);+restore_flag=true;+}+}+/* Set the correct length in the header */pkt->hdr.len=cpu_to_le32(payload_len);
Implement two tests of SOCK_SEQPACKET socket: first sends data by
several 'write()'s and checks that number of 'read()' were same.
Second test checks MSG_TRUNC flag. Cases for connect(), bind(),
etc. are not tested, because it is same as for stream socket.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Commit message updated.
2) Add second test for message bounds.
tools/testing/vsock/util.c | 32 +++++++--
tools/testing/vsock/util.h | 3 +
tools/testing/vsock/vsock_test.c | 116 +++++++++++++++++++++++++++++++
3 files changed, 146 insertions(+), 5 deletions(-)
@@ -84,7 +84,7 @@ void vsock_wait_remote_close(int fd)}/* Connect to <cid, port> and return the file descriptor. */-intvsock_stream_connect(unsignedintcid,unsignedintport)+staticintvsock_connect(unsignedintcid,unsignedintport,inttype){union{structsockaddrsa;
@@ -101,7 +101,7 @@ int vsock_stream_connect(unsigned int cid, unsigned int port)control_expectln("LISTENING");-fd=socket(AF_VSOCK,SOCK_STREAM,0);+fd=socket(AF_VSOCK,type,0);timeout_begin(TIMEOUT);do{
@@ -120,11 +120,21 @@ int vsock_stream_connect(unsigned int cid, unsigned int port)returnfd;}+intvsock_stream_connect(unsignedintcid,unsignedintport)+{+returnvsock_connect(cid,port,SOCK_STREAM);+}++intvsock_seqpacket_connect(unsignedintcid,unsignedintport)+{+returnvsock_connect(cid,port,SOCK_SEQPACKET);+}+/* Listen on <cid, port> and return the first incoming connection. The remote*addressisstoredtoclientaddrp.clientaddrpmaybeNULL.*/-intvsock_stream_accept(unsignedintcid,unsignedintport,-structsockaddr_vm*clientaddrp)+staticintvsock_accept(unsignedintcid,unsignedintport,+structsockaddr_vm*clientaddrp,inttype){union{structsockaddrsa;
@@ -145,7 +155,7 @@ int vsock_stream_accept(unsigned int cid, unsigned int port,intclient_fd;intold_errno;-fd=socket(AF_VSOCK,SOCK_STREAM,0);+fd=socket(AF_VSOCK,type,0);if(bind(fd,&addr.sa,sizeof(addr.svm))<0){perror("bind");
@@ -189,6 +199,18 @@ int vsock_stream_accept(unsigned int cid, unsigned int port,returnclient_fd;}+intvsock_stream_accept(unsignedintcid,unsignedintport,+structsockaddr_vm*clientaddrp)+{+returnvsock_accept(cid,port,clientaddrp,SOCK_STREAM);+}++intvsock_seqpacket_accept(unsignedintcid,unsignedintport,+structsockaddr_vm*clientaddrp)+{+returnvsock_accept(cid,port,clientaddrp,SOCK_SEQPACKET);+}+/* Transmit one byte and check the return value.**expected_ret:
Hi Arseny,
On Thu, May 20, 2021 at 10:13:53PM +0300, Arseny Krasnov wrote:
This patchset implements support of SOCK_SEQPACKET for virtio
transport.
I'll carefully review and test this series next Monday, in the mean time
I think we should have at least an agreement about the changes that
regards virtio-spec before merge this series, to avoid any compatibility
issues.
Do you plan to send a new version of the specification changes?
Thanks,
Stefano
As SOCK_SEQPACKET guarantees to save record boundaries, so to
do it, new bit for field 'flags' was added: SEQ_EOR. This bit is
set to 1 in last RW packet of message.
Now as packets of one socket are not reordered neither on vsock
nor on vhost transport layers, such bit allows to restore original
message on receiver's side. If user's buffer is smaller than message
length, when all out of size data is dropped.
Maximum length of datagram is not limited as in stream socket,
because same credit logic is used. Difference with stream socket is
that user is not woken up until whole record is received or error
occurred. Implementation also supports 'MSG_TRUNC' flags.
Tests also implemented.
Thanks to stsp2@yandex.ru for encouragements and initial design
recommendations.
Arseny Krasnov (18):
af_vsock: update functions for connectible socket
af_vsock: separate wait data loop
af_vsock: separate receive data loop
af_vsock: implement SEQPACKET receive loop
af_vsock: implement send logic for SEQPACKET
af_vsock: rest of SEQPACKET support
af_vsock: update comments for stream sockets
virtio/vsock: set packet's type in virtio_transport_send_pkt_info()
virtio/vsock: simplify credit update function API
virtio/vsock: defines and constants for SEQPACKET
virtio/vsock: dequeue callback for SOCK_SEQPACKET
virtio/vsock: add SEQPACKET receive logic
virtio/vsock: rest of SOCK_SEQPACKET support
virtio/vsock: enable SEQPACKET for transport
vhost/vsock: enable SEQPACKET for transport
vsock/loopback: enable SEQPACKET for transport
vsock_test: add SOCK_SEQPACKET tests
virtio/vsock: update trace event for SEQPACKET
drivers/vhost/vsock.c | 44 +-
include/linux/virtio_vsock.h | 9 +
include/net/af_vsock.h | 7 +
.../events/vsock_virtio_transport_common.h | 5 +-
include/uapi/linux/virtio_vsock.h | 9 +
net/vmw_vsock/af_vsock.c | 465 +++++++++++------
net/vmw_vsock/virtio_transport.c | 25 +
net/vmw_vsock/virtio_transport_common.c | 133 ++++-
net/vmw_vsock/vsock_loopback.c | 11 +
tools/testing/vsock/util.c | 32 +-
tools/testing/vsock/util.h | 3 +
tools/testing/vsock/vsock_test.c | 116 ++++
12 files changed, 672 insertions(+), 187 deletions(-)
v9 -> v10:
General changelog:
- patch for write serialization removed from patchset
- commit messages rephrased
- RFC tag removed
Per patch changelog:
see every patch after '---' line.
v8 -> v9:
General changelog:
- see per patch change log.
Per patch changelog:
see every patch after '---' line.
v7 -> v8:
General changelog:
- whole idea is simplified: channel now considered reliable,
so SEQ_BEGIN, SEQ_END, 'msg_len' and 'msg_id' were removed.
Only thing that is used to mark end of message is bit in
'flags' field of packet header: VIRTIO_VSOCK_SEQ_EOR. Packet
with such bit set to 1 means, that this is last packet of
message.
- POSIX MSG_EOR support is removed, as there is no exact
description how it works.
- all changes to 'include/uapi/linux/virtio_vsock.h' moved
to dedicated patch, as these changes linked with patch to
spec.
- patch 'virtio/vsock: SEQPACKET feature bit support' now merged
to 'virtio/vsock: setup SEQPACKET ops for transport'.
- patch 'vhost/vsock: SEQPACKET feature bit support' now merged
to 'vhost/vsock: setup SEQPACKET ops for transport'.
Per patch changelog:
see every patch after '---' line.
v6 -> v7:
General changelog:
- virtio transport callback for message length now removed
from transport. Length of record is returned by dequeue
callback.
- function which tries to get message length now returns 0
when rx queue is empty. Also length of current message in
progress is set to 0, when message processed or error
happens.
