Page MenuHomeFreeBSD
Differential D34154

Disallow setting of ECN bits with setsockopt()
ClosedPublic
Actions

Authored by rscheff on Feb 3 2022, 10:24 AM.
Tags
None
Referenced Files
Unknown Object (File)
Sun, Feb 8, 2:45 PM
Unknown Object (File)
Sun, Feb 8, 10:27 AM
Unknown Object (File)
Sun, Feb 8, 4:15 AM
Unknown Object (File)
Sun, Feb 8, 12:18 AM
Unknown Object (File)
Sat, Feb 7, 8:02 PM
Unknown Object (File)
Sat, Feb 7, 4:46 PM
Unknown Object (File)
Tue, Feb 3, 3:52 PM
Unknown Object (File)
Mon, Feb 2, 12:07 AM
View All Files
Subscribers
debdrup
imp
melifaro

Details

Reviewers
tuexen
kbowling
glebius
rrs
rgrimes
• hselasky
debdrup
Group Reviewers
manpages
transport
Commits
rG3b0ee680507a: tcp: Prevent setting of ECN bits with setsockopt()
Summary

setsockopt() grants full access to the deprecated
TOS byte. Mask out the ECN codepoint, so that
only the DSCP portion can be adjusted.

Also update the man page examples to deprecate
the use of TOS, and provide DSCP examples instead, and
remove IPTOS_MINCOST to squat the ECT1 codepoint.

Test Plan
// Verifying that TCP will clear out the ECN codepoint
// when setsockopt(IP_TOS) tries to set them

--tolerance_usecs=2000000

// Set sysctl defaults
0.0 `sysctl net.inet.tcp.cc.algorithm=newreno`
+0.1 `sysctl net.inet.tcp.functions_default=freebsd`
+0.1 `sysctl net.inet.tcp.initcwnd_segments=10`
+0.1 `sysctl net.inet.tcp.ecn.enable=0`
+0.1 `sysctl net.inet.tcp.hostcache.purgenow=1`

// Test Server side

// Establish a regular TCP connection.
+0.0 `echo 'server: freebsd tcp dcsp session'`
+0.50 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0.01 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0.01 setsockopt(3, SOL_SOCKET, SO_LINGER, {onoff=1, linger=0}, 8) = 0
+0.01 bind(3, ..., ...) = 0
+0.01 listen(3, 1) = 0
+0.04 <[noecn] S  0:0(0) win 65535 <mss 1000>
+0.00 >[noecn] S.  0:0(0) ack 1 win 65535 <...>
+0.00 <[noecn]  .   1:1(0) ack 1 win 65535
+0.00 accept(3, ..., ...) = 4
+0.01 getsockopt(4, IPPROTO_IP, IP_TOS, [0x0], [4]) = 0
+0.01 setsockopt(4, IPPROTO_IP, IP_TOS, [0xff], 4) = 0
+0.01 getsockopt(4, IPPROTO_IP, IP_TOS, [0xfc], [4]) = 0
+0.01 write(4, ..., 1000) = 1000
+0.00 > (tos 0xfc) P.    1:1001(1000) ack 1 win 65535
+0.00 close(4) = 0
+0.00 > (tos 0xfc) R. 1001:1001(0) ack 1 win 0
+0.00 close(3) = 0

+0.1 `sysctl net.inet.tcp.functions_default=rack`

+0.0 `echo 'server: rack tcp dscp session'`
+0.50 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0.01 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0.01 setsockopt(3, SOL_SOCKET, SO_LINGER, {onoff=1, linger=0}, 8) = 0
+0.01 bind(3, ..., ...) = 0
+0.01 listen(3, 1) = 0
+0.04 <[noecn] S  0:0(0) win 65535 <mss 1000>
+0.00 >[noecn] S.  0:0(0) ack 1 win 65535 <...>
+0.00 <[noecn]  .   1:1(0) ack 1 win 65535
+0.00 accept(3, ..., ...) = 4
+0.01 getsockopt(4, IPPROTO_IP, IP_TOS, [0x0], [4]) = 0
+0.01 setsockopt(4, IPPROTO_IP, IP_TOS, [0xff], 4) = 0
+0.01 getsockopt(4, IPPROTO_IP, IP_TOS, [0xfc], [4]) = 0
+0.01 write(4, ..., 1000) = 1000
+0.00 > (tos 0xfc) P.    1:1001(1000) ack 1 win 65535
+0.00 close(4) = 0
+0.00 > (tos 0xfc) R. 1001:1001(0) ack 1 win 0
+0.00 close(3) = 0


