• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 1991, 1992 Paul Kranenburg <pk@cs.few.eur.nl>
3  * Copyright (c) 1993 Branko Lankester <branko@hacktic.nl>
4  * Copyright (c) 1993, 1994, 1995, 1996 Rick Sladkey <jrs@world.std.com>
5  * Copyright (c) 1996-1999 Wichert Akkerman <wichert@cistron.nl>
6  * Copyright (c) 1999-2018 The strace developers.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. The name of the author may not be used to endorse or promote products
18  *    derived from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include "defs.h"
33 #include <stdarg.h>
34 #include <limits.h>
35 #include <fcntl.h>
36 #include "ptrace.h"
37 #include <signal.h>
38 #include <sys/resource.h>
39 #include <sys/wait.h>
40 #include <sys/stat.h>
41 #ifdef HAVE_PATHS_H
42 # include <paths.h>
43 #endif
44 #include <pwd.h>
45 #include <grp.h>
46 #include <dirent.h>
47 #include <sys/utsname.h>
48 #ifdef HAVE_PRCTL
49 # include <sys/prctl.h>
50 #endif
51 #include <asm/unistd.h>
52 
53 #include "largefile_wrappers.h"
54 #include "number_set.h"
55 #include "scno.h"
56 #include "printsiginfo.h"
57 #include "trace_event.h"
58 #include "xstring.h"
59 
60 /* In some libc, these aren't declared. Do it ourself: */
61 extern char **environ;
62 extern int optind;
63 extern char *optarg;
64 
65 #ifdef USE_LIBUNWIND
66 /* if this is true do the stack trace for every system call */
67 bool stack_trace_enabled;
68 #endif
69 
70 #define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
71 
72 /* Glue for systems without a MMU that cannot provide fork() */
73 #if !defined(HAVE_FORK)
74 # undef NOMMU_SYSTEM
75 # define NOMMU_SYSTEM 1
76 #endif
77 #if NOMMU_SYSTEM
78 # define fork() vfork()
79 #endif
80 
81 const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
82 
83 cflag_t cflag = CFLAG_NONE;
84 unsigned int followfork;
85 unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC
86 				 | PTRACE_O_TRACEEXIT;
87 unsigned int xflag;
88 bool debug_flag;
89 bool Tflag;
90 bool iflag;
91 bool count_wallclock;
92 unsigned int qflag;
93 static unsigned int tflag;
94 static bool rflag;
95 static bool print_pid_pfx;
96 
97 /* -I n */
98 enum {
99 	INTR_NOT_SET        = 0,
100 	INTR_ANYWHERE       = 1, /* don't block/ignore any signals */
101 	INTR_WHILE_WAIT     = 2, /* block fatal signals while decoding syscall. default */
102 	INTR_NEVER          = 3, /* block fatal signals. default if '-o FILE PROG' */
103 	INTR_BLOCK_TSTP_TOO = 4, /* block fatal signals and SIGTSTP (^Z) */
104 	NUM_INTR_OPTS
105 };
106 static int opt_intr;
107 /* We play with signal mask only if this mode is active: */
108 #define interactive (opt_intr == INTR_WHILE_WAIT)
109 
110 /*
111  * daemonized_tracer supports -D option.
112  * With this option, strace forks twice.
113  * Unlike normal case, with -D *grandparent* process exec's,
114  * becoming a traced process. Child exits (this prevents traced process
115  * from having children it doesn't expect to have), and grandchild
116  * attaches to grandparent similarly to strace -p PID.
117  * This allows for more transparent interaction in cases
118  * when process and its parent are communicating via signals,
119  * wait() etc. Without -D, strace process gets lodged in between,
120  * disrupting parent<->child link.