- patches for virtio feature bit moved after patches with
transport ops.
Per patch changelog:
see every patch after '---' line.
v5 -> v6:
General changelog:
- virtio transport specific callbacks which send SEQ_BEGIN or
SEQ_END now hidden inside virtio transport. Only enqueue,
dequeue and record length callbacks are provided by transport.
- virtio feature bit for SEQPACKET socket support introduced:
VIRTIO_VSOCK_F_SEQPACKET.
- 'msg_cnt' field in 'struct virtio_vsock_seq_hdr' renamed to
'msg_id' and used as id.
Per patch changelog:
- 'af_vsock: separate wait data loop':
1) Commit message updated.
2) 'prepare_to_wait()' moved inside while loop(thanks to
Jorgen Hansen).
Marked 'Reviewed-by' with 1), but as 2) I removed R-b.
- 'af_vsock: separate receive data loop': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'af_vsock: implement SEQPACKET receive loop': style fixes.
- 'af_vsock: rest of SEQPACKET support':
1) 'module_put()' added when transport callback check failed.
2) Now only 'seqpacket_allow()' callback called to check
support of SEQPACKET by transport.
- 'af_vsock: update comments for stream sockets': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'virtio/vsock: set packet's type in send':
1) Commit message updated.
2) Parameter 'type' from 'virtio_transport_send_credit_update()'
also removed in this patch instead of in next.
- 'virtio/vsock: dequeue callback for SOCK_SEQPACKET': SEQPACKET
related state wrapped to special struct.
- 'virtio/vsock: update trace event for SEQPACKET': format strings
now not broken by new lines.
v4 -> v5:
- patches reorganized:
1) Setting of packet's type in 'virtio_transport_send_pkt_info()'
is moved to separate patch.
2) Simplifying of 'virtio_transport_send_credit_update()' is
moved to separate patch and before main virtio/vsock patches.
- style problem fixed
- in 'af_vsock: separate receive data loop' extra 'release_sock()'
removed
- added trace event fields for SEQPACKET
- in 'af_vsock: separate wait data loop':
1) 'vsock_wait_data()' removed 'goto out;'
2) Comment for invalid data amount is changed.
- in 'af_vsock: rest of SEQPACKET support', 'new_transport' pointer
check is moved after 'try_module_get()'
- in 'af_vsock: update comments for stream sockets', 'connect-oriented'
replaced with 'connection-oriented'
- in 'loopback/vsock: setup SEQPACKET ops for transport',
'loopback/vsock' replaced with 'vsock/loopback'
v3 -> v4:
- SEQPACKET specific metadata moved from packet header to payload
and called 'virtio_vsock_seq_hdr'
- record integrity check:
1) SEQ_END operation was added, which marks end of record.
2) Both SEQ_BEGIN and SEQ_END carries counter which is incremented
on every marker send.
- af_vsock.c: socket operations for STREAM and SEQPACKET call same
functions instead of having own "gates" differs only by names:
'vsock_seqpacket/stream_getsockopt()' now replaced with
'vsock_connectible_getsockopt()'.
- af_vsock.c: 'seqpacket_dequeue' callback returns error and flag that
record ready. There is no need to return number of copied bytes,
because case when record received successfully is checked at virtio
transport layer, when SEQ_END is processed. Also user doesn't need
number of copied bytes, because 'recv()' from SEQPACKET could return
error, length of users's buffer or length of whole record(both are
known in af_vsock.c).
- af_vsock.c: both wait loops in af_vsock.c(for data and space) moved
to separate functions because now both called from several places.
- af_vsock.c: 'vsock_assign_transport()' checks that 'new_transport'
pointer is not NULL and returns 'ESOCKTNOSUPPORT' instead of 'ENODEV'
if failed to use transport.
- tools/testing/vsock/vsock_test.c: rename tests
v2 -> v3:
- patches reorganized: split for prepare and implementation patches
- local variables are declared in "Reverse Christmas tree" manner
- virtio_transport_common.c: valid leXX_to_cpu() for vsock header
fields access
- af_vsock.c: 'vsock_connectible_*sockopt()' added as shared code
between stream and seqpacket sockets.
- af_vsock.c: loops in '__vsock_*_recvmsg()' refactored.
- af_vsock.c: 'vsock_wait_data()' refactored.
v1 -> v2:
- patches reordered: af_vsock.c related changes now before virtio vsock
- patches reorganized: more small patches, where +/- are not mixed
- tests for SOCK_SEQPACKET added
- all commit messages updated
- af_vsock.c: 'vsock_pre_recv_check()' inlined to
'vsock_connectible_recvmsg()'
- af_vsock.c: 'vsock_assign_transport()' returns ENODEV if transport
was not found
- virtio_transport_common.c: transport callback for seqpacket dequeue
- virtio_transport_common.c: simplified
'virtio_transport_recv_connected()'
- virtio_transport_common.c: send reset on socket and packet type
mismatch.
Signed-off-by: Arseny Krasnov <redacted>
--
2.25.1
Hi Arseny,
On Thu, May 20, 2021 at 10:13:53PM +0300, Arseny Krasnov wrote:
quoted
This patchset implements support of SOCK_SEQPACKET for virtio
transport.
I'll carefully review and test this series next Monday, in the mean time
I think we should have at least an agreement about the changes that
regards virtio-spec before merge this series, to avoid any compatibility
issues.
Do you plan to send a new version of the specification changes?
Thanks,
Stefano
Hello, sorry for long answer. I'm on vacation now, but i plan to send
it in next several days, because with current implementation it is short
Thank You
quoted
As SOCK_SEQPACKET guarantees to save record boundaries, so to
do it, new bit for field 'flags' was added: SEQ_EOR. This bit is
set to 1 in last RW packet of message.
Now as packets of one socket are not reordered neither on vsock
nor on vhost transport layers, such bit allows to restore original
message on receiver's side. If user's buffer is smaller than message
length, when all out of size data is dropped.
Maximum length of datagram is not limited as in stream socket,
because same credit logic is used. Difference with stream socket is
that user is not woken up until whole record is received or error
occurred. Implementation also supports 'MSG_TRUNC' flags.
Tests also implemented.
Thanks to stsp2@yandex.ru for encouragements and initial design
recommendations.
Arseny Krasnov (18):
af_vsock: update functions for connectible socket
af_vsock: separate wait data loop
af_vsock: separate receive data loop
af_vsock: implement SEQPACKET receive loop
af_vsock: implement send logic for SEQPACKET
af_vsock: rest of SEQPACKET support
af_vsock: update comments for stream sockets
virtio/vsock: set packet's type in virtio_transport_send_pkt_info()
virtio/vsock: simplify credit update function API
virtio/vsock: defines and constants for SEQPACKET
virtio/vsock: dequeue callback for SOCK_SEQPACKET
virtio/vsock: add SEQPACKET receive logic
virtio/vsock: rest of SOCK_SEQPACKET support
virtio/vsock: enable SEQPACKET for transport
vhost/vsock: enable SEQPACKET for transport
vsock/loopback: enable SEQPACKET for transport
vsock_test: add SOCK_SEQPACKET tests
virtio/vsock: update trace event for SEQPACKET
drivers/vhost/vsock.c | 44 +-
include/linux/virtio_vsock.h | 9 +
include/net/af_vsock.h | 7 +
.../events/vsock_virtio_transport_common.h | 5 +-
include/uapi/linux/virtio_vsock.h | 9 +
net/vmw_vsock/af_vsock.c | 465 +++++++++++------
net/vmw_vsock/virtio_transport.c | 25 +
net/vmw_vsock/virtio_transport_common.c | 133 ++++-
net/vmw_vsock/vsock_loopback.c | 11 +
tools/testing/vsock/util.c | 32 +-
tools/testing/vsock/util.h | 3 +
tools/testing/vsock/vsock_test.c | 116 ++++
12 files changed, 672 insertions(+), 187 deletions(-)
v9 -> v10:
General changelog:
- patch for write serialization removed from patchset
- commit messages rephrased
- RFC tag removed
Per patch changelog:
see every patch after '---' line.
v8 -> v9:
General changelog:
- see per patch change log.
