| nfstrace(8) | System Manager's Manual | nfstrace(8) |
nfstrace - NFS and CIFS tracing/monitoring/capturing/analyzing tool
nfstrace [ -m drain|live|dump|stat ] [
-i interface ] [ -f filter ]
[ -s 1..65535 ] [ -t Milliseconds ] [ -b
MBytes ] [ -p promisc ]
[ -d in|out|inout ] [ -a path#opt1,opt2=val,... ]
[ -I path ]
[ -O path ] [ -C 'shell command' ] [ -D
MBytes ] [ -E interfaces|plugins ]
[ -M 1..4000 ] [ -Q 1..65535 ] [ -T
true|false ] [ -Z username ] [ -v 0|1|2
]
nfstrace performs live Ethernet 1 Gbps – 10 Gbps packets capturing and helps to determine NFS and CIFS procedures in raw network traffic. Furthermore, it performs filtration, dumping, compression, statistical analysis, visualization and provides the API for custom pluggable analysis modules.
nfstrace captures raw packets from an Ethernet interface using libpcap interface to Linux (LSF) or FreeBSD (BPF) implementations. At the moment it is assumed that libpcap delivers correct TCP and UDP packets. Assembling of IP packets from ethernet frames and IP packets defragmentation are performed in the operating system's kernel.
The application has been tested on the workstations with integrated 1 Gbps NICs (Ethernet 1000baseT/Full).
Currently nfstrace supports the following protocols:
Ethernet > IPv4 | IPv6 > UDP | TCP > NFSv3 | NFSv4 | NFSv4.1 |
CIFSv1 | CIFSv2
nfstrace can operate in four different modes:
- online analysis (--mode=live): performs online capturing, filtration and live analysis of detected NFS procedures using a pluggable analysis module or prints out them to stdout (-T or --trace options);
- online dumping (--mode=dump): performs online traffic capturing, filtration and dumping to the output file (specified with -O or --ofile options);
- offline analysis (--mode=stat): performs offline filtration of the .pcap file that contains previously captured traces and performs analysis using a pluggable analysis module or prints found NFS procedures to stdout (-T or --trace options);
- offline dumping (--mode=drain): performs a reading of traffic from the .pcap file (specified with -I or --ifile options), filtration, dumping to the output .pcap file (specified with -O or --ofile options) and removing all the packets that are not related to NFS procedures.
nfstrace uses libpcap file format for input and output files so any external tool (e.g. Wireshark) can be used in order to inspect filtered traces.
Since nfstrace internally uses libpcap that provides a portable interface to the native system API for capturing network traffic, filtration is performed in the operating system's kernel. nfstrace provides a special option (-f or --filtration) for specifying custom filters in BPF syntax (see pcap-filter(7)).
The default BPF filter in nfstrace is 'port 2049 or port 445', which means that each packet that is delivered to user-space from the kernel satisfies the following conditions: it has IPv4 or IPv6 header and it has TCP and UDP header with source or destination port number equals to 2049 (default NFS port) or 445 (default CIFS port).
The reality is that this filter is very heavy and support of IPv6 is experimental, so if you want to reach faster filtration of IPv4-only traffic we suggest to use the following BPF filter:
nfstrace can usually be run without arguments: in this case default arguments will be used.
Operation Breakdown (OB) analyzer calculates average frequency of NFS and CIFS procedures and computes standard deviation of latency.
$ nfstrace -a libbreakdown.so -h
nfstrace 0.4.0 (Release)
built on Linux-3.16.1-1-generic
by C++ compiler GNU 4.9.1
Usage: ./nfstrace [OPTIONS]...
Usage example:
$ nfstrace -m stat -a libreakdown.so
Breakdown analyzer produces .dat file in the current directory for each detected NFS session that can be visualized using nst.sh and breakdown_nfsv3.plt or breakdown_nfsv4.plt (according to NFS version).
$ nst.sh -a breakdown_nfsv4.plt -d . -p 'breakdown_10.6.137.47:903*.dat'
Watch plugin mimics old nfswatch utility: it monitors NFS and CIFS traffic and displays it in terminal using ncurses. It supports NFSv3, NFSv4, NFSv41, CIFSv1 and CIFSv2.
By default watch plugin will update its screen every second, you can specify another timeout in milliseconds:
$ nfstrace -a libwatch.so#2000
JSON analyzer calculates a total amount of each supported application protocol operation. It accepts TCP-connections on particular TCP-endpoint (host:port), sends a respective JSON to the TCP-client and closes connection. Suggested to be used in live mode.
