1PTRACE(2) Linux Programmer's Manual PTRACE(2)
2
6 ptrace - process trace
7
9 #include <sys/ptrace.h>
10 long ptrace(enum __ptrace_request request, pid_t pid,
12 void *addr, void *data);
13
15 The ptrace() system call provides a means by which one process (the
16 "tracer") may observe and control the execution of another process (the
17 "tracee"), and examine and change the tracee's memory and registers.
18 It is primarily used to implement breakpoint debugging and system call
19 tracing.
20 A tracee first needs to be attached to the tracer. Attachment and sub‐
22 sequent commands are per thread: in a multithreaded process, every
23 thread can be individually attached to a (potentially different)
24 tracer, or left not attached and thus not debugged. Therefore,
25 "tracee" always means "(one) thread", never "a (possibly multithreaded)
26 process". Ptrace commands are always sent to a specific tracee using a
27 call of the form
28 ptrace(PTRACE_foo, pid, ...)
30 where pid is the thread ID of the corresponding Linux thread.
32 (Note that in this page, a "multithreaded process" means a thread group
34 consisting of threads created using the clone(2) CLONE_THREAD flag.)
35 A process can initiate a trace by calling fork(2) and having the
37 resulting child do a PTRACE_TRACEME, followed (typically) by an
38 execve(2). Alternatively, one process may commence tracing another
39 process using PTRACE_ATTACH or PTRACE_SEIZE.
40 While being traced, the tracee will stop each time a signal is deliv‐
42 ered, even if the signal is being ignored. (An exception is SIGKILL,
43 which has its usual effect.) The tracer will be notified at its next
44 call to waitpid(2) (or one of the related "wait" system calls); that
45 call will return a status value containing information that indicates
46 the cause of the stop in the tracee. While the tracee is stopped, the
47 tracer can use various ptrace requests to inspect and modify the
48 tracee. The tracer then causes the tracee to continue, optionally
49 ignoring the delivered signal (or even delivering a different signal
50 instead).
51 If the PTRACE_O_TRACEEXEC option is not in effect, all successful calls
53 to execve(2) by the traced process will cause it to be sent a SIGTRAP
54 signal, giving the parent a chance to gain control before the new pro‐
55 gram begins execution.
56 When the tracer is finished tracing, it can cause the tracee to con‐
58 tinue executing in a normal, untraced mode via PTRACE_DETACH.
59 The value of request determines the action to be performed:
61 PTRACE_TRACEME
63 Indicate that this process is to be traced by its parent. A
64 process probably shouldn't make this request if its parent isn't
65 expecting to trace it. (pid, addr, and data are ignored.)
66 The PTRACE_TRACEME request is used only by the tracee; the
68 remaining requests are used only by the tracer. In the follow‐
69 ing requests, pid specifies the thread ID of the tracee to be
70 acted on. For requests other than PTRACE_ATTACH, PTRACE_SEIZE,
71 PTRACE_INTERRUPT, and PTRACE_KILL, the tracee must be stopped.
72 PTRACE_PEEKTEXT, PTRACE_PEEKDATA
74 Read a word at the address addr in the tracee's memory, return‐
75 ing the word as the result of the ptrace() call. Linux does not
76 have separate text and data address spaces, so these two
77 requests are currently equivalent. (data is ignored; but see
78 NOTES.)
79 PTRACE_PEEKUSER
81 Read a word at offset addr in the tracee's USER area, which
82 holds the registers and other information about the process (see
83 <sys/user.h>). The word is returned as the result of the
84 ptrace() call. Typically, the offset must be word-aligned,
85 though this might vary by architecture. See NOTES. (data is
86 ignored; but see NOTES.)
87 PTRACE_POKETEXT, PTRACE_POKEDATA
89 Copy the word data to the address addr in the tracee's memory.
90 As for PTRACE_PEEKTEXT and PTRACE_PEEKDATA, these two requests
91 are currently equivalent.
92 PTRACE_POKEUSER
94 Copy the word data to offset addr in the tracee's USER area. As
95 for PTRACE_PEEKUSER, the offset must typically be word-aligned.
96 In order to maintain the integrity of the kernel, some modifica‐
97 tions to the USER area are disallowed.
98 PTRACE_GETREGS, PTRACE_GETFPREGS
100 Copy the tracee's general-purpose or floating-point registers,
101 respectively, to the address data in the tracer. See
102 <sys/user.h> for information on the format of this data. (addr
103 is ignored.) Note that SPARC systems have the meaning of data
104 and addr reversed; that is, data is ignored and the registers
105 are copied to the address addr. PTRACE_GETREGS and PTRACE_GETF‐
106 PREGS are not present on all architectures.
107 PTRACE_GETREGSET (since Linux 2.6.34)
109 Read the tracee's registers. addr specifies, in an architec‐
110 ture-dependent way, the type of registers to be read. NT_PRSTA‐
111 TUS (with numerical value 1) usually results in reading of gen‐
112 eral-purpose registers. If the CPU has, for example, floating-
113 point and/or vector registers, they can be retrieved by setting
114 addr to the corresponding NT_foo constant. data points to a
115 struct iovec, which describes the destination buffer's location
116 and length. On return, the kernel modifies iov.len to indicate
117 the actual number of bytes returned.
118 PTRACE_SETREGS, PTRACE_SETFPREGS
120 Modify the tracee's general-purpose or floating-point registers,
121 respectively, from the address data in the tracer. As for
122 PTRACE_POKEUSER, some general-purpose register modifications may
123 be disallowed. (addr is ignored.) Note that SPARC systems have
124 the meaning of data and addr reversed; that is, data is ignored
125 and the registers are copied from the address addr.
126 PTRACE_SETREGS and PTRACE_SETFPREGS are not present on all
127 architectures.
128 PTRACE_SETREGSET (since Linux 2.6.34)
130 Modify the tracee's registers. The meaning of addr and data is
131 analogous to PTRACE_GETREGSET.
132 PTRACE_GETSIGINFO (since Linux 2.3.99-pre6)
134 Retrieve information about the signal that caused the stop.
135 Copy a siginfo_t structure (see sigaction(2)) from the tracee to
136 the address data in the tracer. (addr is ignored.)
137 PTRACE_SETSIGINFO (since Linux 2.3.99-pre6)
139 Set signal information: copy a siginfo_t structure from the
140 address data in the tracer to the tracee. This will affect only
141 signals that would normally be delivered to the tracee and were
142 caught by the tracer. It may be difficult to tell these normal
143 signals from synthetic signals generated by ptrace() itself.
144 (addr is ignored.)
145 PTRACE_PEEKSIGINFO (since Linux 3.10)
147 Retrieve siginfo_t structures without removing signals from a
148 queue. addr points to a ptrace_peeksiginfo_args structure that
149 specifies the ordinal position from which copying of signals
150 should start, and the number of signals to copy. siginfo_t
151 structures are copied into the buffer pointed to by data. The
152 return value contains the number of copied signals (zero indi‐
153 cates that there is no signal corresponding to the specified
154 ordinal position). Within the returned siginfo structures, the
155 si_code field includes information (__SI_CHLD, __SI_FAULT, etc.)
156 that are not otherwise exposed to user space.
157 struct ptrace_peeksiginfo_args {
159 u64 off; /* Ordinal position in queue at which
160 to start copying signals */
161 u32 flags; /* PTRACE_PEEKSIGINFO_SHARED or 0 */
162 s32 nr; /* Number of signals to copy */
163 };
164 Currently, there is only one flag, PTRACE_PEEKSIGINFO_SHARED,
166 for dumping signals from the process-wide signal queue. If this
167 flag is not set, signals are read from the per-thread queue of
168 the specified thread.
169 PTRACE_GETSIGMASK (since Linux 3.11)
171 Place a copy of the mask of blocked signals (see sigprocmask(2))
172 in the buffer pointed to by data, which should be a pointer to a
173 buffer of type sigset_t. The addr argument contains the size of
174 the buffer pointed to by data (i.e., sizeof(sigset_t)).
175 PTRACE_SETSIGMASK (since Linux 3.11)
177 Change the mask of blocked signals (see sigprocmask(2)) to the
178 value specified in the buffer pointed to by data, which should
179 be a pointer to a buffer of type sigset_t. The addr argument
180 contains the size of the buffer pointed to by data (i.e.,
181 sizeof(sigset_t)).
182 PTRACE_SETOPTIONS (since Linux 2.4.6; see BUGS for caveats)
184 Set ptrace options from data. (addr is ignored.) data is
185 interpreted as a bit mask of options, which are specified by the
186 following flags:
187 PTRACE_O_EXITKILL (since Linux 3.8)
189 Send a SIGKILL signal to the tracee if the tracer exits.
190 This option is useful for ptrace jailers that want to
191 ensure that tracees can never escape the tracer's con‐
192 trol.