Per patch changelog:
see every patch after '---' line.
v7 -> v8:
General changelog:
- whole idea is simplified: channel now considered reliable,
so SEQ_BEGIN, SEQ_END, 'msg_len' and 'msg_id' were removed.
Only thing that is used to mark end of message is bit in
'flags' field of packet header: VIRTIO_VSOCK_SEQ_EOR. Packet
with such bit set to 1 means, that this is last packet of
message.
- POSIX MSG_EOR support is removed, as there is no exact
description how it works.
- all changes to 'include/uapi/linux/virtio_vsock.h' moved
to dedicated patch, as these changes linked with patch to
spec.
- patch 'virtio/vsock: SEQPACKET feature bit support' now merged
to 'virtio/vsock: setup SEQPACKET ops for transport'.
- patch 'vhost/vsock: SEQPACKET feature bit support' now merged
to 'vhost/vsock: setup SEQPACKET ops for transport'.
Per patch changelog:
see every patch after '---' line.
v6 -> v7:
General changelog:
- virtio transport callback for message length now removed
from transport. Length of record is returned by dequeue
callback.
- function which tries to get message length now returns 0
when rx queue is empty. Also length of current message in
progress is set to 0, when message processed or error
happens.
- patches for virtio feature bit moved after patches with
transport ops.
Per patch changelog:
see every patch after '---' line.
v5 -> v6:
General changelog:
- virtio transport specific callbacks which send SEQ_BEGIN or
SEQ_END now hidden inside virtio transport. Only enqueue,
dequeue and record length callbacks are provided by transport.
- virtio feature bit for SEQPACKET socket support introduced:
VIRTIO_VSOCK_F_SEQPACKET.
- 'msg_cnt' field in 'struct virtio_vsock_seq_hdr' renamed to
'msg_id' and used as id.
Per patch changelog:
- 'af_vsock: separate wait data loop':
1) Commit message updated.
2) 'prepare_to_wait()' moved inside while loop(thanks to
Jorgen Hansen).
Marked 'Reviewed-by' with 1), but as 2) I removed R-b.
- 'af_vsock: separate receive data loop': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'af_vsock: implement SEQPACKET receive loop': style fixes.
- 'af_vsock: rest of SEQPACKET support':
1) 'module_put()' added when transport callback check failed.
2) Now only 'seqpacket_allow()' callback called to check
support of SEQPACKET by transport.
- 'af_vsock: update comments for stream sockets': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'virtio/vsock: set packet's type in send':
1) Commit message updated.
2) Parameter 'type' from 'virtio_transport_send_credit_update()'
also removed in this patch instead of in next.
- 'virtio/vsock: dequeue callback for SOCK_SEQPACKET': SEQPACKET
related state wrapped to special struct.
- 'virtio/vsock: update trace event for SEQPACKET': format strings
now not broken by new lines.
v4 -> v5:
- patches reorganized:
1) Setting of packet's type in 'virtio_transport_send_pkt_info()'
is moved to separate patch.
2) Simplifying of 'virtio_transport_send_credit_update()' is
moved to separate patch and before main virtio/vsock patches.
- style problem fixed
- in 'af_vsock: separate receive data loop' extra 'release_sock()'
removed
- added trace event fields for SEQPACKET
- in 'af_vsock: separate wait data loop':
1) 'vsock_wait_data()' removed 'goto out;'
2) Comment for invalid data amount is changed.
- in 'af_vsock: rest of SEQPACKET support', 'new_transport' pointer
check is moved after 'try_module_get()'
- in 'af_vsock: update comments for stream sockets', 'connect-oriented'
replaced with 'connection-oriented'
- in 'loopback/vsock: setup SEQPACKET ops for transport',
'loopback/vsock' replaced with 'vsock/loopback'
v3 -> v4:
- SEQPACKET specific metadata moved from packet header to payload
and called 'virtio_vsock_seq_hdr'
- record integrity check:
1) SEQ_END operation was added, which marks end of record.
2) Both SEQ_BEGIN and SEQ_END carries counter which is incremented
on every marker send.
- af_vsock.c: socket operations for STREAM and SEQPACKET call same
functions instead of having own "gates" differs only by names:
'vsock_seqpacket/stream_getsockopt()' now replaced with
'vsock_connectible_getsockopt()'.
- af_vsock.c: 'seqpacket_dequeue' callback returns error and flag that
record ready. There is no need to return number of copied bytes,
because case when record received successfully is checked at virtio
transport layer, when SEQ_END is processed. Also user doesn't need
number of copied bytes, because 'recv()' from SEQPACKET could return
error, length of users's buffer or length of whole record(both are
known in af_vsock.c).
- af_vsock.c: both wait loops in af_vsock.c(for data and space) moved
to separate functions because now both called from several places.
- af_vsock.c: 'vsock_assign_transport()' checks that 'new_transport'
pointer is not NULL and returns 'ESOCKTNOSUPPORT' instead of 'ENODEV'
if failed to use transport.
- tools/testing/vsock/vsock_test.c: rename tests
v2 -> v3:
- patches reorganized: split for prepare and implementation patches
- local variables are declared in "Reverse Christmas tree" manner
- virtio_transport_common.c: valid leXX_to_cpu() for vsock header
fields access
- af_vsock.c: 'vsock_connectible_*sockopt()' added as shared code
between stream and seqpacket sockets.
- af_vsock.c: loops in '__vsock_*_recvmsg()' refactored.
- af_vsock.c: 'vsock_wait_data()' refactored.
v1 -> v2:
- patches reordered: af_vsock.c related changes now before virtio vsock
- patches reorganized: more small patches, where +/- are not mixed
- tests for SOCK_SEQPACKET added
- all commit messages updated
- af_vsock.c: 'vsock_pre_recv_check()' inlined to
'vsock_connectible_recvmsg()'
- af_vsock.c: 'vsock_assign_transport()' returns ENODEV if transport
was not found
- virtio_transport_common.c: transport callback for seqpacket dequeue
- virtio_transport_common.c: simplified
'virtio_transport_recv_connected()'
- virtio_transport_common.c: send reset on socket and packet type
mismatch.
Signed-off-by: Arseny Krasnov <redacted>
--
2.25.1
Hi Arseny,
On Thu, May 20, 2021 at 10:13:53PM +0300, Arseny Krasnov wrote:
quoted
This patchset implements support of SOCK_SEQPACKET for virtio
transport.
I'll carefully review and test this series next Monday, in the mean time
I think we should have at least an agreement about the changes that
regards virtio-spec before merge this series, to avoid any compatibility
issues.
