#include <stdarg.h>
#include <sys/param.h>
#include <fcntl.h>
+#include <signal.h>
#include <sys/resource.h>
#include <sys/wait.h>
#include <sys/stat.h>
#ifdef HAVE_PRCTL
# include <sys/prctl.h>
#endif
-#if defined(IA64)
-# include <asm/ptrace_offsets.h>
-#endif
+
+#include "ptrace.h"
+#include "printsiginfo.h"
+
/* In some libc, these aren't declared. Do it ourself: */
extern char **environ;
extern int optind;
# define fork() vfork()
#endif
+const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
+
cflag_t cflag = CFLAG_NONE;
unsigned int followfork = 0;
-unsigned int ptrace_setoptions = 0;
+unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC;
unsigned int xflag = 0;
-bool need_fork_exec_workarounds = 0;
bool debug_flag = 0;
bool Tflag = 0;
bool iflag = 0;
bool count_wallclock = 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 bool rflag = 0;
static bool print_pid_pfx = 0;
/* Are we "strace PROG" and need to hide everything until execve? */
bool hide_log_until_execve = 0;
-static int exit_code = 0;
+static int exit_code;
static int strace_child = 0;
static int strace_tracer_pid = 0;
#endif /* HAVE_STERRROR */
static void
-usage(FILE *ofp, int exitval)
+usage(void)
{
- fprintf(ofp, "\
-usage: strace [-CdffhiqrtttTvVxxy] [-I n] [-e expr]...\n\
+ printf("\
+usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
[-a column] [-o file] [-s strsize] [-P path]...\n\
-p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
- or: strace -c[df] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
+ or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
-p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
--c -- count time, calls, and errors for each syscall and report summary\n\
--C -- like -c but also print regular output\n\
--w -- summarise syscall latency (default is system time)\n\
--d -- enable debug output to stderr\n\
--D -- run tracer process as a detached grandchild, not as parent\n\
--f -- follow forks, -ff -- with output into separate files\n\
--i -- print instruction pointer at time of syscall\n\
--q -- suppress messages about attaching, detaching, etc.\n\
--r -- print relative timestamp, -t -- absolute timestamp, -tt -- with usecs\n\
--T -- print time spent in each syscall\n\
--v -- verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
--x -- print non-ascii strings in hex, -xx -- print all strings in hex\n\
--y -- print paths associated with file descriptor arguments\n\
--yy -- print ip:port pairs associated with socket file descriptors\n\
--h -- print help message, -V -- print version\n\
--a column -- alignment COLUMN for printing syscall results (default %d)\n\
--b execve -- detach on this syscall\n\
--e expr -- a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
- options: trace, abbrev, verbose, raw, signal, read, write\n\
--I interruptible --\n\
- 1: no signals are blocked\n\
- 2: fatal signals are blocked while decoding syscall (default)\n\
- 3: fatal signals are always blocked (default if '-o FILE PROG')\n\
- 4: fatal signals and SIGTSTP (^Z) are always blocked\n\
- (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
--o file -- send trace output to FILE instead of stderr\n\
--O overhead -- set overhead for tracing syscalls to OVERHEAD usecs\n\
--p pid -- trace process with process id PID, may be repeated\n\
--s strsize -- limit length of print strings to STRSIZE chars (default %d)\n\
--S sortby -- sort syscall counts by: time, calls, name, nothing (default %s)\n\
--u username -- run command as username handling setuid and/or setgid\n\
--E var=val -- put var=val in the environment for command\n\
--E var -- remove var from the environment for command\n\
--P path -- trace accesses to path\n\
+\n\
+Output format:\n\
+ -a column alignment COLUMN for printing syscall results (default %d)\n\
+ -i print instruction pointer at time of syscall\n\
+ -o file send trace output to FILE instead of stderr\n\
+ -q suppress messages about attaching, detaching, etc.\n\
+ -r print relative timestamp\n\
+ -s strsize limit length of print strings to STRSIZE chars (default %d)\n\
+ -t print absolute timestamp\n\
+ -tt print absolute timestamp with usecs\n\
+ -T print time spent in each syscall\n\