193 PTRACE_O_TRACECLONE (since Linux 2.5.46)
195 Stop the tracee at the next clone(2) and automatically
196 start tracing the newly cloned process, which will start
197 with a SIGSTOP, or PTRACE_EVENT_STOP if PTRACE_SEIZE was
198 used. A waitpid(2) by the tracer will return a status
199 value such that
200 status>>8 == (SIGTRAP | (PTRACE_EVENT_CLONE<<8))
202 The PID of the new process can be retrieved with
204 PTRACE_GETEVENTMSG.
205 This option may not catch clone(2) calls in all cases.
207 If the tracee calls clone(2) with the CLONE_VFORK flag,
208 PTRACE_EVENT_VFORK will be delivered instead if
209 PTRACE_O_TRACEVFORK is set; otherwise if the tracee calls
210 clone(2) with the exit signal set to SIGCHLD,
211 PTRACE_EVENT_FORK will be delivered if PTRACE_O_TRACEFORK
212 is set.
213 PTRACE_O_TRACEEXEC (since Linux 2.5.46)
215 Stop the tracee at the next execve(2). A waitpid(2) by
216 the tracer will return a status value such that
217 status>>8 == (SIGTRAP | (PTRACE_EVENT_EXEC<<8))
219 If the execing thread is not a thread group leader, the
221 thread ID is reset to thread group leader's ID before
222 this stop. Since Linux 3.0, the former thread ID can be
223 retrieved with PTRACE_GETEVENTMSG.
224 PTRACE_O_TRACEEXIT (since Linux 2.5.60)
226 Stop the tracee at exit. A waitpid(2) by the tracer will
227 return a status value such that
228 status>>8 == (SIGTRAP | (PTRACE_EVENT_EXIT<<8))
230 The tracee's exit status can be retrieved with
232 PTRACE_GETEVENTMSG.
233 The tracee is stopped early during process exit, when
235 registers are still available, allowing the tracer to see
236 where the exit occurred, whereas the normal exit notifi‐
237 cation is done after the process is finished exiting.
238 Even though context is available, the tracer cannot pre‐
239 vent the exit from happening at this point.
240 PTRACE_O_TRACEFORK (since Linux 2.5.46)
242 Stop the tracee at the next fork(2) and automatically
243 start tracing the newly forked process, which will start
244 with a SIGSTOP, or PTRACE_EVENT_STOP if PTRACE_SEIZE was
245 used. A waitpid(2) by the tracer will return a status
246 value such that
247 status>>8 == (SIGTRAP | (PTRACE_EVENT_FORK<<8))
249 The PID of the new process can be retrieved with
251 PTRACE_GETEVENTMSG.
252 PTRACE_O_TRACESYSGOOD (since Linux 2.4.6)
254 When delivering system call traps, set bit 7 in the sig‐
255 nal number (i.e., deliver SIGTRAP|0x80). This makes it
256 easy for the tracer to distinguish normal traps from
257 those caused by a system call. (PTRACE_O_TRACESYSGOOD
258 may not work on all architectures.)
259 PTRACE_O_TRACEVFORK (since Linux 2.5.46)
261 Stop the tracee at the next vfork(2) and automatically
262 start tracing the newly vforked process, which will start
263 with a SIGSTOP, or PTRACE_EVENT_STOP if PTRACE_SEIZE was
264 used. A waitpid(2) by the tracer will return a status
265 value such that
266 status>>8 == (SIGTRAP | (PTRACE_EVENT_VFORK<<8))
268 The PID of the new process can be retrieved with
270 PTRACE_GETEVENTMSG.
271 PTRACE_O_TRACEVFORKDONE (since Linux 2.5.60)
273 Stop the tracee at the completion of the next vfork(2).
274 A waitpid(2) by the tracer will return a status value
275 such that
276 status>>8 == (SIGTRAP | (PTRACE_EVENT_VFORK_DONE<<8))
278 The PID of the new process can (since Linux 2.6.18) be
280 retrieved with PTRACE_GETEVENTMSG.
281 PTRACE_O_TRACESECCOMP (since Linux 3.5)
283 Stop the tracee when a seccomp(2) SECCOMP_RET_TRACE rule
284 is triggered. A waitpid(2) by the tracer will return a
285 status value such that
286 status>>8 == (SIGTRAP | (PTRACE_EVENT_SECCOMP<<8))
288 While this triggers a PTRACE_EVENT stop, it is similar to
290 a syscall-enter-stop. For details, see the note on
291 PTRACE_EVENT_SECCOMP below. The seccomp event message
292 data (from the SECCOMP_RET_DATA portion of the seccomp
293 filter rule) can be retrieved with PTRACE_GETEVENTMSG.
294 PTRACE_O_SUSPEND_SECCOMP (since Linux 4.3)
296 Suspend the tracee's seccomp protections. This applies
297 regardless of mode, and can be used when the tracee has
298 not yet installed seccomp filters. That is, a valid use
299 case is to suspend a tracee's seccomp protections before
300 they are installed by the tracee, let the tracee install
301 the filters, and then clear this flag when the filters
302 should be resumed. Setting this option requires that the
303 tracer have the CAP_SYS_ADMIN capability, not have any
304 seccomp protections installed, and not have PTRACE_O_SUS‐
305 PEND_SECCOMP set on itself.
306 PTRACE_GETEVENTMSG (since Linux 2.5.46)
308 Retrieve a message (as an unsigned long) about the ptrace event
309 that just happened, placing it at the address data in the
310 tracer. For PTRACE_EVENT_EXIT, this is the tracee's exit sta‐
311 tus. For PTRACE_EVENT_FORK, PTRACE_EVENT_VFORK,
312 PTRACE_EVENT_VFORK_DONE, and PTRACE_EVENT_CLONE, this is the PID
313 of the new process. For PTRACE_EVENT_SECCOMP, this is the sec‐
314 comp(2) filter's SECCOMP_RET_DATA associated with the triggered
315 rule. (addr is ignored.)
316 PTRACE_CONT
318 Restart the stopped tracee process. If data is nonzero, it is
319 interpreted as the number of a signal to be delivered to the
320 tracee; otherwise, no signal is delivered. Thus, for example,
321 the tracer can control whether a signal sent to the tracee is
322 delivered or not. (addr is ignored.)
323 PTRACE_SYSCALL, PTRACE_SINGLESTEP
325 Restart the stopped tracee as for PTRACE_CONT, but arrange for
326 the tracee to be stopped at the next entry to or exit from a
327 system call, or after execution of a single instruction, respec‐
328 tively. (The tracee will also, as usual, be stopped upon
329 receipt of a signal.) From the tracer's perspective, the tracee
330 will appear to have been stopped by receipt of a SIGTRAP. So,
331 for PTRACE_SYSCALL, for example, the idea is to inspect the
332 arguments to the system call at the first stop, then do another
333 PTRACE_SYSCALL and inspect the return value of the system call
334 at the second stop. The data argument is treated as for
335 PTRACE_CONT. (addr is ignored.)
336 PTRACE_SYSEMU, PTRACE_SYSEMU_SINGLESTEP (since Linux 2.6.14)
338 For PTRACE_SYSEMU, continue and stop on entry to the next system
339 call, which will not be executed. See the documentation on
340 syscall-stops below. For PTRACE_SYSEMU_SINGLESTEP, do the same
341 but also singlestep if not a system call. This call is used by
342 programs like User Mode Linux that want to emulate all the
343 tracee's system calls. The data argument is treated as for
344 PTRACE_CONT. The addr argument is ignored. These requests are
345 currently supported only on x86.
346 PTRACE_LISTEN (since Linux 3.4)
348 Restart the stopped tracee, but prevent it from executing. The
349 resulting state of the tracee is similar to a process which has
350 been stopped by a SIGSTOP (or other stopping signal). See the
351 "group-stop" subsection for additional information. PTRACE_LIS‐
352 TEN works only on tracees attached by PTRACE_SEIZE.
353 PTRACE_KILL
355 Send the tracee a SIGKILL to terminate it. (addr and data are
356 ignored.)
357 This operation is deprecated; do not use it! Instead, send a
359 SIGKILL directly using kill(2) or tgkill(2). The problem with
360 PTRACE_KILL is that it requires the tracee to be in signal-
361 delivery-stop, otherwise it may not work (i.e., may complete
362 successfully but won't kill the tracee). By contrast, sending a
363 SIGKILL directly has no such limitation.
364 PTRACE_INTERRUPT (since Linux 3.4)
366 Stop a tracee. If the tracee is running or sleeping in kernel
367 space and PTRACE_SYSCALL is in effect, the system call is inter‐
368 rupted and syscall-exit-stop is reported. (The interrupted sys‐
369 tem call is restarted when the tracee is restarted.) If the
370 tracee was already stopped by a signal and PTRACE_LISTEN was
371 sent to it, the tracee stops with PTRACE_EVENT_STOP and WSTOP‐
372 SIG(status) returns the stop signal. If any other ptrace-stop
373 is generated at the same time (for example, if a signal is sent
374 to the tracee), this ptrace-stop happens. If none of the above
375 applies (for example, if the tracee is running in user space),
376 it stops with PTRACE_EVENT_STOP with WSTOPSIG(status) == SIG‐
377 TRAP. PTRACE_INTERRUPT only works on tracees attached by
378 PTRACE_SEIZE.
379 PTRACE_ATTACH
381 Attach to the process specified in pid, making it a tracee of
382 the calling process. The tracee is sent a SIGSTOP, but will not
383 necessarily have stopped by the completion of this call; use
384 waitpid(2) to wait for the tracee to stop. See the "Attaching
385 and detaching" subsection for additional information. (addr and
386 data are ignored.)
