Skip to content

Commit

Permalink
add the tcpstates tool
Browse files Browse the repository at this point in the history
  • Loading branch information
brendangregg committed Mar 21, 2018
1 parent 130b594 commit bbd9acd
Show file tree
Hide file tree
Showing 4 changed files with 514 additions and 0 deletions.
1 change: 1 addition & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -141,6 +141,7 @@ pair of .c and .py files, and some are directories of files.
- tools/[tcpconnlat](tools/tcpconnlat.py): Trace TCP active connection latency (connect()). [Examples](tools/tcpconnlat_example.txt).
- tools/[tcplife](tools/tcplife.py): Trace TCP sessions and summarize lifespan. [Examples](tools/tcplife_example.txt).
- tools/[tcpretrans](tools/tcpretrans.py): Trace TCP retransmits and TLPs. [Examples](tools/tcpretrans_example.txt).
- tools/[tcpstates](tools/tcpstates.py): Trace TCP session state changes with durations. [Examples](tools/tcpstates_example.txt).
- tools/[tcpsubnet](tools/tcpsubnet.py): Summarize and aggregate TCP send by subnet. [Examples](tools/tcpsubnet_example.txt).
- tools/[tcptop](tools/tcptop.py): Summarize TCP send/recv throughput by host. Top for TCP. [Examples](tools/tcptop_example.txt).
- tools/[tcptracer](tools/tcptracer.py): Trace TCP established connections (connect(), accept(), close()). [Examples](tools/tcptracer_example.txt).
Expand Down
128 changes: 128 additions & 0 deletions man/man8/tcpstates.8
Original file line number Diff line number Diff line change
@@ -0,0 +1,128 @@
.TH tcpstates 8 "2018-03-20" "USER COMMANDS"
.SH NAME
tcpstates \- Trace TCP session state changes with durations. Uses Linux eBPF/bcc.
.SH SYNOPSIS
.B tcpstates [\-h] [\-T] [\-t] [\-w] [\-s] [\-D PORTS] [\-L PORTS]
.SH DESCRIPTION
This tool traces TCP session state changes while tracing, and prints details
including the duration in each state. This can help explain the latency of
TCP connections: whether the time is spent in the ESTABLISHED state (data
transfer), or initialization state (SYN_SENT), etc.

This tool works using the sock:inet_sock_set_state tracepoint, which was
added to Linux 4.16. Linux 4.16 also included extra state transitions so that
all TCP transitions could be observed by this tracepoint.

Only TCP state changes are traced, so it is expected that the
overhead of this tool is much lower than typical send/receive tracing.

Since this uses BPF, only the root user can use this tool.
.SH REQUIREMENTS
CONFIG_BPF and bcc, and the sock:inet_sock_set_state tracepoint.
.SH OPTIONS
.TP
\-h
Print usage message.
.TP
\-s
Comma separated values output (parseable).
.TP
\-t
Include a timestamp column (seconds).
.TP
\-T
Include a time column (HH:MM:SS).
.TP
\-w
Wide column output (fits IPv6 addresses).
.TP
\-L PORTS
Comma-separated list of local ports to trace (filtered in-kernel).
.TP
\-D PORTS
Comma-separated list of destination ports to trace (filtered in-kernel).
.SH EXAMPLES
.TP
Trace all TCP sessions, and show all state changes:
#
.B tcpstates
.TP
Include a timestamp column, and wide column output:
#
.B tcpstates \-tw
.TP
Trace connections to local ports 80 and 81 only:
#
.B tcpstates \-L 80,81
.TP
Trace connections to remote port 80 only:
#
.B tcpstates \-D 80
.SH FIELDS
.TP
TIME
Time of the change, in HH:MM:SS format.
.TP
TIME(s)
Time of the change, in seconds.
.TP
C-PID
The current on-CPU process ID. This may show the process that owns the TCP
session if the state change executes in synchronous process context, else it
is likely to show the kernel (asynchronous state change).
.TP
C-COMM
The current on-CPU process name. This may show the process that owns the TCP
session if the state change executes in synchronous process context, else it
is likely to show the kernel (asynchronous state change).
.TP
IP
IP address family (4 or 6)
.TP
LADDR
Local IP address.
.TP
DADDR
Remote IP address.
.TP
LPORT
Local port.
.TP
DPORT
Destination port.
.TP
OLDSTATE
Previous TCP state.
.TP
NEWSTATE
New TCP state.
.TP
MS
Duration of this state.
.SH OVERHEAD
This traces the kernel TCP set state function, which should be called much
less often than send/receive tracing, and therefore have lower overhead. The
overhead of the tool is relative to the rate of new TCP sessions: if this is
high, over 10,000 per second, then there may be noticeable overhead just to
print out 10k lines of formatted output per second.

You can find out the rate of new TCP sessions using "sar \-n TCP 1", and
adding the active/s and passive/s columns.

As always, test and understand this tools overhead for your types of
workloads before production use.
.SH SOURCE
This is from bcc.
.IP
https://github.com/iovisor/bcc
.PP
Also look in the bcc distribution for a companion _examples.txt file containing
example usage, output, and commentary for this tool.
.SH OS
Linux
.SH STABILITY
Unstable - in development.
.SH AUTHOR
Brendan Gregg
.SH SEE ALSO
tcpaccept(8), tcpconnect(8), tcptop(8), tcplife(8)
Loading

0 comments on commit bbd9acd

Please sign in to comment.