+ -x print non-ascii strings in hex\n\
+ -xx print all strings in hex\n\
+ -y print paths associated with file descriptor arguments\n\
+ -yy print protocol specific information associated with socket file descriptors\n\
+\n\
+Statistics:\n\
+ -c count time, calls, and errors for each syscall and report summary\n\
+ -C like -c but also print regular output\n\
+ -O overhead set overhead for tracing syscalls to OVERHEAD usecs\n\
+ -S sortby sort syscall counts by: time, calls, name, nothing (default %s)\n\
+ -w summarise syscall latency (default is system time)\n\
+\n\
+Filtering:\n\
+ -e expr a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
+ options: trace, abbrev, verbose, raw, signal, read, write\n\
+ -P path trace accesses to path\n\
+\n\
+Tracing:\n\
+ -b execve detach on execve syscall\n\
+ -D run tracer process as a detached grandchild, not as parent\n\
+ -f follow forks\n\
+ -ff follow forks with output into separate files\n\
+ -I interruptible\n\
+ 1: no signals are blocked\n\
+ 2: fatal signals are blocked while decoding syscall (default)\n\
+ 3: fatal signals are always blocked (default if '-o FILE PROG')\n\
+ 4: fatal signals and SIGTSTP (^Z) are always blocked\n\
+ (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
+\n\
+Startup:\n\
+ -E var remove var from the environment for command\n\
+ -E var=val put var=val in the environment for command\n\
+ -p pid trace process with process id PID, may be repeated\n\
+ -u username run command as username handling setuid and/or setgid\n\
+\n\
+Miscellaneous:\n\
+ -d enable debug output to stderr\n\
+ -v verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
+ -h print help message\n\
+ -V print version\n\
"
#ifdef USE_LIBUNWIND
-"-k obtain stack trace between each syscall (experimental)\n\
+" -k obtain stack trace between each syscall (experimental)\n\
"
#endif
/* ancient, no one should use it
-z -- print only succeeding syscalls\n\
*/
, DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
- exit(exitval);
+ exit(0);
}
-static void die(void) __attribute__ ((noreturn));
-static void die(void)
+static void ATTRIBUTE_NORETURN
+die(void)
{
if (strace_tracer_pid == getpid()) {
cflag = 0;
die();
}
+void error_msg_and_help(const char *fmt, ...)
+{
+ if (fmt != NULL) {
+ va_list p;
+ va_start(p, fmt);
+ verror_msg(0, fmt, p);
+ }
+ fprintf(stderr, "Try '%s -h' for more information.\n", progname);
+ die();
+}
+
void perror_msg(const char *fmt, ...)
{
va_list p;
die();
}
-void die_out_of_memory(void)
-{
- static bool recursed = 0;
- if (recursed)
- exit(1);
- recursed = 1;
- error_msg_and_die("Out of memory");
-}
-
static void
error_opt_arg(int opt, const char *arg)
{
- error_msg_and_die("Invalid -%c argument: '%s'", opt, arg);
+ error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
}
-#if USE_SEIZE
+static const char *ptrace_attach_cmd;
+
static int
ptrace_attach_or_seize(int pid)
{
+#if USE_SEIZE
int r;
if (!use_seize)
- return ptrace(PTRACE_ATTACH, pid, 0L, 0L);
- r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long)ptrace_setoptions);
+ return ptrace_attach_cmd = "PTRACE_ATTACH",
+ ptrace(PTRACE_ATTACH, pid, 0L, 0L);
+ r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
if (r)
- return r;
+ return ptrace_attach_cmd = "PTRACE_SEIZE", r;
r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
- return r;
-}
+ return ptrace_attach_cmd = "PTRACE_INTERRUPT", r;
#else
-# define ptrace_attach_or_seize(pid) ptrace(PTRACE_ATTACH, (pid), 0, 0)
+ return ptrace_attach_cmd = "PTRACE_ATTACH",
+ ptrace(PTRACE_ATTACH, pid, 0L, 0L);
#endif
+}
/*
* Used when we want to unblock stopped traced process.
msg = "CONT";
if (op == PTRACE_DETACH)
msg = "DETACH";
-#ifdef PTRACE_LISTEN
if (op == PTRACE_LISTEN)
msg = "LISTEN";
-#endif
/*
* Why curcol != 0? Otherwise sometimes we get this:
*
fcntl(fd, F_SETFD, newflags); /* never fails */
}
-static void kill_save_errno(pid_t pid, int sig)
+static void
+kill_save_errno(pid_t pid, int sig)
{
int saved_errno = errno;
static void
expand_tcbtab(void)
{
- /* Allocate some more TCBs and expand the table.
+ /* Allocate some (more) TCBs (and expand the table).
We don't want to relocate the TCBs because our
callers have pointers and it would be a pain.