387 Permission to perform a PTRACE_ATTACH is governed by a ptrace
389 access mode PTRACE_MODE_ATTACH_REALCREDS check; see below.
390 PTRACE_SEIZE (since Linux 3.4)
392 Attach to the process specified in pid, making it a tracee of
393 the calling process. Unlike PTRACE_ATTACH, PTRACE_SEIZE does
394 not stop the process. Group-stops are reported as
395 PTRACE_EVENT_STOP and WSTOPSIG(status) returns the stop signal.
396 Automatically attached children stop with PTRACE_EVENT_STOP and
397 WSTOPSIG(status) returns SIGTRAP instead of having SIGSTOP sig‐
398 nal delivered to them. execve(2) does not deliver an extra SIG‐
399 TRAP. Only a PTRACE_SEIZEd process can accept PTRACE_INTERRUPT
400 and PTRACE_LISTEN commands. The "seized" behavior just
401 described is inherited by children that are automatically
402 attached using PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, and
403 PTRACE_O_TRACECLONE. addr must be zero. data contains a bit
404 mask of ptrace options to activate immediately.
405 Permission to perform a PTRACE_SEIZE is governed by a ptrace
407 access mode PTRACE_MODE_ATTACH_REALCREDS check; see below.
408 PTRACE_SECCOMP_GET_FILTER (since Linux 4.4)
410 This operation allows the tracer to dump the tracee's classic
411 BPF filters.
412 addr is an integer specifying the index of the filter to be
414 dumped. The most recently installed filter has the index 0. If
415 addr is greater than the number of installed filters, the opera‐
416 tion fails with the error ENOENT.
417 data is either a pointer to a struct sock_filter array that is
419 large enough to store the BPF program, or NULL if the program is
420 not to be stored.
421 Upon success, the return value is the number of instructions in
423 the BPF program. If data was NULL, then this return value can
424 be used to correctly size the struct sock_filter array passed in
425 a subsequent call.
426 This operation fails with the error EACCESS if the caller does
428 not have the CAP_SYS_ADMIN capability or if the caller is in
429 strict or filter seccomp mode. If the filter referred to by
430 addr is not a classic BPF filter, the operation fails with the
431 error EMEDIUMTYPE.
432 This operation is available if the kernel was configured with
434 both the CONFIG_SECCOMP_FILTER and the CONFIG_CHECKPOINT_RESTORE
435 options.
436 PTRACE_DETACH
438 Restart the stopped tracee as for PTRACE_CONT, but first detach
439 from it. Under Linux, a tracee can be detached in this way
440 regardless of which method was used to initiate tracing. (addr
441 is ignored.)
442 PTRACE_GET_THREAD_AREA (since Linux 2.6.0)
444 This operation performs a similar task to get_thread_area(2).
445 It reads the TLS entry in the GDT whose index is given in addr,
446 placing a copy of the entry into the struct user_desc pointed to
447 by data. (By contrast with get_thread_area(2), the entry_number
448 of the struct user_desc is ignored.)
449 PTRACE_SET_THREAD_AREA (since Linux 2.6.0)
451 This operation performs a similar task to set_thread_area(2).
452 It sets the TLS entry in the GDT whose index is given in addr,
453 assigning it the data supplied in the struct user_desc pointed
454 to by data. (By contrast with set_thread_area(2), the
455 entry_number of the struct user_desc is ignored; in other words,
456 this ptrace operation can't be used to allocate a free TLS
457 entry.)
458 Death under ptrace
460 When a (possibly multithreaded) process receives a killing signal (one
461 whose disposition is set to SIG_DFL and whose default action is to kill
462 the process), all threads exit. Tracees report their death to their
463 tracer(s). Notification of this event is delivered via waitpid(2).
464 Note that the killing signal will first cause signal-delivery-stop (on
466 one tracee only), and only after it is injected by the tracer (or after
467 it was dispatched to a thread which isn't traced), will death from the
468 signal happen on all tracees within a multithreaded process. (The term
469 "signal-delivery-stop" is explained below.)
470 SIGKILL does not generate signal-delivery-stop and therefore the tracer
472 can't suppress it. SIGKILL kills even within system calls (syscall-
473 exit-stop is not generated prior to death by SIGKILL). The net effect
474 is that SIGKILL always kills the process (all its threads), even if
475 some threads of the process are ptraced.
476 When the tracee calls _exit(2), it reports its death to its tracer.
478 Other threads are not affected.
479 When any thread executes exit_group(2), every tracee in its thread
481 group reports its death to its tracer.
482 If the PTRACE_O_TRACEEXIT option is on, PTRACE_EVENT_EXIT will happen
484 before actual death. This applies to exits via exit(2), exit_group(2),
485 and signal deaths (except SIGKILL, depending on the kernel version; see
486 BUGS below), and when threads are torn down on execve(2) in a multi‐
487 threaded process.
488 The tracer cannot assume that the ptrace-stopped tracee exists. There
490 are many scenarios when the tracee may die while stopped (such as
491 SIGKILL). Therefore, the tracer must be prepared to handle an ESRCH
492 error on any ptrace operation. Unfortunately, the same error is
493 returned if the tracee exists but is not ptrace-stopped (for commands
494 which require a stopped tracee), or if it is not traced by the process
495 which issued the ptrace call. The tracer needs to keep track of the
496 stopped/running state of the tracee, and interpret ESRCH as "tracee
497 died unexpectedly" only if it knows that the tracee has been observed
498 to enter ptrace-stop. Note that there is no guarantee that wait‐
499 pid(WNOHANG) will reliably report the tracee's death status if a ptrace
500 operation returned ESRCH. waitpid(WNOHANG) may return 0 instead. In
501 other words, the tracee may be "not yet fully dead", but already refus‐
502 ing ptrace requests.
503 The tracer can't assume that the tracee always ends its life by report‐
505 ing WIFEXITED(status) or WIFSIGNALED(status); there are cases where
506 this does not occur. For example, if a thread other than thread group
507 leader does an execve(2), it disappears; its PID will never be seen
508 again, and any subsequent ptrace stops will be reported under the
509 thread group leader's PID.
510 Stopped states
512 A tracee can be in two states: running or stopped. For the purposes of
513 ptrace, a tracee which is blocked in a system call (such as read(2),
514 pause(2), etc.) is nevertheless considered to be running, even if the
515 tracee is blocked for a long time. The state of the tracee after
516 PTRACE_LISTEN is somewhat of a gray area: it is not in any ptrace-stop
517 (ptrace commands won't work on it, and it will deliver waitpid(2) noti‐
518 fications), but it also may be considered "stopped" because it is not
519 executing instructions (is not scheduled), and if it was in group-stop
520 before PTRACE_LISTEN, it will not respond to signals until SIGCONT is
521 received.
522 There are many kinds of states when the tracee is stopped, and in
524 ptrace discussions they are often conflated. Therefore, it is impor‐
525 tant to use precise terms.
526 In this manual page, any stopped state in which the tracee is ready to
528 accept ptrace commands from the tracer is called ptrace-stop. Ptrace-
529 stops can be further subdivided into signal-delivery-stop, group-stop,
530 syscall-stop, PTRACE_EVENT stops, and so on. These stopped states are
531 described in detail below.
532 When the running tracee enters ptrace-stop, it notifies its tracer
534 using waitpid(2) (or one of the other "wait" system calls). Most of
535 this manual page assumes that the tracer waits with:
536 pid = waitpid(pid_or_minus_1, &status, __WALL);
538 Ptrace-stopped tracees are reported as returns with pid greater than 0
540 and WIFSTOPPED(status) true.
541 The __WALL flag does not include the WSTOPPED and WEXITED flags, but
543 implies their functionality.
544 Setting the WCONTINUED flag when calling waitpid(2) is not recommended:
546 the "continued" state is per-process and consuming it can confuse the
547 real parent of the tracee.