121  */
122 static bool daemonized_tracer;
123 
124 #if USE_SEIZE
125 static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP;
126 # define use_seize (post_attach_sigstop == 0)
127 #else
128 # define post_attach_sigstop TCB_IGNORE_ONE_SIGSTOP
129 # define use_seize 0
130 #endif
131 
132 /* Sometimes we want to print only succeeding syscalls. */
133 bool not_failing_only;
134 
135 /* Show path associated with fd arguments */
136 unsigned int show_fd_path;
137 
138 static bool detach_on_execve;
139 
140 static int exit_code;
141 static int strace_child;
142 static int strace_tracer_pid;
143 
144 static const char *username;
145 static uid_t run_uid;
146 static gid_t run_gid;
147 
148 unsigned int max_strlen = DEFAULT_STRLEN;
149 static int acolumn = DEFAULT_ACOLUMN;
150 static char *acolumn_spaces;
151 
152 static const char *outfname;
153 /* If -ff, points to stderr. Else, it's our common output log */
154 static FILE *shared_log;
155 
156 struct tcb *printing_tcp;
157 static struct tcb *current_tcp;
158 
159 static struct tcb **tcbtab;
160 static unsigned int nprocs;
161 static size_t tcbtabsize;
162 
163 #ifndef HAVE_PROGRAM_INVOCATION_NAME
164 char *program_invocation_name;
165 #endif
166 
167 unsigned os_release; /* generated from uname()'s u.release */
168 
169 static void detach(struct tcb *tcp);
170 static void cleanup(void);
171 static void interrupt(int sig);
172 static sigset_t start_set, blocked_set;
173 
174 #ifdef HAVE_SIG_ATOMIC_T
175 static volatile sig_atomic_t interrupted;
176 #else
177 static volatile int interrupted;
178 #endif
179 
180 #ifndef HAVE_STRERROR
181 
182 #if !HAVE_DECL_SYS_ERRLIST
183 extern int sys_nerr;
184 extern char *sys_errlist[];
185 #endif
186 
187 const char *
strerror(int err_no)188 strerror(int err_no)
189 {
190 	static char buf[sizeof("Unknown error %d") + sizeof(int)*3];
191 
192 	if (err_no < 1 || err_no >= sys_nerr) {
193 		xsprintf(buf, "Unknown error %d", err_no);
194 		return buf;
195 	}
196 	return sys_errlist[err_no];
197 }
198 
199 #endif /* HAVE_STERRROR */
200 
201 static void
print_version(void)202 print_version(void)
203 {
204 	static const char features[] =
205 #ifdef USE_LIBUNWIND
206 		" stack-unwind"
207 #endif /* USE_LIBUNWIND */
208 #ifdef USE_DEMANGLE
209 		" stack-demangle"
210 #endif /* USE_DEMANGLE */
211 #if SUPPORTED_PERSONALITIES > 1
212 # if defined HAVE_M32_MPERS
213 		" m32-mpers"
214 # else
215 		" no-m32-mpers"
216 # endif
217 #endif /* SUPPORTED_PERSONALITIES > 1 */
218 #if SUPPORTED_PERSONALITIES > 2
219 # if defined HAVE_MX32_MPERS
220 		" mx32-mpers"
221 # else
222 		" no-mx32-mpers"
223 # endif
224 #endif /* SUPPORTED_PERSONALITIES > 2 */
225 		"";
226 
227 	printf("%s -- version %s\n"
228 	       "Copyright (c) 1991-%s The strace developers <%s>.\n"
229 	       "This is free software; see the source for copying conditions.  There is NO\n"
230 	       "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
231 	       PACKAGE_NAME, PACKAGE_VERSION, COPYRIGHT_YEAR, PACKAGE_URL);
232 	printf("\nOptional features enabled:%s\n",
233 	       features[0] ? features : " (none)");
234 }
235 
236 static void
usage(void)237 usage(void)
238 {
239 	printf("\
240 usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
241               [-a column] [-o file] [-s strsize] [-P path]...\n\
242               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
243    or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
244               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
245 \n\
246 Output format:\n\
247   -a column      alignment COLUMN for printing syscall results (default %d)\n\
248   -i             print instruction pointer at time of syscall\n\
249 "
250 #ifdef USE_LIBUNWIND
251 "\
252   -k             obtain stack trace between each syscall (experimental)\n\
253 "
254 #endif
255 "\
256   -o file        send trace output to FILE instead of stderr\n\
257   -q             suppress messages about attaching, detaching, etc.\n\
258   -r             print relative timestamp\n\
259   -s strsize     limit length of print strings to STRSIZE chars (default %d)\n\
260   -t             print absolute timestamp\n\
261   -tt            print absolute timestamp with usecs\n\
262   -T             print time spent in each syscall\n\
263   -x             print non-ascii strings in hex\n\
264   -xx            print all strings in hex\n\
265   -y             print paths associated with file descriptor arguments\n\
266   -yy            print protocol specific information associated with socket file descriptors\n\
267 \n\
268 Statistics:\n\
269   -c             count time, calls, and errors for each syscall and report summary\n\
270   -C             like -c but also print regular output\n\
271   -O overhead    set overhead for tracing syscalls to OVERHEAD usecs\n\
272   -S sortby      sort syscall counts by: time, calls, name, nothing (default %s)\n\
273   -w             summarise syscall latency (default is system time)\n\
274 \n\
275 Filtering:\n\
276   -e expr        a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