`{
sysctl net.inet.tcp.functions_default=freebsd;
sysctl net.inet.tcp.cc.algorithm=newreno;
sysctl net.inet.tcp.ecn.enable=2;
}`

Diff Detail

Repository
rS FreeBSD src repository - subversion
Lint
Lint Passed
Unit
No Test Coverage
Build Status
Buildable 44268
Build 41156: arc lint + arc unit

Event Timeline

rscheff created this revision.Feb 3 2022, 10:24 AM
Herald added a reviewer: manpages. · View Herald TranscriptFeb 3 2022, 10:24 AM
Herald added a reviewer: transport. · View Herald Transcript
Herald added subscribers: melifaro, imp. · View Herald Transcript
rscheff requested review of this revision.Feb 3 2022, 10:24 AM
Harbormaster completed remote builds in B44268: Diff 102310.Feb 3 2022, 10:24 AM
• hselasky accepted this revision.Feb 3 2022, 10:42 AM

Looks good. Did you check sys/ofed for similar cases?

rscheff added a comment.Feb 3 2022, 11:40 AM

No, I did not. And I don't feel competent enough to commit something there, but I guess this may work:

diff --git a/sys/ofed/drivers/infiniband/core/ib_cma.c b/sys/ofed/drivers/infiniband/core/ib_cma.c
index 586524ac88bb..6d32dadbbef8 100644
--- a/sys/ofed/drivers/infiniband/core/ib_cma.c
+++ b/sys/ofed/drivers/infiniband/core/ib_cma.c
@@ -2497,7 +2497,7 @@ void rdma_set_service_type(struct rdma_cm_id *id, int tos)
        struct rdma_id_private *id_priv;

        id_priv = container_of(id, struct rdma_id_private, id);
-       id_priv->tos = (u8) tos;
+       id_priv->tos = (u8) tos & ~IPTOS_ECN_MASK;
 }
 EXPORT_SYMBOL(rdma_set_service_type);

diff --git a/sys/ofed/drivers/infiniband/core/ib_uverbs_std_types_flow_action.c b/sys/ofed/drivers/infiniband/core/ib_uverbs_std_types_flow_action.c
index 89cd0b313b72..68917c482868 100644
--- a/sys/ofed/drivers/infiniband/core/ib_uverbs_std_types_flow_action.c
+++ b/sys/ofed/drivers/infiniband/core/ib_uverbs_std_types_flow_action.c
@@ -123,7 +123,7 @@ static int parse_esp_ip(enum ib_flow_spec_type proto,
                .src_ip = cpu_to_be32(0xffffffffUL),
                .dst_ip = cpu_to_be32(0xffffffffUL),
                .proto = 0xff,
-               .tos = 0xff,
+               .tos = 0xfc,
                .ttl = 0xff,
                .flags = 0xff,
        };
@@ -132,7 +132,7 @@ static int parse_esp_ip(enum ib_flow_spec_type proto,
                           0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
                .dst_ip = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
                           0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
-               .flow_label = cpu_to_be32(0xffffffffUL),
+               .flow_label = cpu_to_be32(~(IPTOS_ECN_MASK << 20)),
                .next_hdr = 0xff,
                .traffic_class = 0xff,
                .hop_limit = 0xff,
debdrup accepted this revision.Feb 3 2022, 11:49 AM
debdrup added a subscriber: debdrup.

mdoc changes look good to me

• hselasky added a comment.Feb 3 2022, 12:06 PM

@rscheff : OK, I'll take the ibcore part, if any.

tuexen added a comment.Feb 3 2022, 2:18 PM

You are actually changing the semantic of the IP_TOS option. If you mask out the ECN bits, how am I supposed to implement a UDP based protocol supporting ECN?

sys/netinet/tcp_stacks/rack.c
20478

Why is this change needed? If you already ensure that inp_ip_tos is correctly set, this is not necessary. Or am I missing something?

rscheff added a comment.Feb 3 2022, 2:34 PM

I did not consider UDP, actually.

