man.bsd.lv manual page server

Manual Page Search Parameters
PTRACE(2) System Calls Manual PTRACE(2)

ptrace — process tracing and debugging

library “libc”

#include <sys/types.h>
#include <sys/ptrace.h>

int
ptrace(int request, pid_t pid, void *addr, int data);

() provides tracing and debugging facilities. It allows one process (the process) to control another (the process). Most of the time, the traced process runs normally, but when it receives a signal (see sigaction(2)), it stops. The tracing process is expected to notice this via wait(2) or the delivery of a SIGCHLD signal (see siginfo(2)), examine the state of the stopped process, and cause it to terminate or continue as appropriate. ptrace() is the mechanism by which all this happens.

When a process that is traced by a debugger requests and calls execve(2) or any of the routines built on it (such as execv(3)), it will stop before executing the first instruction of the new image and emit SIGTRAP with si_code set to TRAP_EXEC. If a traced program calls execve(2) any setuid or setgid bits on the executable being executed will be ignored.

Program (software) breakpoints are reported with SIGTRAP and the si_code value set to TRAP_BKPT. These breakpoints are machine specific instructions that interrupt the process. In order to put a trap by a tracer into the tracee's program, debugger must violate the PaX MPROTECT restrictions. For details check the security.pax.mprotect.ptrace option described in sysctl(7). When a tracee is interrupted by a trap, the trap is not removed by the kernel and it must be handled by a debugger.

If a program is traced with single steps (PT_STEP) it reports each step with SIGTRAP with si_code set to TRAP_TRACE. This event is not maskable PT_SET_EVENT_MASK.

Child program traps are reported with SIGTRAP and the si_code value set to TRAP_CHLD. These events are by default disabled and can be configured with PT_SET_EVENT_MASK. If this event occurs, check with PT_GET_PROCESS_STATE the details of the process state associated with this event.

Design choices for Debug Register accessors

A debugger might reuse port specific symbols, to help writing portable code as described in the port specific part of the <sys/ptrace.h> header. Among these symbols, there are:

The request argument of ptrace specifies what operation is being performed; the meaning of the rest of the arguments depends on the operation, but except for one special case noted below, all ptrace calls are made by the tracing process, and the pid argument specifies the process ID of the traced process. request can be:

This request is the only one used by the traced process; it declares that the process expects to be traced by its parent. All the other arguments are ignored. If the parent process does not expect to trace the child, it will probably be rather confused by the results; once the traced process stops, it cannot be made to continue except via ().

This call does not stop the process neither emit SIGSTOP to parent.

, PT_READ_D
These requests read a single int of data from the traced process' address space. Traditionally, () has allowed for machines with distinct address spaces for instruction and data, which is why there are two requests: conceptually, PT_READ_I reads from the instruction space and PT_READ_D reads from the data space. In the current NetBSD implementation, these two requests are completely identical. The addr argument specifies the address (in the traced process' virtual address space) at which the read is to be done. This address does not have to meet any alignment constraints. The value read is returned as the return value from ptrace().

These operations return success on incomplete and cancelled byte transfers. New software shall use PT_IO as it allows to check whether a byte transfer was completed.

, PT_WRITE_D
These requests parallel PT_READ_I and PT_READ_D, except that they write rather than read. The data argument supplies the value to be written.

New software shall use PT_IO as it allows to check whether an operation was completed.

