From: Jon Maloy <jmaloy@redhat.com>
To: Stefano Brivio <sbrivio@redhat.com>
Cc: passt-dev@passt.top, lvivier@redhat.com, dgibson@redhat.com
Subject: Re: [PATCH] tcp: move seq_to_tap update to when frame is queued
Date: Fri, 10 May 2024 15:40:41 -0400 [thread overview]
Message-ID: <51413033-7a19-bf44-39da-980e826de896@redhat.com> (raw)
In-Reply-To: <20240510184030.44b57a2f@elisabeth>
On 2024-05-10 12:40, Stefano Brivio wrote:
> On Wed, 8 May 2024 23:00:23 -0400
> Jon Maloy <jmaloy@redhat.com> wrote:
>
>> commit a469fc393fa1 ("tcp, tap: Don't increase tap-side sequence counter for dropped frames")
>> delayed update of conn->seq_to_tap until the moment the corresponding
>> frame has been successfully pushed out. This has the advantage that we
>> immediately can retransmit a buffer that we fail to trasnmit, rather
>> than waiting for the peer side to discover the loss and initiate fast
>> retransmit.
> It's not really fast retransmit, it's a simple retry of the operation
> that didn't succeed. We didn't even transmit.
Ok
>> This approach has turned out to cause a problem with spurious sequence
>> number updates during peer-initiated retransmits, and we have realized
>> it may not be the best way to solve te above issue.
>>
>> We now restore the previous method, by updating the said field at the
>> moment a frame is added to the outqueue. To retain the advantage of fast
>> retansmit
> Same here.
>
>> based on local failure detection, we now scan through the part
>> of the outqueue that had do be dropped, and restore the sequence counter
>> for each affected connection to the most appropriate value.
>>
>> Signed-off-by: Jon Maloy <jmaloy@redhat.com>
>> ---
>> tcp.c | 52 ++++++++++++++++++++++++++++++++++++++++++----------
>> 1 file changed, 42 insertions(+), 10 deletions(-)
>>
>> diff --git a/tcp.c b/tcp.c
>> index 21d0af0..58fdbc9 100644
>> --- a/tcp.c
>> +++ b/tcp.c
>> @@ -412,11 +412,13 @@ static union inany_addr low_rtt_dst[LOW_RTT_TABLE_SIZE];
>>
>> /**
>> * tcp_buf_seq_update - Sequences to update with length of frames once sent
> This is not the case anymore, maybe:
>
> * tcp_conn_old_seq() - Old sequence numbers for connections with pending frames
ok
>> - * @seq: Pointer to sequence number sent to tap-side, to be updated
>> + * @conn: Pointer to connection corresponding to frame. May need update
> Mixed whitespace and tabs. It looks like the connection pointer might
> need to be updated... what about:
>
> * @conn: Pointer to connection for this frame
>
> ?
>
>> + * @seq: Sequence number of the corresponding frame
>> * @len: TCP payload length
> The length is not needed anymore.
Yes. Of course ;-(
>> */
>> struct tcp_buf_seq_update {
>> - uint32_t *seq;
>> + struct tcp_tap_conn *conn;
>> + uint32_t seq;
>> uint16_t len;
>> };
>>
>> @@ -1261,25 +1263,52 @@ static void tcp_flags_flush(const struct ctx *c)
>> tcp4_flags_used = 0;
>> }
>>
>> +/**
>> + * tcp_revert_seq() - Revert affected conn->seq_to_tap after failed transmission
>> + * @seq_update: Array with connection and sequence number data
>> + * @s: Entry corresponding to first dropped frame
>> + * @e: Entry corresponding to last dropped frame
> These are not pointer to the entries, though. They are indices of the
> queued frames.
I had already fixed that.
>
>> + */
>> +static void tcp_revert_seq(struct tcp_buf_seq_update *seq_update, int s, int e)
>> +{
>> + struct tcp_tap_conn *conn;
>> + uint32_t lowest_seq;
>> + int i, ii;
>> +
>> + for (i = s; i < e; i++) {
>> + conn = seq_update[i].conn;
>> + lowest_seq = seq_update[i].seq;
>> +
>> + for (ii = i + 1; ii < e; ii++) {
>> + if (seq_update[ii].conn != conn)
>> + continue;
>> + if (SEQ_GT(lowest_seq, seq_update[ii].seq))
>> + lowest_seq = seq_update[ii].seq;
>> + }
> If I recall correctly, David suggested a simpler approach that avoids
> this O(n^2) scan, based on the observation that 1. the first entry you
> find in the table also has the lowest sequence number (we don't send
> frames out-of-order),
Not so sure about that. We can be in the middle of retransmit.
