1ptrace(2) System Calls Manual ptrace(2)
2
6 ptrace - process trace
7
9 Standard C library (libc, -lc)
10
12 #include <sys/ptrace.h>
13 long ptrace(enum __ptrace_request request, pid_t pid,
15 void *addr, void *data);
16
18 The ptrace() system call provides a means by which one process (the
19 "tracer") may observe and control the execution of another process (the
20 "tracee"), and examine and change the tracee's memory and registers.
21 It is primarily used to implement breakpoint debugging and system call
22 tracing.
23 A tracee first needs to be attached to the tracer. Attachment and sub‐
25 sequent commands are per thread: in a multithreaded process, every
26 thread can be individually attached to a (potentially different)
27 tracer, or left not attached and thus not debugged. Therefore,
28 "tracee" always means "(one) thread", never "a (possibly multithreaded)
29 process". Ptrace commands are always sent to a specific tracee using a
30 call of the form
31 ptrace(PTRACE_foo, pid, ...)
33 where pid is the thread ID of the corresponding Linux thread.
35 (Note that in this page, a "multithreaded process" means a thread group
37 consisting of threads created using the clone(2) CLONE_THREAD flag.)
38 A process can initiate a trace by calling fork(2) and having the re‐
40 sulting child do a PTRACE_TRACEME, followed (typically) by an ex‐
41 ecve(2). Alternatively, one process may commence tracing another
42 process using PTRACE_ATTACH or PTRACE_SEIZE.
43 While being traced, the tracee will stop each time a signal is deliv‐
45 ered, even if the signal is being ignored. (An exception is SIGKILL,
46 which has its usual effect.) The tracer will be notified at its next
47 call to waitpid(2) (or one of the related "wait" system calls); that
48 call will return a status value containing information that indicates
49 the cause of the stop in the tracee. While the tracee is stopped, the
50 tracer can use various ptrace requests to inspect and modify the
51 tracee. The tracer then causes the tracee to continue, optionally ig‐
52 noring the delivered signal (or even delivering a different signal in‐
53 stead).
54 If the PTRACE_O_TRACEEXEC option is not in effect, all successful calls
56 to execve(2) by the traced process will cause it to be sent a SIGTRAP
57 signal, giving the parent a chance to gain control before the new pro‐
58 gram begins execution.
59 When the tracer is finished tracing, it can cause the tracee to con‐
61 tinue executing in a normal, untraced mode via PTRACE_DETACH.
62 The value of request determines the action to be performed:
64 PTRACE_TRACEME
66 Indicate that this process is to be traced by its parent. A
67 process probably shouldn't make this request if its parent isn't
68 expecting to trace it. (pid, addr, and data are ignored.)
69 The PTRACE_TRACEME request is used only by the tracee; the re‐
71 maining requests are used only by the tracer. In the following
72 requests, pid specifies the thread ID of the tracee to be acted
73 on. For requests other than PTRACE_ATTACH, PTRACE_SEIZE,
74 PTRACE_INTERRUPT, and PTRACE_KILL, the tracee must be stopped.
75 PTRACE_PEEKTEXT, PTRACE_PEEKDATA
77 Read a word at the address addr in the tracee's memory, return‐
78 ing the word as the result of the ptrace() call. Linux does not
79 have separate text and data address spaces, so these two re‐
80 quests are currently equivalent. (data is ignored; but see
81 NOTES.)
82 PTRACE_PEEKUSER
84 Read a word at offset addr in the tracee's USER area, which
85 holds the registers and other information about the process (see
86 <sys/user.h>). The word is returned as the result of the
87 ptrace() call. Typically, the offset must be word-aligned,
88 though this might vary by architecture. See NOTES. (data is
89 ignored; but see NOTES.)
90 PTRACE_POKETEXT, PTRACE_POKEDATA
92 Copy the word data to the address addr in the tracee's memory.
93 As for PTRACE_PEEKTEXT and PTRACE_PEEKDATA, these two requests
94 are currently equivalent.
95 PTRACE_POKEUSER
97 Copy the word data to offset addr in the tracee's USER area. As
98 for PTRACE_PEEKUSER, the offset must typically be word-aligned.
99 In order to maintain the integrity of the kernel, some modifica‐
100 tions to the USER area are disallowed.
101 PTRACE_GETREGS, PTRACE_GETFPREGS
103 Copy the tracee's general-purpose or floating-point registers,
104 respectively, to the address data in the tracer. See
105 <sys/user.h> for information on the format of this data. (addr
106 is ignored.) Note that SPARC systems have the meaning of data
107 and addr reversed; that is, data is ignored and the registers
108 are copied to the address addr. PTRACE_GETREGS and PTRACE_GETF‐
109 PREGS are not present on all architectures.
110 PTRACE_GETREGSET (since Linux 2.6.34)
112 Read the tracee's registers. addr specifies, in an architec‐
113 ture-dependent way, the type of registers to be read. NT_PRSTA‐
114 TUS (with numerical value 1) usually results in reading of gen‐
115 eral-purpose registers. If the CPU has, for example, floating-
116 point and/or vector registers, they can be retrieved by setting
117 addr to the corresponding NT_foo constant. data points to a
118 struct iovec, which describes the destination buffer's location
119 and length. On return, the kernel modifies iov.len to indicate
120 the actual number of bytes returned.
121 PTRACE_SETREGS, PTRACE_SETFPREGS
123 Modify the tracee's general-purpose or floating-point registers,
124 respectively, from the address data in the tracer. As for
125 PTRACE_POKEUSER, some general-purpose register modifications may
126 be disallowed. (addr is ignored.) Note that SPARC systems have
127 the meaning of data and addr reversed; that is, data is ignored
128 and the registers are copied from the address addr. PTRACE_SE‐
129 TREGS and PTRACE_SETFPREGS are not present on all architectures.
130 PTRACE_SETREGSET (since Linux 2.6.34)
132 Modify the tracee's registers. The meaning of addr and data is
133 analogous to PTRACE_GETREGSET.
134 PTRACE_GETSIGINFO (since Linux 2.3.99-pre6)
136 Retrieve information about the signal that caused the stop.
137 Copy a siginfo_t structure (see sigaction(2)) from the tracee to
138 the address data in the tracer. (addr is ignored.)
139 PTRACE_SETSIGINFO (since Linux 2.3.99-pre6)
141 Set signal information: copy a siginfo_t structure from the ad‐
142 dress data in the tracer to the tracee. This will affect only
143 signals that would normally be delivered to the tracee and were
144 caught by the tracer. It may be difficult to tell these normal
145 signals from synthetic signals generated by ptrace() itself.
146 (addr is ignored.)
147 PTRACE_PEEKSIGINFO (since Linux 3.10)
149 Retrieve siginfo_t structures without removing signals from a
150 queue. addr points to a ptrace_peeksiginfo_args structure that
151 specifies the ordinal position from which copying of signals
152 should start, and the number of signals to copy. siginfo_t
153 structures are copied into the buffer pointed to by data. The
154 return value contains the number of copied signals (zero indi‐
155 cates that there is no signal corresponding to the specified or‐
156 dinal position). Within the returned siginfo structures, the
157 si_code field includes information (__SI_CHLD, __SI_FAULT, etc.)
158 that are not otherwise exposed to user space.
159 struct ptrace_peeksiginfo_args {
161 u64 off; /* Ordinal position in queue at which
162 to start copying signals */
163 u32 flags; /* PTRACE_PEEKSIGINFO_SHARED or 0 */
164 s32 nr; /* Number of signals to copy */
165 };
166 Currently, there is only one flag, PTRACE_PEEKSIGINFO_SHARED,
168 for dumping signals from the process-wide signal queue. If this
169 flag is not set, signals are read from the per-thread queue of
170 the specified thread.
171 PTRACE_GETSIGMASK (since Linux 3.11)
173 Place a copy of the mask of blocked signals (see sigprocmask(2))
174 in the buffer pointed to by data, which should be a pointer to a
175 buffer of type sigset_t. The addr argument contains the size of
176 the buffer pointed to by data (i.e., sizeof(sigset_t)).
177 PTRACE_SETSIGMASK (since Linux 3.11)
179 Change the mask of blocked signals (see sigprocmask(2)) to the
180 value specified in the buffer pointed to by data, which should
181 be a pointer to a buffer of type sigset_t. The addr argument
182 contains the size of the buffer pointed to by data (i.e.,
183 sizeof(sigset_t)).
184 PTRACE_SETOPTIONS (since Linux 2.4.6; see BUGS for caveats)
186 Set ptrace options from data. (addr is ignored.) data is in‐
187 terpreted as a bit mask of options, which are specified by the
188 following flags:
189 PTRACE_O_EXITKILL (since Linux 3.8)
191 Send a SIGKILL signal to the tracee if the tracer exits.
192 This option is useful for ptrace jailers that want to en‐
193 sure that tracees can never escape the tracer's control.