So tcbtab is a table of pointers. Since we never
free the TCBs, we allocate a single chunk of many. */
- unsigned int i = tcbtabsize;
- struct tcb *newtcbs = calloc(tcbtabsize, sizeof(newtcbs[0]));
- struct tcb **newtab = realloc(tcbtab, tcbtabsize * 2 * sizeof(tcbtab[0]));
- if (!newtab || !newtcbs)
- die_out_of_memory();
- tcbtabsize *= 2;
- tcbtab = newtab;
- while (i < tcbtabsize)
- tcbtab[i++] = newtcbs++;
+ unsigned int new_tcbtabsize, alloc_tcbtabsize;
+ struct tcb *newtcbs;
+
+ if (tcbtabsize) {
+ alloc_tcbtabsize = tcbtabsize;
+ new_tcbtabsize = tcbtabsize * 2;
+ } else {
+ new_tcbtabsize = alloc_tcbtabsize = 1;
+ }
+
+ newtcbs = xcalloc(alloc_tcbtabsize, sizeof(newtcbs[0]));
+ tcbtab = xreallocarray(tcbtab, new_tcbtabsize, sizeof(tcbtab[0]));
+ while (tcbtabsize < new_tcbtabsize)
+ tcbtab[tcbtabsize++] = newtcbs++;
}
static struct tcb *
nprocs++;
if (debug_flag)
- fprintf(stderr, "new tcb for pid %d, active tcbs:%d\n", tcp->pid, nprocs);
+ error_msg("new tcb for pid %d, active tcbs:%d",
+ tcp->pid, nprocs);
return tcp;
}
}
error_msg_and_die("bug in alloctcb");
}
+void *
+get_tcb_priv_data(const struct tcb *tcp)
+{
+ return tcp->_priv_data;
+}
+
+int
+set_tcb_priv_data(struct tcb *tcp, void *const priv_data,
+ void (*const free_priv_data)(void *))
+{
+ if (tcp->_priv_data)
+ return -1;
+
+ tcp->_free_priv_data = free_priv_data;
+ tcp->_priv_data = priv_data;
+
+ return 0;
+}
+
+void
+free_tcb_priv_data(struct tcb *tcp)
+{
+ if (tcp->_priv_data) {
+ if (tcp->_free_priv_data) {
+ tcp->_free_priv_data(tcp->_priv_data);
+ tcp->_free_priv_data = NULL;
+ }
+ tcp->_priv_data = NULL;
+ }
+}
+
static void
droptcb(struct tcb *tcp)
{
if (tcp->pid == 0)
return;
+ free_tcb_priv_data(tcp);
+
#ifdef USE_LIBUNWIND
if (stack_trace_enabled) {
unwind_tcb_fin(tcp);
nprocs--;
if (debug_flag)
- fprintf(stderr, "dropped tcb for pid %d, %d remain\n", tcp->pid, nprocs);
+ error_msg("dropped tcb for pid %d, %d remain",
+ tcp->pid, nprocs);
if (tcp->outf) {
if (followfork >= 2) {
int error;
int status;
- if (tcp->flags & TCB_BPTSET)
- clearbpt(tcp);
-
/*
* Linux wrongly insists the child be stopped
* before detaching. Arghh. We go through hoops
* to make a clean break of things.
