tools/tcptop: rewrite using fentry/fexit for accurate recv statistics #5375
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The original tcptop uses kprobes on tcp_sendmsg and tcp_cleanup_rbuf to
trace TCP-level send and receive statistics. However, the current
receive side tracing via tcp_cleanup_rbuf can result in overcounting due
to the way `copied` is used:
int tcp_recvmsg(...) {
...
int copied = 0;
...
do {
...
tcp_cleanup_rbuf(sk, copied);
...
copied += used;
...
} while (len > 0);
tcp_cleanup_rbuf(sk, copied);
return copied;
...
}
Same data in "copied" might be taken into account several times
Fixing this requires tracing tcp_recvmsg entry/exit, which kretprobe
can't reliably handle due to maxactive limits. fentry/fexit is needed,
but only available on kernels >= 5.5 with BTF.
Move the old version to tools/old/, and introduce a new version using
fentry/fexit.
Reported-by: @sc07kvm <github.com/sc07kvm>
Signed-off-by: Ze Huang <[email protected]>
Replace kprobe/kretprobe with fentry/fexit hooks for tcp_sendmsg. Signed-off-by: Ze Huang <[email protected]>
Rename tcp_sendstat() to tcp_stat() and extend it to support receive-side statistics based on is_send flag. Signed-off-by: Ze Huang <[email protected]>
Replace kprobe on tcp_cleanup_rbuf with fentry/fexit on tcp_recvmsg. Adds a dedicated sock_recv map to track receive-side sockets. Signed-of-by: Ze Huang <[email protected]>
Sutter099
requested review from
brendangregg,
chenhengqi,
ekyooo and
yonghong-song
as code owners
Collaborator
|
Thanks! I checked the goal and also run the new tcptop.py and LGTM. |
sc07kvm
reviewed
Jul 29, 2025
| * we'd much prefer tracepoints once they are available. | ||
| */ | ||
| int kprobe__tcp_cleanup_rbuf(struct pt_regs *ctx, struct sock *sk, int copied) | ||
| KRETFUNC_PROBE(tcp_recvmsg, struct sock *sk, struct msghdr *msg, size_t len, int flags, int *addr_len, int ret) |
There was a problem hiding this comment.
Will only work with 5.19. Since the function signature has changed: torvalds/linux@ec09526
Author
There was a problem hiding this comment.
Indeed, only works for linux over v5.19. Change the comment from v5.5 to v5.19?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR improves the accuracy of TCP receive statistics in
tcptop.Previously,
tcptopused a kprobe ontcp_cleanup_rbufto track receive traffic. However, this function may be called multiple times in a singletcp_recvmsg()call, with an accumulatingcopiedparameter. This can lead to overcounting.This issue was reported by @sc07kvm:
#5352
To resolve this, I followed the suggestion from @yonghong-song:
As a result:
tcptop.pyhas been moved totools/old/fentry/fexithooks is introducedtcp_stat()with shared logictcp_recvmsgwith proper entry/exit coordination