548 Use of the WNOHANG flag may cause waitpid(2) to return 0 ("no wait
550 results available yet") even if the tracer knows there should be a
551 notification. Example:
552 errno = 0;
554 ptrace(PTRACE_CONT, pid, 0L, 0L);
555 if (errno == ESRCH) {
556 /* tracee is dead */
557 r = waitpid(tracee, &status, __WALL | WNOHANG);
558 /* r can still be 0 here! */
559 }
560 The following kinds of ptrace-stops exist: signal-delivery-stops,
562 group-stops, PTRACE_EVENT stops, syscall-stops. They all are reported
563 by waitpid(2) with WIFSTOPPED(status) true. They may be differentiated
564 by examining the value status>>8, and if there is ambiguity in that
565 value, by querying PTRACE_GETSIGINFO. (Note: the WSTOPSIG(status)
566 macro can't be used to perform this examination, because it returns the
567 value (status>>8) & 0xff.)
568 Signal-delivery-stop
570 When a (possibly multithreaded) process receives any signal except
571 SIGKILL, the kernel selects an arbitrary thread which handles the sig‐
572 nal. (If the signal is generated with tgkill(2), the target thread can
573 be explicitly selected by the caller.) If the selected thread is
574 traced, it enters signal-delivery-stop. At this point, the signal is
575 not yet delivered to the process, and can be suppressed by the tracer.
576 If the tracer doesn't suppress the signal, it passes the signal to the
577 tracee in the next ptrace restart request. This second step of signal
578 delivery is called signal injection in this manual page. Note that if
579 the signal is blocked, signal-delivery-stop doesn't happen until the
580 signal is unblocked, with the usual exception that SIGSTOP can't be
581 blocked.
582 Signal-delivery-stop is observed by the tracer as waitpid(2) returning
584 with WIFSTOPPED(status) true, with the signal returned by WSTOPSIG(sta‐
585 tus). If the signal is SIGTRAP, this may be a different kind of
586 ptrace-stop; see the "Syscall-stops" and "execve" sections below for
587 details. If WSTOPSIG(status) returns a stopping signal, this may be a
588 group-stop; see below.
589 Signal injection and suppression
591 After signal-delivery-stop is observed by the tracer, the tracer should
592 restart the tracee with the call
593 ptrace(PTRACE_restart, pid, 0, sig)
595 where PTRACE_restart is one of the restarting ptrace requests. If sig
597 is 0, then a signal is not delivered. Otherwise, the signal sig is
598 delivered. This operation is called signal injection in this manual
599 page, to distinguish it from signal-delivery-stop.
600 The sig value may be different from the WSTOPSIG(status) value: the
602 tracer can cause a different signal to be injected.
603 Note that a suppressed signal still causes system calls to return pre‐
605 maturely. In this case, system calls will be restarted: the tracer
606 will observe the tracee to reexecute the interrupted system call (or
607 restart_syscall(2) system call for a few system calls which use a dif‐
608 ferent mechanism for restarting) if the tracer uses PTRACE_SYSCALL.
609 Even system calls (such as poll(2)) which are not restartable after
610 signal are restarted after signal is suppressed; however, kernel bugs
611 exist which cause some system calls to fail with EINTR even though no
612 observable signal is injected to the tracee.
613 Restarting ptrace commands issued in ptrace-stops other than signal-
615 delivery-stop are not guaranteed to inject a signal, even if sig is
616 nonzero. No error is reported; a nonzero sig may simply be ignored.
617 Ptrace users should not try to "create a new signal" this way: use
618 tgkill(2) instead.
619 The fact that signal injection requests may be ignored when restarting
621 the tracee after ptrace stops that are not signal-delivery-stops is a
622 cause of confusion among ptrace users. One typical scenario is that
623 the tracer observes group-stop, mistakes it for signal-delivery-stop,
624 restarts the tracee with
625 ptrace(PTRACE_restart, pid, 0, stopsig)
627 with the intention of injecting stopsig, but stopsig gets ignored and
629 the tracee continues to run.
630 The SIGCONT signal has a side effect of waking up (all threads of) a
632 group-stopped process. This side effect happens before signal-deliv‐
633 ery-stop. The tracer can't suppress this side effect (it can only sup‐
634 press signal injection, which only causes the SIGCONT handler to not be
635 executed in the tracee, if such a handler is installed). In fact, wak‐
636 ing up from group-stop may be followed by signal-delivery-stop for sig‐
637 nal(s) other than SIGCONT, if they were pending when SIGCONT was deliv‐
638 ered. In other words, SIGCONT may be not the first signal observed by
639 the tracee after it was sent.
640 Stopping signals cause (all threads of) a process to enter group-stop.
642 This side effect happens after signal injection, and therefore can be
643 suppressed by the tracer.
644 In Linux 2.4 and earlier, the SIGSTOP signal can't be injected.
646 PTRACE_GETSIGINFO can be used to retrieve a siginfo_t structure which
648 corresponds to the delivered signal. PTRACE_SETSIGINFO may be used to
649 modify it. If PTRACE_SETSIGINFO has been used to alter siginfo_t, the
650 si_signo field and the sig parameter in the restarting command must
651 match, otherwise the result is undefined.
652 Group-stop
654 When a (possibly multithreaded) process receives a stopping signal, all
655 threads stop. If some threads are traced, they enter a group-stop.
656 Note that the stopping signal will first cause signal-delivery-stop (on
657 one tracee only), and only after it is injected by the tracer (or after
658 it was dispatched to a thread which isn't traced), will group-stop be
659 initiated on all tracees within the multithreaded process. As usual,
660 every tracee reports its group-stop separately to the corresponding
661 tracer.
662 Group-stop is observed by the tracer as waitpid(2) returning with WIF‐
664 STOPPED(status) true, with the stopping signal available via WSTOP‐
665 SIG(status). The same result is returned by some other classes of
666 ptrace-stops, therefore the recommended practice is to perform the call
667 ptrace(PTRACE_GETSIGINFO, pid, 0, &siginfo)
669 The call can be avoided if the signal is not SIGSTOP, SIGTSTP, SIGTTIN,
671 or SIGTTOU; only these four signals are stopping signals. If the
672 tracer sees something else, it can't be a group-stop. Otherwise, the
673 tracer needs to call PTRACE_GETSIGINFO. If PTRACE_GETSIGINFO fails
674 with EINVAL, then it is definitely a group-stop. (Other failure codes
675 are possible, such as ESRCH ("no such process") if a SIGKILL killed the
676 tracee.)
677 If tracee was attached using PTRACE_SEIZE, group-stop is indicated by
679 PTRACE_EVENT_STOP: status>>16 == PTRACE_EVENT_STOP. This allows detec‐
680 tion of group-stops without requiring an extra PTRACE_GETSIGINFO call.
681 As of Linux 2.6.38, after the tracer sees the tracee ptrace-stop and
683 until it restarts or kills it, the tracee will not run, and will not
684 send notifications (except SIGKILL death) to the tracer, even if the
685 tracer enters into another waitpid(2) call.
686 The kernel behavior described in the previous paragraph causes a prob‐
688 lem with transparent handling of stopping signals. If the tracer
689 restarts the tracee after group-stop, the stopping signal is effec‐
690 tively ignored—the tracee doesn't remain stopped, it runs. If the
691 tracer doesn't restart the tracee before entering into the next wait‐
692 pid(2), future SIGCONT signals will not be reported to the tracer; this
693 would cause the SIGCONT signals to have no effect on the tracee.
694 Since Linux 3.4, there is a method to overcome this problem: instead of
696 PTRACE_CONT, a PTRACE_LISTEN command can be used to restart a tracee in
697 a way where it does not execute, but waits for a new event which it can
698 report via waitpid(2) (such as when it is restarted by a SIGCONT).
699 PTRACE_EVENT stops
701 If the tracer sets PTRACE_O_TRACE_* options, the tracee will enter
702 ptrace-stops called PTRACE_EVENT stops.
703 PTRACE_EVENT stops are observed by the tracer as waitpid(2) returning
705 with WIFSTOPPED(status), and WSTOPSIG(status) returns SIGTRAP. An
706 additional bit is set in the higher byte of the status word: the value
707 status>>8 will be
708 (SIGTRAP | PTRACE_EVENT_foo << 8).
710 The following events exist:
712 PTRACE_EVENT_VFORK
714 Stop before return from vfork(2) or clone(2) with the
715 CLONE_VFORK flag. When the tracee is continued after this stop,
716 it will wait for child to exit/exec before continuing its execu‐
717 tion (in other words, the usual behavior on vfork(2)).
718 PTRACE_EVENT_FORK
720 Stop before return from fork(2) or clone(2) with the exit signal
721 set to SIGCHLD.
722 PTRACE_EVENT_CLONE
724 Stop before return from clone(2).
725 PTRACE_EVENT_VFORK_DONE
727 Stop before return from vfork(2) or clone(2) with the
728 CLONE_VFORK flag, but after the child unblocked this tracee by
729 exiting or execing.
730 For all four stops described above, the stop occurs in the parent
732 (i.e., the tracee), not in the newly created thread.