194 PTRACE_O_TRACECLONE (since Linux 2.5.46)
196 Stop the tracee at the next clone(2) and automatically
197 start tracing the newly cloned process, which will start
198 with a SIGSTOP, or PTRACE_EVENT_STOP if PTRACE_SEIZE was
199 used. A waitpid(2) by the tracer will return a status
200 value such that
201 status>>8 == (SIGTRAP | (PTRACE_EVENT_CLONE<<8))
203 The PID of the new process can be retrieved with
205 PTRACE_GETEVENTMSG.
206 This option may not catch clone(2) calls in all cases.
208 If the tracee calls clone(2) with the CLONE_VFORK flag,
209 PTRACE_EVENT_VFORK will be delivered instead if
210 PTRACE_O_TRACEVFORK is set; otherwise if the tracee calls
211 clone(2) with the exit signal set to SIGCHLD,
212 PTRACE_EVENT_FORK will be delivered if PTRACE_O_TRACEFORK
213 is set.
214 PTRACE_O_TRACEEXEC (since Linux 2.5.46)
216 Stop the tracee at the next execve(2). A waitpid(2) by
217 the tracer will return a status value such that
218 status>>8 == (SIGTRAP | (PTRACE_EVENT_EXEC<<8))
220 If the execing thread is not a thread group leader, the
222 thread ID is reset to thread group leader's ID before
223 this stop. Since Linux 3.0, the former thread ID can be
224 retrieved with PTRACE_GETEVENTMSG.
225 PTRACE_O_TRACEEXIT (since Linux 2.5.60)
227 Stop the tracee at exit. A waitpid(2) by the tracer will
228 return a status value such that
229 status>>8 == (SIGTRAP | (PTRACE_EVENT_EXIT<<8))
231 The tracee's exit status can be retrieved with
233 PTRACE_GETEVENTMSG.
234 The tracee is stopped early during process exit, when
236 registers are still available, allowing the tracer to see
237 where the exit occurred, whereas the normal exit notifi‐
238 cation is done after the process is finished exiting.
239 Even though context is available, the tracer cannot pre‐
240 vent the exit from happening at this point.
241 PTRACE_O_TRACEFORK (since Linux 2.5.46)
243 Stop the tracee at the next fork(2) and automatically
244 start tracing the newly forked process, which will start
245 with a SIGSTOP, or PTRACE_EVENT_STOP if PTRACE_SEIZE was
246 used. A waitpid(2) by the tracer will return a status
247 value such that
248 status>>8 == (SIGTRAP | (PTRACE_EVENT_FORK<<8))
250 The PID of the new process can be retrieved with
252 PTRACE_GETEVENTMSG.
253 PTRACE_O_TRACESYSGOOD (since Linux 2.4.6)
255 When delivering system call traps, set bit 7 in the sig‐
256 nal number (i.e., deliver SIGTRAP|0x80). This makes it
257 easy for the tracer to distinguish normal traps from
258 those caused by a system call.
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 re‐
329 ceipt 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 ar‐
332 guments 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_SET_SYSCALL (since Linux 2.6.16)
338 When in syscall-enter-stop, change the number of the system call
339 that is about to be executed to the number specified in the data
340 argument. The addr argument is ignored. This request is cur‐
341 rently supported only on arm (and arm64, though only for back‐
342 wards compatibility), but most other architectures have other
343 means of accomplishing this (usually by changing the register
344 that the userland code passed the system call number in).
345 PTRACE_SYSEMU, PTRACE_SYSEMU_SINGLESTEP (since Linux 2.6.14)
347 For PTRACE_SYSEMU, continue and stop on entry to the next system
348 call, which will not be executed. See the documentation on
349 syscall-stops below. For PTRACE_SYSEMU_SINGLESTEP, do the same
350 but also singlestep if not a system call. This call is used by
351 programs like User Mode Linux that want to emulate all the
352 tracee's system calls. The data argument is treated as for
353 PTRACE_CONT. The addr argument is ignored. These requests are
354 currently supported only on x86.
355 PTRACE_LISTEN (since Linux 3.4)
357 Restart the stopped tracee, but prevent it from executing. The
358 resulting state of the tracee is similar to a process which has
359 been stopped by a SIGSTOP (or other stopping signal). See the
360 "group-stop" subsection for additional information. PTRACE_LIS‐
361 TEN works only on tracees attached by PTRACE_SEIZE.
362 PTRACE_KILL
364 Send the tracee a SIGKILL to terminate it. (addr and data are
365 ignored.)
366 This operation is deprecated; do not use it! Instead, send a
368 SIGKILL directly using kill(2) or tgkill(2). The problem with
369 PTRACE_KILL is that it requires the tracee to be in signal-de‐
370 livery-stop, otherwise it may not work (i.e., may complete suc‐
371 cessfully but won't kill the tracee). By contrast, sending a
372 SIGKILL directly has no such limitation.
373 PTRACE_INTERRUPT (since Linux 3.4)
375 Stop a tracee. If the tracee is running or sleeping in kernel
376 space and PTRACE_SYSCALL is in effect, the system call is inter‐
377 rupted and syscall-exit-stop is reported. (The interrupted sys‐
378 tem call is restarted when the tracee is restarted.) If the
379 tracee was already stopped by a signal and PTRACE_LISTEN was
380 sent to it, the tracee stops with PTRACE_EVENT_STOP and WSTOP‐
381 SIG(status) returns the stop signal. If any other ptrace-stop
382 is generated at the same time (for example, if a signal is sent
383 to the tracee), this ptrace-stop happens. If none of the above
384 applies (for example, if the tracee is running in user space),
385 it stops with PTRACE_EVENT_STOP with WSTOPSIG(status) == SIG‐
386 TRAP. PTRACE_INTERRUPT only works on tracees attached by
387 PTRACE_SEIZE.
388 PTRACE_ATTACH
390 Attach to the process specified in pid, making it a tracee of
391 the calling process. The tracee is sent a SIGSTOP, but will not
392 necessarily have stopped by the completion of this call; use
393 waitpid(2) to wait for the tracee to stop. See the "Attaching
394 and detaching" subsection for additional information. (addr and
395 data are ignored.)
396 Permission to perform a PTRACE_ATTACH is governed by a ptrace
398 access mode PTRACE_MODE_ATTACH_REALCREDS check; see below.
399 PTRACE_SEIZE (since Linux 3.4)
401 Attach to the process specified in pid, making it a tracee of
402 the calling process. Unlike PTRACE_ATTACH, PTRACE_SEIZE does
403 not stop the process. Group-stops are reported as
404 PTRACE_EVENT_STOP and WSTOPSIG(status) returns the stop signal.
405 Automatically attached children stop with PTRACE_EVENT_STOP and
406 WSTOPSIG(status) returns SIGTRAP instead of having SIGSTOP sig‐
407 nal delivered to them. execve(2) does not deliver an extra SIG‐
408 TRAP. Only a PTRACE_SEIZEd process can accept PTRACE_INTERRUPT
409 and PTRACE_LISTEN commands. The "seized" behavior just de‐
410 scribed is inherited by children that are automatically attached
411 using PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, and
412 PTRACE_O_TRACECLONE. addr must be zero. data contains a bit
413 mask of ptrace options to activate immediately.
414 Permission to perform a PTRACE_SEIZE is governed by a ptrace ac‐
416 cess mode PTRACE_MODE_ATTACH_REALCREDS check; see below.
417 PTRACE_SECCOMP_GET_FILTER (since Linux 4.4)
419 This operation allows the tracer to dump the tracee's classic
420 BPF filters.
421 addr is an integer specifying the index of the filter to be
423 dumped. The most recently installed filter has the index 0. If
424 addr is greater than the number of installed filters, the opera‐
425 tion fails with the error ENOENT.
426 data is either a pointer to a struct sock_filter array that is
428 large enough to store the BPF program, or NULL if the program is
429 not to be stored.
430 Upon success, the return value is the number of instructions in
432 the BPF program. If data was NULL, then this return value can
433 be used to correctly size the struct sock_filter array passed in
434 a subsequent call.
435 This operation fails with the error EACCES if the caller does
437 not have the CAP_SYS_ADMIN capability or if the caller is in
438 strict or filter seccomp mode. If the filter referred to by
439 addr is not a classic BPF filter, the operation fails with the
440 error EMEDIUMTYPE.
441 This operation is available if the kernel was configured with
443 both the CONFIG_SECCOMP_FILTER and the CONFIG_CHECKPOINT_RESTORE
444 options.
445 PTRACE_DETACH
447 Restart the stopped tracee as for PTRACE_CONT, but first detach
448 from it. Under Linux, a tracee can be detached in this way re‐
449 gardless of which method was used to initiate tracing. (addr is
450 ignored.)