*/
-#if defined(SPARC)
-# undef PTRACE_DETACH
-# define PTRACE_DETACH PTRACE_SUNDETACH
-#endif
if (!(tcp->flags & TCB_ATTACHED))
goto drop;
}
sig = WSTOPSIG(status);
if (debug_flag)
- fprintf(stderr, "detach wait: event:%d sig:%d\n",
- (unsigned)status >> 16, sig);
+ error_msg("detach wait: event:%d sig:%d",
+ (unsigned)status >> 16, sig);
if (use_seize) {
unsigned event = (unsigned)status >> 16;
if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) {
drop:
if (!qflag && (tcp->flags & TCB_ATTACHED))
- fprintf(stderr, "Process %u detached\n", tcp->pid);
+ error_msg("Process %u detached", tcp->pid);
droptcb(tcp);
}
}
}
+static void
+attach_tcb(struct tcb *const tcp)
+{
+ if (ptrace_attach_or_seize(tcp->pid) < 0) {
+ perror_msg("attach: ptrace(%s, %d)",
+ ptrace_attach_cmd, tcp->pid);
+ droptcb(tcp);
+ return;
+ }
+
+ tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
+ newoutf(tcp);
+ if (debug_flag)
+ error_msg("attach to pid %d (main) succeeded", tcp->pid);
+
+ char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
+ DIR *dir;
+ unsigned int ntid = 0, nerr = 0;
+
+ if (followfork && tcp->pid != strace_child &&
+ sprintf(procdir, "/proc/%d/task", tcp->pid) > 0 &&
+ (dir = opendir(procdir)) != NULL) {
+ struct_dirent *de;
+
+ while ((de = read_dir(dir)) != NULL) {
+ if (de->d_fileno == 0)
+ continue;
+
+ int tid = string_to_uint(de->d_name);
+ if (tid <= 0 || tid == tcp->pid)
+ continue;
+
+ ++ntid;
+ if (ptrace_attach_or_seize(tid) < 0) {
+ ++nerr;
+ if (debug_flag)
+ perror_msg("attach: ptrace(%s, %d)",
+ ptrace_attach_cmd, tid);
+ continue;
+ }
+ if (debug_flag)
+ error_msg("attach to pid %d succeeded", tid);
+
+ struct tcb *tid_tcp = alloctcb(tid);
+ tid_tcp->flags |= TCB_ATTACHED | TCB_STARTUP |
+ post_attach_sigstop;
+ newoutf(tid_tcp);
+ }
+
+ closedir(dir);
+ }
+
+ if (!qflag) {
+ if (ntid > nerr)
+ error_msg("Process %u attached"
+ " with %u threads",
+ tcp->pid, ntid - nerr + 1);
+ else
+ error_msg("Process %u attached",
+ tcp->pid);
+ }
+}
+
static void
startup_attach(void)
{
+ pid_t parent_pid = strace_tracer_pid;
unsigned int tcbi;
struct tcb *tcp;
if (tcp->flags & TCB_ATTACHED)
continue; /* no, we already attached it */
- if (followfork && !daemonized_tracer) {
- char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
- DIR *dir;
-
- sprintf(procdir, "/proc/%d/task", tcp->pid);
- dir = opendir(procdir);
- if (dir != NULL) {
- unsigned int ntid = 0, nerr = 0;
- struct_dirent *de;
-
- while ((de = read_dir(dir)) != NULL) {
- struct tcb *cur_tcp;
- int tid;
-
- if (de->d_fileno == 0)
- continue;
- /* we trust /proc filesystem */
- tid = atoi(de->d_name);
- if (tid <= 0)
- continue;
- ++ntid;
- if (ptrace_attach_or_seize(tid) < 0) {
- ++nerr;
- if (debug_flag)
- fprintf(stderr, "attach to pid %d failed\n", tid);
- continue;
- }
- if (debug_flag)
- fprintf(stderr, "attach to pid %d succeeded\n", tid);
- cur_tcp = tcp;
- if (tid != tcp->pid)
- cur_tcp = alloctcb(tid);
- cur_tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
- newoutf(cur_tcp);
- }
- closedir(dir);
- if (interactive) {
- sigprocmask(SIG_SETMASK, &empty_set, NULL);
- if (interrupted)
- goto ret;
- sigprocmask(SIG_BLOCK, &blocked_set, NULL);
- }
- ntid -= nerr;
- if (ntid == 0) {
- perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
- droptcb(tcp);
- continue;
- }
- if (!qflag) {
- fprintf(stderr, ntid > 1
-? "Process %u attached with %u threads\n"
-: "Process %u attached\n",
- tcp->pid, ntid);
- }
- if (!(tcp->flags & TCB_ATTACHED)) {
- /* -p PID, we failed to attach to PID itself
- * but did attach to some of its sibling threads.
- * Drop PID's tcp.
- */
- droptcb(tcp);
- }
- continue;
- } /* if (opendir worked) */
- } /* if (-f) */
- if (ptrace_attach_or_seize(tcp->pid) < 0) {
- perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
+ if (tcp->pid == parent_pid || tcp->pid == strace_tracer_pid) {
+ errno = EPERM;
+ perror_msg("attach: pid %d", tcp->pid);
droptcb(tcp);
continue;
}
- tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
- newoutf(tcp);
- if (debug_flag)
- fprintf(stderr, "attach to pid %d (main) succeeded\n", tcp->pid);
- if (daemonized_tracer) {
- /*
- * Make parent go away.
- * Also makes grandparent's wait() unblock.
- */
- kill(getppid(), SIGKILL);
- }
+ attach_tcb(tcp);
- if (!qflag)
- fprintf(stderr,
- "Process %u attached\n",
- tcp->pid);
+ if (interactive) {
+ sigprocmask(SIG_SETMASK, &empty_set, NULL);
+ if (interrupted)
+ goto ret;
+ sigprocmask(SIG_BLOCK, &blocked_set, NULL);
+ }
} /* for each tcbtab[] */
+ if (daemonized_tracer) {
+ /*
+ * Make parent go away.