733 PTRACE_GETEVENTMSG can be used to retrieve the new thread's ID.
734 PTRACE_EVENT_EXEC
736 Stop before return from execve(2). Since Linux 3.0,
737 PTRACE_GETEVENTMSG returns the former thread ID.
738 PTRACE_EVENT_EXIT
740 Stop before exit (including death from exit_group(2)), signal
741 death, or exit caused by execve(2) in a multithreaded process.
742 PTRACE_GETEVENTMSG returns the exit status. Registers can be
743 examined (unlike when "real" exit happens). The tracee is still
744 alive; it needs to be PTRACE_CONTed or PTRACE_DETACHed to finish
745 exiting.
746 PTRACE_EVENT_STOP
748 Stop induced by PTRACE_INTERRUPT command, or group-stop, or ini‐
749 tial ptrace-stop when a new child is attached (only if attached
750 using PTRACE_SEIZE).
751 PTRACE_EVENT_SECCOMP
753 Stop triggered by a seccomp(2) rule on tracee syscall entry when
754 PTRACE_O_TRACESECCOMP has been set by the tracer. The seccomp
755 event message data (from the SECCOMP_RET_DATA portion of the
756 seccomp filter rule) can be retrieved with PTRACE_GETEVENTMSG.
757 The semantics of this stop are described in detail in a separate
758 section below.
759 PTRACE_GETSIGINFO on PTRACE_EVENT stops returns SIGTRAP in si_signo,
761 with si_code set to (event<<8) | SIGTRAP.
762 Syscall-stops
764 If the tracee was restarted by PTRACE_SYSCALL or PTRACE_SYSEMU, the
765 tracee enters syscall-enter-stop just prior to entering any system call
766 (which will not be executed if the restart was using PTRACE_SYSEMU,
767 regardless of any change made to registers at this point or how the
768 tracee is restarted after this stop). No matter which method caused
769 the syscall-entry-stop, if the tracer restarts the tracee with
770 PTRACE_SYSCALL, the tracee enters syscall-exit-stop when the system
771 call is finished, or if it is interrupted by a signal. (That is, sig‐
772 nal-delivery-stop never happens between syscall-enter-stop and syscall-
773 exit-stop; it happens after syscall-exit-stop.). If the tracee is con‐
774 tinued using any other method (including PTRACE_SYSEMU), no syscall-
775 exit-stop occurs. Note that all mentions PTRACE_SYSEMU apply equally
776 to PTRACE_SYSEMU_SINGLESTEP.
777 However, even if the tracee was continued using PTRACE_SYSCALL , it is
779 not guaranteed that the next stop will be a syscall-exit-stop. Other
780 possibilities are that the tracee may stop in a PTRACE_EVENT stop
781 (including seccomp stops), exit (if it entered _exit(2) or
782 exit_group(2)), be killed by SIGKILL, or die silently (if it is a
783 thread group leader, the execve(2) happened in another thread, and that
784 thread is not traced by the same tracer; this situation is discussed
785 later).
786 Syscall-enter-stop and syscall-exit-stop are observed by the tracer as
788 waitpid(2) returning with WIFSTOPPED(status) true, and WSTOPSIG(status)
789 giving SIGTRAP. If the PTRACE_O_TRACESYSGOOD option was set by the
790 tracer, then WSTOPSIG(status) will give the value (SIGTRAP | 0x80).
791 Syscall-stops can be distinguished from signal-delivery-stop with SIG‐
793 TRAP by querying PTRACE_GETSIGINFO for the following cases:
794 si_code <= 0
796 SIGTRAP was delivered as a result of a user-space action, for
797 example, a system call (tgkill(2), kill(2), sigqueue(3), etc.),
798 expiration of a POSIX timer, change of state on a POSIX message
799 queue, or completion of an asynchronous I/O request.
800 si_code == SI_KERNEL (0x80)
802 SIGTRAP was sent by the kernel.
803 si_code == SIGTRAP or si_code == (SIGTRAP|0x80)
805 This is a syscall-stop.
806 However, syscall-stops happen very often (twice per system call), and
808 performing PTRACE_GETSIGINFO for every syscall-stop may be somewhat
809 expensive.
810 Some architectures allow the cases to be distinguished by examining
812 registers. For example, on x86, rax == -ENOSYS in syscall-enter-stop.
813 Since SIGTRAP (like any other signal) always happens after syscall-
814 exit-stop, and at this point rax almost never contains -ENOSYS, the
815 SIGTRAP looks like "syscall-stop which is not syscall-enter-stop"; in
816 other words, it looks like a "stray syscall-exit-stop" and can be
817 detected this way. But such detection is fragile and is best avoided.
818 Using the PTRACE_O_TRACESYSGOOD option is the recommended method to
820 distinguish syscall-stops from other kinds of ptrace-stops, since it is
821 reliable and does not incur a performance penalty.
822 Syscall-enter-stop and syscall-exit-stop are indistinguishable from
824 each other by the tracer. The tracer needs to keep track of the
825 sequence of ptrace-stops in order to not misinterpret syscall-enter-
826 stop as syscall-exit-stop or vice versa. In general, a syscall-enter-
827 stop is always followed by syscall-exit-stop, PTRACE_EVENT stop, or the
828 tracee's death; no other kinds of ptrace-stop can occur in between.
829 However, note that seccomp stops (see below) can cause syscall-exit-
830 stops, without preceding syscall-entry-stops. If seccomp is in use,
831 care needs to be taken not to misinterpret such stops as syscall-entry-
832 stops.
833 If after syscall-enter-stop, the tracer uses a restarting command other
835 than PTRACE_SYSCALL, syscall-exit-stop is not generated.
836 PTRACE_GETSIGINFO on syscall-stops returns SIGTRAP in si_signo, with
838 si_code set to SIGTRAP or (SIGTRAP|0x80).
839 PTRACE_EVENT_SECCOMP stops (Linux 3.5 to 4.7)
841 The behavior of PTRACE_EVENT_SECCOMP stops and their interaction with
842 other kinds of ptrace stops has changed between kernel versions. This
843 documents the behavior from their introduction until Linux 4.7 (inclu‐
844 sive). The behavior in later kernel versions is documented in the next
845 section.
846 A PTRACE_EVENT_SECCOMP stop occurs whenever a SECCOMP_RET_TRACE rule is
848 triggered. This is independent of which methods was used to restart
849 the system call. Notably, seccomp still runs even if the tracee was
850 restarted using PTRACE_SYSEMU and this system call is unconditionally
851 skipped.
852 Restarts from this stop will behave as if the stop had occurred right
854 before the system call in question. In particular, both PTRACE_SYSCALL
855 and PTRACE_SYSEMU will normally cause a subsequent syscall-entry-stop.
856 However, if after the PTRACE_EVENT_SECCOMP the system call number is
857 negative, both the syscall-entry-stop and the system call itself will
858 be skipped. This means that if the system call number is negative
859 after a PTRACE_EVENT_SECCOMP and the tracee is restarted using
860 PTRACE_SYSCALL, the next observed stop will be a syscall-exit-stop,
861 rather than the syscall-entry-stop that might have been expected.
862 PTRACE_EVENT_SECCOMP stops (since Linux 4.8)
864 Starting with Linux 4.8, the PTRACE_EVENT_SECCOMP stop was reordered to
865 occur between syscall-entry-stop and syscall-exit-stop. Note that sec‐
866 comp no longer runs (and no PTRACE_EVENT_SECCOMP will be reported) if
867 the system call is skipped due to PTRACE_SYSEMU.
868 Functionally, a PTRACE_EVENT_SECCOMP stop functions comparably to a
870 syscall-entry-stop (i.e., continuations using PTRACE_SYSCALL will cause
871 syscall-exit-stops, the system call number may be changed and any other
872 modified registers are visible to the to-be-executed system call as
873 well). Note that there may be, but need not have been a preceding
874 syscall-entry-stop.
875 After a PTRACE_EVENT_SECCOMP stop, seccomp will be rerun, with a SEC‐
877 COMP_RET_TRACE rule now functioning the same as a SECCOMP_RET_ALLOW.
878 Specifically, this means that if registers are not modified during the
879 PTRACE_EVENT_SECCOMP stop, the system call will then be allowed.
880 PTRACE_SINGLESTEP stops
882 [Details of these kinds of stops are yet to be documented.]
883 Informational and restarting ptrace commands
885 Most ptrace commands (all except PTRACE_ATTACH, PTRACE_SEIZE,
886 PTRACE_TRACEME, PTRACE_INTERRUPT, and PTRACE_KILL) require the tracee
887 to be in a ptrace-stop, otherwise they fail with ESRCH.