451 PTRACE_GET_THREAD_AREA (since Linux 2.6.0)
453 This operation performs a similar task to get_thread_area(2).
454 It reads the TLS entry in the GDT whose index is given in addr,
455 placing a copy of the entry into the struct user_desc pointed to
456 by data. (By contrast with get_thread_area(2), the entry_number
457 of the struct user_desc is ignored.)
458 PTRACE_SET_THREAD_AREA (since Linux 2.6.0)
460 This operation performs a similar task to set_thread_area(2).
461 It sets the TLS entry in the GDT whose index is given in addr,
462 assigning it the data supplied in the struct user_desc pointed
463 to by data. (By contrast with set_thread_area(2), the en‐
464 try_number of the struct user_desc is ignored; in other words,
465 this ptrace operation can't be used to allocate a free TLS en‐
466 try.)
467 PTRACE_GET_SYSCALL_INFO (since Linux 5.3)
469 Retrieve information about the system call that caused the stop.
470 The information is placed into the buffer pointed by the data
471 argument, which should be a pointer to a buffer of type struct
472 ptrace_syscall_info. The addr argument contains the size of the
473 buffer pointed to by the data argument (i.e., sizeof(struct
474 ptrace_syscall_info)). The return value contains the number of
475 bytes available to be written by the kernel. If the size of the
476 data to be written by the kernel exceeds the size specified by
477 the addr argument, the output data is truncated.
478 The ptrace_syscall_info structure contains the following fields:
480 struct ptrace_syscall_info {
482 __u8 op; /* Type of system call stop */
483 __u32 arch; /* AUDIT_ARCH_* value; see seccomp(2) */
484 __u64 instruction_pointer; /* CPU instruction pointer */
485 __u64 stack_pointer; /* CPU stack pointer */
486 union {
487 struct { /* op == PTRACE_SYSCALL_INFO_ENTRY */
488 __u64 nr; /* System call number */
489 __u64 args[6]; /* System call arguments */
490 } entry;
491 struct { /* op == PTRACE_SYSCALL_INFO_EXIT */
492 __s64 rval; /* System call return value */
493 __u8 is_error; /* System call error flag;
494 Boolean: does rval contain
495 an error value (-ERRCODE) or
496 a nonerror return value? */
497 } exit;
498 struct { /* op == PTRACE_SYSCALL_INFO_SECCOMP */
499 __u64 nr; /* System call number */
500 __u64 args[6]; /* System call arguments */
501 __u32 ret_data; /* SECCOMP_RET_DATA portion
502 of SECCOMP_RET_TRACE
503 return value */
504 } seccomp;
505 };
506 };
507 The op, arch, instruction_pointer, and stack_pointer fields are
509 defined for all kinds of ptrace system call stops. The rest of
510 the structure is a union; one should read only those fields that
511 are meaningful for the kind of system call stop specified by the
512 op field.
513 The op field has one of the following values (defined in
515 <linux/ptrace.h>) indicating what type of stop occurred and
516 which part of the union is filled:
517 PTRACE_SYSCALL_INFO_ENTRY
519 The entry component of the union contains information re‐
520 lating to a system call entry stop.
521 PTRACE_SYSCALL_INFO_EXIT
523 The exit component of the union contains information re‐
524 lating to a system call exit stop.
525 PTRACE_SYSCALL_INFO_SECCOMP
527 The seccomp component of the union contains information
528 relating to a PTRACE_EVENT_SECCOMP stop.
529 PTRACE_SYSCALL_INFO_NONE
531 No component of the union contains relevant information.
532 In case of system call entry or exit stops, the data returned by
534 PTRACE_GET_SYSCALL_INFO is limited to type
535 PTRACE_SYSCALL_INFO_NONE unless PTRACE_O_TRACESYSGOOD option is
536 set before the corresponding system call stop has occurred.
537 Death under ptrace
539 When a (possibly multithreaded) process receives a killing signal (one
540 whose disposition is set to SIG_DFL and whose default action is to kill
541 the process), all threads exit. Tracees report their death to their
542 tracer(s). Notification of this event is delivered via waitpid(2).
543 Note that the killing signal will first cause signal-delivery-stop (on
545 one tracee only), and only after it is injected by the tracer (or after
546 it was dispatched to a thread which isn't traced), will death from the
547 signal happen on all tracees within a multithreaded process. (The term
548 "signal-delivery-stop" is explained below.)
549 SIGKILL does not generate signal-delivery-stop and therefore the tracer
551 can't suppress it. SIGKILL kills even within system calls (syscall-
552 exit-stop is not generated prior to death by SIGKILL). The net effect
553 is that SIGKILL always kills the process (all its threads), even if
554 some threads of the process are ptraced.
555 When the tracee calls _exit(2), it reports its death to its tracer.
557 Other threads are not affected.
558 When any thread executes exit_group(2), every tracee in its thread
560 group reports its death to its tracer.
561 If the PTRACE_O_TRACEEXIT option is on, PTRACE_EVENT_EXIT will happen
563 before actual death. This applies to exits via exit(2), exit_group(2),
564 and signal deaths (except SIGKILL, depending on the kernel version; see
565 BUGS below), and when threads are torn down on execve(2) in a multi‐
566 threaded process.
567 The tracer cannot assume that the ptrace-stopped tracee exists. There
569 are many scenarios when the tracee may die while stopped (such as
570 SIGKILL). Therefore, the tracer must be prepared to handle an ESRCH
571 error on any ptrace operation. Unfortunately, the same error is re‐
572 turned if the tracee exists but is not ptrace-stopped (for commands
573 which require a stopped tracee), or if it is not traced by the process
574 which issued the ptrace call. The tracer needs to keep track of the
575 stopped/running state of the tracee, and interpret ESRCH as "tracee
576 died unexpectedly" only if it knows that the tracee has been observed
577 to enter ptrace-stop. Note that there is no guarantee that wait‐
578 pid(WNOHANG) will reliably report the tracee's death status if a ptrace
579 operation returned ESRCH. waitpid(WNOHANG) may return 0 instead. In
580 other words, the tracee may be "not yet fully dead", but already refus‐
581 ing ptrace requests.
582 The tracer can't assume that the tracee always ends its life by report‐
584 ing WIFEXITED(status) or WIFSIGNALED(status); there are cases where
585 this does not occur. For example, if a thread other than thread group
586 leader does an execve(2), it disappears; its PID will never be seen
587 again, and any subsequent ptrace stops will be reported under the
588 thread group leader's PID.
589 Stopped states
591 A tracee can be in two states: running or stopped. For the purposes of
592 ptrace, a tracee which is blocked in a system call (such as read(2),
593 pause(2), etc.) is nevertheless considered to be running, even if the
594 tracee is blocked for a long time. The state of the tracee after
595 PTRACE_LISTEN is somewhat of a gray area: it is not in any ptrace-stop
596 (ptrace commands won't work on it, and it will deliver waitpid(2) noti‐
597 fications), but it also may be considered "stopped" because it is not
598 executing instructions (is not scheduled), and if it was in group-stop
599 before PTRACE_LISTEN, it will not respond to signals until SIGCONT is
600 received.
601 There are many kinds of states when the tracee is stopped, and in
603 ptrace discussions they are often conflated. Therefore, it is impor‐
604 tant to use precise terms.
605 In this manual page, any stopped state in which the tracee is ready to
607 accept ptrace commands from the tracer is called ptrace-stop. Ptrace-
608 stops can be further subdivided into signal-delivery-stop, group-stop,
609 syscall-stop, PTRACE_EVENT stops, and so on. These stopped states are
610 described in detail below.
611 When the running tracee enters ptrace-stop, it notifies its tracer us‐
613 ing waitpid(2) (or one of the other "wait" system calls). Most of this
614 manual page assumes that the tracer waits with:
615 pid = waitpid(pid_or_minus_1, &status, __WALL);
617 Ptrace-stopped tracees are reported as returns with pid greater than 0
619 and WIFSTOPPED(status) true.
620 The __WALL flag does not include the WSTOPPED and WEXITED flags, but
622 implies their functionality.
623 Setting the WCONTINUED flag when calling waitpid(2) is not recommended:
625 the "continued" state is per-process and consuming it can confuse the
626 real parent of the tracee.
627 Use of the WNOHANG flag may cause waitpid(2) to return 0 ("no wait re‐
629 sults available yet") even if the tracer knows there should be a noti‐
630 fication. Example:
631 errno = 0;
633 ptrace(PTRACE_CONT, pid, 0L, 0L);
634 if (errno == ESRCH) {
635 /* tracee is dead */
636 r = waitpid(tracee, &status, __WALL | WNOHANG);
637 /* r can still be 0 here! */
638 }
639 The following kinds of ptrace-stops exist: signal-delivery-stops,
641 group-stops, PTRACE_EVENT stops, syscall-stops. They all are reported
642 by waitpid(2) with WIFSTOPPED(status) true. They may be differentiated
643 by examining the value status>>8, and if there is ambiguity in that
644 value, by querying PTRACE_GETSIGINFO. (Note: the WSTOPSIG(status)
645 macro can't be used to perform this examination, because it returns the
646 value (status>>8) & 0xff.)