+ * Also makes grandparent's wait() unblock.
+ */
+ kill(parent_pid, SIGKILL);
+ strace_child = 0;
+ }
+
ret:
if (interactive)
sigprocmask(SIG_SETMASK, &empty_set, NULL);
char *pathname;
};
static struct exec_params params_for_tracee;
-static void __attribute__ ((noinline, noreturn))
+
+static void ATTRIBUTE_NOINLINE ATTRIBUTE_NORETURN
exec_or_die(void)
{
struct exec_params *params = ¶ms_for_tracee;
perror_msg_and_die("exec");
}
+/*
+ * Open a dummy descriptor for use as a placeholder.
+ * The descriptor is O_RDONLY with FD_CLOEXEC flag set.
+ * A read attempt from such descriptor ends with EOF,
+ * a write attempt is rejected with EBADF.
+ */
+static int
+open_dummy_desc(void)
+{
+ int fds[2];
+
+ if (pipe(fds))
+ perror_msg_and_die("pipe");
+ close(fds[1]);
+ set_cloexec_flag(fds[0]);
+ return fds[0];
+}
+
+/* placeholder fds status for stdin and stdout */
+static bool fd_is_placeholder[2];
+
+/*
+ * Ensure that all standard file descriptors are open by opening placeholder
+ * file descriptors for those standard file descriptors that are not open.
+ *
+ * The information which descriptors have been made open is saved
+ * in fd_is_placeholder for later use.
+ */
+static void
+ensure_standard_fds_opened(void)
+{
+ int fd;
+
+ while ((fd = open_dummy_desc()) <= 2) {
+ if (fd == 2)
+ break;
+ fd_is_placeholder[fd] = true;
+ }
+
+ if (fd > 2)
+ close(fd);
+}
+
+/*
+ * Redirect stdin and stdout unless they have been opened earlier
+ * by ensure_standard_fds_opened as placeholders.
+ */
+static void
+redirect_standard_fds(void)
+{
+ int i;
+
+ /*
+ * It might be a good idea to redirect stderr as well,
+ * but we sometimes need to print error messages.
+ */
+ for (i = 0; i <= 1; ++i) {
+ if (!fd_is_placeholder[i]) {
+ close(i);
+ open_dummy_desc();
+ }
+ }
+}
+
static void
startup_child(char **argv)
{
struct_stat statbuf;
const char *filename;
+ size_t filename_len;
char pathname[PATH_MAX];
int pid;
struct tcb *tcp;
filename = argv[0];
+ filename_len = strlen(filename);
+
+ if (filename_len > sizeof(pathname) - 1) {
+ errno = ENAMETOOLONG;
+ perror_msg_and_die("exec");
+ }
if (strchr(filename, '/')) {
- if (strlen(filename) > sizeof pathname - 1) {
- errno = ENAMETOOLONG;
- perror_msg_and_die("exec");
- }
strcpy(pathname, filename);
}
#ifdef USE_DEBUGGING_EXEC
}
if (len && pathname[len - 1] != '/')
pathname[len++] = '/';
+ if (filename_len + len > sizeof(pathname) - 1)
+ continue;
strcpy(pathname + len, filename);
if (stat_file(pathname, &statbuf) == 0 &&
/* Accept only regular files
(statbuf.st_mode & 0111))
break;
}
+ if (!path || !*path)
+ pathname[0] = '\0';
}
if (stat_file(pathname, &statbuf) < 0) {
perror_msg_and_die("Can't stat '%s'", filename);
* On NOMMU, can be safely freed only after execve in tracee.
* It's hard to know when that happens, so we just leak it.
*/
- params_for_tracee.pathname = NOMMU_SYSTEM ? strdup(pathname) : pathname;
+ params_for_tracee.pathname = NOMMU_SYSTEM ? xstrdup(pathname) : pathname;
#if defined HAVE_PRCTL && defined PR_SET_PTRACER && defined PR_SET_PTRACER_ANY
if (daemonized_tracer)
if (ptrace_attach_or_seize(pid)) {
kill_save_errno(pid, SIGKILL);
- perror_msg_and_die("Can't attach to %d", pid);
+ perror_msg_and_die("attach: ptrace(%s, %d)",
+ ptrace_attach_cmd, pid);
}
if (!NOMMU_SYSTEM)
kill(pid, SIGCONT);
newoutf(tcp);
}
else {
- /* With -D, we are *child* here, IOW: different pid. Fetch it: */
+ /* With -D, we are *child* here, the tracee is our parent. */
+ strace_child = strace_tracer_pid;
strace_tracer_pid = getpid();
- /* The tracee is our parent: */
- pid = getppid();
- alloctcb(pid);
+ alloctcb(strace_child);
/* attaching will be done later, by startup_attach */
/* note: we don't do newoutf(tcp) here either! */
* to create a genuine separate stack and execute on it.