Available options
Example of use
Start nfstrace in one console:
$ sudo nfstrace -i eth0 -a
analyzers/libjson.so#host=localhost
...
Make a TCP-request to nfstrace in another console to fetch current statistics:
$ telnet localhost 8888
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
{
The following command demonstrates available options of the application and plugged analysis modules (attached with --analysis or -a options). Note that you can pass more than one module here.
nfstrace --help --analysis=libjson.so
The following command will run nfstrace in online analysis mode (specified with --mode or -m options) without a pluggable analysis module. It will capture NFS and CIFS traffic transferred over TCP or UDP with source or destination port number equals to 2049 and will simply print them out to stdout (-T or --trace options). Capturing ends when nfstrace receives SIGINT (Control-C).
Note that capturing from network interface requires superuser privileges.
The following command demonstrates running nfstrace in online analysis mode. Just like in the previous example it will capture NFS and CIFS traffic transferred over TCP or UDP with source or destination port number equals to 2049 and then it will perform Operation Breakdown analysis using pluggable analysis module libbreakdown.so.
The following example demonstrates running nfstrace in online dumping and offline analysis modes.
At first nfstrace will capture NFS traffic transferred over TCP or UDP with source or destination port number equals to 2049 and will dump captured packets to dump.pcap file (specified with --ofile or -O options).
At the second run nfstrace will perform offline Operation Breakdown analysis using pluggable analysis module libbreakdown.so.
# Analyse dump.pcap using libbreakdown.so
nfstrace -m stat --ifile=dump.pcap -a libbreakdown.so
The following example demonstrates running nfstrace in online dumping and offline analysis modes. Since dump file can easily exhaust disk space, compression makes sense.
At first nfstrace will capture NFS traffic transferred over TCP or UDP with source or destination port number equals to 2049 and will dump captured packets to dump.pcap file.
Note that compression is done by the external tool (executed in script passed with --command or -C options) and it will be executed when capturing is done. The output file can be inspected using some external tool.
At the second run nfstrace will perform offline analysis. Again, the external tool (bzcat in this example) is used in order to decompress previously saved dump. nfstrace will read stdin (note the -I – option) and perform offline analysis using Operation Breakdown analyzer.
# Extract dump.pcap from dump.pcap.bz2 to stdin.
# Read stdin and analyze data with libbreakdown.so module.
bzcat dump.pcap.bz2 | nfstrace -m stat -I - -a libbreakdown.so
This example is similar to the previous one except one thing: output dump file can be very huge and cause problems in some situations, so nfstrace provides the ability to split it into parts.
At first nfstrace will be invoked in online dumping mode. Everything is similar to the previous example except -D (--dump-size) option: it specifies the size limit in MBytes, so dump file will be split according to this value.
At the second run nfstrace will perform offline analysis of captured packets using Operation Breakdown analyzer.
Please note that only the first dump file has the pcap header.
# get list of parts in the right order:
# dump.pcap.bz2
# dump.pcap-1.bz2
# dump.pcap-2.bz2
parts=$(ls dump.pcap*.bz2 | sort -n -t - -k 2)
# Extract dump.pcap from main dump.pcap.bz2 and parts to
stdin.
# Read stdin and analyze data with libbreakdown.so module.
bzcat $parts | nfstrace --mode=stat -I - --analysis=libbreakdown.so
This example demonstrates the ability to plot graphical representation of data collected by Operation Breakdown analyzer.
nst.sh is a shell script that collects data generated by analyzers and passes it to gnuplot script specified with -a option.
breakdown.plt is a gnuplot script that understands output data format of Operation Breakdown analyzer and generates .png files with plots.
Note that gnuplot must be installed.
# Generate plot according to *.dat files generated by
# libbreakdown.so analyzer.
nst.sh -a breakdown.plt -d . -p 'breakdown*.dat' -v
Any problems, bugs, questions, desirable enhancements, etc. should be sent to <nfstrace@epam.com>
Vitali Adamenka <Vitali_Adamenka@epam.com>
Yauheni Azaranka <Yauheni_Azaranka@epam.com>
Alexey Costroma <Alexey_Costroma@epam.com>
Dzianis Huznou <Dzianis_Huznou@epam.com>
Pavel Karneliuk <Pavel_Karneliuk@epam.com>
Andrey Kuznetsov <Andrey_Kuznetsov@epam.com>
Mikhail Litvinets <Mikhail_Litvinets@epam.com>
Ilya Storozhilov <Ilya_Storozhilov@epam.com>
| 2022-03-18 | version 0.4.3 |