277      options:    trace, abbrev, verbose, raw, signal, read, write, fault\n\
278   -P path        trace accesses to path\n\
279 \n\
280 Tracing:\n\
281   -b execve      detach on execve syscall\n\
282   -D             run tracer process as a detached grandchild, not as parent\n\
283   -f             follow forks\n\
284   -ff            follow forks with output into separate files\n\
285   -I interruptible\n\
286      1:          no signals are blocked\n\
287      2:          fatal signals are blocked while decoding syscall (default)\n\
288      3:          fatal signals are always blocked (default if '-o FILE PROG')\n\
289      4:          fatal signals and SIGTSTP (^Z) are always blocked\n\
290                  (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
291 \n\
292 Startup:\n\
293   -E var         remove var from the environment for command\n\
294   -E var=val     put var=val in the environment for command\n\
295   -p pid         trace process with process id PID, may be repeated\n\
296   -u username    run command as username handling setuid and/or setgid\n\
297 \n\
298 Miscellaneous:\n\
299   -d             enable debug output to stderr\n\
300   -v             verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
301   -h             print help message\n\
302   -V             print version\n\
303 "
304 /* ancient, no one should use it
305 -F -- attempt to follow vforks (deprecated, use -f)\n\
306  */
307 /* this is broken, so don't document it
308 -z -- print only succeeding syscalls\n\
309  */
310 , DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
311 	exit(0);
312 }
313 
314 void ATTRIBUTE_NORETURN
die(void)315 die(void)
316 {
317 	if (strace_tracer_pid == getpid()) {
318 		cflag = 0;
319 		cleanup();
320 		exit(1);
321 	}
322 
323 	_exit(1);
324 }
325 
326 static void
error_opt_arg(int opt,const char * arg)327 error_opt_arg(int opt, const char *arg)
328 {
329 	error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
330 }
331 
332 static const char *ptrace_attach_cmd;
333 
334 static int
ptrace_attach_or_seize(int pid)335 ptrace_attach_or_seize(int pid)
336 {
337 #if USE_SEIZE
338 	int r;
339 	if (!use_seize)
340 		return ptrace_attach_cmd = "PTRACE_ATTACH",
341 		       ptrace(PTRACE_ATTACH, pid, 0L, 0L);
342 	r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
343 	if (r)
344 		return ptrace_attach_cmd = "PTRACE_SEIZE", r;
345 	r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
346 	return ptrace_attach_cmd = "PTRACE_INTERRUPT", r;
347 #else
348 		return ptrace_attach_cmd = "PTRACE_ATTACH",
349 		       ptrace(PTRACE_ATTACH, pid, 0L, 0L);
350 #endif
351 }
352 
353 /*
354  * Used when we want to unblock stopped traced process.
355  * Should be only used with PTRACE_CONT, PTRACE_DETACH and PTRACE_SYSCALL.
356  * Returns 0 on success or if error was ESRCH
357  * (presumably process was killed while we talk to it).
358  * Otherwise prints error message and returns -1.
359  */
360 static int
ptrace_restart(const unsigned int op,struct tcb * const tcp,unsigned int sig)361 ptrace_restart(const unsigned int op, struct tcb *const tcp, unsigned int sig)
362 {
363 	int err;
364 	const char *msg;
365 
366 	errno = 0;
367 	ptrace(op, tcp->pid, 0L, (unsigned long) sig);
368 	err = errno;
369 	if (!err)
370 		return 0;
371 
372 	switch (op) {
373 		case PTRACE_CONT:
374 			msg = "CONT";
375 			break;
376 		case PTRACE_DETACH:
377 			msg = "DETACH";
378 			break;
379 		case PTRACE_LISTEN:
380 			msg = "LISTEN";
381 			break;
382 		default:
383 			msg = "SYSCALL";
384 	}
385 
386 	/*
387 	 * Why curcol != 0? Otherwise sometimes we get this:
388 	 *
389 	 * 10252 kill(10253, SIGKILL)              = 0
390 	 *  <ptrace(SYSCALL,10252):No such process>10253 ...next decode...
391 	 *
392 	 * 10252 died after we retrieved syscall exit data,
393 	 * but before we tried to restart it. Log looks ugly.
394 	 */
395 	if (current_tcp && current_tcp->curcol != 0) {
396 		tprintf(" <ptrace(%s):%s>\n", msg, strerror(err));
397 		line_ended();
398 	}
399 	if (err == ESRCH)
400 		return 0;
401 	errno = err;
402 	perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%u)", msg, tcp->pid, sig);
403 	return -1;
404 }
405 
406 static void
set_cloexec_flag(int fd)407 set_cloexec_flag(int fd)
408 {
409 	int flags, newflags;
410 
411 	flags = fcntl(fd, F_GETFD);
412 	if (flags < 0) {
413 		/* Can happen only if fd is bad.
414 		 * Should never happen: if it does, we have a bug
415 		 * in the caller. Therefore we just abort
416 		 * instead of propagating the error.
417 		 */
418 		perror_msg_and_die("fcntl(%d, F_GETFD)", fd);
419 	}
420 
421 	newflags = flags | FD_CLOEXEC;
422 	if (flags == newflags)
423 		return;
424 
425 	fcntl(fd, F_SETFD, newflags); /* never fails */
426 }
427 
428 static void
kill_save_errno(pid_t pid,int sig)429 kill_save_errno(pid_t pid, int sig)
430 {
431 	int saved_errno = errno;
432 
433 	(void) kill(pid, sig);
434 	errno = saved_errno;
435 }
436 
437 /*
438  * When strace is setuid executable, we have to swap uids
439  * before and after filesystem and process management operations.