647 Signal-delivery-stop
649 When a (possibly multithreaded) process receives any signal except
650 SIGKILL, the kernel selects an arbitrary thread which handles the sig‐
651 nal. (If the signal is generated with tgkill(2), the target thread can
652 be explicitly selected by the caller.) If the selected thread is
653 traced, it enters signal-delivery-stop. At this point, the signal is
654 not yet delivered to the process, and can be suppressed by the tracer.
655 If the tracer doesn't suppress the signal, it passes the signal to the
656 tracee in the next ptrace restart request. This second step of signal
657 delivery is called signal injection in this manual page. Note that if
658 the signal is blocked, signal-delivery-stop doesn't happen until the
659 signal is unblocked, with the usual exception that SIGSTOP can't be
660 blocked.
661 Signal-delivery-stop is observed by the tracer as waitpid(2) returning
663 with WIFSTOPPED(status) true, with the signal returned by WSTOPSIG(sta‐
664 tus). If the signal is SIGTRAP, this may be a different kind of
665 ptrace-stop; see the "Syscall-stops" and "execve" sections below for
666 details. If WSTOPSIG(status) returns a stopping signal, this may be a
667 group-stop; see below.
668 Signal injection and suppression
670 After signal-delivery-stop is observed by the tracer, the tracer should
671 restart the tracee with the call
672 ptrace(PTRACE_restart, pid, 0, sig)
674 where PTRACE_restart is one of the restarting ptrace requests. If sig
676 is 0, then a signal is not delivered. Otherwise, the signal sig is de‐
677 livered. This operation is called signal injection in this manual
678 page, to distinguish it from signal-delivery-stop.
679 The sig value may be different from the WSTOPSIG(status) value: the
681 tracer can cause a different signal to be injected.
682 Note that a suppressed signal still causes system calls to return pre‐
684 maturely. In this case, system calls will be restarted: the tracer
685 will observe the tracee to reexecute the interrupted system call (or
686 restart_syscall(2) system call for a few system calls which use a dif‐
687 ferent mechanism for restarting) if the tracer uses PTRACE_SYSCALL.
688 Even system calls (such as poll(2)) which are not restartable after
689 signal are restarted after signal is suppressed; however, kernel bugs
690 exist which cause some system calls to fail with EINTR even though no
691 observable signal is injected to the tracee.
692 Restarting ptrace commands issued in ptrace-stops other than signal-de‐
694 livery-stop are not guaranteed to inject a signal, even if sig is non‐
695 zero. No error is reported; a nonzero sig may simply be ignored.
696 Ptrace users should not try to "create a new signal" this way: use
697 tgkill(2) instead.
698 The fact that signal injection requests may be ignored when restarting
700 the tracee after ptrace stops that are not signal-delivery-stops is a
701 cause of confusion among ptrace users. One typical scenario is that
702 the tracer observes group-stop, mistakes it for signal-delivery-stop,
703 restarts the tracee with
704 ptrace(PTRACE_restart, pid, 0, stopsig)
706 with the intention of injecting stopsig, but stopsig gets ignored and
708 the tracee continues to run.
709 The SIGCONT signal has a side effect of waking up (all threads of) a
711 group-stopped process. This side effect happens before signal-deliv‐
712 ery-stop. The tracer can't suppress this side effect (it can only sup‐
713 press signal injection, which only causes the SIGCONT handler to not be
714 executed in the tracee, if such a handler is installed). In fact, wak‐
715 ing up from group-stop may be followed by signal-delivery-stop for sig‐
716 nal(s) other than SIGCONT, if they were pending when SIGCONT was deliv‐
717 ered. In other words, SIGCONT may be not the first signal observed by
718 the tracee after it was sent.
719 Stopping signals cause (all threads of) a process to enter group-stop.
721 This side effect happens after signal injection, and therefore can be
722 suppressed by the tracer.
723 In Linux 2.4 and earlier, the SIGSTOP signal can't be injected.
725 PTRACE_GETSIGINFO can be used to retrieve a siginfo_t structure which
727 corresponds to the delivered signal. PTRACE_SETSIGINFO may be used to
728 modify it. If PTRACE_SETSIGINFO has been used to alter siginfo_t, the
729 si_signo field and the sig parameter in the restarting command must
730 match, otherwise the result is undefined.
731 Group-stop
733 When a (possibly multithreaded) process receives a stopping signal, all
734 threads stop. If some threads are traced, they enter a group-stop.
735 Note that the stopping signal will first cause signal-delivery-stop (on
736 one tracee only), and only after it is injected by the tracer (or after
737 it was dispatched to a thread which isn't traced), will group-stop be
738 initiated on all tracees within the multithreaded process. As usual,
739 every tracee reports its group-stop separately to the corresponding
740 tracer.
741 Group-stop is observed by the tracer as waitpid(2) returning with WIF‐
743 STOPPED(status) true, with the stopping signal available via WSTOP‐
744 SIG(status). The same result is returned by some other classes of
745 ptrace-stops, therefore the recommended practice is to perform the call
746 ptrace(PTRACE_GETSIGINFO, pid, 0, &siginfo)
748 The call can be avoided if the signal is not SIGSTOP, SIGTSTP, SIGTTIN,
750 or SIGTTOU; only these four signals are stopping signals. If the
751 tracer sees something else, it can't be a group-stop. Otherwise, the
752 tracer needs to call PTRACE_GETSIGINFO. If PTRACE_GETSIGINFO fails
753 with EINVAL, then it is definitely a group-stop. (Other failure codes
754 are possible, such as ESRCH ("no such process") if a SIGKILL killed the
755 tracee.)
756 If tracee was attached using PTRACE_SEIZE, group-stop is indicated by
758 PTRACE_EVENT_STOP: status>>16 == PTRACE_EVENT_STOP. This allows detec‐
759 tion of group-stops without requiring an extra PTRACE_GETSIGINFO call.
760 As of Linux 2.6.38, after the tracer sees the tracee ptrace-stop and
762 until it restarts or kills it, the tracee will not run, and will not
763 send notifications (except SIGKILL death) to the tracer, even if the
764 tracer enters into another waitpid(2) call.
765 The kernel behavior described in the previous paragraph causes a prob‐
767 lem with transparent handling of stopping signals. If the tracer
768 restarts the tracee after group-stop, the stopping signal is effec‐
769 tively ignored—the tracee doesn't remain stopped, it runs. If the
770 tracer doesn't restart the tracee before entering into the next wait‐
771 pid(2), future SIGCONT signals will not be reported to the tracer; this
772 would cause the SIGCONT signals to have no effect on the tracee.
773 Since Linux 3.4, there is a method to overcome this problem: instead of
775 PTRACE_CONT, a PTRACE_LISTEN command can be used to restart a tracee in
776 a way where it does not execute, but waits for a new event which it can
777 report via waitpid(2) (such as when it is restarted by a SIGCONT).
778 PTRACE_EVENT stops
780 If the tracer sets PTRACE_O_TRACE_* options, the tracee will enter
781 ptrace-stops called PTRACE_EVENT stops.
782 PTRACE_EVENT stops are observed by the tracer as waitpid(2) returning
784 with WIFSTOPPED(status), and WSTOPSIG(status) returns SIGTRAP (or for
785 PTRACE_EVENT_STOP, returns the stopping signal if tracee is in a group-
786 stop). An additional bit is set in the higher byte of the status word:
787 the value status>>8 will be
788 ((PTRACE_EVENT_foo<<8) | SIGTRAP).
790 The following events exist:
792 PTRACE_EVENT_VFORK
794 Stop before return from vfork(2) or clone(2) with the
795 CLONE_VFORK flag. When the tracee is continued after this stop,
796 it will wait for child to exit/exec before continuing its execu‐
797 tion (in other words, the usual behavior on vfork(2)).
798 PTRACE_EVENT_FORK
800 Stop before return from fork(2) or clone(2) with the exit signal
801 set to SIGCHLD.
802 PTRACE_EVENT_CLONE
804 Stop before return from clone(2).
805 PTRACE_EVENT_VFORK_DONE
807 Stop before return from vfork(2) or clone(2) with the
808 CLONE_VFORK flag, but after the child unblocked this tracee by
809 exiting or execing.
810 For all four stops described above, the stop occurs in the parent
812 (i.e., the tracee), not in the newly created thread.
813 PTRACE_GETEVENTMSG can be used to retrieve the new thread's ID.
814 PTRACE_EVENT_EXEC
816 Stop before return from execve(2). Since Linux 3.0,
817 PTRACE_GETEVENTMSG returns the former thread ID.
818 PTRACE_EVENT_EXIT
820 Stop before exit (including death from exit_group(2)), signal
821 death, or exit caused by execve(2) in a multithreaded process.
