linux의 strace 명령어(CLI)는 시스템 콜과 시그널에 대해 추적할 수 있는 명령어 입니다.

기본 사용법

print("hello world")
# 다양한 옵션을 가지고 실행한 python 프로그램
$> strace -tt -T -o hello.log ./
hello world

# 출력한 hello.log 파일을 보면 실제 수행된 system call을 확인 할 수 있습니다.
$> cat hello.log
16:15:23.942523 execve("./", ["./"], 0x7ffd453cab30 /* 119 vars */) = 0 <0.000605>
16:15:23.943465 brk(NULL)               = 0x5609554ff000 <0.000021>
16:15:23.943543 arch_prctl(0x3001 /* ARCH_??? */, 0x7fffeaff96a0) = -1 EINVAL (Invalid argument) <0.000020>
16:15:23.943723 access("/etc/", R_OK) = -1 ENOENT (No such file or directory) <0.000025>
16:15:23.943831 openat(AT_FDCWD, "/etc/", O_RDONLY|O_CLOEXEC) = 3 <0.000029>
16:15:23.990166 read(3, "\nprint(\"hello world\")\n", 4096) = 22 <0.000023>
16:15:23.992196 write(1, "hello world\n", 12) = 12 <0.000032>
16:15:23.996285 munmap(0x7f20e205a000, 16384) = 0 <0.000033>
16:15:23.996388 exit_group(0)           = ?
16:15:23.997530 +++ exited with 0 +++

strace help

$> strace --help
Usage: strace [-ACdffhikqqrtttTvVwxxyyzZ] [-I N] [-b execve] [-e EXPR]...
              [-a COLUMN] [-o FILE] [-s STRSIZE] [-X FORMAT] [-O OVERHEAD]
              [-S SORTBY] [-P PATH]... [-p PID]... [-U COLUMNS] [--seccomp-bpf]
              { -p PID | [-DDD] [-E VAR=VAL]... [-u USERNAME] PROG [ARGS] }
   or: strace -c[dfwzZ] [-I N] [-b execve] [-e EXPR]... [-O OVERHEAD]
              [-S SORTBY] [-P PATH]... [-p PID]... [-U COLUMNS] [--seccomp-bpf]
              { -p PID | [-DDD] [-E VAR=VAL]... [-u USERNAME] PROG [ARGS] }

  -e EXPR        a qualifying expression: OPTION=[!]all or OPTION=[!]VAL1[,VAL2]...
     options:    trace, abbrev, verbose, raw, signal, read, write, fault,
                 inject, status, quiet, kvm, decode-fds

  -E VAR=VAL, --env=VAR=VAL
                 put VAR=VAL in the environment for command
  -E VAR, --env=VAR
                 remove VAR from the environment for command
  -p PID, --attach=PID
                 trace process with process id PID, may be repeated
                 run command as USERNAME handling setuid and/or setgid
  --argv0=NAME   set PROG argv[0] to NAME

  -b execve, --detach-on=execve
                 detach on execve syscall
  -D, --daemonize[=grandchild]
                 run tracer process as a grandchild, not as a parent
  -DD, --daemonize=pgroup
                 run tracer process in a separate process group
  -DDD, --daemonize=session
                 run tracer process in a separate session
  -f, --follow-forks
                 follow forks
  -ff, --follow-forks --output-separately
                 follow forks with output into separate files
     1, anywhere:   no signals are blocked
     2, waiting:    fatal signals are blocked while decoding syscall (default)
     3, never:      fatal signals are always blocked (default if '-o FILE PROG')
     4, never_tstp: fatal signals and SIGTSTP (^Z) are always blocked
                    (useful to make 'strace -o FILE PROG' not stop on ^Z)

  -e trace=[!][?]{{SYSCALL|GROUP|all|/REGEX}[@64|@32|@x32]|none},
                 trace only specified syscalls.
     groups:     %clock, %creds, %desc, %file, %fstat, %fstatfs %ipc, %lstat,
                 %memory, %net, %process, %pure, %signal, %stat, %%stat,
                 %statfs, %%statfs
  -e signal=SET, --signal=SET
                 trace only the specified set of signals
                 print only the signals from SET
  -e status=SET, --status=SET
                 print only system calls with the return statuses in SET
     statuses:   successful, failed, unfinished, unavailable, detached
  -e trace-fds=SET, --trace-fds=SET
                 trace operations on file descriptors from SET
  -P PATH, --trace-path=PATH
                 trace accesses to PATH
  -z, --successful-only
                 print only syscalls that returned without an error code
  -Z, --failed-only
                 print only syscalls that returned with an error code

