2020-07-30 13:48:05 |
Amir Tzin |
bug |
|
|
added bug |
2020-07-30 14:00:08 |
Ubuntu Kernel Bot |
linux (Ubuntu): status |
New |
Incomplete |
|
2020-07-30 21:03:09 |
Jeff Lane |
linux (Ubuntu): status |
Incomplete |
In Progress |
|
2020-07-30 21:03:12 |
Jeff Lane |
linux (Ubuntu): importance |
Undecided |
Medium |
|
2020-07-30 21:03:14 |
Jeff Lane |
linux (Ubuntu): assignee |
|
Jeff Lane (bladernr) |
|
2020-07-30 21:04:45 |
Jeff Lane |
nominated for series |
|
Ubuntu Focal |
|
2020-07-30 21:04:45 |
Jeff Lane |
bug task added |
|
linux (Ubuntu Focal) |
|
2020-07-30 21:40:08 |
Jeff Lane |
linux (Ubuntu Focal): status |
New |
In Progress |
|
2020-07-30 21:40:11 |
Jeff Lane |
linux (Ubuntu Focal): importance |
Undecided |
Medium |
|
2020-07-30 21:40:14 |
Jeff Lane |
linux (Ubuntu Focal): assignee |
|
Jeff Lane (bladernr) |
|
2020-07-31 01:27:24 |
Jeff Lane |
description |
The bellow upstream commit fixes a bug in Ktls feature.
It is applied cleanly above the ubuntu-focal tree and passed basic sanity testing.
We would like it to be backported ubuntu-focal
Thanks,
Amir
commit 41b14fb8724d5a4b382a63cb4a1a61880347ccb8
Author: Tariq Toukan <tariqt@mellanox.com>
Date: Mon Jun 22 23:26:04 2020 +0300
net: Do not clear the sock TX queue in sk_set_socket()
Clearing the sock TX queue in sk_set_socket() might cause unexpected
out-of-order transmit when called from sock_orphan(), as outstanding
packets can pick a different TX queue and bypass the ones already queued.
This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.
Remove the call to sk_tx_queue_clear() in sk_set_socket(), and add it
explicitly only where needed.
Fixes: e022f0b4a03f ("net: Introduce sk_tx_queue_mapping")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net> |
[IMPACT]
Clearing the sock TX queue in sk_set_socket() might cause unexpected
out-of-order transmit when called from sock_orphan(), as outstanding
packets can pick a different TX queue and bypass the ones already queued.
This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.
Remove the call to sk_tx_queue_clear() in sk_set_socket(), and add it
explicitly only where needed.
[FIXES]
e022f0b4a03f "net: Introduce sk_tx_queue_mapping"
This cleanly cherry picks into 5.4 from 5.8.
It can be checked out in my branch here:
https://git.launchpad.net/~bladernr/ubuntu/+source/linux/+git/focal 1889625-mxl-ktls-bugfix
[REGRESSION RISK]
[TEST] |
|
2020-08-13 13:08:49 |
Amir Tzin |
description |
[IMPACT]
Clearing the sock TX queue in sk_set_socket() might cause unexpected
out-of-order transmit when called from sock_orphan(), as outstanding
packets can pick a different TX queue and bypass the ones already queued.
This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.
Remove the call to sk_tx_queue_clear() in sk_set_socket(), and add it
explicitly only where needed.
[FIXES]
e022f0b4a03f "net: Introduce sk_tx_queue_mapping"
This cleanly cherry picks into 5.4 from 5.8.
It can be checked out in my branch here:
https://git.launchpad.net/~bladernr/ubuntu/+source/linux/+git/focal 1889625-mxl-ktls-bugfix
[REGRESSION RISK]
[TEST] |
[IMPACT]
Clearing the sock TX queue in sk_set_socket() might cause unexpected
out-of-order transmit when called from sock_orphan(), as outstanding
packets can pick a different TX queue and bypass the ones already queued.
This is undesired in general. More specifically, it breaks the in-order
scheduling property guarantee for device-offloaded TLS sockets.
Remove the call to sk_tx_queue_clear() in sk_set_socket(), and add it
explicitly only where needed.
[FIXES]
e022f0b4a03f "net: Introduce sk_tx_queue_mapping"
This cleanly cherry picks into 5.4 from 5.8.
It can be checked out in my branch here:
https://git.launchpad.net/~bladernr/ubuntu/+source/linux/+git/focal 1889625-mxl-ktls-bugfix
[REGRESSION RISK]
low!
[TEST]
reproducing the bug is not trivial. in general terms:
nic: ConnectX6-dx with crypto enabled
send intense encrypted tcp traffic with tls offload between many clients and one server.
* clients may run on the same machine.
* clients continuously opens and closes connection to server
at some point decryption errores might arise on some of the clients. |
|
2020-09-17 19:56:07 |
Jeff Lane |
linux (Ubuntu): status |
In Progress |
Invalid |
|
2020-09-17 19:56:11 |
Jeff Lane |
linux (Ubuntu Focal): status |
In Progress |
Invalid |
|