440  */
441 static void
swap_uid(void)442 swap_uid(void)
443 {
444 	int euid = geteuid(), uid = getuid();
445 
446 	if (euid != uid && setreuid(euid, uid) < 0) {
447 		perror_msg_and_die("setreuid");
448 	}
449 }
450 
451 static FILE *
strace_fopen(const char * path)452 strace_fopen(const char *path)
453 {
454 	FILE *fp;
455 
456 	swap_uid();
457 	fp = fopen_for_output(path, "w");
458 	if (!fp)
459 		perror_msg_and_die("Can't fopen '%s'", path);
460 	swap_uid();
461 	set_cloexec_flag(fileno(fp));
462 	return fp;
463 }
464 
465 static int popen_pid;
466 
467 #ifndef _PATH_BSHELL
468 # define _PATH_BSHELL "/bin/sh"
469 #endif
470 
471 /*
472  * We cannot use standard popen(3) here because we have to distinguish
473  * popen child process from other processes we trace, and standard popen(3)
474  * does not export its child's pid.
475  */
476 static FILE *
strace_popen(const char * command)477 strace_popen(const char *command)
478 {
479 	FILE *fp;
480 	int pid;
481 	int fds[2];
482 
483 	swap_uid();
484 	if (pipe(fds) < 0)
485 		perror_msg_and_die("pipe");
486 
487 	set_cloexec_flag(fds[1]); /* never fails */
488 
489 	pid = vfork();
490 	if (pid < 0)
491 		perror_msg_and_die("vfork");
492 
493 	if (pid == 0) {
494 		/* child */
495 		close(fds[1]);
496 		if (fds[0] != 0) {
497 			if (dup2(fds[0], 0))
498 				perror_msg_and_die("dup2");
499 			close(fds[0]);
500 		}
501 		execl(_PATH_BSHELL, "sh", "-c", command, NULL);
502 		perror_msg_and_die("Can't execute '%s'", _PATH_BSHELL);
503 	}
504 
505 	/* parent */
506 	popen_pid = pid;
507 	close(fds[0]);
508 	swap_uid();
509 	fp = fdopen(fds[1], "w");
510 	if (!fp)
511 		perror_msg_and_die("fdopen");
512 	return fp;
513 }
514 
515 static void
outf_perror(const struct tcb * const tcp)516 outf_perror(const struct tcb * const tcp)
517 {
518 	if (tcp->outf == stderr)
519 		return;
520 
521 	/* This is ugly, but we don't store separate file names */
522 	if (followfork >= 2)
523 		perror_msg("%s.%u", outfname, tcp->pid);
524 	else
525 		perror_msg("%s", outfname);
526 }
527 
528 ATTRIBUTE_FORMAT((printf, 1, 0))
529 static void
tvprintf(const char * const fmt,va_list args)530 tvprintf(const char *const fmt, va_list args)
531 {
532 	if (current_tcp) {
533 		int n = vfprintf(current_tcp->outf, fmt, args);
534 		if (n < 0) {
535 			/* very unlikely due to vfprintf buffering */
536 			outf_perror(current_tcp);
537 		} else
538 			current_tcp->curcol += n;
539 	}
540 }
541 
542 void
tprintf(const char * fmt,...)543 tprintf(const char *fmt, ...)
544 {
545 	va_list args;
546 	va_start(args, fmt);
547 	tvprintf(fmt, args);
548 	va_end(args);
549 }
550 
551 #ifndef HAVE_FPUTS_UNLOCKED
552 # define fputs_unlocked fputs
553 #endif
554 
555 void
tprints(const char * str)556 tprints(const char *str)
557 {
558 	if (current_tcp) {
559 		int n = fputs_unlocked(str, current_tcp->outf);
560 		if (n >= 0) {
561 			current_tcp->curcol += strlen(str);
562 			return;
563 		}
564 		/* very unlikely due to fputs_unlocked buffering */
565 		outf_perror(current_tcp);
566 	}
567 }
568 
569 void
tprints_comment(const char * const str)570 tprints_comment(const char *const str)
571 {
572 	if (str && *str)
573 		tprintf(" /* %s */", str);
574 }
575 
576 void
tprintf_comment(const char * fmt,...)577 tprintf_comment(const char *fmt, ...)
