This post appeared originally in our sysadvent series and has been moved here following the discontinuation of the sysadvent microsite

This post appeared originally in our sysadvent series and has been moved here following the discontinuation of the sysadvent microsite
There will come a time when you will find yourself asking “What the heck is that process doing?”. To uncover the mysteries behind the behaviour of a process, we have a tool called strace
.
The program strace
is very handy when you want to debug the execution of a program. It catches and states all the system calls performed called by a process. It will also catch and state any inter-process signals received by this process.
Let’s dive into some examples.
strace ls
..is the simple straight-forward way to use it. The output might look something like:
execve("/bin/ls", ["ls"], [/* 32 vars */]) = 0
brk(0) = 0x1117000
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fe7c5bc2000
access("/etc/ld.so.preload", R_OK) = -1 ENOENT (No such file or directory)
open("/etc/ld.so.cache", O_RDONLY|O_CLOEXEC) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=69243, ...}) = 0
mmap(NULL, 69243, PROT_READ, MAP_PRIVATE, 3, 0) = 0x7fe7c5bb1000
close(3) = 0
open("/lib64/libselinux.so.1", O_RDONLY|O_CLOEXEC) = 3
read(3, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\3\0>\0\1\0\0\0\240d\0\0\0\0\0\0"..., 832) = 832
fstat(3, {st_mode=S_IFREG|0755, st_size=142136, ...}) = 0
mmap(NULL, 2242712, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0x7fe7c5780000
mprotect(0x7fe7c57a1000, 2093056, PROT_NONE) = 0
The output looks a bit messy, but it can provide very useful information like which files does this program use, what is this program doing right now or why does this program not read the configuration file - does it even look for it?
You can run strace
on a process that’s already running. Use -p
, and provide the Process ID (PID
):
$ pidof urxvtd
772
$ strace -p 772
The output presented to you is similar to the example above. Notice that you can only trace a process you have access rights to. Multiple -p
options will also trace these processes with a limit of 32 processes that strace
can attach to.
Use -f
to trace child processes as they are created by currently traced processes. This is useful for debugging a program that spawns children. strace
will prepend the PID of the traced process output:
$ strace -f man strace
25470 execve("/usr/bin/man", ["man", "strace"], [/* 18 vars */]) = 0
...
strace -e open ls
The e
-flag, along with the call open
, displays only open system calls of the ls
command, which outputs something along these lines:
open("/lib64/libdl.so.2", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libattr.so.1", O_RDONLY|O_CLOEXEC) = 3
open("/lib64/libpthread.so.0", O_RDONLY|O_CLOEXEC) = 3
open("/proc/filesystems", O_RDONLY) = 3
open("/usr/lib/locale/locale-archive", O_RDONLY|O_CLOEXEC) = 3
ghost ghost.zip
+++ exited with 0 +++
With the above example, close to the end, you can see two things – ghost
and ghost.zip
, a folder and file, which is the actual output of the command ls
.
A few of the more handy options available after -e trace=
are:
What if you want to trace multiple system calls in one command though? The option -e trace=
can take a comma separated set of system calls as argument:
strace -e trace=open,read ls
Which outputs:
open("/lib64/libattr.so.1", O_RDONLY|O_CLOEXEC) = 3
read(3, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\3\0>\0\1\0\0\0\320\23\0\0\0\0\0\0"..., 832) = 832
open("/lib64/libpthread.so.0", O_RDONLY|O_CLOEXEC) = 3
read(3, "\177ELF\2\1\1\0\0\0\0\0\0\0\0\0\3\0>\0\1\0\0\0\340m\0\0\0\0\0\0"..., 832) = 832
open("/proc/filesystems", O_RDONLY) = 3
read(3, "nodev\tsysfs\nnodev\trootfs\nnodev\tr"..., 1024) = 356
read(3, "", 1024) = 0
open("/usr/lib/locale/locale-archive", O_RDONLY|O_CLOEXEC) = 3
ghost ghost.zip
+++ exited with 0 +++
The option -e trace=file
is particularly useful. The file
argument tells strace to trace all system calls that
takes a file name as an argument. Think of file
as a preset set of system calls.
If you’d like to write the output of strace
to a file, add a -o
:
strace -o ls.txt ls
When you print the contents of ls.txt
, you’ll see that it contains only the output from strace
, and not from the command ls
:
$ cat.ls.txt
...
write(1, "18.png\t\t Apps Declarations d"..., 164) = 164
write(1, "4238158452532.pdf bin\t Document"..., 175) = 175
close(1) = 0
munmap(0x7f993acb6000, 4096) = 0
close(2) = 0
exit_group(0) = ?
write(1, "18.png\t\t Apps Declarations d"..., 164) = 164
write(1, "4238158452532.pdf bin\t Document"..., 175) = 175
close(1) = 0
munmap(0x7f993acb6000, 4096) = 0
close(2) = 0
exit_group(0) = ?
+++ exited with 0 +++
If you add a -t
to the command, you’ll be able to add a timestamp to each printed line. You can add up to 3 -t
. The more you add, the more verbose and detailed the timestamp will be.
One -t
shows you seconds:
$ strace -t ls
...
11:52:20 close(2) = 0
11:52:20 exit_group(0) = ?
11:52:20 +++ exited with 0 +++
Adding a second -t
will display microseconds:
$ strace -tt ls
...
11:53:07.593382 close(2) = 0
11:53:07.593500 exit_group(0) = ?
11:53:07.593651 +++ exited with 0 +++
A third -t
will include the microseconds and the leading portion will be printed as the number of seconds since the epoch.
$ strace -ttt ls
...
1390042410.543785 close(2) = 0
1390042410.543906 exit_group(0) = ?
1390042410.544077 +++ exited with 0 +++
If you’d like to find out the execution time of each call, -r
will certainly do the trick:
$ strace -r ls
...
0.000105 close(2) = 0
0.000115 exit_group(0) = ?
0.000146 +++ exited with 0 +++
Would you like to have a summary of calls, time, and errors for each system call? -c
does this:
$ strace -c ls
ghost ghost.zip
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
100.00 0.000054 11 5 mprotect
0.00 0.000000 0 4 read
0.00 0.000000 0 1 write
0.00 0.000000 0 6 open
0.00 0.000000 0 9 close
0.00 0.000000 0 1 execve
0.00 0.000000 0 1 1 access
0.00 0.000000 0 3 brk
0.00 0.000000 0 2 ioctl
0.00 0.000000 0 2 munmap
0.00 0.000000 0 15 mmap2
0.00 0.000000 0 7 fstat64
0.00 0.000000 0 2 getdents64
0.00 0.000000 0 1 set_thread_area
0.00 0.000000 0 1 openat
------ ----------- ----------- --------- --------- ----------------
100.00 0.000054 60 1 total
This option is very useful when trying to find out why a program is running slow.
Open source i offentlig sektor - utmaningar, möjligheter och vägen framåt.
Denna artikel beskriver processen och lärdomarna från att släppa ett API som öppen källkod inom offentlig sektor. Projektet, som utvecklades för digitala nationella prov (“DNP”), visar hur öppen källkod kan stärka samarbete, transparens och innovation. Artikeln lyfter fram både möjligheter och utmaningar – från säkerhet och juridiska aspekter till kulturellt motstånd – och ger insikter för andra myndigheter som överväger liknande initiativ.
Slutsatsen är att öppen källkod ... [continue reading]