822 PTRACE_GETEVENTMSG returns the exit status. Registers can be
823 examined (unlike when "real" exit happens). The tracee is still
824 alive; it needs to be PTRACE_CONTed or PTRACE_DETACHed to finish
825 exiting.
826 PTRACE_EVENT_STOP
828 Stop induced by PTRACE_INTERRUPT command, or group-stop, or ini‐
829 tial ptrace-stop when a new child is attached (only if attached
830 using PTRACE_SEIZE).
831 PTRACE_EVENT_SECCOMP
833 Stop triggered by a seccomp(2) rule on tracee syscall entry when
834 PTRACE_O_TRACESECCOMP has been set by the tracer. The seccomp
835 event message data (from the SECCOMP_RET_DATA portion of the
836 seccomp filter rule) can be retrieved with PTRACE_GETEVENTMSG.
837 The semantics of this stop are described in detail in a separate
838 section below.
839 PTRACE_GETSIGINFO on PTRACE_EVENT stops returns SIGTRAP in si_signo,
841 with si_code set to (event<<8) | SIGTRAP.
842 Syscall-stops
844 If the tracee was restarted by PTRACE_SYSCALL or PTRACE_SYSEMU, the
845 tracee enters syscall-enter-stop just prior to entering any system call
846 (which will not be executed if the restart was using PTRACE_SYSEMU, re‐
847 gardless of any change made to registers at this point or how the
848 tracee is restarted after this stop). No matter which method caused
849 the syscall-entry-stop, if the tracer restarts the tracee with
850 PTRACE_SYSCALL, the tracee enters syscall-exit-stop when the system
851 call is finished, or if it is interrupted by a signal. (That is, sig‐
852 nal-delivery-stop never happens between syscall-enter-stop and syscall-
853 exit-stop; it happens after syscall-exit-stop.). If the tracee is con‐
854 tinued using any other method (including PTRACE_SYSEMU), no syscall-
855 exit-stop occurs. Note that all mentions PTRACE_SYSEMU apply equally
856 to PTRACE_SYSEMU_SINGLESTEP.
857 However, even if the tracee was continued using PTRACE_SYSCALL, it is
859 not guaranteed that the next stop will be a syscall-exit-stop. Other
860 possibilities are that the tracee may stop in a PTRACE_EVENT stop (in‐
861 cluding seccomp stops), exit (if it entered _exit(2) or exit_group(2)),
862 be killed by SIGKILL, or die silently (if it is a thread group leader,
863 the execve(2) happened in another thread, and that thread is not traced
864 by the same tracer; this situation is discussed later).
865 Syscall-enter-stop and syscall-exit-stop are observed by the tracer as
867 waitpid(2) returning with WIFSTOPPED(status) true, and WSTOPSIG(status)
868 giving SIGTRAP. If the PTRACE_O_TRACESYSGOOD option was set by the
869 tracer, then WSTOPSIG(status) will give the value (SIGTRAP | 0x80).
870 Syscall-stops can be distinguished from signal-delivery-stop with SIG‐
872 TRAP by querying PTRACE_GETSIGINFO for the following cases:
873 si_code <= 0
875 SIGTRAP was delivered as a result of a user-space action, for
876 example, a system call (tgkill(2), kill(2), sigqueue(3), etc.),
877 expiration of a POSIX timer, change of state on a POSIX message
878 queue, or completion of an asynchronous I/O request.
879 si_code == SI_KERNEL (0x80)
881 SIGTRAP was sent by the kernel.
882 si_code == SIGTRAP or si_code == (SIGTRAP|0x80)
884 This is a syscall-stop.
885 However, syscall-stops happen very often (twice per system call), and
887 performing PTRACE_GETSIGINFO for every syscall-stop may be somewhat ex‐
888 pensive.
889 Some architectures allow the cases to be distinguished by examining
891 registers. For example, on x86, rax == -ENOSYS in syscall-enter-stop.
892 Since SIGTRAP (like any other signal) always happens after syscall-
893 exit-stop, and at this point rax almost never contains -ENOSYS, the
894 SIGTRAP looks like "syscall-stop which is not syscall-enter-stop"; in
895 other words, it looks like a "stray syscall-exit-stop" and can be de‐
896 tected this way. But such detection is fragile and is best avoided.
897 Using the PTRACE_O_TRACESYSGOOD option is the recommended method to
899 distinguish syscall-stops from other kinds of ptrace-stops, since it is
900 reliable and does not incur a performance penalty.
901 Syscall-enter-stop and syscall-exit-stop are indistinguishable from
903 each other by the tracer. The tracer needs to keep track of the se‐
904 quence of ptrace-stops in order to not misinterpret syscall-enter-stop
905 as syscall-exit-stop or vice versa. In general, a syscall-enter-stop
906 is always followed by syscall-exit-stop, PTRACE_EVENT stop, or the
907 tracee's death; no other kinds of ptrace-stop can occur in between.
908 However, note that seccomp stops (see below) can cause syscall-exit-
909 stops, without preceding syscall-entry-stops. If seccomp is in use,
910 care needs to be taken not to misinterpret such stops as syscall-entry-
911 stops.
912 If after syscall-enter-stop, the tracer uses a restarting command other
914 than PTRACE_SYSCALL, syscall-exit-stop is not generated.
915 PTRACE_GETSIGINFO on syscall-stops returns SIGTRAP in si_signo, with
917 si_code set to SIGTRAP or (SIGTRAP|0x80).
918 PTRACE_EVENT_SECCOMP stops (Linux 3.5 to Linux 4.7)
920 The behavior of PTRACE_EVENT_SECCOMP stops and their interaction with
921 other kinds of ptrace stops has changed between kernel versions. This
922 documents the behavior from their introduction until Linux 4.7 (inclu‐
923 sive). The behavior in later kernel versions is documented in the next
924 section.
925 A PTRACE_EVENT_SECCOMP stop occurs whenever a SECCOMP_RET_TRACE rule is
927 triggered. This is independent of which methods was used to restart
928 the system call. Notably, seccomp still runs even if the tracee was
929 restarted using PTRACE_SYSEMU and this system call is unconditionally
930 skipped.
931 Restarts from this stop will behave as if the stop had occurred right
933 before the system call in question. In particular, both PTRACE_SYSCALL
934 and PTRACE_SYSEMU will normally cause a subsequent syscall-entry-stop.
935 However, if after the PTRACE_EVENT_SECCOMP the system call number is
936 negative, both the syscall-entry-stop and the system call itself will
937 be skipped. This means that if the system call number is negative af‐
938 ter a PTRACE_EVENT_SECCOMP and the tracee is restarted using
939 PTRACE_SYSCALL, the next observed stop will be a syscall-exit-stop,
940 rather than the syscall-entry-stop that might have been expected.
941 PTRACE_EVENT_SECCOMP stops (since Linux 4.8)
943 Starting with Linux 4.8, the PTRACE_EVENT_SECCOMP stop was reordered to
944 occur between syscall-entry-stop and syscall-exit-stop. Note that sec‐
945 comp no longer runs (and no PTRACE_EVENT_SECCOMP will be reported) if
946 the system call is skipped due to PTRACE_SYSEMU.
947 Functionally, a PTRACE_EVENT_SECCOMP stop functions comparably to a
949 syscall-entry-stop (i.e., continuations using PTRACE_SYSCALL will cause
950 syscall-exit-stops, the system call number may be changed and any other
951 modified registers are visible to the to-be-executed system call as
952 well). Note that there may be, but need not have been a preceding
953 syscall-entry-stop.
954 After a PTRACE_EVENT_SECCOMP stop, seccomp will be rerun, with a SEC‐
956 COMP_RET_TRACE rule now functioning the same as a SECCOMP_RET_ALLOW.
957 Specifically, this means that if registers are not modified during the
958 PTRACE_EVENT_SECCOMP stop, the system call will then be allowed.
959 PTRACE_SINGLESTEP stops
961 [Details of these kinds of stops are yet to be documented.]
962 Informational and restarting ptrace commands
964 Most ptrace commands (all except PTRACE_ATTACH, PTRACE_SEIZE,
965 PTRACE_TRACEME, PTRACE_INTERRUPT, and PTRACE_KILL) require the tracee
966 to be in a ptrace-stop, otherwise they fail with ESRCH.
967 When the tracee is in ptrace-stop, the tracer can read and write data
969 to the tracee using informational commands. These commands leave the
970 tracee in ptrace-stopped state:
971 ptrace(PTRACE_PEEKTEXT/PEEKDATA/PEEKUSER, pid, addr, 0);
973 ptrace(PTRACE_POKETEXT/POKEDATA/POKEUSER, pid, addr, long_val);
974 ptrace(PTRACE_GETREGS/GETFPREGS, pid, 0, &struct);
975 ptrace(PTRACE_SETREGS/SETFPREGS, pid, 0, &struct);
976 ptrace(PTRACE_GETREGSET, pid, NT_foo, &iov);
977 ptrace(PTRACE_SETREGSET, pid, NT_foo, &iov);
978 ptrace(PTRACE_GETSIGINFO, pid, 0, &siginfo);
979 ptrace(PTRACE_SETSIGINFO, pid, 0, &siginfo);
980 ptrace(PTRACE_GETEVENTMSG, pid, 0, &long_var);
981 ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_flags);
982 Note that some errors are not reported. For example, setting signal
984 information (siginfo) may have no effect in some ptrace-stops, yet the
985 call may succeed (return 0 and not set errno); querying
986 PTRACE_GETEVENTMSG may succeed and return some random value if current
987 ptrace-stop is not documented as returning a meaningful event message.