The traced process continues execution. addr is an address specifying the place where execution is to be resumed (a new value for the program counter), or (void *)1 to indicate that execution is to pick up where it left off. data provides a signal number to be delivered to the traced process as it resumes execution, or 0 if no signal is to be sent. If a negative value is supplied, that is the negative of the LWP ID of the thread to be resumed, and only that thread executes.
The traced process terminates, as if PT_CONTINUE has been used with SIGKILL given as the signal to be delivered. However, unlike PT_CONTINUE, PT_KILL can be used on a non-stopped tracee. The addr and data arguments are ignored.
The traced process stops, as if kill has been used with SIGSTOP given as the signal to be delivered. wait(2) will report the child (again) as stopped even if it was stopped before. The addr and data arguments are ignored. Unlike PT_CONTINUE call with SIGSTOP, PT_STOP works both on running and stopped processes.
This request allows a process to gain control of an otherwise unrelated process and begin tracing it. It does not need any cooperation from the to-be-traced process. In this case, pid specifies the process ID of the to-be-traced process, and the other two arguments are ignored. This request requires that the target process must have the same real UID as the tracing process, and that it must not be executing a setuid or setgid executable. (If the tracing process is running as root, these restrictions do not apply.)

The tracing process will see the newly-traced process stop and may then control it as if it had been traced all along. It means that the SIGSTOP signal is emitted to tracer. It is different behavior to the one from PT_TRACE_ME.

Three other restrictions apply to all tracing processes, even those running as root. First, no process may trace a system process. Second, no process may trace the process running init(8). Third, if a process has its root directory set with chroot(2), it may not trace another process unless that process' root directory is at or below the tracing process' root.

This request is like PT_CONTINUE, except that after it succeeds, the traced process is no longer traced and continues execution normally.
This request is a more general interface that can be used instead of PT_READ_D, PT_WRITE_D, PT_READ_I, and PT_WRITE_I. The I/O request is encoded in a struct ptrace_io_desc defined as: 
struct ptrace_io_desc {
	int	piod_op;
	void	*piod_offs;
	void	*piod_addr;
	size_t	piod_len;
};

where piod_offs is the offset within the traced process where the I/O operation should take place, piod_addr is the buffer in the tracing process, and piod_len is the length of the I/O request. The piod_op field specifies which type of I/O operation to perform. Possible values are:

 
 
 
 
 

See the description of PT_READ_I for the difference between I and D spaces.

The PIOD_READ_AUXV operation can be used to read from the ELF auxiliary vector. The piod_offs argument sets the offset within the tracee's vector. To read from the beginning of it, this value must be set to 0 and cast to (void *).

A pointer to the I/O descriptor is passed in the addr argument to (). On return, the piod_len field in the I/O descriptor will be updated with the actual number of bytes transferred. If the requested I/O could not be successfully performed, ptrace() will return -1 and set errno.

This interface returns success for partial and cancelled byte transfers. For an interrupted transfer, a user shall check whether occurred at least a single of the following two conditions: piod_len == 0 and set errno. Successful but incomplete byte transfers shall be restarted in the place where they were stopped.

Makes the process specified in the pid pid generate a core dump. The addr argument should contain the name of the core file to be generated and the data argument should contain the length of the core filename.
Returns information about a thread from the list of threads for the process specified in the pid argument. The addr argument should contain a struct ptrace_lwpinfo defined as: 
struct ptrace_lwpinfo {
	lwpid_t pl_lwpid;
	int pl_event;
};

where pl_lwpid contains a thread LWP ID. Information is returned for the thread following the one with the specified ID in the process thread list, or for the first thread if pl_lwpid is 0. Upon return pl_lwpid contains the LWP ID of the thread that was found, or 0 if there is no thread after the one whose LWP ID was supplied in the call. pl_event contains the event that stopped the thread. Possible values are:

 
 
 

The data argument should contain “sizeof(struct ptrace_lwpinfo)”.

Stops a process before and after executing each system call. Otherwise this operation is the same as PT_CONTINUE.
Intercept and ignore a system call before it has been executed, for use with PT_SYSCALL. This operation shall be called for syscall entry trap from PT_SYSCALL. To resume execution after intercepting the system call, another PT_SYSCALL shall be used.
This request can be used to specify which events in the traced process should be reported to the tracing process. These events are specified in a struct ptrace_event defined as: 
typedef struct ptrace_event {
	int	pe_set_event;
} ptrace_event_t;

pe_set_event is the set of events to be reported. This set is formed by OR'ing together the following values:

PTRACE_FORK
Report fork(2).
PTRACE_VFORK
Report vfork(2).
PTRACE_VFORK_DONE
Report parent resumed after vfork(2).
PTRACE_LWP_CREATE
Report thread birth.
PTRACE_LWP_EXIT
Report thread termination.
PTRACE_POSIX_SPAWN
Report posix_spawn(3).

