Remove support for systems without PTRACE_SETOPTIONS

Assume that the kernel is v2.5.46 or newer, i.e. PTRACE_SETOPTIONS
and PTRACE_O_TRACESYSGOOD|PTRACE_O_TRACEEXEC|PTRACE_O_TRACECLONE
are universally available.

This change removes all code that implemented post-execve SIGTRAP
handling and fork/vfork/clone->CLONE_PTRACE substitution.

* defs.h (TCB_BPTSET, TCB_WAITEXECVE): Remove macros.
(need_fork_exec_workarounds, setbpt, clearbpt): Remove declarations.
* strace.c (need_fork_exec_workarounds,
test_ptrace_setoptions_followfork, test_ptrace_setoptions_for_all):
Remove.
(syscall_trap_sig): Set to (SIGTRAP | 0x80).
(ptrace_setoptions): Set to (PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC).
(detach): Do not test for TCB_BPTSET.
(init): Do not call test_ptrace_setoptions_followfork and
test_ptrace_setoptions_for_all.  Do not test for TCB_BPTSET.
* syscall.c (syscall_fixup_on_sysenter, internal_fork, internal_exec,
syscall_fixup_for_fork_exec, syscall_fixup_on_sysexit): Remove.
(trace_syscall_entering): Do not test for TCB_WAITEXECVE.  Do not call
syscall_fixup_on_sysenter and syscall_fixup_for_fork_exec.
(trace_syscall_exiting): Do not call syscall_fixup_on_sysexit and
syscall_fixup_for_fork_exec.
[IA64] (ia64_ia32mode): Make static.
* linux/ia64/arch_regs.h (ia64_ia32mode): Remove declaration.
* util.c: Do not include "syscall.h".
(arg_setup, get_arg0, get_arg1, set_arg0, set_arg1, restore_arg0,
restore_arg1, arg_finish_change, change_syscall, setbpt, clearbpt):
Remove.
* tests/ptrace_setoptions.test: Remove.
* tests/Makefile.am (TESTS): Remove it.
This commit is contained in:
Дмитрий Левин 2015-02-08 13:05:53 +00:00
parent 1c5feadf75
commit 23ce9e48b5
7 changed files with 17 additions and 949 deletions

41
defs.h
View File

@ -274,41 +274,13 @@ struct tcb {
* are limited to trace(), this condition is never observed in trace_syscall() * are limited to trace(), this condition is never observed in trace_syscall()
* and below. * and below.
* The bit is cleared after all syscall exit processing is done. * The bit is cleared after all syscall exit processing is done.
* User-generated SIGTRAPs and post-execve SIGTRAP make it necessary
* to be very careful and NOT set TCB_INSYSCALL bit when they are encountered.
* TCB_WAITEXECVE bit is used for this purpose (see below).
* *
* Use entering(tcp) / exiting(tcp) to check this bit to make code more readable. * Use entering(tcp) / exiting(tcp) to check this bit to make code more readable.
*/ */
#define TCB_INSYSCALL 0x04 #define TCB_INSYSCALL 0x04
#define TCB_ATTACHED 0x08 /* We attached to it already */ #define TCB_ATTACHED 0x08 /* We attached to it already */
#define TCB_BPTSET 0x10 /* "Breakpoint" set after fork(2) */ #define TCB_REPRINT 0x10 /* We should reprint this syscall on exit */
#define TCB_REPRINT 0x20 /* We should reprint this syscall on exit */ #define TCB_FILTERED 0x20 /* This system call has been filtered out */
#define TCB_FILTERED 0x40 /* This system call has been filtered out */
/*
* x86 does not need TCB_WAITEXECVE.
* It can detect post-execve SIGTRAP by looking at eax/rax.
* See "not a syscall entry (eax = %ld)\n" message.
*
* Note! On new kernels (about 2.5.46+), we use PTRACE_O_TRACEEXEC, which
* suppresses post-execve SIGTRAP. If you are adding a new arch which is
* only supported by newer kernels, you most likely don't need to define
* TCB_WAITEXECVE!
*/
#if defined(ALPHA) \
|| defined(SPARC) || defined(SPARC64) \
|| defined(POWERPC) \
|| defined(IA64) \
|| defined(HPPA) \
|| defined(SH) || defined(SH64) \
|| defined(S390) || defined(S390X) \
|| defined(ARM) \
|| defined(MIPS)
/* This tracee has entered into execve syscall. Expect post-execve SIGTRAP
* to happen. (When it is detected, tracee is continued and this bit is cleared.)
*/
# define TCB_WAITEXECVE 0x80
#endif
/* qualifier flags */ /* qualifier flags */
#define QUAL_TRACE 0x001 /* this system call should be traced */ #define QUAL_TRACE 0x001 /* this system call should be traced */
@ -405,7 +377,6 @@ extern bool hide_log_until_execve;
/* are we filtering traces based on paths? */ /* are we filtering traces based on paths? */
extern const char **paths_selected; extern const char **paths_selected;
#define tracing_paths (paths_selected != NULL) #define tracing_paths (paths_selected != NULL)
extern bool need_fork_exec_workarounds;
extern unsigned xflag; extern unsigned xflag;
extern unsigned followfork; extern unsigned followfork;
#ifdef USE_LIBUNWIND #ifdef USE_LIBUNWIND
@ -455,14 +426,6 @@ extern int upeek(int pid, long, long *);
#if defined(SPARC) || defined(SPARC64) || defined(IA64) || defined(SH) #if defined(SPARC) || defined(SPARC64) || defined(IA64) || defined(SH)
extern long getrval2(struct tcb *); extern long getrval2(struct tcb *);
#endif #endif
/*
* On Linux, "setbpt" is a misnomer: we don't set a breakpoint
* (IOW: no poking in user's text segment),
* instead we change fork/vfork/clone into clone(CLONE_PTRACE).
* On newer kernels, we use PTRACE_O_TRACECLONE/TRACE[V]FORK instead.
*/
extern int setbpt(struct tcb *);
extern int clearbpt(struct tcb *);
extern const char *signame(const int); extern const char *signame(const int);
extern void pathtrace_select(const char *); extern void pathtrace_select(const char *);

