NET: Make ts_recent_stamp unsigned
PATCH 4/5 NET: Make ts_recent_stamp unsigned
The get_seconds() function returns an unsigned long. To prevent incorrect comparison results between values saved in ts_recent_stamp and later invocations of get_seconds(), change the type of ts_recent_stamp to match the return type of get_seconds().
---
include/linux/tcp.h | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
include/linux/tcp.h +1 -1
@@ -206,7 +206,7 @@ struct tcp_sack_block {struct tcp_options_received {/* PAWS/RTTM data */ -long ts_recent_stamp;/* Time we stored ts_recent (for aging) */ +unsigned long ts_recent_stamp;/* Time we stored ts_recent (for aging) */ u32 ts_recent;/* Time stamp to echo next */ u32 rcv_tsval;/* Time stamp value */ u32 rcv_tsecr;/* Time stamp echo reply */
From: Chuck Lever <[email protected]>
Date: Tue, 23 Oct 2007 11:44:28 -0400
Quoted 7 lines
The
get_seconds()function returns an unsigned long. To prevent incorrect comparison results between values saved ints_recent_stampand later invocations ofget_seconds(), change the type ofts_recent_stampto match the return type ofget_seconds().Signed-off-by Chuck Lever <[email protected]>Cc <[email protected]>
I see two potential problems with this patch:
1) If you update struct tcp_options_received you should also update struct tcp_timewait_sock similarly.
The fact that you missed this suggests that you didn't grep the tree to see how else this variable is used and this makes me extra concerned about this patch's correctness.
2) There are computations in the TCP stack using this member that probably care about the signedness, such as:
net/ipv4/tcp_ipv4.c: get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {include/net/tcp.h: if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)include/net/tcp.h: if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
We should make sure we understand what is expected here, and why it would still be correct after making both ts_recent_stamp members unsigned.
Thanks.
Quoted 11 lines · David Miller
From: Chuck Lever <[email protected]>
Date: Tue, 23 Oct 2007 11:44:28 -0400Quoted 7 lines
The
get_seconds()function returns an unsigned long. To prevent incorrect comparison results between values saved ints_recent_stampand later invocations ofget_seconds(), change the type ofts_recent_stampto match the return type ofget_seconds().Signed-off-by Chuck Lever <[email protected]>Cc <[email protected]>
Quoted 8 lines
I see two potential problems with this patch:
1) If you update struct
tcp_options_receivedyou should also update structtcp_timewait_socksimilarly.The fact that you missed this suggests that you didn't grep the tree to see how else this variable is used and this makes me extra concerned about this patch's correctness.
Perhaps the result of wishful thinking on my part. I was hoping for a small and self-contained change.
Quoted 10 lines
2) There are computations in the TCP stack using this member that probably care about the signedness, such as:
net/ipv4/tcp_ipv4.c:get_seconds()-tcptw->tw_ts_recent_stamp> 1))) {include/net/tcp.h: if (get_seconds()>=rx_opt->ts_recent_stamp+TCP_PAWS_24DAYS)include/net/tcp.h: if (get_seconds()>=rx_opt->ts_recent_stamp+TCP_PAWS_24DAYS)We should make sure we understand what is expected here, and why it would still be correct after making both
ts_recent_stampmembers unsigned.
Agreed.
I wonder how one could construct a series of mixed case time stamp comparisons *on purpose* (and without documentation of this assumption) that produces consistently correct results.
From the invocations of get_seconds() that I sampled, the design of these comparisons seems to assume that both sides of the comparison are non-negative. However, they do not seem to account for time crossing zero.
begin:vcard
fn:Chuck Lever
n:Lever;Chuck
org:Oracle Corporation;Corporate Architecture: Linux Projects Group adr:;;1015 Granger Avenue;Ann Arbor;MI;48104;USA email;internet:chuck dot lever at nospam oracle dot com title:Principal Member of Staff
tel;work:+1 248 614 5091
x-mozilla-html:FALSE
version:2.1
end:vcard