578 {
579 	if (!fmt || !*fmt)
580 		return;
581 
582 	va_list args;
583 	va_start(args, fmt);
584 	tprints(" /* ");
585 	tvprintf(fmt, args);
586 	tprints(" */");
587 	va_end(args);
588 }
589 
590 static void
flush_tcp_output(const struct tcb * const tcp)591 flush_tcp_output(const struct tcb *const tcp)
592 {
593 	if (fflush(tcp->outf))
594 		outf_perror(tcp);
595 }
596 
597 void
line_ended(void)598 line_ended(void)
599 {
600 	if (current_tcp) {
601 		current_tcp->curcol = 0;
602 		flush_tcp_output(current_tcp);
603 	}
604 	if (printing_tcp) {
605 		printing_tcp->curcol = 0;
606 		printing_tcp = NULL;
607 	}
608 }
609 
610 void
set_current_tcp(const struct tcb * tcp)611 set_current_tcp(const struct tcb *tcp)
612 {
613 	current_tcp = (struct tcb *) tcp;
614 
615 	/* Sync current_personality and stuff */
616 	if (current_tcp)
617 		set_personality(current_tcp->currpers);
618 }
619 
620 void
printleader(struct tcb * tcp)621 printleader(struct tcb *tcp)
622 {
623 	/* If -ff, "previous tcb we printed" is always the same as current,
624 	 * because we have per-tcb output files.
625 	 */
626 	if (followfork >= 2)
627 		printing_tcp = tcp;
628 
629 	if (printing_tcp) {
630 		set_current_tcp(printing_tcp);
631 		if (printing_tcp->curcol != 0 && (followfork < 2 || printing_tcp == tcp)) {
632 			/*
633 			 * case 1: we have a shared log (i.e. not -ff), and last line
634 			 * wasn't finished (same or different tcb, doesn't matter).
635 			 * case 2: split log, we are the same tcb, but our last line
636 			 * didn't finish ("SIGKILL nuked us after syscall entry" etc).
637 			 */
638 			tprints(" <unfinished ...>\n");
639 			printing_tcp->curcol = 0;
640 		}
641 	}
642 
643 	printing_tcp = tcp;
644 	set_current_tcp(tcp);
645 	current_tcp->curcol = 0;
646 
647 	if (print_pid_pfx)
648 		tprintf("%-5d ", tcp->pid);
649 	else if (nprocs > 1 && !outfname)
650 		tprintf("[pid %5u] ", tcp->pid);
651 
652 	if (tflag) {
653 		char str[sizeof("HH:MM:SS")];
654 		struct timeval tv, dtv;
655 		static struct timeval otv;
656 
657 		gettimeofday(&tv, NULL);
658 		if (rflag) {
659 			if (otv.tv_sec == 0)
660 				otv = tv;
661 			tv_sub(&dtv, &tv, &otv);
662 			tprintf("%6ld.%06ld ",
663 				(long) dtv.tv_sec, (long) dtv.tv_usec);
664 			otv = tv;
665 		} else if (tflag > 2) {
666 			tprintf("%ld.%06ld ",
667 				(long) tv.tv_sec, (long) tv.tv_usec);
668 		} else {
669 			time_t local = tv.tv_sec;
670 			strftime(str, sizeof(str), "%T", localtime(&local));
671 			if (tflag > 1)
672 				tprintf("%s.%06ld ", str, (long) tv.tv_usec);
673 			else
674 				tprintf("%s ", str);
675 		}
676 	}
677 	if (iflag)
678 		print_pc(tcp);
679 }
680 
681 void
tabto(void)682 tabto(void)
683 {
684 	if (current_tcp->curcol < acolumn)
685 		tprints(acolumn_spaces + current_tcp->curcol);
686 }
687 
688 /* Should be only called directly *after successful attach* to a tracee.
689  * Otherwise, "strace -oFILE -ff -p<nonexistant_pid>"
690  * may create bogus empty FILE.<nonexistant_pid>, and then die.
691  */
692 static void
newoutf(struct tcb * tcp)693 newoutf(struct tcb *tcp)
694 {
695 	tcp->outf = shared_log; /* if not -ff mode, the same file is for all */
696 	if (followfork >= 2) {
697 		char name[PATH_MAX];
698 		xsprintf(name, "%s.%u", outfname, tcp->pid);
699 		tcp->outf = strace_fopen(name);
700 	}
701 }
702 
703 static void
expand_tcbtab(void)704 expand_tcbtab(void)
705 {
706 	/* Allocate some (more) TCBs (and expand the table).
707 	   We don't want to relocate the TCBs because our
708 	   callers have pointers and it would be a pain.