888 When the tracee is in ptrace-stop, the tracer can read and write data
890 to the tracee using informational commands. These commands leave the
891 tracee in ptrace-stopped state:
892 ptrace(PTRACE_PEEKTEXT/PEEKDATA/PEEKUSER, pid, addr, 0);
894 ptrace(PTRACE_POKETEXT/POKEDATA/POKEUSER, pid, addr, long_val);
895 ptrace(PTRACE_GETREGS/GETFPREGS, pid, 0, &struct);
896 ptrace(PTRACE_SETREGS/SETFPREGS, pid, 0, &struct);
897 ptrace(PTRACE_GETREGSET, pid, NT_foo, &iov);
898 ptrace(PTRACE_SETREGSET, pid, NT_foo, &iov);
899 ptrace(PTRACE_GETSIGINFO, pid, 0, &siginfo);
900 ptrace(PTRACE_SETSIGINFO, pid, 0, &siginfo);
901 ptrace(PTRACE_GETEVENTMSG, pid, 0, &long_var);
902 ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_flags);
903 Note that some errors are not reported. For example, setting signal
905 information (siginfo) may have no effect in some ptrace-stops, yet the
906 call may succeed (return 0 and not set errno); querying
907 PTRACE_GETEVENTMSG may succeed and return some random value if current
908 ptrace-stop is not documented as returning a meaningful event message.
909 The call
911 ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_flags);
913 affects one tracee. The tracee's current flags are replaced. Flags
915 are inherited by new tracees created and "auto-attached" via active
916 PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, or PTRACE_O_TRACECLONE
917 options.
918 Another group of commands makes the ptrace-stopped tracee run. They
920 have the form:
921 ptrace(cmd, pid, 0, sig);
923 where cmd is PTRACE_CONT, PTRACE_LISTEN, PTRACE_DETACH, PTRACE_SYSCALL,
925 PTRACE_SINGLESTEP, PTRACE_SYSEMU, or PTRACE_SYSEMU_SINGLESTEP. If the
926 tracee is in signal-delivery-stop, sig is the signal to be injected (if
927 it is nonzero). Otherwise, sig may be ignored. (When restarting a
928 tracee from a ptrace-stop other than signal-delivery-stop, recommended
929 practice is to always pass 0 in sig.)
930 Attaching and detaching
932 A thread can be attached to the tracer using the call
933 ptrace(PTRACE_ATTACH, pid, 0, 0);
935 or
937 ptrace(PTRACE_SEIZE, pid, 0, PTRACE_O_flags);
939 PTRACE_ATTACH sends SIGSTOP to this thread. If the tracer wants this
941 SIGSTOP to have no effect, it needs to suppress it. Note that if other
942 signals are concurrently sent to this thread during attach, the tracer
943 may see the tracee enter signal-delivery-stop with other signal(s)
944 first! The usual practice is to reinject these signals until SIGSTOP
945 is seen, then suppress SIGSTOP injection. The design bug here is that
946 a ptrace attach and a concurrently delivered SIGSTOP may race and the
947 concurrent SIGSTOP may be lost.
948 Since attaching sends SIGSTOP and the tracer usually suppresses it,
950 this may cause a stray EINTR return from the currently executing system
951 call in the tracee, as described in the "Signal injection and suppres‐
952 sion" section.
953 Since Linux 3.4, PTRACE_SEIZE can be used instead of PTRACE_ATTACH.
955 PTRACE_SEIZE does not stop the attached process. If you need to stop
956 it after attach (or at any other time) without sending it any signals,
957 use PTRACE_INTERRUPT command.
958 The request
960 ptrace(PTRACE_TRACEME, 0, 0, 0);
962 turns the calling thread into a tracee. The thread continues to run
964 (doesn't enter ptrace-stop). A common practice is to follow the
965 PTRACE_TRACEME with
966 raise(SIGSTOP);
968 and allow the parent (which is our tracer now) to observe our signal-
970 delivery-stop.
971 If the PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, or PTRACE_O_TRACECLONE
973 options are in effect, then children created by, respectively, vfork(2)
974 or clone(2) with the CLONE_VFORK flag, fork(2) or clone(2) with the
975 exit signal set to SIGCHLD, and other kinds of clone(2), are automati‐
976 cally attached to the same tracer which traced their parent. SIGSTOP
977 is delivered to the children, causing them to enter signal-delivery-
978 stop after they exit the system call which created them.
979 Detaching of the tracee is performed by:
981 ptrace(PTRACE_DETACH, pid, 0, sig);
983 PTRACE_DETACH is a restarting operation; therefore it requires the
985 tracee to be in ptrace-stop. If the tracee is in signal-delivery-stop,
986 a signal can be injected. Otherwise, the sig parameter may be silently
987 ignored.
988 If the tracee is running when the tracer wants to detach it, the usual
990 solution is to send SIGSTOP (using tgkill(2), to make sure it goes to
991 the correct thread), wait for the tracee to stop in signal-delivery-
992 stop for SIGSTOP and then detach it (suppressing SIGSTOP injection). A
993 design bug is that this can race with concurrent SIGSTOPs. Another
994 complication is that the tracee may enter other ptrace-stops and needs
995 to be restarted and waited for again, until SIGSTOP is seen. Yet
996 another complication is to be sure that the tracee is not already
997 ptrace-stopped, because no signal delivery happens while it is—not even
998 SIGSTOP.
999 If the tracer dies, all tracees are automatically detached and
1001 restarted, unless they were in group-stop. Handling of restart from
1002 group-stop is currently buggy, but the "as planned" behavior is to
1003 leave tracee stopped and waiting for SIGCONT. If the tracee is
1004 restarted from signal-delivery-stop, the pending signal is injected.
1005 execve(2) under ptrace
1007 When one thread in a multithreaded process calls execve(2), the kernel
1008 destroys all other threads in the process, and resets the thread ID of
1009 the execing thread to the thread group ID (process ID). (Or, to put
1010 things another way, when a multithreaded process does an execve(2), at
1011 completion of the call, it appears as though the execve(2) occurred in
1012 the thread group leader, regardless of which thread did the execve(2).)
1013 This resetting of the thread ID looks very confusing to tracers:
1014 * All other threads stop in PTRACE_EVENT_EXIT stop, if the
1016 PTRACE_O_TRACEEXIT option was turned on. Then all other threads
1017 except the thread group leader report death as if they exited via
1018 _exit(2) with exit code 0.
1019 * The execing tracee changes its thread ID while it is in the
1021 execve(2). (Remember, under ptrace, the "pid" returned from wait‐
1022 pid(2), or fed into ptrace calls, is the tracee's thread ID.) That
1023 is, the tracee's thread ID is reset to be the same as its process
1024 ID, which is the same as the thread group leader's thread ID.
1025 * Then a PTRACE_EVENT_EXEC stop happens, if the PTRACE_O_TRACEEXEC
1027 option was turned on.
1028 * If the thread group leader has reported its PTRACE_EVENT_EXIT stop
1030 by this time, it appears to the tracer that the dead thread leader
1031 "reappears from nowhere". (Note: the thread group leader does not
1032 report death via WIFEXITED(status) until there is at least one other
1033 live thread. This eliminates the possibility that the tracer will
1034 see it dying and then reappearing.) If the thread group leader was
1035 still alive, for the tracer this may look as if thread group leader
1036 returns from a different system call than it entered, or even
1037 "returned from a system call even though it was not in any system
1038 call". If the thread group leader was not traced (or was traced by
1039 a different tracer), then during execve(2) it will appear as if it
1040 has become a tracee of the tracer of the execing tracee.
1041 All of the above effects are the artifacts of the thread ID change in
1043 the tracee.
1044 The PTRACE_O_TRACEEXEC option is the recommended tool for dealing with
1046 this situation. First, it enables PTRACE_EVENT_EXEC stop, which occurs
1047 before execve(2) returns. In this stop, the tracer can use
1048 PTRACE_GETEVENTMSG to retrieve the tracee's former thread ID. (This
1049 feature was introduced in Linux 3.0.) Second, the PTRACE_O_TRACEEXEC
1050 option disables legacy SIGTRAP generation on execve(2).
1051 When the tracer receives PTRACE_EVENT_EXEC stop notification, it is
1053 guaranteed that except this tracee and the thread group leader, no
1054 other threads from the process are alive.
1055 On receiving the PTRACE_EVENT_EXEC stop notification, the tracer should
1057 clean up all its internal data structures describing the threads of
1058 this process, and retain only one data structure—one which describes
1059 the single still running tracee, with
1060 thread ID == thread group ID == process ID.
1062 Example: two threads call execve(2) at the same time:
1064 *** we get syscall-enter-stop in thread 1: **
1066 PID1 execve("/bin/foo", "foo" <unfinished ...>
1067 *** we issue PTRACE_SYSCALL for thread 1 **
1068 *** we get syscall-enter-stop in thread 2: **
1069 PID2 execve("/bin/bar", "bar" <unfinished ...>
1070 *** we issue PTRACE_SYSCALL for thread 2 **
1071 *** we get PTRACE_EVENT_EXEC for PID0, we issue PTRACE_SYSCALL **
1072 *** we get syscall-exit-stop for PID0: **
1073 PID0 <... execve resumed> ) = 0
1074 If the PTRACE_O_TRACEEXEC option is not in effect for the execing
1076 tracee, and if the tracee was PTRACE_ATTACHed rather that
1077 PTRACE_SEIZEd, the kernel delivers an extra SIGTRAP to the tracee after
1078 execve(2) returns. This is an ordinary signal (similar to one which
1079 can be generated by kill -TRAP), not a special kind of ptrace-stop.