View File

@ -1,4 +1,2 @@
#include <asm/ptrace_offsets.h> #include <asm/ptrace_offsets.h>
#include <asm/rse.h> #include <asm/rse.h>
extern bool ia64_ia32mode;

253
strace.c
View File

@ -74,18 +74,17 @@ bool stack_trace_enabled = false;
# define fork() vfork() # define fork() vfork()
#endif #endif
const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
cflag_t cflag = CFLAG_NONE; cflag_t cflag = CFLAG_NONE;
unsigned int followfork = 0; unsigned int followfork = 0;
unsigned int ptrace_setoptions = 0; unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC;
unsigned int xflag = 0; unsigned int xflag = 0;
bool need_fork_exec_workarounds = 0;
bool debug_flag = 0; bool debug_flag = 0;
bool Tflag = 0; bool Tflag = 0;
bool iflag = 0; bool iflag = 0;
bool count_wallclock = 0; bool count_wallclock = 0;
unsigned int qflag = 0; unsigned int qflag = 0;
/* Which WSTOPSIG(status) value marks syscall traps? */
static unsigned int syscall_trap_sig = SIGTRAP;
static unsigned int tflag = 0; static unsigned int tflag = 0;
static bool rflag = 0; static bool rflag = 0;
static bool print_pid_pfx = 0; static bool print_pid_pfx = 0;
@ -347,7 +346,7 @@ ptrace_attach_or_seize(int pid)
int r; int r;
if (!use_seize) if (!use_seize)
return ptrace(PTRACE_ATTACH, pid, 0L, 0L); return ptrace(PTRACE_ATTACH, pid, 0L, 0L);
r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long)ptrace_setoptions); r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
if (r) if (r)
return r; return r;
r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L); r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
@ -766,9 +765,6 @@ detach(struct tcb *tcp)
int error; int error;
int status; int status;
if (tcp->flags & TCB_BPTSET)
clearbpt(tcp);
/* /*
* Linux wrongly insists the child be stopped * Linux wrongly insists the child be stopped
* before detaching. Arghh. We go through hoops * before detaching. Arghh. We go through hoops
@ -1321,219 +1317,6 @@ startup_child(char **argv)
} }
} }
/*
* Test whether the kernel support PTRACE_O_TRACECLONE et al options.
* First fork a new child, call ptrace with PTRACE_SETOPTIONS on it,
* and then see which options are supported by the kernel.
*/
static int
test_ptrace_setoptions_followfork(void)
{
int pid, expected_grandchild = 0, found_grandchild = 0;
const unsigned int test_options = PTRACE_O_TRACECLONE |
PTRACE_O_TRACEFORK |
PTRACE_O_TRACEVFORK;
/* Need fork for test. NOMMU has no forks */
if (NOMMU_SYSTEM)
goto worked; /* be bold, and pretend that test succeeded */
pid = fork();
if (pid < 0)
perror_msg_and_die("fork");
if (pid == 0) {
pid = getpid();
if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0)
perror_msg_and_die("%s: PTRACE_TRACEME doesn't work",
__func__);
kill_save_errno(pid, SIGSTOP);
if (fork() < 0)
perror_msg_and_die("fork");
_exit(0);
}
while (1) {
int status, tracee_pid;
errno = 0;
tracee_pid = wait(&status);
if (tracee_pid <= 0) {
if (errno == EINTR)
continue;
if (errno == ECHILD)
break;
kill_save_errno(pid, SIGKILL);
perror_msg_and_die("%s: unexpected wait result %d",
__func__, tracee_pid);
}
if (WIFEXITED(status)) {
if (WEXITSTATUS(status)) {
if (tracee_pid != pid)
kill_save_errno(pid, SIGKILL);
error_msg_and_die("%s: unexpected exit status %u",
__func__, WEXITSTATUS(status));
}
continue;
}
if (WIFSIGNALED(status)) {
if (tracee_pid != pid)
kill_save_errno(pid, SIGKILL);
error_msg_and_die("%s: unexpected signal %u",
__func__, WTERMSIG(status));
}
if (!WIFSTOPPED(status)) {
if (tracee_pid != pid)
kill_save_errno(tracee_pid, SIGKILL);
kill_save_errno(pid, SIGKILL);
error_msg_and_die("%s: unexpected wait status %x",
__func__, status);
}
if (tracee_pid != pid) {
found_grandchild = tracee_pid;
if (ptrace(PTRACE_CONT, tracee_pid, 0, 0) < 0) {
kill_save_errno(tracee_pid, SIGKILL);
kill_save_errno(pid, SIGKILL);
perror_msg_and_die("PTRACE_CONT doesn't work");
}
continue;
}
switch (WSTOPSIG(status)) {
case SIGSTOP:
if (ptrace(PTRACE_SETOPTIONS, pid, 0, test_options) < 0
&& errno != EINVAL && errno != EIO)
perror_msg("PTRACE_SETOPTIONS");
break;
case SIGTRAP:
if (status >> 16 == PTRACE_EVENT_FORK) {
long msg = 0;
if (ptrace(PTRACE_GETEVENTMSG, pid,
NULL, (long) &msg) == 0)
expected_grandchild = msg;
}
break;
}
if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0) {
kill_save_errno(pid, SIGKILL);
perror_msg_and_die("PTRACE_SYSCALL doesn't work");
}
}
if (expected_grandchild && expected_grandchild == found_grandchild) {