709 	   So tcbtab is a table of pointers.  Since we never
710 	   free the TCBs, we allocate a single chunk of many.  */
711 	size_t old_tcbtabsize;
712 	struct tcb *newtcbs;
713 	struct tcb **tcb_ptr;
714 
715 	old_tcbtabsize = tcbtabsize;
716 
717 	tcbtab = xgrowarray(tcbtab, &tcbtabsize, sizeof(tcbtab[0]));
718 	newtcbs = xcalloc(tcbtabsize - old_tcbtabsize, sizeof(newtcbs[0]));
719 
720 	for (tcb_ptr = tcbtab + old_tcbtabsize;
721 	    tcb_ptr < tcbtab + tcbtabsize; tcb_ptr++, newtcbs++)
722 		*tcb_ptr = newtcbs;
723 }
724 
725 static struct tcb *
alloctcb(int pid)726 alloctcb(int pid)
727 {
728 	unsigned int i;
729 	struct tcb *tcp;
730 
731 	if (nprocs == tcbtabsize)
732 		expand_tcbtab();
733 
734 	for (i = 0; i < tcbtabsize; i++) {
735 		tcp = tcbtab[i];
736 		if (!tcp->pid) {
737 			memset(tcp, 0, sizeof(*tcp));
738 			tcp->pid = pid;
739 #if SUPPORTED_PERSONALITIES > 1
740 			tcp->currpers = current_personality;
741 #endif
742 
743 #ifdef USE_LIBUNWIND
744 			if (stack_trace_enabled)
745 				unwind_tcb_init(tcp);
746 #endif
747 
748 			nprocs++;
749 			debug_msg("new tcb for pid %d, active tcbs:%d",
750 				  tcp->pid, nprocs);
751 			return tcp;
752 		}
753 	}
754 	error_msg_and_die("bug in alloctcb");
755 }
756 
757 void *
get_tcb_priv_data(const struct tcb * tcp)758 get_tcb_priv_data(const struct tcb *tcp)
759 {
760 	return tcp->_priv_data;
761 }
762 
763 int
set_tcb_priv_data(struct tcb * tcp,void * const priv_data,void (* const free_priv_data)(void *))764 set_tcb_priv_data(struct tcb *tcp, void *const priv_data,
765 		  void (*const free_priv_data)(void *))
766 {
767 	if (tcp->_priv_data)
768 		return -1;
769 
770 	tcp->_free_priv_data = free_priv_data;
771 	tcp->_priv_data = priv_data;
772 
773 	return 0;
774 }
775 
776 void
free_tcb_priv_data(struct tcb * tcp)777 free_tcb_priv_data(struct tcb *tcp)
778 {
779 	if (tcp->_priv_data) {
780 		if (tcp->_free_priv_data) {
781 			tcp->_free_priv_data(tcp->_priv_data);
782 			tcp->_free_priv_data = NULL;
783 		}
784 		tcp->_priv_data = NULL;
785 	}
786 }
787 
788 static void
droptcb(struct tcb * tcp)789 droptcb(struct tcb *tcp)
790 {
791 	if (tcp->pid == 0)
792 		return;
793 
794 	int p;
795 	for (p = 0; p < SUPPORTED_PERSONALITIES; ++p)
796 		free(tcp->inject_vec[p]);
797 
798 	free_tcb_priv_data(tcp);
799 
800 #ifdef USE_LIBUNWIND
801 	if (stack_trace_enabled) {
802 		unwind_tcb_fin(tcp);
803 	}
804 #endif
805 
806 	nprocs--;
807 	debug_msg("dropped tcb for pid %d, %d remain", tcp->pid, nprocs);
808 
809 	if (tcp->outf) {
810 		if (followfork >= 2) {
811 			if (tcp->curcol != 0)
812 				fprintf(tcp->outf, " <detached ...>\n");
813 			fclose(tcp->outf);
814 		} else {
815 			if (printing_tcp == tcp && tcp->curcol != 0)
816 				fprintf(tcp->outf, " <detached ...>\n");
817 			flush_tcp_output(tcp);
818 		}
819 	}
820 
821 	if (current_tcp == tcp)
822 		set_current_tcp(NULL);
823 	if (printing_tcp == tcp)
824 		printing_tcp = NULL;
825 
826 	memset(tcp, 0, sizeof(*tcp));
827 }
828 
829 /* Detach traced process.
830  * Never call DETACH twice on the same process as both unattached and
831  * attached-unstopped processes give the same ESRCH.  For unattached process we
832  * would SIGSTOP it and wait for its SIGSTOP notification forever.
833  */
834 static void
detach(struct tcb * tcp)835 detach(struct tcb *tcp)
836 {
837 	int error;
838 	int status;
839 
840 	/*
841 	 * Linux wrongly insists the child be stopped
842 	 * before detaching.  Arghh.  We go through hoops
843 	 * to make a clean break of things.
844 	 */
845 
846 	if (!(tcp->flags & TCB_ATTACHED))
847 		goto drop;
848 
849 	/* We attached but possibly didn't see the expected SIGSTOP.
850 	 * We must catch exactly one as otherwise the detached process
851 	 * would be left stopped (process state T).