The fork(2) and vfork(2) events can occur with clone(2). The PTRACE_FORK value means that process gives birth to its child without pending on its termination or execve(2) operation. If enabled, the child is also traced by the debugger and SIGTRAP is generated twice, first for the parent and second for the child. The PTRACE_VFORK event is the same as PTRACE_FORK, but the parent blocks after giving birth to the child. The PTRACE_VFORK_DONE event can be used to report unblocking of the parent.

() on NetBSD directly creates the child process without intermediate fork. The PTRACE_POSIX_SPAWN event semantics are the same as PTRACE_FORK, but the child is reported with implied execution of a file.

A pointer to this structure is passed in addr. The data argument should be set to sizeof(struct ptrace_event).

This request can be used to determine which events in the traced process will be reported. The information is read into the struct ptrace_event pointed to by addr. The data argument should be set to sizeof(struct ptrace_event).
This request reads the state information associated with the event that stopped the traced process. The information is reported in a struct ptrace_state defined as: 
typedef struct ptrace_state {
	int	pe_report_event;
	pid_t	pe_other_pid;
} ptrace_state_t;

A pointer to this structure is passed in addr. The data argument should be set to sizeof(struct ptrace_state).

This request can be used to specify signal information emitted to tracee. This signal information is specified in struct ptrace_siginfo defined as: 
typedef struct ptrace_siginfo {
	siginfo_t	psi_siginfo;
	lwpid_t		psi_lwpid;
} ptrace_siginfo_t;

Where psi_siginfo is the set to signal information structure. The psi_lwpid field describes LWP address of the signal. Value 0 means the whole process (route signal to all LWPs).

A pointer to this structure is passed in addr. The data argument should be set to sizeof(struct ptrace_siginfo).

In order to pass faked signal to the tracee, the signal type must match the signal passed to the process with PT_CONTINUE or PT_SYSCALL.

This request can be used to determine signal information that was received by a debugger (see siginfo(2)). The information is read into the struct ptrace_siginfo pointed to by addr. The data argument should be set to sizeof(struct ptrace_siginfo).
Allow execution of a specified thread, change its state from suspended to continued. The addr argument is unused. The data argument specifies the LWP ID.

This call is equivalent to _lwp_continue(2) called by a traced process. This call does not change the general process state from stopped to continued.

Prevent execution of a specified thread, change its state from continued to suspended. The addr argument is unused. The data argument specifies the requested LWP ID.

This call is equivalent to _lwp_suspend(2) called by a traced process. This call does not change the general process state from continued to stopped.

Additionally, the following requests exist but are not available on all machine architectures. The file <machine/ptrace.h> lists which requests exist on a given machine.