worked:
ptrace_setoptions |= test_options;
if (debug_flag)
fprintf(stderr, "ptrace_setoptions = %#x\n",
ptrace_setoptions);
return 0;
}
error_msg("Test for PTRACE_O_TRACECLONE failed, "
"giving up using this feature.");
return 1;
}
/*
* Test whether the kernel support PTRACE_O_TRACESYSGOOD.
* First fork a new child, call ptrace(PTRACE_SETOPTIONS) on it,
* and then see whether it will stop with (SIGTRAP | 0x80).
*
* Use of this option enables correct handling of user-generated SIGTRAPs,
* and SIGTRAPs generated by special instructions such as int3 on x86:
# compile with: gcc -nostartfiles -nostdlib -o int3 int3.S
_start: .globl _start
int3
movl $42, %ebx
movl $1, %eax
int $0x80
*/
static int
test_ptrace_setoptions_for_all(void)
{
const unsigned int test_options = PTRACE_O_TRACESYSGOOD |
PTRACE_O_TRACEEXEC;
int pid;
int it_worked = 0;
/* Need fork for test. NOMMU has no forks */
if (NOMMU_SYSTEM)
goto worked; /* be bold, and pretend that test succeeded */
pid = fork();
if (pid < 0)
perror_msg_and_die("fork");
if (pid == 0) {
pid = getpid();
if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0)
/* Note: exits with exitcode 1 */
perror_msg_and_die("%s: PTRACE_TRACEME doesn't work",
__func__);
kill(pid, SIGSTOP);
_exit(0); /* parent should see entry into this syscall */
}
while (1) {
int status, tracee_pid;
errno = 0;
tracee_pid = wait(&status);
if (tracee_pid <= 0) {
if (errno == EINTR)
continue;
kill_save_errno(pid, SIGKILL);
perror_msg_and_die("%s: unexpected wait result %d",
__func__, tracee_pid);
}
if (WIFEXITED(status)) {
if (WEXITSTATUS(status) == 0)
break;
error_msg_and_die("%s: unexpected exit status %u",
__func__, WEXITSTATUS(status));
}
if (WIFSIGNALED(status)) {
error_msg_and_die("%s: unexpected signal %u",
__func__, WTERMSIG(status));
}
if (!WIFSTOPPED(status)) {
kill(pid, SIGKILL);
error_msg_and_die("%s: unexpected wait status %x",
__func__, status);
}
if (WSTOPSIG(status) == SIGSTOP) {
/*
* We don't check "options aren't accepted" error.
* If it happens, we'll never get (SIGTRAP | 0x80),
* and thus will decide to not use the option.
* IOW: the outcome of the test will be correct.
*/
if (ptrace(PTRACE_SETOPTIONS, pid, 0L, test_options) < 0
&& errno != EINVAL && errno != EIO)
perror_msg("PTRACE_SETOPTIONS");
}
if (WSTOPSIG(status) == (SIGTRAP | 0x80)) {
it_worked = 1;
}
if (ptrace(PTRACE_SYSCALL, pid, 0L, 0L) < 0) {
kill_save_errno(pid, SIGKILL);
perror_msg_and_die("PTRACE_SYSCALL doesn't work");
}
}
if (it_worked) {
worked:
syscall_trap_sig = (SIGTRAP | 0x80);
ptrace_setoptions |= test_options;
if (debug_flag)
fprintf(stderr, "ptrace_setoptions = %#x\n",
ptrace_setoptions);
return 0;
}
error_msg("Test for PTRACE_O_TRACESYSGOOD failed, "
"giving up using this feature.");
return 1;
}
#if USE_SEIZE #if USE_SEIZE
static void static void
test_ptrace_seize(void) test_ptrace_seize(void)
@ -1871,14 +1654,12 @@ init(int argc, char *argv[])
run_gid = getgid(); run_gid = getgid();
} }
/*
* On any reasonably recent Linux kernel (circa about 2.5.46)
* need_fork_exec_workarounds should stay 0 after these tests:
*/
/*need_fork_exec_workarounds = 0; - already is */
if (followfork) if (followfork)
need_fork_exec_workarounds = test_ptrace_setoptions_followfork(); ptrace_setoptions |= PTRACE_O_TRACECLONE |
need_fork_exec_workarounds |= test_ptrace_setoptions_for_all(); PTRACE_O_TRACEFORK |
PTRACE_O_TRACEVFORK;
if (debug_flag)
fprintf(stderr, "ptrace_setoptions = %#x\n", ptrace_setoptions);
test_ptrace_seize(); test_ptrace_seize();
/* Check if they want to redirect the output. */ /* Check if they want to redirect the output. */
@ -2226,20 +2007,7 @@ startup_tcb(struct tcb *tcp)
tcp->flags &= ~TCB_STARTUP; tcp->flags &= ~TCB_STARTUP;
if (tcp->flags & TCB_BPTSET) { if (!use_seize) {
/*
* One example is a breakpoint inherited from
* parent through fork().
*/
if (clearbpt(tcp) < 0) {
/* Pretty fatal */
droptcb(tcp);
exit_code = 1;
return false;
}
}
if (!use_seize && ptrace_setoptions) {
if (debug_flag) if (debug_flag)
fprintf(stderr, "setting opts 0x%x on pid %d\n", fprintf(stderr, "setting opts 0x%x on pid %d\n",
ptrace_setoptions, tcp->pid); ptrace_setoptions, tcp->pid);
@ -2460,7 +2228,6 @@ show_stopsig:
/* /*
* This should be syscall entry or exit. * This should be syscall entry or exit.
* (Or it still can be that pesky post-execve SIGTRAP!)
* Handle it. * Handle it.
*/ */
if (trace_syscall(tcp) < 0) { if (trace_syscall(tcp) < 0) {