1080 Employing PTRACE_GETSIGINFO for this signal returns si_code set to 0
1081 (SI_USER). This signal may be blocked by signal mask, and thus may be
1082 delivered (much) later.
1083 Usually, the tracer (for example, strace(1)) would not want to show
1085 this extra post-execve SIGTRAP signal to the user, and would suppress
1086 its delivery to the tracee (if SIGTRAP is set to SIG_DFL, it is a
1087 killing signal). However, determining which SIGTRAP to suppress is not
1088 easy. Setting the PTRACE_O_TRACEEXEC option or using PTRACE_SEIZE and
1089 thus suppressing this extra SIGTRAP is the recommended approach.
1090 Real parent
1092 The ptrace API (ab)uses the standard UNIX parent/child signaling over
1093 waitpid(2). This used to cause the real parent of the process to stop
1094 receiving several kinds of waitpid(2) notifications when the child
1095 process is traced by some other process.
1096 Many of these bugs have been fixed, but as of Linux 2.6.38 several
1098 still exist; see BUGS below.
1099 As of Linux 2.6.38, the following is believed to work correctly:
1101 * exit/death by signal is reported first to the tracer, then, when the
1103 tracer consumes the waitpid(2) result, to the real parent (to the
1104 real parent only when the whole multithreaded process exits). If
1105 the tracer and the real parent are the same process, the report is
1106 sent only once.
1107
1109 On success, the PTRACE_PEEK* requests return the requested data (but
1110 see NOTES), the PTRACE_SECCOMP_GET_FILTER request returns the number of
1111 instructions in the BPF program, and other requests return zero.
1112 On error, all requests return -1, and errno is set appropriately.
1114 Since the value returned by a successful PTRACE_PEEK* request may be
1115 -1, the caller must clear errno before the call, and then check it
1116 afterward to determine whether or not an error occurred.
1117
1119 EBUSY (i386 only) There was an error with allocating or freeing a
1120 debug register.
1121 EFAULT There was an attempt to read from or write to an invalid area in
1123 the tracer's or the tracee's memory, probably because the area
1124 wasn't mapped or accessible. Unfortunately, under Linux, dif‐
1125 ferent variations of this fault will return EIO or EFAULT more
1126 or less arbitrarily.
1127 EINVAL An attempt was made to set an invalid option.
1129 EIO request is invalid, or an attempt was made to read from or write
1131 to an invalid area in the tracer's or the tracee's memory, or
1132 there was a word-alignment violation, or an invalid signal was
1133 specified during a restart request.
1134 EPERM The specified process cannot be traced. This could be because
1136 the tracer has insufficient privileges (the required capability
1137 is CAP_SYS_PTRACE); unprivileged processes cannot trace pro‐
1138 cesses that they cannot send signals to or those running set-
1139 user-ID/set-group-ID programs, for obvious reasons. Alterna‐
1140 tively, the process may already be being traced, or (on kernels
1141 before 2.6.26) be init(1) (PID 1).
1142 ESRCH The specified process does not exist, or is not currently being
1144 traced by the caller, or is not stopped (for requests that
1145 require a stopped tracee).
1146
1148 SVr4, 4.3BSD.
1149
1151 Although arguments to ptrace() are interpreted according to the proto‐
1152 type given, glibc currently declares ptrace() as a variadic function
1153 with only the request argument fixed. It is recommended to always sup‐
1154 ply four arguments, even if the requested operation does not use them,
1155 setting unused/ignored arguments to 0L or (void *) 0.
1156 In Linux kernels before 2.6.26, init(1), the process with PID 1, may
1158 not be traced.
1159 A tracees parent continues to be the tracer even if that tracer calls
1161 execve(2).
1162 The layout of the contents of memory and the USER area are quite oper‐
1164 ating-system- and architecture-specific. The offset supplied, and the
1165 data returned, might not entirely match with the definition of struct
1166 user.
1167 The size of a "word" is determined by the operating-system variant
1169 (e.g., for 32-bit Linux it is 32 bits).
1170 This page documents the way the ptrace() call works currently in Linux.
1172 Its behavior differs significantly on other flavors of UNIX. In any
1173 case, use of ptrace() is highly specific to the operating system and
1174 architecture.
1175 Ptrace access mode checking
1177 Various parts of the kernel-user-space API (not just ptrace() opera‐
1178 tions), require so-called "ptrace access mode" checks, whose outcome
1179 determines whether an operation is permitted (or, in a few cases,
1180 causes a "read" operation to return sanitized data). These checks are
1181 performed in cases where one process can inspect sensitive information
1182 about, or in some cases modify the state of, another process. The
1183 checks are based on factors such as the credentials and capabilities of
1184 the two processes, whether or not the "target" process is dumpable, and
1185 the results of checks performed by any enabled Linux Security Module
1186 (LSM)—for example, SELinux, Yama, or Smack—and by the commoncap LSM
1187 (which is always invoked).
1188 Prior to Linux 2.6.27, all access checks were of a single type. Since
1190 Linux 2.6.27, two access mode levels are distinguished:
1191 PTRACE_MODE_READ
1193 For "read" operations or other operations that are less danger‐
1194 ous, such as: get_robust_list(2); kcmp(2); reading
1195 /proc/[pid]/auxv, /proc/[pid]/environ, or /proc/[pid]/stat; or
1196 readlink(2) of a /proc/[pid]/ns/* file.
1197 PTRACE_MODE_ATTACH
1199 For "write" operations, or other operations that are more dan‐
1200 gerous, such as: ptrace attaching (PTRACE_ATTACH) to another
1201 process or calling process_vm_writev(2). (PTRACE_MODE_ATTACH
1202 was effectively the default before Linux 2.6.27.)
1203 Since Linux 4.5, the above access mode checks are combined (ORed) with
1205 one of the following modifiers:
1206 PTRACE_MODE_FSCREDS
1208 Use the caller's filesystem UID and GID (see credentials(7)) or
1209 effective capabilities for LSM checks.
1210 PTRACE_MODE_REALCREDS
1212 Use the caller's real UID and GID or permitted capabilities for
1213 LSM checks. This was effectively the default before Linux 4.5.
1214 Because combining one of the credential modifiers with one of the
1216 aforementioned access modes is typical, some macros are defined in the
1217 kernel sources for the combinations:
1218 PTRACE_MODE_READ_FSCREDS
1220 Defined as PTRACE_MODE_READ | PTRACE_MODE_FSCREDS.
1221 PTRACE_MODE_READ_REALCREDS
1223 Defined as PTRACE_MODE_READ | PTRACE_MODE_REALCREDS.
1224 PTRACE_MODE_ATTACH_FSCREDS
1226 Defined as PTRACE_MODE_ATTACH | PTRACE_MODE_FSCREDS.
1227 PTRACE_MODE_ATTACH_REALCREDS
1229 Defined as PTRACE_MODE_ATTACH | PTRACE_MODE_REALCREDS.
1230 One further modifier can be ORed with the access mode:
1232 PTRACE_MODE_NOAUDIT (since Linux 3.3)
1234 Don't audit this access mode check. This modifier is employed
1235 for ptrace access mode checks (such as checks when reading
1236 /proc/[pid]/stat) that merely cause the output to be filtered or
1237 sanitized, rather than causing an error to be returned to the
1238 caller. In these cases, accessing the file is not a security
1239 violation and there is no reason to generate a security audit
1240 record. This modifier suppresses the generation of such an
1241 audit record for the particular access check.
1242 Note that all of the PTRACE_MODE_* constants described in this subsec‐
1244 tion are kernel-internal, and not visible to user space. The constant
1245 names are mentioned here in order to label the various kinds of ptrace
1246 access mode checks that are performed for various system calls and
1247 accesses to various pseudofiles (e.g., under /proc). These names are
1248 used in other manual pages to provide a simple shorthand for labeling
1249 the different kernel checks.
1250 The algorithm employed for ptrace access mode checking determines
1252 whether the calling process is allowed to perform the corresponding
1253 action on the target process. (In the case of opening /proc/[pid]
1254 files, the "calling process" is the one opening the file, and the
1255 process with the corresponding PID is the "target process".) The algo‐
1256 rithm is as follows:
1257 1. If the calling thread and the target thread are in the same thread
1259 group, access is always allowed.
1260 2. If the access mode specifies PTRACE_MODE_FSCREDS, then, for the
1262 check in the next step, employ the caller's filesystem UID and GID.