Do you plan to send a new version of the specification changes?
Thanks,
Stefano
Hello, sorry for long answer. I'm on vacation now, but i plan to send
it in next several days, because with current implementation it is short
Thank You
As SOCK_SEQPACKET guarantees to save record boundaries, so to
do it, new bit for field 'flags' was added: SEQ_EOR. This bit is
set to 1 in last RW packet of message.
Now as packets of one socket are not reordered neither on vsock
nor on vhost transport layers, such bit allows to restore original
message on receiver's side. If user's buffer is smaller than message
length, when all out of size data is dropped.
Maximum length of datagram is not limited as in stream socket,
because same credit logic is used. Difference with stream socket is
that user is not woken up until whole record is received or error
occurred. Implementation also supports 'MSG_TRUNC' flags.
Tests also implemented.
Thanks to stsp2@yandex.ru for encouragements and initial design
recommendations.
Arseny Krasnov (18):
af_vsock: update functions for connectible socket
af_vsock: separate wait data loop
af_vsock: separate receive data loop
af_vsock: implement SEQPACKET receive loop
af_vsock: implement send logic for SEQPACKET
af_vsock: rest of SEQPACKET support
af_vsock: update comments for stream sockets
virtio/vsock: set packet's type in virtio_transport_send_pkt_info()
virtio/vsock: simplify credit update function API
virtio/vsock: defines and constants for SEQPACKET
virtio/vsock: dequeue callback for SOCK_SEQPACKET
virtio/vsock: add SEQPACKET receive logic
virtio/vsock: rest of SOCK_SEQPACKET support
virtio/vsock: enable SEQPACKET for transport
vhost/vsock: enable SEQPACKET for transport
vsock/loopback: enable SEQPACKET for transport
vsock_test: add SOCK_SEQPACKET tests
virtio/vsock: update trace event for SEQPACKET
drivers/vhost/vsock.c | 44 +-
include/linux/virtio_vsock.h | 9 +
include/net/af_vsock.h | 7 +
.../events/vsock_virtio_transport_common.h | 5 +-
include/uapi/linux/virtio_vsock.h | 9 +
net/vmw_vsock/af_vsock.c | 465 +++++++++++------
net/vmw_vsock/virtio_transport.c | 25 +
net/vmw_vsock/virtio_transport_common.c | 133 ++++-
net/vmw_vsock/vsock_loopback.c | 11 +
tools/testing/vsock/util.c | 32 +-
tools/testing/vsock/util.h | 3 +
tools/testing/vsock/vsock_test.c | 116 ++++
12 files changed, 672 insertions(+), 187 deletions(-)
v9 -> v10:
General changelog:
- patch for write serialization removed from patchset
- commit messages rephrased
- RFC tag removed
Per patch changelog:
see every patch after '---' line.
v8 -> v9:
General changelog:
- see per patch change log.
Per patch changelog:
see every patch after '---' line.
v7 -> v8:
General changelog:
- whole idea is simplified: channel now considered reliable,
so SEQ_BEGIN, SEQ_END, 'msg_len' and 'msg_id' were removed.
Only thing that is used to mark end of message is bit in
'flags' field of packet header: VIRTIO_VSOCK_SEQ_EOR. Packet
with such bit set to 1 means, that this is last packet of
message.
- POSIX MSG_EOR support is removed, as there is no exact
description how it works.
- all changes to 'include/uapi/linux/virtio_vsock.h' moved
to dedicated patch, as these changes linked with patch to
spec.
- patch 'virtio/vsock: SEQPACKET feature bit support' now merged
to 'virtio/vsock: setup SEQPACKET ops for transport'.
- patch 'vhost/vsock: SEQPACKET feature bit support' now merged
to 'vhost/vsock: setup SEQPACKET ops for transport'.
Per patch changelog:
see every patch after '---' line.
v6 -> v7:
General changelog:
- virtio transport callback for message length now removed
from transport. Length of record is returned by dequeue
callback.
- function which tries to get message length now returns 0
when rx queue is empty. Also length of current message in
progress is set to 0, when message processed or error
happens.
- patches for virtio feature bit moved after patches with
transport ops.
Per patch changelog:
see every patch after '---' line.
v5 -> v6:
General changelog:
- virtio transport specific callbacks which send SEQ_BEGIN or
SEQ_END now hidden inside virtio transport. Only enqueue,
dequeue and record length callbacks are provided by transport.
- virtio feature bit for SEQPACKET socket support introduced:
VIRTIO_VSOCK_F_SEQPACKET.
- 'msg_cnt' field in 'struct virtio_vsock_seq_hdr' renamed to
'msg_id' and used as id.
Per patch changelog:
- 'af_vsock: separate wait data loop':
1) Commit message updated.
2) 'prepare_to_wait()' moved inside while loop(thanks to
Jorgen Hansen).
Marked 'Reviewed-by' with 1), but as 2) I removed R-b.
- 'af_vsock: separate receive data loop': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'af_vsock: implement SEQPACKET receive loop': style fixes.
- 'af_vsock: rest of SEQPACKET support':
1) 'module_put()' added when transport callback check failed.
2) Now only 'seqpacket_allow()' callback called to check
support of SEQPACKET by transport.
- 'af_vsock: update comments for stream sockets': commit message
updated.
Marked 'Reviewed-by' with that fix.
- 'virtio/vsock: set packet's type in send':
1) Commit message updated.
2) Parameter 'type' from 'virtio_transport_send_credit_update()'
also removed in this patch instead of in next.
- 'virtio/vsock: dequeue callback for SOCK_SEQPACKET': SEQPACKET
related state wrapped to special struct.
- 'virtio/vsock: update trace event for SEQPACKET': format strings
now not broken by new lines.
v4 -> v5:
- patches reorganized:
1) Setting of packet's type in 'virtio_transport_send_pkt_info()'
is moved to separate patch.
2) Simplifying of 'virtio_transport_send_credit_update()' is
moved to separate patch and before main virtio/vsock patches.
- style problem fixed
- in 'af_vsock: separate receive data loop' extra 'release_sock()'
removed
- added trace event fields for SEQPACKET
- in 'af_vsock: separate wait data loop':
1) 'vsock_wait_data()' removed 'goto out;'
2) Comment for invalid data amount is changed.
- in 'af_vsock: rest of SEQPACKET support', 'new_transport' pointer
check is moved after 'try_module_get()'
- in 'af_vsock: update comments for stream sockets', 'connect-oriented'
replaced with 'connection-oriented'
- in 'loopback/vsock: setup SEQPACKET ops for transport',
'loopback/vsock' replaced with 'vsock/loopback'
v3 -> v4:
- SEQPACKET specific metadata moved from packet header to payload
and called 'virtio_vsock_seq_hdr'
- record integrity check:
1) SEQ_END operation was added, which marks end of record.
2) Both SEQ_BEGIN and SEQ_END carries counter which is incremented
on every marker send.
- af_vsock.c: socket operations for STREAM and SEQPACKET call same
functions instead of having own "gates" differs only by names:
'vsock_seqpacket/stream_getsockopt()' now replaced with
'vsock_connectible_getsockopt()'.
- af_vsock.c: 'seqpacket_dequeue' callback returns error and flag that
record ready. There is no need to return number of copied bytes,
because case when record received successfully is checked at virtio
transport layer, when SEQ_END is processed. Also user doesn't need
number of copied bytes, because 'recv()' from SEQPACKET could return
error, length of users's buffer or length of whole record(both are
known in af_vsock.c).