213
syscall.c
View File

@ -702,10 +702,10 @@ static struct iovec x86_io = {
# define ARCH_REGS_FOR_GETREGSET x86_regs_union # define ARCH_REGS_FOR_GETREGSET x86_regs_union
# define ARCH_IOVEC_FOR_GETREGSET x86_io # define ARCH_IOVEC_FOR_GETREGSET x86_io
#elif defined(IA64) #elif defined(IA64)
bool ia64_ia32mode = 0; /* not static */ static bool ia64_ia32mode;
static long ia64_r8, ia64_r10; static long ia64_r8, ia64_r10;
#elif defined(POWERPC) #elif defined(POWERPC)
struct pt_regs ppc_regs; struct pt_regs ppc_regs; /* not static */
# define ARCH_REGS_FOR_GETREGS ppc_regs # define ARCH_REGS_FOR_GETREGS ppc_regs
#elif defined(M68K) #elif defined(M68K)
static long m68k_d0; static long m68k_d0;
@ -771,7 +771,7 @@ static long sh64_r9;
#elif defined(CRISV10) || defined(CRISV32) #elif defined(CRISV10) || defined(CRISV32)
static long cris_r10; static long cris_r10;
#elif defined(TILE) #elif defined(TILE)
struct pt_regs tile_regs; struct pt_regs tile_regs; /* not static */
# define ARCH_REGS_FOR_GETREGS tile_regs # define ARCH_REGS_FOR_GETREGS tile_regs
#elif defined(MICROBLAZE) #elif defined(MICROBLAZE)
static long microblaze_r3; static long microblaze_r3;
@ -1607,173 +1607,6 @@ is_negated_errno(kernel_ulong_t val)
return val >= max; return val >= max;
} }
/* Called at each syscall entry.
* Returns:
* 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
* 1: ok, continue in trace_syscall_entering().
* other: error, trace_syscall_entering() should print error indicator
* ("????" etc) and bail out.
*/
static int
syscall_fixup_on_sysenter(struct tcb *tcp)
{
/* Do we have post-execve SIGTRAP suppressed? */
if (ptrace_setoptions & PTRACE_O_TRACEEXEC)
return 1;
/*
* No, unfortunately. Apply -ENOSYS heuristics.
* We don't have to workaround SECCOMP_RET_ERRNO side effects
* because any kernel with SECCOMP_RET_ERRNO support surely
* implements PTRACE_O_TRACEEXEC.
*/
#if defined(I386)
if (i386_regs.eax != -ENOSYS) {
if (debug_flag)
fprintf(stderr, "not a syscall entry (eax = %ld)\n",
i386_regs.eax);
return 0;
}
#elif defined(X86_64) || defined(X32)
if (x86_io.iov_len == sizeof(i386_regs)) {
if ((int) i386_regs.eax != -ENOSYS) {
if (debug_flag)
fprintf(stderr,
"not a syscall entry (eax = %d)\n",
(int) i386_regs.eax);
return 0;
}
} else {
if ((long long) x86_64_regs.rax != -ENOSYS) {
if (debug_flag)
fprintf(stderr,
"not a syscall entry (rax = %lld)\n",
(long long) x86_64_regs.rax);
return 0;
}
}
#elif defined(M68K)
/* TODO? Eliminate upeek's in arches below like we did in x86 */
if (upeek(tcp->pid, 4*PT_D0, &m68k_d0) < 0)
return -1;
if (m68k_d0 != -ENOSYS) {
if (debug_flag)
fprintf(stderr, "not a syscall entry (d0 = %ld)\n", m68k_d0);
return 0;
}
#elif defined(IA64)
if (upeek(tcp->pid, PT_R10, &ia64_r10) < 0)
return -1;
if (upeek(tcp->pid, PT_R8, &ia64_r8) < 0)
return -1;
if (ia64_ia32mode && ia64_r8 != -ENOSYS) {
if (debug_flag)
fprintf(stderr, "not a syscall entry (r8 = %ld)\n", ia64_r8);
return 0;
}
#elif defined(CRISV10) || defined(CRISV32)
if (upeek(tcp->pid, 4*PT_R10, &cris_r10) < 0)
return -1;
if (cris_r10 != -ENOSYS) {
if (debug_flag)
fprintf(stderr, "not a syscall entry (r10 = %ld)\n", cris_r10);
return 0;
}
#elif defined(MICROBLAZE)
if (upeek(tcp->pid, 3 * 4, &microblaze_r3) < 0)
return -1;
if (microblaze_r3 != -ENOSYS) {
if (debug_flag)
fprintf(stderr, "not a syscall entry (r3 = %ld)\n", microblaze_r3);
return 0;
}
#endif
return 1;
}
static void
internal_fork(struct tcb *tcp)
{
#if defined S390 || defined S390X || defined CRISV10 || defined CRISV32
# define ARG_FLAGS 1
#else
# define ARG_FLAGS 0
#endif
#ifndef CLONE_UNTRACED
# define CLONE_UNTRACED 0x00800000
#endif
if ((ptrace_setoptions
& (PTRACE_O_TRACECLONE | PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK))