Of course, if I continue to flush the queue just before retransmit,
which I didn't intend to do, this will be true.
> and that 2. you'll never revert to a higher
> sequence number (the two lines below take care of that).
>
> That is, you could just scan the table once, and if you find a sequence
> number that's lower than the current sequence stored for the connection,
> store it.
Yes, I can do that, and it will work even without flushing the queue.
I missed that aspect of David's description.
>
>> +
>> + if (SEQ_GT(conn->seq_to_tap, lowest_seq))
>> + conn->seq_to_tap = lowest_seq;
>> + }
>> +}
>> +
>> /**
>> * tcp_payload_flush() - Send out buffers for segments with data
>> * @c: Execution context
>> */
>> static void tcp_payload_flush(const struct ctx *c)
>> {
>> - unsigned i;
>> size_t m;
>>
>> m = tap_send_frames(c, &tcp6_l2_iov[0][0], TCP_NUM_IOVS,
>> tcp6_payload_used);
>> - for (i = 0; i < m; i++)
>> - *tcp6_seq_update[i].seq += tcp6_seq_update[i].len;
>> + if (m != tcp6_payload_used)
>> + tcp_revert_seq(tcp6_seq_update, m, tcp6_payload_used);
>> tcp6_payload_used = 0;
>>
>> m = tap_send_frames(c, &tcp4_l2_iov[0][0], TCP_NUM_IOVS,
>> tcp4_payload_used);
>> - for (i = 0; i < m; i++)
>> - *tcp4_seq_update[i].seq += tcp4_seq_update[i].len;
>> + if (m != tcp4_payload_used)
>> + tcp_revert_seq(tcp4_seq_update, m, tcp4_payload_used);
>> tcp4_payload_used = 0;
>> }
>>
>> @@ -2129,10 +2158,11 @@ static int tcp_sock_consume(const struct tcp_tap_conn *conn, uint32_t ack_seq)
>> static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>> ssize_t dlen, int no_csum, uint32_t seq)
>> {
>> - uint32_t *seq_update = &conn->seq_to_tap;
>> struct iovec *iov;
>> size_t l4len;
>>
>> + conn->seq_to_tap = seq;
> This is the sequence number for the frame we're sending (start of this
> frame), but not the current byte sequence sent to the "tap" (end of
> this frame), which would be seq + dlen, I think.
Already noticed during my testing and fixed. Strangely enough, it still
worked well for a while :-(
>
>> +
>> if (CONN_V4(conn)) {
>> struct iovec *iov_prev = tcp4_l2_iov[tcp4_payload_used - 1];
>> const uint16_t *check = NULL;
>> @@ -2142,7 +2172,8 @@ static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>> check = &iph->check;
>> }
>>
>> - tcp4_seq_update[tcp4_payload_used].seq = seq_update;
>> + tcp4_seq_update[tcp4_payload_used].conn = conn;
>> + tcp4_seq_update[tcp4_payload_used].seq = seq;
>> tcp4_seq_update[tcp4_payload_used].len = dlen;
>>
>> iov = tcp4_l2_iov[tcp4_payload_used++];
>> @@ -2151,7 +2182,8 @@ static void tcp_data_to_tap(const struct ctx *c, struct tcp_tap_conn *conn,
>> if (tcp4_payload_used > TCP_FRAMES_MEM - 1)
>> tcp_payload_flush(c);
>> } else if (CONN_V6(conn)) {
>> - tcp6_seq_update[tcp6_payload_used].seq = seq_update;
>> + tcp6_seq_update[tcp6_payload_used].conn = conn;
>> + tcp6_seq_update[tcp6_payload_used].seq = seq;
>> tcp6_seq_update[tcp6_payload_used].len = dlen;
>>
>> iov = tcp6_l2_iov[tcp6_payload_used++];
I will fix the loop and repost shortly.
///jon
next prev parent reply other threads:[~2024-05-10 19:40 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-09 3:00 [PATCH] tcp: move seq_to_tap update to when frame is queued Jon Maloy
2024-05-10 16:40 ` Stefano Brivio
2024-05-10 19:40 ` Jon Maloy [this message]
2024-05-13 1:32 ` David Gibson
2024-05-13 1:03 ` David Gibson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=51413033-7a19-bf44-39da-980e826de896@redhat.com \
--to=jmaloy@redhat.com \
--cc=dgibson@redhat.com \
--cc=lvivier@redhat.com \
--cc=passt-dev@passt.top \
--cc=sbrivio@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
Code repositories for project(s) associated with this public inbox
https://passt.top/passt
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for IMAP folder(s).