Re: git: feeb19201dca - main - tcp: improve consistency of KASSERTs in tcp_sack.c

> On 7. Aug 2025, at 00:29, Gleb Smirnoff <glebius@FreeBSD.org> wrote:
> 
> On Wed, Aug 06, 2025 at 08:30:30AM +0000, Michael Tuexen wrote:
> M>     When panicing, don't print the condition, which was violated,
> M>     but the condition which holds at the time of the panic.
> M>     
> M>     Reviewed by:            Nick Banks
> M>     MFC after:              1 week
> M>     Sponsored by:           Netflix, Inc.
> M>     Differential Revision:  https://reviews.freebsd.org/D51726
> M> ---
> M>  sys/netinet/tcp_sack.c | 30 +++++++++++++++++-------------
> M>  1 file changed, 17 insertions(+), 13 deletions(-)
> M> 
> M> diff --git a/sys/netinet/tcp_sack.c b/sys/netinet/tcp_sack.c
> M> index 4405098a8620..f48e60207cc2 100644
> M> --- a/sys/netinet/tcp_sack.c
> M> +++ b/sys/netinet/tcp_sack.c
> M> @@ -283,7 +283,7 @@ tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
> M>   INP_WLOCK_ASSERT(tptoinpcb(tp));
> M>  
> M>   /* Check arguments. */
> M> - KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("rcv_start <= rcv_end"));
> M> + KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("SEG_GT(rcv_start, rcv_end)"));
> M>  
> M>   if ((rcv_start == rcv_end) &&
> M>      (tp->rcv_numsacks >= 1) &&
> M> @@ -498,8 +498,8 @@ tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole)
> M>   tp->snd_numholes--;
> M>   atomic_subtract_int(&V_tcp_sack_globalholes, 1);
> M>  
> M> - KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0"));
> M> - KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0"));
> M> + KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes < 0"));
> M> + KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes < 0"));
> 
> IMHO, better just through MPASS() instead of self-repetitive KASSERT.
Up to now, tcp_sack.c does not use MPASS(). So I wanted to stay consistent..
I just go confused when looking at the panic messages from tcp_sack.c.
Some messages printed the condition which should hold, some the condition
which holds.

Best regards
Michael
> 
> -- 
> Gleb Smirnoff