== (PTRACE_O_TRACECLONE | PTRACE_O_TRACEFORK | PTRACE_O_TRACEVFORK))
return;
if (!followfork)
return;
if (entering(tcp)) {
/*
* We won't see the new child if clone is called with
* CLONE_UNTRACED, so we keep the same logic with that option
* and don't trace it.
*/
if ((tcp->s_ent->sys_func == sys_clone)
&& (tcp->u_arg[ARG_FLAGS] & CLONE_UNTRACED)
)
return;
setbpt(tcp);
} else {
if (tcp->flags & TCB_BPTSET)
clearbpt(tcp);
}
}
#if defined(TCB_WAITEXECVE)
static void
internal_exec(struct tcb *tcp)
{
/* Maybe we have post-execve SIGTRAP suppressed? */
if (ptrace_setoptions & PTRACE_O_TRACEEXEC)
return; /* yes, no need to do anything */
if (exiting(tcp) && syserror(tcp))
/* Error in execve, no post-execve SIGTRAP expected */
tcp->flags &= ~TCB_WAITEXECVE;
else
tcp->flags |= TCB_WAITEXECVE;
}
#endif
static void
syscall_fixup_for_fork_exec(struct tcb *tcp)
{
/*
* We must always trace a few critical system calls in order to
* correctly support following forks in the presence of tracing
* qualifiers.
*/
int (*func)();
func = tcp->s_ent->sys_func;
if ( sys_fork == func
|| sys_clone == func
) {
internal_fork(tcp);
return;
}
#if defined(TCB_WAITEXECVE)
if ( sys_execve == func
# if defined(SPARC) || defined(SPARC64)
|| sys_execv == func
# endif
) {
internal_exec(tcp);
return;
}
#endif
}
/* Return -1 on error or 1 on success (never 0!) */ /* Return -1 on error or 1 on success (never 0!) */
static int static int
get_syscall_args(struct tcb *tcp) get_syscall_args(struct tcb *tcp)
@ -2008,24 +1841,11 @@ trace_syscall_entering(struct tcb *tcp)
{ {
int res, scno_good; int res, scno_good;
#if defined TCB_WAITEXECVE
if (tcp->flags & TCB_WAITEXECVE) {
/* This is the post-execve SIGTRAP. */
tcp->flags &= ~TCB_WAITEXECVE;
return 0;
}
#endif
scno_good = res = (get_regs_error ? -1 : get_scno(tcp)); scno_good = res = (get_regs_error ? -1 : get_scno(tcp));
if (res == 0) if (res == 0)
return res; return res;
if (res == 1) { if (res == 1)
res = syscall_fixup_on_sysenter(tcp); res = get_syscall_args(tcp);
if (res == 0)
return res;
if (res == 1)
res = get_syscall_args(tcp);
}
if (res != 1) { if (res != 1) {
printleader(tcp); printleader(tcp);
@ -2068,9 +1888,6 @@ trace_syscall_entering(struct tcb *tcp)
} }
#endif #endif
if (need_fork_exec_workarounds)
syscall_fixup_for_fork_exec(tcp);
if (!(tcp->qual_flg & QUAL_TRACE) if (!(tcp->qual_flg & QUAL_TRACE)
|| (tracing_paths && !pathtrace_match(tcp)) || (tracing_paths && !pathtrace_match(tcp))
) { ) {
@ -2170,23 +1987,6 @@ get_syscall_result(struct tcb *tcp)
return 1; return 1;
} }
/* Called at each syscall exit */
static void
syscall_fixup_on_sysexit(struct tcb *tcp)
{
#if defined(S390) || defined(S390X)
if ((tcp->flags & TCB_WAITEXECVE)
&& (s390_gpr2 == -ENOSYS || s390_gpr2 == tcp->scno)) {
/*
* Return from execve.
* Fake a return value of zero. We leave the TCB_WAITEXECVE
* flag set for the post-execve SIGTRAP to see and reset.
*/
s390_gpr2 = 0;
}
#endif
}
/* Returns: /* Returns:
* 1: ok, continue in trace_syscall_exiting(). * 1: ok, continue in trace_syscall_exiting().
* -1: error, trace_syscall_exiting() should print error indicator * -1: error, trace_syscall_exiting() should print error indicator
@ -2510,10 +2310,7 @@ trace_syscall_exiting(struct tcb *tcp)
#endif #endif
res = (get_regs_error ? -1 : get_syscall_result(tcp)); res = (get_regs_error ? -1 : get_syscall_result(tcp));
if (res == 1) { if (res == 1) {
syscall_fixup_on_sysexit(tcp); /* never fails */
get_error(tcp); /* never fails */ get_error(tcp); /* never fails */
if (need_fork_exec_workarounds)
syscall_fixup_for_fork_exec(tcp);
if (filtered(tcp) || hide_log_until_execve) if (filtered(tcp) || hide_log_until_execve)
goto ret; goto ret;
} }