- af_vsock.c: both wait loops in af_vsock.c(for data and space) moved
to separate functions because now both called from several places.
- af_vsock.c: 'vsock_assign_transport()' checks that 'new_transport'
pointer is not NULL and returns 'ESOCKTNOSUPPORT' instead of 'ENODEV'
if failed to use transport.
- tools/testing/vsock/vsock_test.c: rename tests
v2 -> v3:
- patches reorganized: split for prepare and implementation patches
- local variables are declared in "Reverse Christmas tree" manner
- virtio_transport_common.c: valid leXX_to_cpu() for vsock header
fields access
- af_vsock.c: 'vsock_connectible_*sockopt()' added as shared code
between stream and seqpacket sockets.
- af_vsock.c: loops in '__vsock_*_recvmsg()' refactored.
- af_vsock.c: 'vsock_wait_data()' refactored.
v1 -> v2:
- patches reordered: af_vsock.c related changes now before virtio vsock
- patches reorganized: more small patches, where +/- are not mixed
- tests for SOCK_SEQPACKET added
- all commit messages updated
- af_vsock.c: 'vsock_pre_recv_check()' inlined to
'vsock_connectible_recvmsg()'
- af_vsock.c: 'vsock_assign_transport()' returns ENODEV if transport
was not found
- virtio_transport_common.c: transport callback for seqpacket dequeue
- virtio_transport_common.c: simplified
'virtio_transport_recv_connected()'
- virtio_transport_common.c: send reset on socket and packet type
mismatch.
Signed-off-by: Arseny Krasnov <redacted>
--
2.25.1
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
On Thu, May 20, 2021 at 10:16:36PM +0300, Arseny Krasnov wrote:
To make SEQPACKET socket functional, socket ops was added
for SEQPACKET type and such type of socket was allowed
to create.
If you need to resend, I think is better to use the present in the
commit message.
Maybe you can rephrase something like this:
"Add socket ops for SEQPACKET type and .seqpacket_allow() callback
to query transports if they support SEQPACKET"
On Tue, May 25, 2021 at 11:22:09AM +0300, Arseny Krasnov wrote:
On 23.05.2021 15:14, Arseny Krasnov wrote:
quoted
On 21.05.2021 10:55, Stefano Garzarella wrote:
quoted
Hi Arseny,
On Thu, May 20, 2021 at 10:13:53PM +0300, Arseny Krasnov wrote:
quoted
This patchset implements support of SOCK_SEQPACKET for virtio
transport.
I'll carefully review and test this series next Monday, in the mean time
I think we should have at least an agreement about the changes that
regards virtio-spec before merge this series, to avoid any compatibility
issues.
Do you plan to send a new version of the specification changes?
Thanks,
Stefano
Hello, sorry for long answer. I'm on vacation now, but i plan to send
it in next several days, because with current implementation it is short
Thank You
Yep, sure.
About this series I think is better to split in two series since it
became very long. Patchwork [1] also complains here [2].
You can send a first series with patches from 1 to 7. These patches are
reviewed by me and can go regardless of the discussion of the VIRTIO
specifications.
Maybe you can also add the patch with the test to this first series.
Please specify in the cover letter that the implementation for virtio
devices is under development and will be sent later.
When it will be merged in the net-next tree, you can post the second
part with the rest of the series that implements SEQPACKET for virtio
devices, possibly after we received an agreement for the specifications.
Please use the "net-next" tag and take a look at
Documentation/networking/netdev-FAQ.rst about netdev development.
Anyway, in the next days (hopefully tomorrow) I'll review the rest of
the series related to virtio devices and spec.
Thanks,
Stefano
[1]
https://patchwork.kernel.org/project/netdevbpf/list/?series=486011&state=*
[2]
https://patchwork.kernel.org/project/netdevbpf/patch/20210520191449.1270723-1-arseny.krasnov@kaspersky.com/
On Tue, May 25, 2021 at 11:22:09AM +0300, Arseny Krasnov wrote:
quoted
On 23.05.2021 15:14, Arseny Krasnov wrote:
quoted
On 21.05.2021 10:55, Stefano Garzarella wrote:
quoted
Hi Arseny,
On Thu, May 20, 2021 at 10:13:53PM +0300, Arseny Krasnov wrote:
quoted
This patchset implements support of SOCK_SEQPACKET for virtio
transport.
I'll carefully review and test this series next Monday, in the mean time
I think we should have at least an agreement about the changes that
regards virtio-spec before merge this series, to avoid any compatibility
issues.
Do you plan to send a new version of the specification changes?
Thanks,
Stefano
Hello, sorry for long answer. I'm on vacation now, but i plan to send
it in next several days, because with current implementation it is short
Thank You
Yep, sure.
About this series I think is better to split in two series since it
became very long. Patchwork [1] also complains here [2].
You can send a first series with patches from 1 to 7. These patches are
reviewed by me and can go regardless of the discussion of the VIRTIO
specifications.
Ok, i'll send it on next week.
Maybe you can also add the patch with the test to this first series.
Please specify in the cover letter that the implementation for virtio
devices is under development and will be sent later.
When it will be merged in the net-next tree, you can post the second
part with the rest of the series that implements SEQPACKET for virtio
devices, possibly after we received an agreement for the specifications.
Please use the "net-next" tag and take a look at
Documentation/networking/netdev-FAQ.rst about netdev development.
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted hunk
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
}
EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
+static u16 virtio_transport_get_type(struct sock *sk)
+{
+ if (sk->sk_type == SOCK_STREAM)
+ return VIRTIO_VSOCK_TYPE_STREAM;
+ else
+ return VIRTIO_VSOCK_TYPE_SEQPACKET;
+}
+
/* This function can only be used on connecting/connected sockets,
* since a socket assigned to a transport is required.
*
struct virtio_vsock_pkt, list);
/* If there is space in the last packet queued, we copy the
- * new packet in its buffer.
+ * new packet in its buffer(except SEQPACKET case, when we
+ * also check that last packet is not last packet of previous
+ * record).
Is better to explain why we don't do this for SEQPACKET, something like this:
/* If there is space in the last packet queued, we copy the
* new packet in its buffer.
* We avoid this if the last packet queued has
* VIRTIO_VSOCK_SEQ_EOR set, because it is the delimiter
* of SEQPACKET record, so `pkt` is the first packet
* of a new record.
*/
On Thu, May 20, 2021 at 10:18:37PM +0300, Arseny Krasnov wrote:
quoted hunk
Small updates to make SOCK_SEQPACKET work:
1) Send SHUTDOWN on socket close for SEQPACKET type.
2) Set SEQPACKET packet type during send.
3) Set 'VIRTIO_VSOCK_SEQ_EOR' bit in flags for last
packet of message.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
2) Commit message updated.
3) Add check for socket type when setting SEQ_EOR bit.
include/linux/virtio_vsock.h | 4 ++++
net/vmw_vsock/virtio_transport_common.c | 18 ++++++++++++++++--
2 files changed, 20 insertions(+), 2 deletions(-)
On Thu, May 20, 2021 at 10:18:57PM +0300, Arseny Krasnov wrote:
To make transport work with SOCK_SEQPACKET two updates were
added:
Present is better, and you can also mention that we enable it only if
the feature is negotiated with the device.
quoted hunk
1) SOCK_SEQPACKET ops for virtio transport and 'seqpacket_allow()'
callback.