988 The call
990 ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_flags);
992 affects one tracee. The tracee's current flags are replaced. Flags
994 are inherited by new tracees created and "auto-attached" via active
995 PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, or PTRACE_O_TRACECLONE op‐
996 tions.
997 Another group of commands makes the ptrace-stopped tracee run. They
999 have the form:
1000 ptrace(cmd, pid, 0, sig);
1002 where cmd is PTRACE_CONT, PTRACE_LISTEN, PTRACE_DETACH, PTRACE_SYSCALL,
1004 PTRACE_SINGLESTEP, PTRACE_SYSEMU, or PTRACE_SYSEMU_SINGLESTEP. If the
1005 tracee is in signal-delivery-stop, sig is the signal to be injected (if
1006 it is nonzero). Otherwise, sig may be ignored. (When restarting a
1007 tracee from a ptrace-stop other than signal-delivery-stop, recommended
1008 practice is to always pass 0 in sig.)
1009 Attaching and detaching
1011 A thread can be attached to the tracer using the call
1012 ptrace(PTRACE_ATTACH, pid, 0, 0);
1014 or
1016 ptrace(PTRACE_SEIZE, pid, 0, PTRACE_O_flags);
1018 PTRACE_ATTACH sends SIGSTOP to this thread. If the tracer wants this
1020 SIGSTOP to have no effect, it needs to suppress it. Note that if other
1021 signals are concurrently sent to this thread during attach, the tracer
1022 may see the tracee enter signal-delivery-stop with other signal(s)
1023 first! The usual practice is to reinject these signals until SIGSTOP
1024 is seen, then suppress SIGSTOP injection. The design bug here is that
1025 a ptrace attach and a concurrently delivered SIGSTOP may race and the
1026 concurrent SIGSTOP may be lost.
1027 Since attaching sends SIGSTOP and the tracer usually suppresses it,
1029 this may cause a stray EINTR return from the currently executing system
1030 call in the tracee, as described in the "Signal injection and suppres‐
1031 sion" section.
1032 Since Linux 3.4, PTRACE_SEIZE can be used instead of PTRACE_ATTACH.
1034 PTRACE_SEIZE does not stop the attached process. If you need to stop
1035 it after attach (or at any other time) without sending it any signals,
1036 use PTRACE_INTERRUPT command.
1037 The request
1039 ptrace(PTRACE_TRACEME, 0, 0, 0);
1041 turns the calling thread into a tracee. The thread continues to run
1043 (doesn't enter ptrace-stop). A common practice is to follow the
1044 PTRACE_TRACEME with
1045 raise(SIGSTOP);
1047 and allow the parent (which is our tracer now) to observe our signal-
1049 delivery-stop.
1050 If the PTRACE_O_TRACEFORK, PTRACE_O_TRACEVFORK, or PTRACE_O_TRACECLONE
1052 options are in effect, then children created by, respectively, vfork(2)
1053 or clone(2) with the CLONE_VFORK flag, fork(2) or clone(2) with the
1054 exit signal set to SIGCHLD, and other kinds of clone(2), are automati‐
1055 cally attached to the same tracer which traced their parent. SIGSTOP
1056 is delivered to the children, causing them to enter signal-delivery-
1057 stop after they exit the system call which created them.
1058 Detaching of the tracee is performed by:
1060 ptrace(PTRACE_DETACH, pid, 0, sig);
1062 PTRACE_DETACH is a restarting operation; therefore it requires the
1064 tracee to be in ptrace-stop. If the tracee is in signal-delivery-stop,
1065 a signal can be injected. Otherwise, the sig parameter may be silently
1066 ignored.
1067 If the tracee is running when the tracer wants to detach it, the usual
1069 solution is to send SIGSTOP (using tgkill(2), to make sure it goes to
1070 the correct thread), wait for the tracee to stop in signal-delivery-
1071 stop for SIGSTOP and then detach it (suppressing SIGSTOP injection). A
1072 design bug is that this can race with concurrent SIGSTOPs. Another
1073 complication is that the tracee may enter other ptrace-stops and needs
1074 to be restarted and waited for again, until SIGSTOP is seen. Yet an‐
1075 other complication is to be sure that the tracee is not already ptrace-
1076 stopped, because no signal delivery happens while it is—not even
1077 SIGSTOP.
1078 If the tracer dies, all tracees are automatically detached and
1080 restarted, unless they were in group-stop. Handling of restart from
1081 group-stop is currently buggy, but the "as planned" behavior is to
1082 leave tracee stopped and waiting for SIGCONT. If the tracee is
1083 restarted from signal-delivery-stop, the pending signal is injected.
1084 execve(2) under ptrace
1086 When one thread in a multithreaded process calls execve(2), the kernel
1087 destroys all other threads in the process, and resets the thread ID of
1088 the execing thread to the thread group ID (process ID). (Or, to put
1089 things another way, when a multithreaded process does an execve(2), at
1090 completion of the call, it appears as though the execve(2) occurred in
1091 the thread group leader, regardless of which thread did the execve(2).)
1092 This resetting of the thread ID looks very confusing to tracers:
1093 • All other threads stop in PTRACE_EVENT_EXIT stop, if the
1095 PTRACE_O_TRACEEXIT option was turned on. Then all other threads ex‐
1096 cept the thread group leader report death as if they exited via
1097 _exit(2) with exit code 0.
1098 • The execing tracee changes its thread ID while it is in the ex‐
1100 ecve(2). (Remember, under ptrace, the "pid" returned from wait‐
1101 pid(2), or fed into ptrace calls, is the tracee's thread ID.) That
1102 is, the tracee's thread ID is reset to be the same as its process
1103 ID, which is the same as the thread group leader's thread ID.
1104 • Then a PTRACE_EVENT_EXEC stop happens, if the PTRACE_O_TRACEEXEC op‐
1106 tion was turned on.
1107 • If the thread group leader has reported its PTRACE_EVENT_EXIT stop
1109 by this time, it appears to the tracer that the dead thread leader
1110 "reappears from nowhere". (Note: the thread group leader does not
1111 report death via WIFEXITED(status) until there is at least one other
1112 live thread. This eliminates the possibility that the tracer will
1113 see it dying and then reappearing.) If the thread group leader was
1114 still alive, for the tracer this may look as if thread group leader
1115 returns from a different system call than it entered, or even "re‐
1116 turned from a system call even though it was not in any system
1117 call". If the thread group leader was not traced (or was traced by
1118 a different tracer), then during execve(2) it will appear as if it
1119 has become a tracee of the tracer of the execing tracee.
1120 All of the above effects are the artifacts of the thread ID change in
1122 the tracee.
1123 The PTRACE_O_TRACEEXEC option is the recommended tool for dealing with
1125 this situation. First, it enables PTRACE_EVENT_EXEC stop, which occurs
1126 before execve(2) returns. In this stop, the tracer can use
1127 PTRACE_GETEVENTMSG to retrieve the tracee's former thread ID. (This
1128 feature was introduced in Linux 3.0.) Second, the PTRACE_O_TRACEEXEC
1129 option disables legacy SIGTRAP generation on execve(2).
1130 When the tracer receives PTRACE_EVENT_EXEC stop notification, it is
1132 guaranteed that except this tracee and the thread group leader, no
1133 other threads from the process are alive.
1134 On receiving the PTRACE_EVENT_EXEC stop notification, the tracer should
1136 clean up all its internal data structures describing the threads of
1137 this process, and retain only one data structure—one which describes
1138 the single still running tracee, with
1139 thread ID == thread group ID == process ID.
1141 Example: two threads call execve(2) at the same time:
1143 *** we get syscall-enter-stop in thread 1: **
1145 PID1 execve("/bin/foo", "foo" <unfinished ...>
1146 *** we issue PTRACE_SYSCALL for thread 1 **
1147 *** we get syscall-enter-stop in thread 2: **
1148 PID2 execve("/bin/bar", "bar" <unfinished ...>
1149 *** we issue PTRACE_SYSCALL for thread 2 **
1150 *** we get PTRACE_EVENT_EXEC for PID0, we issue PTRACE_SYSCALL **
1151 *** we get syscall-exit-stop for PID0: **
1152 PID0 <... execve resumed> ) = 0
1153 If the PTRACE_O_TRACEEXEC option is not in effect for the execing
1155 tracee, and if the tracee was PTRACE_ATTACHed rather that
1156 PTRACE_SEIZEd, the kernel delivers an extra SIGTRAP to the tracee after
1157 execve(2) returns. This is an ordinary signal (similar to one which
1158 can be generated by kill -TRAP), not a special kind of ptrace-stop.