View File

@ -41,7 +41,6 @@ stack_fcall_SOURCES = stack-fcall.c \
stack-fcall-0.c stack-fcall-1.c stack-fcall-2.c stack-fcall-3.c stack-fcall-0.c stack-fcall-1.c stack-fcall-2.c stack-fcall-3.c
TESTS = \ TESTS = \
ptrace_setoptions.test \
strace-f.test \ strace-f.test \
qual_syscall.test \ qual_syscall.test \
bexecve.test \ bexecve.test \

View File

@ -1,32 +0,0 @@
#!/bin/sh
# Ensure that strace tests kernel PTRACE_O_TRACECLONE
# and PTRACE_O_TRACESYSGOOD support properly.
. "${srcdir=.}/init.sh"
check_prog grep
[ "$(uname -s)" = Linux ] ||
skip_ 'The kernel is not a Linux kernel'
case "$(uname -r)" in
2.[6-9]*|2.[1-5][0-9]*|[3-9].*|[12][0-9]*) ;;
*) skip_ 'The kernel is not Linux 2.6.* or newer' ;;
esac
$STRACE -df -enone / > /dev/null 2> $LOG
grep -F -x 'ptrace_setoptions = 0xe' $LOG > /dev/null || {
cat $LOG
fail_ 'strace -f failed to recognize proper kernel PTRACE_O_TRACECLONE support'
}
grep -F -x 'ptrace_setoptions = 0x1f' $LOG > /dev/null || {
cat $LOG
fail_ 'strace -f failed to recognize proper kernel PTRACE_O_TRACESYSGOOD support'
}
$STRACE -d -enone / > /dev/null 2> $LOG
grep -F -x 'ptrace_setoptions = 0x11' $LOG > /dev/null || {
cat $LOG
fail_ 'strace failed to recognize proper kernel PTRACE_O_TRACESYSGOOD support'
}