2) Handling of SEQPACKET bit: guest tries to negotiate it with vhost.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'virtio_has_feature()' to check feature bit.
2) Move assignment to 'seqpacket_allow' before 'rcu_assign_pointer()'.
net/vmw_vsock/virtio_transport.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
On Thu, May 20, 2021 at 10:19:13PM +0300, Arseny Krasnov wrote:
Please describe better the changes included in this patch in the first
part of the commit message.
quoted hunk
As vhost places data in buffers of guest's rx queue, keep SEQ_EOR
bit set only when last piece of data is copied. Otherwise we get
sequence packets for one socket in guest's rx queue with SEQ_EOR bit
set. Also remove ignore of non-stream type of packets, handle SEQPACKET
feature bit.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Move 'restore_flag' handling to 'payload_len' calculation
block.
drivers/vhost/vsock.c | 44 +++++++++++++++++++++++++++++++++++++++----
1 file changed, 40 insertions(+), 4 deletions(-)
/* If the packet is greater than the space available in the
* buffer, we split it using multiple buffers.
*/
- if (payload_len > iov_len - sizeof(pkt->hdr))
+ if (payload_len > iov_len - sizeof(pkt->hdr)) {
payload_len = iov_len - sizeof(pkt->hdr);
Please, add a comment here to explain why we need this.
quoted hunk
+ if (le32_to_cpu(pkt->hdr.flags) &
VIRTIO_VSOCK_SEQ_EOR) {
+ pkt->hdr.flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
+ restore_flag = true;
+ }
+ }
+
/* Set the correct length in the header */
pkt->hdr.len = cpu_to_le32(payload_len);
Maybe we can restore the flag only if we are queueing again the same
packet, I mean in the `if (pkt->off < pkt->len) {` branch below.
What do you think?
On Thu, May 20, 2021 at 10:19:50PM +0300, Arseny Krasnov wrote:
Implement two tests of SOCK_SEQPACKET socket: first sends data by
several 'write()'s and checks that number of 'read()' were same.
Second test checks MSG_TRUNC flag. Cases for connect(), bind(),
etc. are not tested, because it is same as for stream socket.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Commit message updated.
2) Add second test for message bounds.
This patch LGTM, but I'll review better with the next version, running
also the test suite on my VMs.
Thanks,
Stefano
}
/* Connect to <cid, port> and return the file descriptor. */
-int vsock_stream_connect(unsigned int cid, unsigned int port)
+static int vsock_connect(unsigned int cid, unsigned int port, int type)
{
union {
struct sockaddr sa;
@@ -101,7 +101,7 @@ int vsock_stream_connect(unsigned int cid, unsigned int port)
@@ -120,11 +120,21 @@ int vsock_stream_connect(unsigned int cid, unsigned int port)
return fd;
}
+int vsock_stream_connect(unsigned int cid, unsigned int port)
+{
+ return vsock_connect(cid, port, SOCK_STREAM);
+}
+
+int vsock_seqpacket_connect(unsigned int cid, unsigned int port)
+{
+ return vsock_connect(cid, port, SOCK_SEQPACKET);
+}
+
/* Listen on <cid, port> and return the first incoming connection. The remote
* address is stored to clientaddrp. clientaddrp may be NULL.
*/
-int vsock_stream_accept(unsigned int cid, unsigned int port,
- struct sockaddr_vm *clientaddrp)
+static int vsock_accept(unsigned int cid, unsigned int port,
+ struct sockaddr_vm *clientaddrp, int type)
{
union {
struct sockaddr sa;
@@ -145,7 +155,7 @@ int vsock_stream_accept(unsigned int cid, unsigned int port,
int client_fd;
int old_errno;
- fd = socket(AF_VSOCK, SOCK_STREAM, 0);
+ fd = socket(AF_VSOCK, type, 0);
if (bind(fd, &addr.sa, sizeof(addr.svm)) < 0) {
perror("bind");
@@ -189,6 +199,18 @@ int vsock_stream_accept(unsigned int cid, unsigned int port,
return client_fd;
}
+int vsock_stream_accept(unsigned int cid, unsigned int port,
+ struct sockaddr_vm *clientaddrp)
+{
+ return vsock_accept(cid, port, clientaddrp, SOCK_STREAM);
+}
+
+int vsock_seqpacket_accept(unsigned int cid, unsigned int port,
+ struct sockaddr_vm *clientaddrp)
+{
+ return vsock_accept(cid, port, clientaddrp, SOCK_SEQPACKET);
+}
+
/* Transmit one byte and check the return value.
*
* expected_ret:
void init_signals(void);
unsigned int parse_cid(const char *str);
int vsock_stream_connect(unsigned int cid, unsigned int port);
+int vsock_seqpacket_connect(unsigned int cid, unsigned int port);
int vsock_stream_accept(unsigned int cid, unsigned int port,
struct sockaddr_vm *clientaddrp);
+int vsock_seqpacket_accept(unsigned int cid, unsigned int port,
+ struct sockaddr_vm *clientaddrp);
void vsock_wait_remote_close(int fd);
void send_byte(int fd, int expected_ret, int flags);
void recv_byte(int fd, int expected_ret, int flags);
I think we should fixe the indentation here (e.g. following show_op):
#define show_type(val) \
__print_symbolic(val, \
{ VIRTIO_VSOCK_TYPE_STREAM, "STREAM" }, \
{ VIRTIO_VSOCK_TYPE_SEQPACKET, "SEQPACKET" })
Thanks,
Stefano
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error, we still need
to free packet. It is possible to do something like this:
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
if (dequeued_len < 0)
break;
On Thu, May 20, 2021 at 10:19:13PM +0300, Arseny Krasnov wrote:
Please describe better the changes included in this patch in the first
part of the commit message.
quoted
As vhost places data in buffers of guest's rx queue, keep SEQ_EOR
bit set only when last piece of data is copied. Otherwise we get
sequence packets for one socket in guest's rx queue with SEQ_EOR bit
set. Also remove ignore of non-stream type of packets, handle SEQPACKET
feature bit.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Move 'restore_flag' handling to 'payload_len' calculation
block.
drivers/vhost/vsock.c | 44 +++++++++++++++++++++++++++++++++++++++----
1 file changed, 40 insertions(+), 4 deletions(-)
/* If the packet is greater than the space available in the
* buffer, we split it using multiple buffers.
*/
- if (payload_len > iov_len - sizeof(pkt->hdr))
+ if (payload_len > iov_len - sizeof(pkt->hdr)) {
payload_len = iov_len - sizeof(pkt->hdr);
Please, add a comment here to explain why we need this.
quoted
+ if (le32_to_cpu(pkt->hdr.flags) &
VIRTIO_VSOCK_SEQ_EOR) {
+ pkt->hdr.flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
+ restore_flag = true;
+ }
+ }
+
/* Set the correct length in the header */
pkt->hdr.len = cpu_to_le32(payload_len);
Maybe we can restore the flag only if we are queueing again the same
packet, I mean in the `if (pkt->off < pkt->len) {` branch below.
What do you think?
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error, we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
It is possible to do something like this:
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
if (dequeued_len < 0)
break;
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
quoted hunk
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
quoted
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error, we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are unneeded,
as channel considedered lossless. What do You think?