*/
}
-}
-
-/*
- * 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;
+ /*
+ * A case where straced process is part of a pipe:
+ * { sleep 1; yes | head -n99999; } | strace -o/dev/null sh -c 'exec <&-; sleep 9'
+ * If strace won't close its fd#0, closing it in tracee is not enough:
+ * the pipe is still open, it has a reader. Thus, "head" will not get its
+ * SIGPIPE at once, on the first write.
+ *
+ * Preventing it by redirecting strace's stdin/out.
+ * (Don't leave fds 0 and 1 closed, this is bad practice: future opens
+ * will reuse them, unexpectedly making a newly opened object "stdin").
+ */
+ redirect_standard_fds();
}
#if USE_SEIZE
if (ptrace(PTRACE_SEIZE, pid, 0, 0) == 0) {
post_attach_sigstop = 0; /* this sets use_seize to 1 */
} else if (debug_flag) {
- fprintf(stderr, "PTRACE_SEIZE doesn't work\n");
+ error_msg("PTRACE_SEIZE doesn't work");
}
kill(pid, SIGKILL);
* Don't want main() to inline us and defeat the reason
* we have a separate function.
*/
-static void __attribute__ ((noinline))
+static void ATTRIBUTE_NOINLINE
init(int argc, char *argv[])
{
- struct tcb *tcp;
int c, i;
int optF = 0;
- unsigned int tcbi;
struct sigaction sa;
progname = argv[0] ? argv[0] : "strace";
os_release = get_os_release();
- /* Allocate the initial tcbtab. */
- tcbtabsize = argc; /* Surely enough for all -p args. */
- tcbtab = calloc(tcbtabsize, sizeof(tcbtab[0]));
- if (!tcbtab)
- die_out_of_memory();
- tcp = calloc(tcbtabsize, sizeof(*tcp));
- if (!tcp)
- die_out_of_memory();
- for (tcbi = 0; tcbi < tcbtabsize; tcbi++)
- tcbtab[tcbi] = tcp++;
-
shared_log = stderr;
set_sortby(DEFAULT_SORTBY);
set_personality(DEFAULT_PERSONALITY);
break;
case 'c':
if (cflag == CFLAG_BOTH) {
- error_msg_and_die("-c and -C are mutually exclusive");
+ error_msg_and_help("-c and -C are mutually exclusive");
}
cflag = CFLAG_ONLY_STATS;
break;
case 'C':
if (cflag == CFLAG_ONLY_STATS) {
- error_msg_and_die("-c and -C are mutually exclusive");
+ error_msg_and_help("-c and -C are mutually exclusive");
}
cflag = CFLAG_BOTH;
break;
followfork++;
break;
case 'h':
- usage(stdout, 0);
+ usage();
break;
case 'i':
iflag = 1;
qualify(optarg);
break;
case 'o':
- outfname = strdup(optarg);
+ outfname = xstrdup(optarg);
break;
case 'O':
i = string_to_uint(optarg);
set_sortby(optarg);
break;
case 'u':
- username = strdup(optarg);
+ username = xstrdup(optarg);
break;
#ifdef USE_LIBUNWIND
case 'k':
error_opt_arg(c, optarg);
break;
default:
- usage(stderr, 1);
+ error_msg_and_help(NULL);
break;
}
}
argv += optind;
/* argc -= optind; - no need, argc is not used below */
- acolumn_spaces = malloc(acolumn + 1);
- if (!acolumn_spaces)
- die_out_of_memory();
+ acolumn_spaces = xmalloc(acolumn + 1);
memset(acolumn_spaces, ' ', acolumn);
acolumn_spaces[acolumn] = '\0';
- /* Must have PROG [ARGS], or -p PID. Not both. */
- if (!argv[0] == !nprocs)
- usage(stderr, 1);
+ if (!argv[0] && !nprocs) {
+ error_msg_and_help("must have PROG [ARGS] or -p PID");
+ }
- if (nprocs != 0 && daemonized_tracer) {
- error_msg_and_die("-D and -p are mutually exclusive");
+ if (!argv[0] && daemonized_tracer) {
+ error_msg_and_help("PROG [ARGS] must be specified with -D");
}
if (!followfork)
followfork = optF;
if (followfork >= 2 && cflag) {
- error_msg_and_die("(-c or -C) and -ff are mutually exclusive");
+ error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
}
if (count_wallclock && !cflag) {
- error_msg_and_die("-w must be given with (-c or -C)");
+ error_msg_and_help("-w must be given with (-c or -C)");
}
if (cflag == CFLAG_ONLY_STATS) {
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)
- need_fork_exec_workarounds = test_ptrace_setoptions_followfork();
- need_fork_exec_workarounds |= test_ptrace_setoptions_for_all();
+ ptrace_setoptions |= PTRACE_O_TRACECLONE |
+ PTRACE_O_TRACEFORK |
+ PTRACE_O_TRACEVFORK;
+ if (debug_flag)
+ error_msg("ptrace_setoptions = %#x", ptrace_setoptions);
test_ptrace_seize();
+ /*
+ * Is something weird with our stdin and/or stdout -
+ * for example, may they be not open? In this case,
+ * ensure that none of the future opens uses them.