852 	 */
853 	if (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)
854 		goto wait_loop;
855 
856 	error = ptrace(PTRACE_DETACH, tcp->pid, 0, 0);
857 	if (!error) {
858 		/* On a clear day, you can see forever. */
859 		goto drop;
860 	}
861 	if (errno != ESRCH) {
862 		/* Shouldn't happen. */
863 		perror_func_msg("ptrace(PTRACE_DETACH,%u)", tcp->pid);
864 		goto drop;
865 	}
866 	/* ESRCH: process is either not stopped or doesn't exist. */
867 	if (my_tkill(tcp->pid, 0) < 0) {
868 		if (errno != ESRCH)
869 			/* Shouldn't happen. */
870 			perror_func_msg("tkill(%u,0)", tcp->pid);
871 		/* else: process doesn't exist. */
872 		goto drop;
873 	}
874 	/* Process is not stopped, need to stop it. */
875 	if (use_seize) {
876 		/*
877 		 * With SEIZE, tracee can be in group-stop already.
878 		 * In this state sending it another SIGSTOP does nothing.
879 		 * Need to use INTERRUPT.
880 		 * Testcase: trying to ^C a "strace -p <stopped_process>".
881 		 */
882 		error = ptrace(PTRACE_INTERRUPT, tcp->pid, 0, 0);
883 		if (!error)
884 			goto wait_loop;
885 		if (errno != ESRCH)
886 			perror_func_msg("ptrace(PTRACE_INTERRUPT,%u)", tcp->pid);
887 	} else {
888 		error = my_tkill(tcp->pid, SIGSTOP);
889 		if (!error)
890 			goto wait_loop;
891 		if (errno != ESRCH)
892 			perror_func_msg("tkill(%u,SIGSTOP)", tcp->pid);
893 	}
894 	/* Either process doesn't exist, or some weird error. */
895 	goto drop;
896 
897  wait_loop:
898 	/* We end up here in three cases:
899 	 * 1. We sent PTRACE_INTERRUPT (use_seize case)
900 	 * 2. We sent SIGSTOP (!use_seize)
901 	 * 3. Attach SIGSTOP was already pending (TCB_IGNORE_ONE_SIGSTOP set)
902 	 */
903 	for (;;) {
904 		unsigned int sig;
905 		if (waitpid(tcp->pid, &status, __WALL) < 0) {
906 			if (errno == EINTR)
907 				continue;
908 			/*
909 			 * if (errno == ECHILD) break;
910 			 * ^^^  WRONG! We expect this PID to exist,
911 			 * and want to emit a message otherwise:
912 			 */
913 			perror_func_msg("waitpid(%u)", tcp->pid);
914 			break;
915 		}
916 		if (!WIFSTOPPED(status)) {
917 			/*
918 			 * Tracee exited or was killed by signal.
919 			 * We shouldn't normally reach this place:
920 			 * we don't want to consume exit status.
921 			 * Consider "strace -p PID" being ^C-ed:
922 			 * we want merely to detach from PID.
923 			 *
924 			 * However, we _can_ end up here if tracee
925 			 * was SIGKILLed.
926 			 */
927 			break;
928 		}
929 		sig = WSTOPSIG(status);
930 		debug_msg("detach wait: event:%d sig:%d",
931 			  (unsigned) status >> 16, sig);
932 		if (use_seize) {
933 			unsigned event = (unsigned)status >> 16;
934 			if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) {
935 				/*
936 				 * sig == SIGTRAP: PTRACE_INTERRUPT stop.
937 				 * sig == other: process was already stopped
938 				 * with this stopping sig (see tests/detach-stopped).
939 				 * Looks like re-injecting this sig is not necessary
940 				 * in DETACH for the tracee to remain stopped.
941 				 */
942 				sig = 0;
943 			}
944 			/*
945 			 * PTRACE_INTERRUPT is not guaranteed to produce
946 			 * the above event if other ptrace-stop is pending.
947 			 * See tests/detach-sleeping testcase:
948 			 * strace got SIGINT while tracee is sleeping.
949 			 * We sent PTRACE_INTERRUPT.
950 			 * We see syscall exit, not PTRACE_INTERRUPT stop.
951 			 * We won't get PTRACE_INTERRUPT stop
952 			 * if we would CONT now. Need to DETACH.
953 			 */
954 			if (sig == syscall_trap_sig)
955 				sig = 0;
956 			/* else: not sure in which case we can be here.
957 			 * Signal stop? Inject it while detaching.
958 			 */
959 			ptrace_restart(PTRACE_DETACH, tcp, sig);
960 			break;
961 		}
962 		/* Note: this check has to be after use_seize check */
963 		/* (else, in use_seize case SIGSTOP will be mistreated) */
964 		if (sig == SIGSTOP) {
965 			/* Detach, suppressing SIGSTOP */
966 			ptrace_restart(PTRACE_DETACH, tcp, 0);
967 			break;
968 		}
969 		if (sig == syscall_trap_sig)
970 			sig = 0;
971 		/* Can't detach just yet, may need to wait for SIGSTOP */
972 		error = ptrace_restart(PTRACE_CONT, tcp, sig);
973 		if (error < 0) {
974 			/* Should not happen.