Thank You
quoted
It is possible to do something like this:
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
if (dequeued_len < 0)
break;
On Fri, Jun 04, 2021 at 09:00:14PM +0300, Arseny Krasnov wrote:
On 04.06.2021 18:06, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
quoted
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
return err;
}
+static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
+ size_t len, int flags)
+{
+ const struct vsock_transport *transport;
+ bool msg_ready;
+ struct vsock_sock *vsk;
+ ssize_t record_len;
+ long timeout;
+ int err = 0;
+ DEFINE_WAIT(wait);
+
+ vsk = vsock_sk(sk);
+ transport = vsk->transport;
+
+ timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
+ msg_ready = false;
+ record_len = 0;
+
+ while (1) {
+ ssize_t fragment_len;
+
+ if (vsock_wait_data(sk, &wait, timeout, NULL, 0) <= 0) {
+ /* In case of any loop break(timeout, signal
+ * interrupt or shutdown), we report user that
+ * nothing was copied.
+ */
+ err = 0;
Why we report that nothing was copied?
What happen to the bytes already copied in `msg`?
Seems i need to return result of vsock_wait_data()...
I'm not sure.
My biggest concern is if we reach timeout or get a signal while waiting
for the other pieces of a message.
I believe that we should not start copying a message if we have not
received all the fragments. Otherwise we have this problem.
When we are sure that we have all the pieces, then we should copy them
without interrupting.
IIRC this was done in previous versions.
Stefano
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
Maybe we should start using skbuffs for seqpackets as well, but that
might take some time, so that might be okay for now.
Thanks,
Stefano
On Fri, Jun 04, 2021 at 09:00:14PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:06, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
quoted
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
return err;
}
+static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
+ size_t len, int flags)
+{
+ const struct vsock_transport *transport;
+ bool msg_ready;
+ struct vsock_sock *vsk;
+ ssize_t record_len;
+ long timeout;
+ int err = 0;
+ DEFINE_WAIT(wait);
+
+ vsk = vsock_sk(sk);
+ transport = vsk->transport;
+
+ timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
+ msg_ready = false;
+ record_len = 0;
+
+ while (1) {
+ ssize_t fragment_len;
+
+ if (vsock_wait_data(sk, &wait, timeout, NULL, 0) <= 0) {
+ /* In case of any loop break(timeout, signal
+ * interrupt or shutdown), we report user that
+ * nothing was copied.
+ */
+ err = 0;
Why we report that nothing was copied?
What happen to the bytes already copied in `msg`?
Seems i need to return result of vsock_wait_data()...
I'm not sure.
My biggest concern is if we reach timeout or get a signal while waiting
for the other pieces of a message.
I believe that we should not start copying a message if we have not
received all the fragments. Otherwise we have this problem.
When we are sure that we have all the pieces, then we should copy them
without interrupting.
IIRC this was done in previous versions.
As i remember, previous versions also returned 0, because i thought,
that for interrupted read we can copy piece of data to user's buffer,
but we must return that nothing copied or error. In this way user
won't read part of message, because syscall returned that there is
nothing to copy. So as i understand, it is not enough - user's buffer
must be touched only when whole message is copied?
On Mon, Jun 07, 2021 at 02:29:28PM +0300, Arseny Krasnov wrote:
On 07.06.2021 13:48, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:00:14PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:06, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:16:08PM +0300, Arseny Krasnov wrote:
quoted
Add receive loop for SEQPACKET. It looks like receive loop for
STREAM, but there are differences:
1) It doesn't call notify callbacks.
2) It doesn't care about 'SO_SNDLOWAT' and 'SO_RCVLOWAT' values, because
there is no sense for these values in SEQPACKET case.
3) It waits until whole record is received or error is found during
receiving.
4) It processes and sets 'MSG_TRUNC' flag.
So to avoid extra conditions for two types of socket inside one loop, two
independent functions were created.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
include/net/af_vsock.h | 4 +++
net/vmw_vsock/af_vsock.c | 72 +++++++++++++++++++++++++++++++++++++++-
2 files changed, 75 insertions(+), 1 deletion(-)
return err;
}
+static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
+ size_t len, int flags)
+{
+ const struct vsock_transport *transport;
+ bool msg_ready;
+ struct vsock_sock *vsk;
+ ssize_t record_len;
+ long timeout;
+ int err = 0;
+ DEFINE_WAIT(wait);
+
+ vsk = vsock_sk(sk);
+ transport = vsk->transport;
+
+ timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
+ msg_ready = false;
+ record_len = 0;
+
+ while (1) {
+ ssize_t fragment_len;
+
+ if (vsock_wait_data(sk, &wait, timeout, NULL, 0) <= 0) {
+ /* In case of any loop break(timeout, signal
+ * interrupt or shutdown), we report user that
+ * nothing was copied.
+ */
+ err = 0;
Why we report that nothing was copied?
What happen to the bytes already copied in `msg`?
Seems i need to return result of vsock_wait_data()...
I'm not sure.
My biggest concern is if we reach timeout or get a signal while waiting
for the other pieces of a message.
I believe that we should not start copying a message if we have not
received all the fragments. Otherwise we have this problem.
When we are sure that we have all the pieces, then we should copy them
without interrupting.
IIRC this was done in previous versions.
As i remember, previous versions also returned 0, because i thought,
that for interrupted read we can copy piece of data to user's buffer,
but we must return that nothing copied or error. In this way user
This can also be fine, but we should remove packet form the rx_queue
only when we are sure that we delivered the entire message.
won't read part of message, because syscall returned that there is
nothing to copy. So as i understand, it is not enough - user's buffer
must be touched only when whole message is copied?
The important thing is to not remove packets from the rx_queue unless we
are sure that everything went well and we are returning the entire
message to the user.
Stefano
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
Thank You
Maybe we should start using skbuffs for seqpackets as well, but that
might take some time, so that might be okay for now.
Thanks,
Stefano
On Mon, Jun 07, 2021 at 04:18:38PM +0300, Arseny Krasnov wrote:
On 07.06.2021 14:04, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
I think the best approach could be to send credit updates when we remove
them from the rx_queue.
Stefano
On Mon, Jun 07, 2021 at 04:18:38PM +0300, Arseny Krasnov wrote:
quoted
On 07.06.2021 14:04, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
I think the best approach could be to send credit updates when we remove
them from the rx_queue.
In that case, it will be impossible to send message bigger than size of rx buffer
(e.g. credit allowed size), because packet will be queued without credit update
reply until credit allowed reach 0.
Thank You
On Tue, Jun 08, 2021 at 12:40:39PM +0300, Arseny Krasnov wrote:
On 08.06.2021 11:23, Stefano Garzarella wrote:
quoted
On Mon, Jun 07, 2021 at 04:18:38PM +0300, Arseny Krasnov wrote:
quoted
On 07.06.2021 14:04, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
I think the best approach could be to send credit updates when we remove
them from the rx_queue.
In that case, it will be impossible to send message bigger than size of rx buffer
(e.g. credit allowed size), because packet will be queued without credit update
reply until credit allowed reach 0.
Yep, but I think it is a reasonable limit for a datagram socket.
Maybe we can add a check on the TX side, since we know this value and
return an error to the user.