1159 Employing PTRACE_GETSIGINFO for this signal returns si_code set to 0
1160 (SI_USER). This signal may be blocked by signal mask, and thus may be
1161 delivered (much) later.
1162 Usually, the tracer (for example, strace(1)) would not want to show
1164 this extra post-execve SIGTRAP signal to the user, and would suppress
1165 its delivery to the tracee (if SIGTRAP is set to SIG_DFL, it is a
1166 killing signal). However, determining which SIGTRAP to suppress is not
1167 easy. Setting the PTRACE_O_TRACEEXEC option or using PTRACE_SEIZE and
1168 thus suppressing this extra SIGTRAP is the recommended approach.
1169 Real parent
1171 The ptrace API (ab)uses the standard UNIX parent/child signaling over
1172 waitpid(2). This used to cause the real parent of the process to stop
1173 receiving several kinds of waitpid(2) notifications when the child
1174 process is traced by some other process.
1175 Many of these bugs have been fixed, but as of Linux 2.6.38 several
1177 still exist; see BUGS below.
1178 As of Linux 2.6.38, the following is believed to work correctly:
1180 • exit/death by signal is reported first to the tracer, then, when the
1182 tracer consumes the waitpid(2) result, to the real parent (to the
1183 real parent only when the whole multithreaded process exits). If
1184 the tracer and the real parent are the same process, the report is
1185 sent only once.
1186
1188 On success, the PTRACE_PEEK* requests return the requested data (but
1189 see NOTES), the PTRACE_SECCOMP_GET_FILTER request returns the number of
1190 instructions in the BPF program, the PTRACE_GET_SYSCALL_INFO request
1191 returns the number of bytes available to be written by the kernel, and
1192 other requests return zero.
1193 On error, all requests return -1, and errno is set to indicate the er‐
1195 ror. Since the value returned by a successful PTRACE_PEEK* request may
1196 be -1, the caller must clear errno before the call, and then check it
1197 afterward to determine whether or not an error occurred.
1198
1200 EBUSY (i386 only) There was an error with allocating or freeing a de‐
1201 bug register.
1202 EFAULT There was an attempt to read from or write to an invalid area in
1204 the tracer's or the tracee's memory, probably because the area
1205 wasn't mapped or accessible. Unfortunately, under Linux, dif‐
1206 ferent variations of this fault will return EIO or EFAULT more
1207 or less arbitrarily.
1208 EINVAL An attempt was made to set an invalid option.
1210 EIO request is invalid, or an attempt was made to read from or write
1212 to an invalid area in the tracer's or the tracee's memory, or
1213 there was a word-alignment violation, or an invalid signal was
1214 specified during a restart request.
1215 EPERM The specified process cannot be traced. This could be because
1217 the tracer has insufficient privileges (the required capability
1218 is CAP_SYS_PTRACE); unprivileged processes cannot trace pro‐
1219 cesses that they cannot send signals to or those running set-
1220 user-ID/set-group-ID programs, for obvious reasons. Alterna‐
1221 tively, the process may already be being traced, or (before
1222 Linux 2.6.26) be init(1) (PID 1).
1223 ESRCH The specified process does not exist, or is not currently being
1225 traced by the caller, or is not stopped (for requests that re‐
1226 quire a stopped tracee).
1227
1229 None.
1230
1232 SVr4, 4.3BSD.
1233 Before Linux 2.6.26, init(1), the process with PID 1, may not be
1235 traced.
1236
1238 Although arguments to ptrace() are interpreted according to the proto‐
1239 type given, glibc currently declares ptrace() as a variadic function
1240 with only the request argument fixed. It is recommended to always sup‐
1241 ply four arguments, even if the requested operation does not use them,
1242 setting unused/ignored arguments to 0L or (void *) 0.
1243 A tracees parent continues to be the tracer even if that tracer calls
1245 execve(2).
1246 The layout of the contents of memory and the USER area are quite oper‐
1248 ating-system- and architecture-specific. The offset supplied, and the
1249 data returned, might not entirely match with the definition of struct
1250 user.
1251 The size of a "word" is determined by the operating-system variant
1253 (e.g., for 32-bit Linux it is 32 bits).
1254 This page documents the way the ptrace() call works currently in Linux.
1256 Its behavior differs significantly on other flavors of UNIX. In any
1257 case, use of ptrace() is highly specific to the operating system and
1258 architecture.
1259 Ptrace access mode checking
1261 Various parts of the kernel-user-space API (not just ptrace() opera‐
1262 tions), require so-called "ptrace access mode" checks, whose outcome
1263 determines whether an operation is permitted (or, in a few cases,
1264 causes a "read" operation to return sanitized data). These checks are
1265 performed in cases where one process can inspect sensitive information
1266 about, or in some cases modify the state of, another process. The
1267 checks are based on factors such as the credentials and capabilities of
1268 the two processes, whether or not the "target" process is dumpable, and
1269 the results of checks performed by any enabled Linux Security Module
1270 (LSM)—for example, SELinux, Yama, or Smack—and by the commoncap LSM
1271 (which is always invoked).
1272 Prior to Linux 2.6.27, all access checks were of a single type. Since
1274 Linux 2.6.27, two access mode levels are distinguished:
1275 PTRACE_MODE_READ
1277 For "read" operations or other operations that are less danger‐
1278 ous, such as: get_robust_list(2); kcmp(2); reading
1279 /proc/pid/auxv, /proc/pid/environ, or /proc/pid/stat; or read‐
1280 link(2) of a /proc/pid/ns/* file.
1281 PTRACE_MODE_ATTACH
1283 For "write" operations, or other operations that are more dan‐
1284 gerous, such as: ptrace attaching (PTRACE_ATTACH) to another
1285 process or calling process_vm_writev(2). (PTRACE_MODE_ATTACH
1286 was effectively the default before Linux 2.6.27.)
1287 Since Linux 4.5, the above access mode checks are combined (ORed) with
1289 one of the following modifiers:
1290 PTRACE_MODE_FSCREDS
1292 Use the caller's filesystem UID and GID (see credentials(7)) or
1293 effective capabilities for LSM checks.
1294 PTRACE_MODE_REALCREDS
1296 Use the caller's real UID and GID or permitted capabilities for
1297 LSM checks. This was effectively the default before Linux 4.5.
1298 Because combining one of the credential modifiers with one of the
1300 aforementioned access modes is typical, some macros are defined in the
1301 kernel sources for the combinations:
1302 PTRACE_MODE_READ_FSCREDS
1304 Defined as PTRACE_MODE_READ | PTRACE_MODE_FSCREDS.
1305 PTRACE_MODE_READ_REALCREDS
1307 Defined as PTRACE_MODE_READ | PTRACE_MODE_REALCREDS.
1308 PTRACE_MODE_ATTACH_FSCREDS
1310 Defined as PTRACE_MODE_ATTACH | PTRACE_MODE_FSCREDS.
1311 PTRACE_MODE_ATTACH_REALCREDS
1313 Defined as PTRACE_MODE_ATTACH | PTRACE_MODE_REALCREDS.
1314 One further modifier can be ORed with the access mode:
1316 PTRACE_MODE_NOAUDIT (since Linux 3.3)
1318 Don't audit this access mode check. This modifier is employed
1319 for ptrace access mode checks (such as checks when reading
1320 /proc/pid/stat) that merely cause the output to be filtered or
1321 sanitized, rather than causing an error to be returned to the
1322 caller. In these cases, accessing the file is not a security
1323 violation and there is no reason to generate a security audit
1324 record. This modifier suppresses the generation of such an au‐
1325 dit record for the particular access check.
1326 Note that all of the PTRACE_MODE_* constants described in this subsec‐
1328 tion are kernel-internal, and not visible to user space. The constant
1329 names are mentioned here in order to label the various kinds of ptrace
1330 access mode checks that are performed for various system calls and ac‐
1331 cesses to various pseudofiles (e.g., under /proc). These names are
1332 used in other manual pages to provide a simple shorthand for labeling
1333 the different kernel checks.
1334 The algorithm employed for ptrace access mode checking determines
1336 whether the calling process is allowed to perform the corresponding ac‐
1337 tion on the target process. (In the case of opening /proc/pid files,
1338 the "calling process" is the one opening the file, and the process with
1339 the corresponding PID is the "target process".) The algorithm is as
1340 follows:
1341 (1) If the calling thread and the target thread are in the same thread
1343 group, access is always allowed.