Execution continues as in request PT_CONTINUE; however as soon as possible after execution of at least one instruction, execution stops again. If the data argument is greater than 0, it contains the LWP ID of the thread to be stepped, and any other threads are continued. If the data argument is less than zero, it contains the negative of the LWP ID of the thread to be stepped, and only that thread executes.
This request will turn on single stepping of the specified thread. addr is unused. data specifies the LWP ID of the thread to be stepped. The execution does not continue until PT_CONTINUE is issued. This request permits combining single-stepping with sending signals and PT_SYSCALL.
This request will turn off single stepping of the specified thread. addr is unused. data specifies the LWP ID of the thread to disable single-stepping.
This request reads the traced process' machine registers into the struct reg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read.
This request is the converse of PT_GETREGS; it loads the traced process' machine registers from the struct reg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written.
This request reads the traced process' floating-point registers into the struct fpreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read.
This request is the converse of PT_GETFPREGS; it loads the traced process' floating-point registers from the struct fpreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written.
This request reads the traced process' debug registers into the struct dbreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read.
This request is the converse of PT_GETDBREGS; it loads the traced process' debug registers from the struct dbreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written.
This request reads the traced process' XMM registers into the struct xmmregs (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read.
This request is the converse of PT_GETXMMREGS; it loads the traced process' XMM registers from the struct xmmregs (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written.
This request reads the traced process' vector registers into the struct vreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read.
This request is the converse of PT_GETVECREGS; it loads the traced process' vector registers from the struct vreg (defined in <machine/reg.h>) pointed to by addr. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written.
This request reads the traced process' FPU extended state into the struct xstate (defined in <machine/cpu_extended_state.h>). addr should be a pointer to struct iovec (defined in <sys/uio.h>) specifying the pointer to the aforementioned struct as iov_base and its size as iov_len. The data argument contains the LWP ID of the thread whose registers are to be read. If zero is supplied, the first thread of the process is read. The struct will be filled up to the specified iov_len. The caller needs to check the xs_rfbm bitmap in order to determine which fields were provided by the CPU, and may check xs_xstate_bv to determine which component states were changed from the initial state.
This request is the converse of PT_GETXSTATE; it loads the traced process' extended FPU state from the struct xstate (defined in <machine/cpu_extended_state.h>). addr should be a pointer to struct iovec (defined in <sys/uio.h>) specifying the pointer to the aforementioned struct as iov_base and its size as iov_len. The data argument contains the LWP ID of the thread whose registers are to be written. If zero is supplied, the first thread of the process is written. The xs_rfbm field of the supplied xstate specifies which state components are to be updated. Other components (fields) will be ignored. The xs_xstate_bv field specifies whether component state should be set to provided values (when 1) or reset to unitialized (when 0). The request will fail if xs_xstate_bv is not a subset of xs_rfbm, or any of the specified components is not supported by the CPU or kernel (i.e., not returned by PT_GETXSTATE).

Some requests can cause ptrace() to return -1 as a non-error value; to disambiguate, errno can be set to 0 before the call and checked afterwards. The possible errors are:

[]
Process is currently exec'ing and cannot be traced.
[]
  • was attempted on a process that was already being traced.
  • A request attempted to manipulate a process that was being traced by some process other than the one making the request.
  • A request (other than PT_ATTACH) specified a process that wasn't stopped.
[]
An attempt to unstop a process with locked threads.
[]
  • A process attempted to use PT_ATTACH on itself.
  • The request was not a legal request on this machine architecture.
  • The signal number (in data) to PT_CONTINUE was neither 0 nor a legal signal number.
  • , PT_SETREGS, PT_GETFPREGS, PT_SETFPREGS, PT_GETXSTATE, or PT_SETXSTATE was attempted on a process with no valid register set. (This is normally true only of system processes.)
  • A process attempted to set Program Counter to 0 in PT_CONTINUE, PT_SYSCALL or PT_DETACH with vm.user_va0_disable set to 1.
  • attempted to set state components not supported by the kernel, or xs_xstate_bv was not a subset of xs_rfbm.
[]
  • A request (other than PT_ATTACH) attempted to manipulate a process that wasn't being traced at all.
  • An attempt was made to use PT_ATTACH on a process in violation of the requirements listed under PT_ATTACH above.
[]
No process having the specified process ID exists.

sigaction(2), signal(7)

The ptrace() function appeared in Version 7 AT&T UNIX.

On the SPARC, the PC is set to the provided PC value for PT_CONTINUE and similar calls, but the NPC is set willy-nilly to 4 greater than the PC value. Using PT_GETREGS and PT_SETREGS to modify the PC, passing (void *)1 to ptrace(), should be able to sidestep this.

PT_SET_SIGINFO, PT_RESUME and PT_SUSPEND can change the image of process returned by PT_LWPINFO.

October 9, 2019 NetBSD-9.2