+ *
+ * This was seen in the wild when /proc/sys/kernel/core_pattern
+ * was set to "|/bin/strace -o/tmp/LOG PROG":
+ * kernel runs coredump helper with fd#0 open but fd#1 closed (!),
+ * therefore LOG gets opened to fd#1, and fd#1 is closed by
+ * "don't hold up stdin/out open" code soon after.
+ */
+ ensure_standard_fds_opened();
+
/* Check if they want to redirect the output. */
if (outfname) {
/* See if they want to pipe the output. */
* when using popen, so prohibit it.
*/
if (followfork >= 2)
- error_msg_and_die("Piping the output and -ff are mutually exclusive");
+ error_msg_and_help("piping the output and -ff are mutually exclusive");
shared_log = strace_popen(outfname + 1);
}
else if (followfork < 2)
}
if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
- char *buf = malloc(BUFSIZ);
- if (!buf)
- die_out_of_memory();
+ char *buf = xmalloc(BUFSIZ);
setvbuf(shared_log, buf, _IOLBF, BUFSIZ);
}
if (outfname && argv[0]) {
if (!opt_intr)
opt_intr = INTR_NEVER;
- qflag = 1;
+ if (!qflag)
+ qflag = 1;
}
if (!opt_intr)
opt_intr = INTR_WHILE_WAIT;
/* argv[0] -pPID -oFILE Default interactive setting
- * yes 0 0 INTR_WHILE_WAIT
+ * yes * 0 INTR_WHILE_WAIT
* no 1 0 INTR_WHILE_WAIT
- * yes 0 1 INTR_NEVER
+ * yes * 1 INTR_NEVER
* no 1 1 INTR_WHILE_WAIT
*/
if (!tcp->pid)
continue;
if (debug_flag)
- fprintf(stderr,
- "cleanup: looking at pid %u\n", tcp->pid);
+ error_msg("cleanup: looking at pid %u", tcp->pid);
if (tcp->pid == strace_child) {
kill(tcp->pid, SIGCONT);
kill(tcp->pid, fatal_sig);
}
static void
-print_debug_info(const int pid, const int status)
+print_debug_info(const int pid, int status)
{
const unsigned int event = (unsigned int) status >> 16;
char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16];
e = "STOP";
sprintf(evbuf, ",EVENT_%s (%u)", e, event);
}
- fprintf(stderr, " [wait(0x%06x) = %u] %s%s\n", status, pid, buf, evbuf);
+ error_msg("[wait(0x%06x) = %u] %s%s", status, pid, buf, evbuf);
}
static struct tcb *
-maybe_allocate_tcb(const int pid, const int status)
+maybe_allocate_tcb(const int pid, int status)
{
if (!WIFSTOPPED(status)) {
if (detach_on_execve && pid == strace_child) {
tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
newoutf(tcp);
if (!qflag)
- fprintf(stderr, "Process %d attached\n", pid);
+ error_msg("Process %d attached", pid);
return tcp;
} else {
/* This can happen if a clone call used
* CLONE_PTRACE itself.