Thanks,
Stefano
On Tue, Jun 08, 2021 at 12:40:39PM +0300, Arseny Krasnov wrote:
quoted
On 08.06.2021 11:23, Stefano Garzarella wrote:
quoted
On Mon, Jun 07, 2021 at 04:18:38PM +0300, Arseny Krasnov wrote:
quoted
On 07.06.2021 14:04, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
I think the best approach could be to send credit updates when we remove
them from the rx_queue.
In that case, it will be impossible to send message bigger than size of rx buffer
(e.g. credit allowed size), because packet will be queued without credit update
reply until credit allowed reach 0.
Yep, but I think it is a reasonable limit for a datagram socket.
Maybe we can add a check on the TX side, since we know this value and
return an error to the user.
E.g., to before sending message using SEQPACKET socket,
i need to call setsockopt with SO_VM_SOCKETS_BUFFER_MAX_SIZE/
SO_VM_SOCKETS_BUFFER_SIZE params to setup maximum message size,
if user tries to send message bigger than it, return -EMSGSIZE ?
Thank You
On Tue, Jun 08, 2021 at 01:24:58PM +0300, Arseny Krasnov wrote:
On 08.06.2021 13:19, Stefano Garzarella wrote:
quoted
On Tue, Jun 08, 2021 at 12:40:39PM +0300, Arseny Krasnov wrote:
quoted
On 08.06.2021 11:23, Stefano Garzarella wrote:
quoted
On Mon, Jun 07, 2021 at 04:18:38PM +0300, Arseny Krasnov wrote:
quoted
On 07.06.2021 14:04, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 09:03:26PM +0300, Arseny Krasnov wrote:
quoted
On 04.06.2021 18:03, Stefano Garzarella wrote:
quoted
On Fri, Jun 04, 2021 at 04:12:23PM +0300, Arseny Krasnov wrote:
quoted
On 03.06.2021 17:45, Stefano Garzarella wrote:
quoted
On Thu, May 20, 2021 at 10:17:58PM +0300, Arseny Krasnov wrote:
quoted
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <redacted>
---
v9 -> v10:
1) Number of dequeued bytes incremented even in case when
user's buffer is full.
2) Use 'msg_data_left()' instead of direct access to 'msg_hdr'.
3) Rename variable 'err' to 'dequeued_len', in case of error
it has negative value.
include/linux/virtio_vsock.h | 5 ++
net/vmw_vsock/virtio_transport_common.c | 65 +++++++++++++++++++++++++
2 files changed, 70 insertions(+)
+ size_t bytes_to_copy;
+ size_t pkt_len;
+
+ pkt = list_first_entry(&vvs->rx_queue, struct virtio_vsock_pkt, list);
+ pkt_len = (size_t)le32_to_cpu(pkt->hdr.len);
+ bytes_to_copy = min(user_buf_len, pkt_len);
+
+ if (bytes_to_copy) {
+ /* sk_lock is held by caller so no one else can dequeue.
+ * Unlock rx_lock since memcpy_to_msg() may sleep.
+ */
+ spin_unlock_bh(&vvs->rx_lock);
+
+ if (memcpy_to_msg(msg, pkt->buf, bytes_to_copy))
+ dequeued_len = -EINVAL;
I think here is better to return the error returned by memcpy_to_msg(),
as we do in the other place where we use memcpy_to_msg().
I mean something like this:
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err)
dequeued_len = err;
Maybe here we can simply break the cycle if we have an error:
if (dequeued_len < 0)
break;
Or we can refactor a bit, simplifying the while() condition and also the
code in this way (not tested):
while (!*msg_ready && !list_empty(&vvs->rx_queue)) {
...
if (bytes_to_copy) {
int err;
/* ...
*/
spin_unlock_bh(&vvs->rx_lock);
err = memcpy_to_msgmsg, pkt->buf, bytes_to_copy);
if (err) {
dequeued_len = err;
goto out;
}
spin_lock_bh(&vvs->rx_lock);
user_buf_len -= bytes_to_copy;
}
dequeued_len += pkt_len;
if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR)
*msg_ready = true;
virtio_transport_dec_rx_pkt(vvs, pkt);
list_del(&pkt->list);
virtio_transport_free_pkt(pkt);
}
out:
spin_unlock_bh(&vvs->rx_lock);
virtio_transport_send_credit_update(vsk);
return dequeued_len;
}
I think we can't do 'goto out' or break, because in case of error,
we still need
to free packet.
Didn't we have code that remove packets from a previous message?
I don't see it anymore.
For example if we have 10 packets queued for a message (the 10th
packet
has the EOR flag) and the memcpy_to_msg() fails on the 2nd packet, with
you proposal we are freeing only the first 2 packets, the rest is there
and should be freed when reading the next message, but I don't see that
code.
The same can happen if the recvmsg syscall is interrupted. In that case
we report that nothing was copied, but we freed the first N packets, so
they are lost but the other packets are still in the queue.
Please check also the patch where we implemented
__vsock_seqpacket_recvmsg().
I thinks we should free packets only when we are sure we copied them to
the user space.
Hm, yes, this is problem. To solve it i can restore previous approach
with seqbegin/seqend. In that case i can detect unfinished record and
drop it's packets. Seems seqbegin will be a bit like
VIRTIO_VSOCK_SEQ_EOR in flags
field of header(e.g. VIRTIO_VSOCK_SEQ_BEGIN). Message id and length are
unneeded,
as channel considedered lossless. What do You think?
I think VIRTIO_VSOCK_SEQ_BEGIN is redundant, using only EOR should be
fine.
When we receive EOR we know that this is the last packet on this message
and the next packet will be the first of a new message.
What we should do is check that we have all the fragments of a packet
and return them all together, otherwise we have to say we have nothing.
For example as we process packets from the vitqueue and queue them in
the rx_queue we could use a counter of how many EORs are in the
rx_queue, which we decrease in virtio_transport_seqpacket_do_dequeue()
when we copied all the fragments.
If the counter is 0, we don't remove anything from the queue and
virtio_transport_seqpacket_do_dequeue() returns 0.
So .seqpacket_dequeue should return 0 if there is not at least one
complete message, or return the entire message. A partial message should
never return.
What do you think?
I like it, i've implemented this approach in some early pre v1 versions.
But in this case, credit update logic will be changed - in current implementation
(both seqpacket and stream) credit update reply is sent when data is copied
to user's buffer(e.g. we copy data somewhere, free packet and ready to process
new packet). But if we don't touch user's buffer and keeping incoming packet in rx queue
until whole record is ready, when to send credit update?
I think the best approach could be to send credit updates when we remove
them from the rx_queue.
In that case, it will be impossible to send message bigger than size of rx buffer
(e.g. credit allowed size), because packet will be queued without credit update
reply until credit allowed reach 0.
Yep, but I think it is a reasonable limit for a datagram socket.
Maybe we can add a check on the TX side, since we know this value and
return an error to the user.
E.g., to before sending message using SEQPACKET socket,
i need to call setsockopt with SO_VM_SOCKETS_BUFFER_MAX_SIZE/
SO_VM_SOCKETS_BUFFER_SIZE params to setup maximum message size,
if user tries to send message bigger than it, return -EMSGSIZE ?
Yep, I mean the receiver side must set it (IIRC default is 256K).
In the transmitter side we can check it using `vvs->peer_buf_alloc` and
return the error.
Stefano