424
util.c
View File

@ -1257,427 +1257,3 @@ upeek(int pid, long off, long *res)
*res = val; *res = val;
return 0; return 0;
} }
/* Note! On new kernels (about 2.5.46+), we use PTRACE_O_TRACECLONE
* and PTRACE_O_TRACE[V]FORK for tracing children.
* If you are adding a new arch which is only supported by newer kernels,
* you most likely don't need to add any code below
* beside a dummy "return 0" block in change_syscall().
*/
/*
* These #if's are huge, please indent them correctly.
* It's easy to get confused otherwise.
*/
#include "syscall.h"
#ifndef CLONE_PTRACE
# define CLONE_PTRACE 0x00002000
#endif
#ifndef CLONE_VFORK
# define CLONE_VFORK 0x00004000
#endif
#ifndef CLONE_VM
# define CLONE_VM 0x00000100
#endif
#ifdef IA64
typedef unsigned long *arg_setup_state;
static int
arg_setup(struct tcb *tcp, arg_setup_state *state)
{
unsigned long cfm, sof, sol;
long bsp;
if (ia64_ia32mode) {
/* Satisfy a false GCC warning. */
*state = NULL;
return 0;
}
if (upeek(tcp->pid, PT_AR_BSP, &bsp) < 0)
return -1;
if (upeek(tcp->pid, PT_CFM, (long *) &cfm) < 0)
return -1;
sof = (cfm >> 0) & 0x7f;
sol = (cfm >> 7) & 0x7f;
bsp = (long) ia64_rse_skip_regs((unsigned long *) bsp, -sof + sol);
*state = (unsigned long *) bsp;
return 0;
}
# define arg_finish_change(tcp, state) 0
static int
get_arg0(struct tcb *tcp, arg_setup_state *state, long *valp)
{
int ret;
if (ia64_ia32mode)
ret = upeek(tcp->pid, PT_R11, valp);
else
ret = umoven(tcp,
(unsigned long) ia64_rse_skip_regs(*state, 0),
sizeof(long), (void *) valp);
return ret;
}
static int
get_arg1(struct tcb *tcp, arg_setup_state *state, long *valp)
{
int ret;
if (ia64_ia32mode)
ret = upeek(tcp->pid, PT_R9, valp);
else
ret = umoven(tcp,
(unsigned long) ia64_rse_skip_regs(*state, 1),
sizeof(long), (void *) valp);
return ret;
}
static int
set_arg0(struct tcb *tcp, arg_setup_state *state, long val)
{
int req = PTRACE_POKEDATA;
void *ap;
if (ia64_ia32mode) {
ap = (void *) (intptr_t) PT_R11; /* r11 == EBX */
req = PTRACE_POKEUSER;
} else
ap = ia64_rse_skip_regs(*state, 0);
errno = 0;
ptrace(req, tcp->pid, ap, val);
return errno ? -1 : 0;
}
static int
set_arg1(struct tcb *tcp, arg_setup_state *state, long val)
{
int req = PTRACE_POKEDATA;
void *ap;
if (ia64_ia32mode) {
ap = (void *) (intptr_t) PT_R9; /* r9 == ECX */
req = PTRACE_POKEUSER;
} else
ap = ia64_rse_skip_regs(*state, 1);
errno = 0;
ptrace(req, tcp->pid, ap, val);
return errno ? -1 : 0;
}
/* ia64 does not return the input arguments from functions (and syscalls)
according to ia64 RSE (Register Stack Engine) behavior. */
# define restore_arg0(tcp, state, val) ((void) (state), 0)
# define restore_arg1(tcp, state, val) ((void) (state), 0)
#elif defined(SPARC) || defined(SPARC64)
# if defined(SPARC64)
# undef PTRACE_GETREGS
# define PTRACE_GETREGS PTRACE_GETREGS64
# undef PTRACE_SETREGS
# define PTRACE_SETREGS PTRACE_SETREGS64
# endif
typedef struct pt_regs arg_setup_state;
# define arg_setup(tcp, state) \
(ptrace(PTRACE_GETREGS, (tcp)->pid, (char *) (state), 0))
# define arg_finish_change(tcp, state) \
(ptrace(PTRACE_SETREGS, (tcp)->pid, (char *) (state), 0))
# define get_arg0(tcp, state, valp) (*(valp) = (state)->u_regs[U_REG_O0], 0)
# define get_arg1(tcp, state, valp) (*(valp) = (state)->u_regs[U_REG_O1], 0)
# define set_arg0(tcp, state, val) ((state)->u_regs[U_REG_O0] = (val), 0)
# define set_arg1(tcp, state, val) ((state)->u_regs[U_REG_O1] = (val), 0)
# define restore_arg0(tcp, state, val) 0
#else /* other architectures */
# if defined S390 || defined S390X
/* Note: this is only true for the `clone' system call, which handles
arguments specially. We could as well say that its first two arguments
are swapped relative to other architectures, but that would just be
another #ifdef in the calls. */
# define arg0_offset PT_GPR3
# define arg1_offset PT_ORIGGPR2
# define restore_arg0(tcp, state, val) ((void) (state), 0)
# define restore_arg1(tcp, state, val) ((void) (state), 0)
# define arg0_index 1
# define arg1_index 0
# elif defined(ALPHA) || defined(MIPS)
# define arg0_offset REG_A0
# define arg1_offset (REG_A0+1)
# elif defined(POWERPC)
# define arg0_offset (sizeof(unsigned long)*PT_R3)
# define arg1_offset (sizeof(unsigned long)*PT_R4)
# define restore_arg0(tcp, state, val) ((void) (state), 0)
# elif defined(HPPA)
# define arg0_offset PT_GR26
# define arg1_offset (PT_GR26-4)
# elif defined(X86_64) || defined(X32)
# define arg0_offset ((long)(8*(current_personality ? RBX : RDI)))
# define arg1_offset ((long)(8*(current_personality ? RCX : RSI)))
# elif defined(SH)
# define arg0_offset (4*(REG_REG0+4))
# define arg1_offset (4*(REG_REG0+5))
# elif defined(SH64)
/* ABI defines arg0 & 1 in r2 & r3 */
# define arg0_offset (REG_OFFSET+16)
# define arg1_offset (REG_OFFSET+24)
# define restore_arg0(tcp, state, val) 0
# elif defined CRISV10 || defined CRISV32
# define arg0_offset (4*PT_R11)
# define arg1_offset (4*PT_ORIG_R10)
# define restore_arg0(tcp, state, val) 0
# define restore_arg1(tcp, state, val) 0
# define arg0_index 1
# define arg1_index 0
# else
# define arg0_offset 0
# define arg1_offset 4
# if defined ARM
# define restore_arg0(tcp, state, val) 0
# endif
# endif
typedef int arg_setup_state;
# define arg_setup(tcp, state) (0)
# define arg_finish_change(tcp, state) 0
# define get_arg0(tcp, cookie, valp) (upeek((tcp)->pid, arg0_offset, (valp)))
# define get_arg1(tcp, cookie, valp) (upeek((tcp)->pid, arg1_offset, (valp)))
static int
set_arg0(struct tcb *tcp, void *cookie, long val)
{
return ptrace(PTRACE_POKEUSER, tcp->pid, (char*)arg0_offset, val);
}
static int
set_arg1(struct tcb *tcp, void *cookie, long val)
{
return ptrace(PTRACE_POKEUSER, tcp->pid, (char*)arg1_offset, val);
}
#endif /* architectures */
#ifndef restore_arg0
# define restore_arg0(tcp, state, val) set_arg0((tcp), (state), (val))
#endif
#ifndef restore_arg1
# define restore_arg1(tcp, state, val) set_arg1((tcp), (state), (val))
#endif