Output format:
  -a COLUMN, --columns=COLUMN
                 alignment COLUMN for printing syscall results (default 40)
  -e abbrev=SET, --abbrev=SET
                 abbreviate output for the syscalls in SET
  -e verbose=SET, --verbose=SET
                 dereference structures for the syscall in SET
  -e raw=SET, --raw=SET
                 print undecoded arguments for the syscalls in SET
  -e read=SET, --read=SET
                 dump the data read from the file descriptors in SET
  -e write=SET, --write=SET
                 dump the data written to the file descriptors in SET
  -e quiet=SET, --quiet=SET
                 suppress various informational messages
     messages:   attach, exit, path-resolution, personality, thread-execve
  -e kvm=vcpu, --kvm=vcpu
                 print exit reason of kvm vcpu
  -e decode-fds=SET, --decode-fds=SET
                 what kinds of file descriptor information details to decode
     details:    dev (device major/minor for block/char device files)
                 path (file path),
                 pidfd (associated PID for pidfds),
                 socket (protocol-specific information for socket descriptors),
                 signalfd (signal masks for signalfds)
  -i, --instruction-pointer
                 print instruction pointer at time of syscall
  -k, --stack-traces
                 obtain stack trace between each syscall
  -n, --syscall-number
                 print syscall number
  -o FILE, --output=FILE
                 send trace output to FILE instead of stderr
  -A, --output-append-mode
                 open the file provided in the -o option in append mode
                 output into separate files (by appending pid to file names)
  -q, --quiet=attach,personality
                 suppress messages about attaching, detaching, etc.
  -qq, --quiet=attach,personality,exit
                 suppress messages about process exit status as well.
  -qqq, --quiet=all
                 suppress all suppressible messages.
  -r, --relative-timestamps[=PRECISION]
                 print relative timestamp
     precision:  one of s, ms, us, ns; default is microseconds
  -s STRSIZE, --string-limit=STRSIZE
                 limit length of print strings to STRSIZE chars (default 32)
                 set the format of absolute timestamps
     format:     none, time, or unix; default is time
     precision:  one of s, ms, us, ns; default is seconds
  -t, --absolute-timestamps[=time]
                 print absolute timestamp
  -tt, --absolute-timestamps=[time,]us
                 print absolute timestamp with usecs
  -ttt, --absolute-timestamps=unix,us
                 print absolute UNIX time with usecs
  -T, --syscall-times[=PRECISION]
                 print time spent in each syscall
     precision:  one of s, ms, us, ns; default is microseconds
  -v, --no-abbrev
                 verbose mode: print entities unabbreviated
                 use hex instead of octal in escape sequences
  -x, --strings-in-hex=non-ascii
                 print non-ASCII strings in hex
  -xx, --strings-in-hex[=all]
                 print all strings in hex
  -X FORMAT, --const-print-style=FORMAT
                 set the FORMAT for printing of named constants and flags
     formats:    raw, abbrev, verbose
  -y, --decode-fds[=path]
                 print paths associated with file descriptor arguments
  -yy, --decode-fds=all
                 print all available information associated with file
                 descriptors in addition to paths
                 print PIDs in strace's namespace, too
  -Y, --decode-pids=comm
                 print command names associated with PIDs

  -c, --summary-only
                 count time, calls, and errors for each syscall and report
  -C, --summary  like -c, but also print the regular output
  -O OVERHEAD[UNIT], --summary-syscall-overhead=OVERHEAD[UNIT]
                 set overhead for tracing syscalls to OVERHEAD UNITs
     units:      one of s, ms, us, ns; default is microseconds
  -S SORTBY, --summary-sort-by=SORTBY
                 sort syscall counts by: time, min-time, max-time, avg-time,
                 calls, errors, name, nothing (default time)
  -U COLUMNS, --summary-columns=COLUMNS
                 show specific columns in the summary report: comma-separated
                 list of time-percent, total-time, min-time, max-time,
                 avg-time, calls, errors, name
                 (default time-percent,total-time,avg-time,calls,errors,name)
  -w, --summary-wall-clock
                 summarise syscall latency (default is system time)

Stop condition:
                 Detach all tracees after tracing LIMIT syscalls

  -e inject=SET[:error=ERRNO|:retval=VALUE][:signal=SIG][:syscall=SYSCALL]
                 perform syscall tampering for the syscalls in SET
     delay:      microseconds or NUMBER{s|ms|us|ns}
     when:       FIRST[..LAST][+[STEP]]
  -e fault=SET[:error=ERRNO][:when=WHEN], --fault=SET[:error=ERRNO][:when=WHEN]
                 synonym for -e inject with default ERRNO set to ENOSYS.

  -d, --debug    enable debug output to stderr
  -h, --help     print help message
  --seccomp-bpf  enable seccomp-bpf filtering
                 show strace tips, tricks, and tweaks on exit
     id:         non-negative integer or random; default is random
     format:     none, compact, full; default is compact
  -V, --version  print version


$> tldr strace


  Troubleshooting tool for tracing system calls.
  More information:

  - Start tracing a specific process by its PID:
    strace -p pid

  - Trace a process and filter output by system call:
    strace -p pid -e system_call_name

  - Count time, calls, and errors for each system call and report a summary on program exit:
    strace -p pid -c

  - Show the time spent in every system call:
    strace -p pid -T

  - Start tracing a program by executing it:
    strace program

  - Start tracing file operations of a program:
    strace -e trace=file program