*/
- ptrace(PTRACE_CONT, pid, (char *) 0, 0);
+ ptrace(PTRACE_CONT, pid, NULL, 0);
error_msg("Stop of unknown pid %u seen, PTRACE_CONTed it", pid);
return NULL;
}
}
static void
-print_signalled(struct tcb *tcp, const int pid, const int status)
+print_signalled(struct tcb *tcp, const int pid, int status)
{
if (pid == strace_child) {
exit_code = 0x100 | WTERMSIG(status);
}
static void
-print_exited(struct tcb *tcp, const int pid, const int status)
+print_exited(struct tcb *tcp, const int pid, int status)
{
if (pid == strace_child) {
exit_code = WEXITSTATUS(status);
printleader(tcp);
if (si) {
tprintf("--- %s ", signame(sig));
- printsiginfo(si, verbose(tcp));
+ printsiginfo(si);
tprints(" ---\n");
} else
tprintf("--- stopped by %s ---\n", signame(sig));
}
}
-static bool
+static void
startup_tcb(struct tcb *tcp)
{
if (debug_flag)
- fprintf(stderr, "pid %d has TCB_STARTUP, initializing it\n",
- tcp->pid);
+ error_msg("pid %d has TCB_STARTUP, initializing it", tcp->pid);
tcp->flags &= ~TCB_STARTUP;
- if (tcp->flags & TCB_BPTSET) {
- /*
- * 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 (!use_seize) {
if (debug_flag)
- fprintf(stderr, "setting opts 0x%x on pid %d\n",
- ptrace_setoptions, tcp->pid);
+ error_msg("setting opts 0x%x on pid %d",
+ ptrace_setoptions, tcp->pid);
if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) {
if (errno != ESRCH) {
/* Should never happen, really */
}
}
}
-
- return true;
}
/* Returns true iff the main trace loop has to continue. */
if (interrupted)
return false;
- if (popen_pid != 0 && nprocs == 0)
- return false;
+ /*
+ * Used to exit simply when nprocs hits zero, but in this testcase:
+ * int main() { _exit(!!fork()); }
+ * under strace -f, parent sometimes (rarely) manages
+ * to exit before we see the first stop of the child,
+ * and we are losing track of it:
+ * 19923 clone(...) = 19924
+ * 19923 exit_group(1) = ?
+ * 19923 +++ exited with 1 +++
+ * Exiting only when wait() returns ECHILD works better.
+ */
+ if (popen_pid != 0) {
+ /* However, if -o|logger is in use, we can't do that.
+ * Can work around that by double-forking the logger,
+ * but that loses the ability to wait for its completion
+ * on exit. Oh well...
+ */
+ if (nprocs == 0)
+ return false;
+ }
if (interactive)
sigprocmask(SIG_SETMASK, &empty_set, NULL);
return true;
}
- clear_regs();
if (WIFSTOPPED(status))
get_regs(pid);
+ else
+ clear_regs();
event = (unsigned int) status >> 16;
/* Is this the very first time we see this tracee stopped? */
if (tcp->flags & TCB_STARTUP) {
- if (!startup_tcb(tcp))
- return false;
+ startup_tcb(tcp);
+ if (get_scno(tcp) == 1)
+ tcp->s_prev_ent = tcp->s_ent;
}
sig = WSTOPSIG(status);
*/
if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) {
if (debug_flag)
- fprintf(stderr, "ignored SIGSTOP on pid %d\n", tcp->pid);
+ error_msg("ignored SIGSTOP on pid %d", tcp->pid);
tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP;
goto restart_tracee_with_sig_0;
}
if (sig != syscall_trap_sig) {
- siginfo_t si;
+ siginfo_t si = {};
/*
* True if tracee is stopped by signal
/*
* This should be syscall entry or exit.
- * (Or it still can be that pesky post-execve SIGTRAP!)
* Handle it.
*/
if (trace_syscall(tcp) < 0) {
{
init(argc, argv);
- /*
- * Run main tracing loop.
- *
- * Used to be "while (nprocs != 0)", but in this testcase:
- * int main() { _exit(!!fork()); }
- * under strace -f, parent sometimes (rarely) manages
- * to exit before we see the first stop of the child,
- * and we are losing track of it:
- * 19923 clone(...) = 19924
- * 19923 exit_group(1) = ?
- * 19923 +++ exited with 1 +++
- * Waiting for ECHILD works better.
- * (However, if -o|logger is in use, we can't do that.
- * Can work around that by double-forking the logger,
- * but that loses the ability to wait for its completion on exit.
- * Oh well...)
- */
+ exit_code = !nprocs;
+
while (trace())
;