#ifndef arg0_index
# define arg0_index 0
# define arg1_index 1
#endif
static int
change_syscall(struct tcb *tcp, arg_setup_state *state, int new)
{
#if defined(I386)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(ORIG_EAX * 4), new) < 0)
return -1;
return 0;
#elif defined(X86_64)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(ORIG_RAX * 8), new) < 0)
return -1;
return 0;
#elif defined(X32)
/* setbpt/clearbpt never used: */
/* X32 is only supported since about linux-3.0.30 */
#elif defined(POWERPC)
if (ptrace(PTRACE_POKEUSER, tcp->pid,
(char*)(sizeof(unsigned long)*PT_R0), new) < 0)
return -1;
return 0;
#elif defined(S390) || defined(S390X)
/* s390 linux after 2.4.7 has a hook in entry.S to allow this */
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(PT_GPR2), new) < 0)
return -1;
return 0;
#elif defined(M68K)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(4*PT_ORIG_D0), new) < 0)
return -1;
return 0;
#elif defined(SPARC) || defined(SPARC64)
state->u_regs[U_REG_G1] = new;
return 0;
#elif defined(MIPS)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(REG_V0), new) < 0)
return -1;
return 0;
#elif defined(ALPHA)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(REG_A3), new) < 0)
return -1;
return 0;
#elif defined(AVR32)
/* setbpt/clearbpt never used: */
/* AVR32 is only supported since about linux-2.6.19 */
#elif defined(BFIN)
/* setbpt/clearbpt never used: */
/* Blackfin is only supported since about linux-2.6.23 */
#elif defined(IA64)
if (ia64_ia32mode) {
switch (new) {
case 2:
break; /* x86 SYS_fork */
case SYS_clone:
new = 120;
break;
default:
fprintf(stderr, "%s: unexpected syscall %d\n",
__FUNCTION__, new);
return -1;
}
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(PT_R1), new) < 0)
return -1;
} else if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(PT_R15), new) < 0)
return -1;
return 0;
#elif defined(HPPA)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(PT_GR20), new) < 0)
return -1;
return 0;
#elif defined(SH)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(4*(REG_REG0+3)), new) < 0)
return -1;
return 0;
#elif defined(SH64)
/* Top half of reg encodes the no. of args n as 0x1n.
Assume 0 args as kernel never actually checks... */
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(REG_SYSCALL),
0x100000 | new) < 0)
return -1;
return 0;
#elif defined(CRISV10) || defined(CRISV32)
if (ptrace(PTRACE_POKEUSER, tcp->pid, (char*)(4*PT_R9), new) < 0)
return -1;
return 0;
#elif defined(ARM)
/* Some kernels support this, some (pre-2.6.16 or so) don't. */
# ifndef PTRACE_SET_SYSCALL
# define PTRACE_SET_SYSCALL 23
# endif
if (ptrace(PTRACE_SET_SYSCALL, tcp->pid, 0, new & 0xffff) != 0)
return -1;
return 0;
#elif defined(AARCH64)
/* setbpt/clearbpt never used: */
/* AARCH64 is only supported since about linux-3.0.31 */
#elif defined(TILE)
/* setbpt/clearbpt never used: */
/* Tilera CPUs are only supported since about linux-2.6.34 */
#elif defined(MICROBLAZE)
/* setbpt/clearbpt never used: */
/* microblaze is only supported since about linux-2.6.30 */
#elif defined(OR1K)
/* never reached; OR1K is only supported by kernels since 3.1.0. */
#elif defined(METAG)
/* setbpt/clearbpt never used: */
/* Meta is only supported since linux-3.7 */
#elif defined(XTENSA)
/* setbpt/clearbpt never used: */
/* Xtensa is only supported since linux 2.6.13 */
#elif defined(ARC)
/* setbpt/clearbpt never used: */
/* ARC only supported since 3.9 */
#else
#warning Do not know how to handle change_syscall for this architecture
#endif /* architecture */
return -1;
}
int
setbpt(struct tcb *tcp)
{
static int clone_scno[SUPPORTED_PERSONALITIES] = { SYS_clone };
arg_setup_state state;
if (tcp->flags & TCB_BPTSET) {
fprintf(stderr, "PANIC: TCB already set in pid %u\n", tcp->pid);
return -1;
}
/*
* It's a silly kludge to initialize this with a search at runtime.
* But it's better than maintaining another magic thing in the
* godforsaken tables.
*/
if (clone_scno[current_personality] == 0) {
unsigned int i;
for (i = 0; i < nsyscalls; ++i)
if (sysent[i].sys_func == sys_clone) {
clone_scno[current_personality] = i;
break;
}
}
if (tcp->s_ent->sys_func == sys_fork) {
if (arg_setup(tcp, &state) < 0
|| get_arg0(tcp, &state, &tcp->inst[0]) < 0
|| get_arg1(tcp, &state, &tcp->inst[1]) < 0
|| change_syscall(tcp, &state,
clone_scno[current_personality]) < 0
|| set_arg0(tcp, &state, CLONE_PTRACE|SIGCHLD) < 0
|| set_arg1(tcp, &state, 0) < 0
|| arg_finish_change(tcp, &state) < 0)
return -1;
tcp->u_arg[arg0_index] = CLONE_PTRACE|SIGCHLD;
tcp->u_arg[arg1_index] = 0;
tcp->flags |= TCB_BPTSET;
return 0;
}
if (tcp->s_ent->sys_func == sys_clone) {
/* ia64 calls directly `clone (CLONE_VFORK | CLONE_VM)'
contrary to x86 vfork above. Even on x86 we turn the
vfork semantics into plain fork - each application must not
depend on the vfork specifics according to POSIX. We would
hang waiting for the parent resume otherwise. We need to
clear also CLONE_VM but only in the CLONE_VFORK case as
otherwise we would break pthread_create. */
long new_arg0 = (tcp->u_arg[arg0_index] | CLONE_PTRACE);
if (new_arg0 & CLONE_VFORK)
new_arg0 &= ~(unsigned long)(CLONE_VFORK | CLONE_VM);
if (arg_setup(tcp, &state) < 0
|| set_arg0(tcp, &state, new_arg0) < 0
|| arg_finish_change(tcp, &state) < 0)
return -1;
tcp->inst[0] = tcp->u_arg[arg0_index];
tcp->inst[1] = tcp->u_arg[arg1_index];
tcp->flags |= TCB_BPTSET;
return 0;
}
fprintf(stderr, "PANIC: setbpt for syscall %ld on %u???\n",
tcp->scno, tcp->pid);
return -1;
}
int
clearbpt(struct tcb *tcp)
{
arg_setup_state state;
if (arg_setup(tcp, &state) < 0
|| change_syscall(tcp, &state, tcp->scno) < 0
|| restore_arg0(tcp, &state, tcp->inst[0]) < 0
|| restore_arg1(tcp, &state, tcp->inst[1]) < 0
|| arg_finish_change(tcp, &state))
if (errno != ESRCH)
return -1;
tcp->flags &= ~TCB_BPTSET;
return 0;
}