1344 (2) If the access mode specifies PTRACE_MODE_FSCREDS, then, for the
1346 check in the next step, employ the caller's filesystem UID and
1347 GID. (As noted in credentials(7), the filesystem UID and GID al‐
1348 most always have the same values as the corresponding effective
1349 IDs.)
1350 Otherwise, the access mode specifies PTRACE_MODE_REALCREDS, so use
1352 the caller's real UID and GID for the checks in the next step.
1353 (Most APIs that check the caller's UID and GID use the effective
1354 IDs. For historical reasons, the PTRACE_MODE_REALCREDS check uses
1355 the real IDs instead.)
1356 (3) Deny access if neither of the following is true:
1358 • The real, effective, and saved-set user IDs of the target match
1360 the caller's user ID, and the real, effective, and saved-set
1361 group IDs of the target match the caller's group ID.
1362 • The caller has the CAP_SYS_PTRACE capability in the user name‐
1364 space of the target.
1365 (4) Deny access if the target process "dumpable" attribute has a value
1367 other than 1 (SUID_DUMP_USER; see the discussion of
1368 PR_SET_DUMPABLE in prctl(2)), and the caller does not have the
1369 CAP_SYS_PTRACE capability in the user namespace of the target
1370 process.
1371 (5) The kernel LSM security_ptrace_access_check() interface is invoked
1373 to see if ptrace access is permitted. The results depend on the
1374 LSM(s). The implementation of this interface in the commoncap LSM
1375 performs the following steps:
1376 (5.1) If the access mode includes PTRACE_MODE_FSCREDS, then use
1378 the caller's effective capability set in the following
1379 check; otherwise (the access mode specifies PTRACE_MODE_RE‐
1380 ALCREDS, so) use the caller's permitted capability set.
1381 (5.2) Deny access if neither of the following is true:
1383 • The caller and the target process are in the same user
1385 namespace, and the caller's capabilities are a superset
1386 of the target process's permitted capabilities.
1387 • The caller has the CAP_SYS_PTRACE capability in the tar‐
1389 get process's user namespace.
1390 Note that the commoncap LSM does not distinguish between
1392 PTRACE_MODE_READ and PTRACE_MODE_ATTACH.
1393 (6) If access has not been denied by any of the preceding steps, then
1395 access is allowed.
1396 /proc/sys/kernel/yama/ptrace_scope
1398 On systems with the Yama Linux Security Module (LSM) installed (i.e.,
1399 the kernel was configured with CONFIG_SECURITY_YAMA), the
1400 /proc/sys/kernel/yama/ptrace_scope file (available since Linux 3.4) can
1401 be used to restrict the ability to trace a process with ptrace() (and
1402 thus also the ability to use tools such as strace(1) and gdb(1)). The
1403 goal of such restrictions is to prevent attack escalation whereby a
1404 compromised process can ptrace-attach to other sensitive processes
1405 (e.g., a GPG agent or an SSH session) owned by the user in order to
1406 gain additional credentials that may exist in memory and thus expand
1407 the scope of the attack.
1408 More precisely, the Yama LSM limits two types of operations:
1410 • Any operation that performs a ptrace access mode PTRACE_MODE_ATTACH
1412 check—for example, ptrace() PTRACE_ATTACH. (See the "Ptrace access
1413 mode checking" discussion above.)
1414 • ptrace() PTRACE_TRACEME.
1416 A process that has the CAP_SYS_PTRACE capability can update the
1418 /proc/sys/kernel/yama/ptrace_scope file with one of the following val‐
1419 ues:
1420 0 ("classic ptrace permissions")
1422 No additional restrictions on operations that perform
1423 PTRACE_MODE_ATTACH checks (beyond those imposed by the commoncap
1424 and other LSMs).
1425 The use of PTRACE_TRACEME is unchanged.
1427 1 ("restricted ptrace") [default value]
1429 When performing an operation that requires a PTRACE_MODE_ATTACH
1430 check, the calling process must either have the CAP_SYS_PTRACE
1431 capability in the user namespace of the target process or it
1432 must have a predefined relationship with the target process. By
1433 default, the predefined relationship is that the target process
1434 must be a descendant of the caller.
1435 A target process can employ the prctl(2) PR_SET_PTRACER opera‐
1437 tion to declare an additional PID that is allowed to perform
1438 PTRACE_MODE_ATTACH operations on the target. See the kernel
1439 source file Documentation/admin-guide/LSM/Yama.rst (or Documen‐
1440 tation/security/Yama.txt before Linux 4.13) for further details.
1441 The use of PTRACE_TRACEME is unchanged.
1443 2 ("admin-only attach")
1445 Only processes with the CAP_SYS_PTRACE capability in the user
1446 namespace of the target process may perform PTRACE_MODE_ATTACH
1447 operations or trace children that employ PTRACE_TRACEME.
1448 3 ("no attach")
1450 No process may perform PTRACE_MODE_ATTACH operations or trace
1451 children that employ PTRACE_TRACEME.
1452 Once this value has been written to the file, it cannot be
1454 changed.
1455 With respect to values 1 and 2, note that creating a new user namespace
1457 effectively removes the protection offered by Yama. This is because a
1458 process in the parent user namespace whose effective UID matches the
1459 UID of the creator of a child namespace has all capabilities (including
1460 CAP_SYS_PTRACE) when performing operations within the child user name‐
1461 space (and further-removed descendants of that namespace). Conse‐
1462 quently, when a process tries to use user namespaces to sandbox itself,
1463 it inadvertently weakens the protections offered by the Yama LSM.
1464 C library/kernel differences
1466 At the system call level, the PTRACE_PEEKTEXT, PTRACE_PEEKDATA, and
1467 PTRACE_PEEKUSER requests have a different API: they store the result at
1468 the address specified by the data parameter, and the return value is
1469 the error flag. The glibc wrapper function provides the API given in
1470 DESCRIPTION above, with the result being returned via the function re‐
1471 turn value.
1472
1474 On hosts with Linux 2.6 kernel headers, PTRACE_SETOPTIONS is declared
1475 with a different value than the one for Linux 2.4. This leads to ap‐
1476 plications compiled with Linux 2.6 kernel headers failing when run on
1477 Linux 2.4. This can be worked around by redefining PTRACE_SETOPTIONS
1478 to PTRACE_OLDSETOPTIONS, if that is defined.
1479 Group-stop notifications are sent to the tracer, but not to real par‐
1481 ent. Last confirmed on 2.6.38.6.
1482 If a thread group leader is traced and exits by calling _exit(2), a
1484 PTRACE_EVENT_EXIT stop will happen for it (if requested), but the sub‐
1485 sequent WIFEXITED notification will not be delivered until all other
1486 threads exit. As explained above, if one of other threads calls ex‐
1487 ecve(2), the death of the thread group leader will never be reported.
1488 If the execed thread is not traced by this tracer, the tracer will
1489 never know that execve(2) happened. One possible workaround is to
1490 PTRACE_DETACH the thread group leader instead of restarting it in this
1491 case. Last confirmed on 2.6.38.6.
1492 A SIGKILL signal may still cause a PTRACE_EVENT_EXIT stop before actual
1494 signal death. This may be changed in the future; SIGKILL is meant to
1495 always immediately kill tasks even under ptrace. Last confirmed on
1496 Linux 3.13.
1497 Some system calls return with EINTR if a signal was sent to a tracee,
1499 but delivery was suppressed by the tracer. (This is very typical oper‐
1500 ation: it is usually done by debuggers on every attach, in order to not
1501 introduce a bogus SIGSTOP). As of Linux 3.2.9, the following system
1502 calls are affected (this list is likely incomplete): epoll_wait(2), and
1503 read(2) from an inotify(7) file descriptor. The usual symptom of this
1504 bug is that when you attach to a quiescent process with the command
1505 strace -p <process-ID>
1507 then, instead of the usual and expected one-line output such as
1509 restart_syscall(<... resuming interrupted call ...>_
1511 or
1513 select(6, [5], NULL, [5], NULL_
1515 ('_' denotes the cursor position), you observe more than one line. For
1517 example:
1518 clock_gettime(CLOCK_MONOTONIC, {15370, 690928118}) = 0
1520 epoll_wait(4,_
1521 What is not visible here is that the process was blocked in
1523 epoll_wait(2) before strace(1) has attached to it. Attaching caused
1524 epoll_wait(2) to return to user space with the error EINTR. In this
1525 particular case, the program reacted to EINTR by checking the current
1526 time, and then executing epoll_wait(2) again. (Programs which do not
1527 expect such "stray" EINTR errors may behave in an unintended way upon
1528 an strace(1) attach.)
1529 Contrary to the normal rules, the glibc wrapper for ptrace() can set
1531 errno to zero.
1532
1534 gdb(1), ltrace(1), strace(1), clone(2), execve(2), fork(2), gettid(2),
1535 prctl(2), seccomp(2), sigaction(2), tgkill(2), vfork(2), waitpid(2),
1536 exec(3), capabilities(7), signal(7)
1537Linux man-pages 6.04 2023-03-30 ptrace(2)