Jason Xing wrote:

On Mon, Sep 30, 2024 at 6:39 PM Willem de Bruijn
[off-list ref] wrote:

quoted

Jason Xing wrote:

quoted

From: Jason Xing <kernelxing@tencent.com>

Even though this case is unlikely to happen, we have to avoid such
a case occurring at an earlier point: the sk_rmem_alloc could get
increased because of inserting more and more skbs into the errqueue
when calling __skb_complete_tx_timestamp(). This bad case would stop
the socket transmitting soon.

It is up to the application to read from the error queue frequently
enough and/or increase SO_RCVBUF.

Sure thing. If we test it without setting SOF_TIMESTAMPING_SOFTWARE on
the loopback, it will soon stop. That's the reason why I tried to add
the restriction just in case.

I don't follow at all.

That bit does not affect the core issue: that the application is not
clearing its error queue quickly enough.
 

quoted

quoted

Signed-off-by: Jason Xing <kernelxing@tencent.com>
---
 net/core/sock.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/net/core/sock.c b/net/core/sock.c
index fe87f9bd8f16..4bddd6f62e4f 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c

@@ -905,6 +905,10 @@ int sock_set_timestamping(struct sock *sk, int optname,
      if (val & ~SOF_TIMESTAMPING_MASK)
              return -EINVAL;

+     if (val & SOF_TIMESTAMPING_TX_RECORD_MASK &&
+         !(val & SOF_TIMESTAMPING_SOFTWARE))
+             return -EINVAL;
+

This breaks hardware timestamping

Yes, and sorry about that. I'll fix this.

As is I don't understand the purpose of this patch. Please do not
just resubmit with a change, but explain the problem and suggested
solution first.

quoted

quoted

      if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
          !(val & SOF_TIMESTAMPING_OPT_ID))
              return -EINVAL;
--
2.37.3