1263 (As noted in credentials(7), the filesystem UID and GID almost
1264 always have the same values as the corresponding effective IDs.)
1265 Otherwise, the access mode specifies PTRACE_MODE_REALCREDS, so use
1267 the caller's real UID and GID for the checks in the next step.
1268 (Most APIs that check the caller's UID and GID use the effective
1269 IDs. For historical reasons, the PTRACE_MODE_REALCREDS check uses
1270 the real IDs instead.)
1271 3. Deny access if neither of the following is true:
1273 · The real, effective, and saved-set user IDs of the target match
1275 the caller's user ID, and the real, effective, and saved-set group
1276 IDs of the target match the caller's group ID.
1277 · The caller has the CAP_SYS_PTRACE capability in the user namespace
1279 of the target.
1280 4. Deny access if the target process "dumpable" attribute has a value
1282 other than 1 (SUID_DUMP_USER; see the discussion of PR_SET_DUMPABLE
1283 in prctl(2)), and the caller does not have the CAP_SYS_PTRACE capa‐
1284 bility in the user namespace of the target process.
1285 5. The kernel LSM security_ptrace_access_check() interface is invoked
1287 to see if ptrace access is permitted. The results depend on the
1288 LSM(s). The implementation of this interface in the commoncap LSM
1289 performs the following steps:
1290 a) If the access mode includes PTRACE_MODE_FSCREDS, then use the
1292 caller's effective capability set in the following check; other‐
1293 wise (the access mode specifies PTRACE_MODE_REALCREDS, so) use
1294 the caller's permitted capability set.
1295 b) Deny access if neither of the following is true:
1297 · The caller and the target process are in the same user names‐
1299 pace, and the caller's capabilities are a proper superset of
1300 the target process's permitted capabilities.
1301 · The caller has the CAP_SYS_PTRACE capability in the target
1303 process's user namespace.
1304 Note that the commoncap LSM does not distinguish between
1306 PTRACE_MODE_READ and PTRACE_MODE_ATTACH.
1307 6. If access has not been denied by any of the preceding steps, then
1309 access is allowed.
1310 /proc/sys/kernel/yama/ptrace_scope
1312 On systems with the Yama Linux Security Module (LSM) installed (i.e.,
1313 the kernel was configured with CONFIG_SECURITY_YAMA), the
1314 /proc/sys/kernel/yama/ptrace_scope file (available since Linux 3.4) can
1315 be used to restrict the ability to trace a process with ptrace() (and
1316 thus also the ability to use tools such as strace(1) and gdb(1)). The
1317 goal of such restrictions is to prevent attack escalation whereby a
1318 compromised process can ptrace-attach to other sensitive processes
1319 (e.g., a GPG agent or an SSH session) owned by the user in order to
1320 gain additional credentials that may exist in memory and thus expand
1321 the scope of the attack.
1322 More precisely, the Yama LSM limits two types of operations:
1324 * Any operation that performs a ptrace access mode PTRACE_MODE_ATTACH
1326 check—for example, ptrace() PTRACE_ATTACH. (See the "Ptrace access
1327 mode checking" discussion above.)
1328 * ptrace() PTRACE_TRACEME.
1330 A process that has the CAP_SYS_PTRACE capability can update the
1332 /proc/sys/kernel/yama/ptrace_scope file with one of the following val‐
1333 ues:
1334 0 ("classic ptrace permissions")
1336 No additional restrictions on operations that perform
1337 PTRACE_MODE_ATTACH checks (beyond those imposed by the commoncap
1338 and other LSMs).
1339 The use of PTRACE_TRACEME is unchanged.
1341 1 ("restricted ptrace") [default value]
1343 When performing an operation that requires a PTRACE_MODE_ATTACH
1344 check, the calling process must either have the CAP_SYS_PTRACE
1345 capability in the user namespace of the target process or it
1346 must have a predefined relationship with the target process. By
1347 default, the predefined relationship is that the target process
1348 must be a descendant of the caller.
1349 A target process can employ the prctl(2) PR_SET_PTRACER opera‐
1351 tion to declare an additional PID that is allowed to perform
1352 PTRACE_MODE_ATTACH operations on the target. See the kernel
1353 source file Documentation/admin-guide/LSM/Yama.rst (or Documen‐
1354 tation/security/Yama.txt before Linux 4.13) for further details.
1355 The use of PTRACE_TRACEME is unchanged.
1357 2 ("admin-only attach")
1359 Only processes with the CAP_SYS_PTRACE capability in the user
1360 namespace of the target process may perform PTRACE_MODE_ATTACH
1361 operations or trace children that employ PTRACE_TRACEME.
1362 3 ("no attach")
1364 No process may perform PTRACE_MODE_ATTACH operations or trace
1365 children that employ PTRACE_TRACEME.
1366 Once this value has been written to the file, it cannot be
1368 changed.
1369 With respect to values 1 and 2, note that creating a new user namespace
1371 effectively removes the protection offered by Yama. This is because a
1372 process in the parent user namespace whose effective UID matches the
1373 UID of the creator of a child namespace has all capabilities (including
1374 CAP_SYS_PTRACE) when performing operations within the child user names‐
1375 pace (and further-removed descendants of that namespace). Conse‐
1376 quently, when a process tries to use user namespaces to sandbox itself,
1377 it inadvertently weakens the protections offered by the Yama LSM.
1378 C library/kernel differences
1380 At the system call level, the PTRACE_PEEKTEXT, PTRACE_PEEKDATA, and
1381 PTRACE_PEEKUSER requests have a different API: they store the result at
1382 the address specified by the data parameter, and the return value is
1383 the error flag. The glibc wrapper function provides the API given in
1384 DESCRIPTION above, with the result being returned via the function
1385 return value.
1386
1388 On hosts with 2.6 kernel headers, PTRACE_SETOPTIONS is declared with a
1389 different value than the one for 2.4. This leads to applications com‐
1390 piled with 2.6 kernel headers failing when run on 2.4 kernels. This
1391 can be worked around by redefining PTRACE_SETOPTIONS to PTRACE_OLDSE‐
1392 TOPTIONS, if that is defined.
1393 Group-stop notifications are sent to the tracer, but not to real par‐
1395 ent. Last confirmed on 2.6.38.6.
1396 If a thread group leader is traced and exits by calling _exit(2), a
1398 PTRACE_EVENT_EXIT stop will happen for it (if requested), but the sub‐
1399 sequent WIFEXITED notification will not be delivered until all other
1400 threads exit. As explained above, if one of other threads calls
1401 execve(2), the death of the thread group leader will never be reported.
1402 If the execed thread is not traced by this tracer, the tracer will
1403 never know that execve(2) happened. One possible workaround is to
1404 PTRACE_DETACH the thread group leader instead of restarting it in this
1405 case. Last confirmed on 2.6.38.6.
1406 A SIGKILL signal may still cause a PTRACE_EVENT_EXIT stop before actual
1408 signal death. This may be changed in the future; SIGKILL is meant to
1409 always immediately kill tasks even under ptrace. Last confirmed on
1410 Linux 3.13.
1411 Some system calls return with EINTR if a signal was sent to a tracee,
1413 but delivery was suppressed by the tracer. (This is very typical oper‐
1414 ation: it is usually done by debuggers on every attach, in order to not
1415 introduce a bogus SIGSTOP). As of Linux 3.2.9, the following system
1416 calls are affected (this list is likely incomplete): epoll_wait(2), and
1417 read(2) from an inotify(7) file descriptor. The usual symptom of this
1418 bug is that when you attach to a quiescent process with the command
1419 strace -p <process-ID>
1421 then, instead of the usual and expected one-line output such as
1423 restart_syscall(<... resuming interrupted call ...>_
1425 or
1427 select(6, [5], NULL, [5], NULL_
1429 ('_' denotes the cursor position), you observe more than one line. For
1431 example:
1432 clock_gettime(CLOCK_MONOTONIC, {15370, 690928118}) = 0
1434 epoll_wait(4,_
1435 What is not visible here is that the process was blocked in
1437 epoll_wait(2) before strace(1) has attached to it. Attaching caused
1438 epoll_wait(2) to return to user space with the error EINTR. In this
1439 particular case, the program reacted to EINTR by checking the current
1440 time, and then executing epoll_wait(2) again. (Programs which do not
1441 expect such "stray" EINTR errors may behave in an unintended way upon
1442 an strace(1) attach.)
1443
1445 gdb(1), ltrace(1), strace(1), clone(2), execve(2), fork(2), gettid(2),
1446 prctl(2), seccomp(2), sigaction(2), tgkill(2), vfork(2), waitpid(2),
1447 exec(3), capabilities(7), signal(7)
1448
1450 This page is part of release 4.16 of the Linux man-pages project. A
1451 description of the project, information about reporting bugs, and the
1452 latest version of this page, can be found at
1453 https://www.kernel.org/doc/man-pages/.
1454Linux 2018-04-30 PTRACE(2)