Then let me revoke the code change in ip_ctloutput, and just go with the clearing of the current pkthdr ecn codepoint from D34148 plus perhaps something similar to the rack_set_sockopt for the base stack, such than regular sessions do not end up getting IP ECN codepoints, as D34148 only clears the ecn codepoints on (TCP-) ECN enabled sessions, but leaves non-ECN sessions unperturbed.

Does that sound reasonable?

sys/netinet/tcp_stacks/rack.c
20478

I am not sure why this is in both places - other than RACK maybe caching the ip-> header rather than building up a new one from the inp-> struct?

tuexen added a comment.Feb 3 2022, 2:36 PM
In D34154#772560, @rscheff wrote:

I did not consider UDP, actually.

Then let me revoke the code change in ip_ctloutput, and just go with the clearing of the current pkthdr ecn codepoint from D34148 plus perhaps something similar to the rack_set_sockopt for the base stack, such than regular sessions do not end up getting IP ECN codepoints, as D34148 only clears the ecn codepoints on (TCP-) ECN enabled sessions, but leaves non-ECN sessions unperturbed.

Does that sound reasonable?

Yes. In SCTP we do https://cgit.freebsd.org/src/tree/sys/netinet/sctp_output.c#n4069

rscheff updated this revision to Diff 102324.Feb 3 2022, 3:14 PM
  • only mask out ECN codepoint for TCP, leave other protos unperturbed
Harbormaster completed remote builds in B44273: Diff 102324.Feb 3 2022, 3:14 PM
rscheff edited the test plan for this revision. (Show Details)Feb 3 2022, 3:18 PM
rscheff updated this revision to Diff 102329.Feb 3 2022, 4:01 PM
  • only mask out ECN codepoint for TCP, leave other protos unperturbed
  • no need to parse the optval again
Harbormaster completed remote builds in B44275: Diff 102329.Feb 3 2022, 4:01 PM
tuexen added inline comments.Feb 3 2022, 4:02 PM
share/man/man4/ip.4
96

Isn't the old text still correct?

Maybe you want to add a sentence that setting the ECN bits on a socket using a transport protocol implementing ECN has no effect.

106

Maybe keeping the variable name 'tos' makes sense?

tuexen added inline comments.Feb 3 2022, 4:03 PM
sys/netinet/tcp_usrreq.c
1764 ↗(On Diff #102329)

This change looks good to me.

rscheff updated this revision to Diff 102332.Feb 3 2022, 4:12 PM
  • update the ip.4 man page, to discourage the use of a tos field,
Harbormaster completed remote builds in B44277: Diff 102332.Feb 3 2022, 4:12 PM
tuexen added inline comments.Feb 3 2022, 4:22 PM
share/man/man4/ip.4
96

OK.

106

I think keeping the variable name as tos makes sense. You are providing the TOS byte, not only the DSCP codepoint.

rscheff updated this revision to Diff 102333.Feb 3 2022, 4:43 PM
rscheff marked an inline comment as done.
  • restore "tos" variable name
Harbormaster completed remote builds in B44278: Diff 102333.Feb 3 2022, 4:43 PM
rscheff marked an inline comment as done.Feb 3 2022, 4:53 PM
rscheff added inline comments.
share/man/man4/ip.4
106

The examples at least should move ahead and not use historic field names for variable names IMHO.

tuexen accepted this revision.Feb 3 2022, 5:40 PM
This revision is now accepted and ready to land.Feb 3 2022, 5:40 PM
Closed by commit rG3b0ee680507a: tcp: Prevent setting of ECN bits with setsockopt() (authored by rscheff). · Explain WhyFeb 3 2022, 7:12 PM
This revision was automatically updated to reflect the committed changes.
rscheff added a commit: rG3b0ee680507a: tcp: Prevent setting of ECN bits with setsockopt().

Revision Contents
Changeset List

PathSize
share/
man/
man4/
8 lines
sys/
netinet/
2 lines
2 lines
tcp_stacks/
2 lines

Diff 102310

View Options

share/man/man4/ip.4

Loading...
View Options

sys/netinet/ip.h

Loading...
View Options

sys/netinet/ip_output.c

Loading...
View Options

sys/netinet/tcp_stacks/rack.c

Loading...