975 			 * Note: ptrace_restart returns 0 on ESRCH, so it's not it.
976 			 * ptrace_restart already emitted error message.
977 			 */
978 			break;
979 		}
980 	}
981 
982  drop:
983 	if (!qflag && (tcp->flags & TCB_ATTACHED))
984 		error_msg("Process %u detached", tcp->pid);
985 
986 	droptcb(tcp);
987 }
988 
989 static void
process_opt_p_list(char * opt)990 process_opt_p_list(char *opt)
991 {
992 	while (*opt) {
993 		/*
994 		 * We accept -p PID,PID; -p "`pidof PROG`"; -p "`pgrep PROG`".
995 		 * pidof uses space as delim, pgrep uses newline. :(
996 		 */
997 		int pid;
998 		char *delim = opt + strcspn(opt, "\n\t ,");
999 		char c = *delim;
1000 
1001 		*delim = '\0';
1002 		pid = string_to_uint(opt);
1003 		if (pid <= 0) {
1004 			error_msg_and_die("Invalid process id: '%s'", opt);
1005 		}
1006 		if (pid == strace_tracer_pid) {
1007 			error_msg_and_die("I'm sorry, I can't let you do that, Dave.");
1008 		}
1009 		*delim = c;
1010 		alloctcb(pid);
1011 		if (c == '\0')
1012 			break;
1013 		opt = delim + 1;
1014 	}
1015 }
1016 
1017 static void
attach_tcb(struct tcb * const tcp)1018 attach_tcb(struct tcb *const tcp)
1019 {
1020 	if (ptrace_attach_or_seize(tcp->pid) < 0) {
1021 		perror_msg("attach: ptrace(%s, %d)",
1022 			   ptrace_attach_cmd, tcp->pid);
1023 		droptcb(tcp);
1024 		return;
1025 	}
1026 
1027 	tcp->flags |= TCB_ATTACHED | TCB_GRABBED | TCB_STARTUP |
1028 		      post_attach_sigstop;
1029 	newoutf(tcp);
1030 	debug_msg("attach to pid %d (main) succeeded", tcp->pid);
1031 
1032 	static const char task_path[] = "/proc/%d/task";
1033 	char procdir[sizeof(task_path) + sizeof(int) * 3];
1034 	DIR *dir;
1035 	unsigned int ntid = 0, nerr = 0;
1036 
1037 	if (followfork && tcp->pid != strace_child &&
1038 	    xsprintf(procdir, task_path, tcp->pid) > 0 &&
1039 	    (dir = opendir(procdir)) != NULL) {
1040 		struct_dirent *de;
1041 
1042 		while ((de = read_dir(dir)) != NULL) {
1043 			if (de->d_fileno == 0)
1044 				continue;
1045 
1046 			int tid = string_to_uint(de->d_name);
1047 			if (tid <= 0 || tid == tcp->pid)
1048 				continue;
1049 
1050 			++ntid;
1051 			if (ptrace_attach_or_seize(tid) < 0) {
1052 				++nerr;
1053 				debug_perror_msg("attach: ptrace(%s, %d)",
1054 						 ptrace_attach_cmd, tid);
1055 				continue;
1056 			}
1057 			debug_msg("attach to pid %d succeeded", tid);
1058 
1059 			struct tcb *tid_tcp = alloctcb(tid);
1060 			tid_tcp->flags |= TCB_ATTACHED | TCB_GRABBED |
1061 					  TCB_STARTUP | post_attach_sigstop;
1062 			newoutf(tid_tcp);
1063 		}
1064 
1065 		closedir(dir);
1066 	}
1067 
1068 	if (!qflag) {
1069 		if (ntid > nerr)
1070 			error_msg("Process %u attached"
1071 				  " with %u threads",
1072 				  tcp->pid, ntid - nerr + 1);
1073 		else
1074 			error_msg("Process %u attached",
1075 				  tcp->pid);
1076 	}
1077 }
1078 
1079 static void
startup_attach(void)1080 startup_attach(void)
1081 {
1082 	pid_t parent_pid = strace_tracer_pid;
1083 	unsigned int tcbi;
1084 	struct tcb *tcp;
1085 
1086 	/*
1087 	 * Block user interruptions as we would leave the traced
1088 	 * process stopped (process state T) if we would terminate in
1089 	 * between PTRACE_ATTACH and wait4() on SIGSTOP.
1090 	 * We rely on cleanup() from this point on.
1091 	 */
1092 	if (interactive)
1093 		sigprocmask(SIG_SETMASK, &blocked_set, NULL);
1094 
1095 	if (daemonized_tracer) {
1096 		pid_t pid = fork();
1097 		if (pid < 0)
1098 			perror_func_msg_and_die("fork");
1099 
1100 		if (pid) { /* parent */
1101 			/*
1102 			 * Wait for grandchild to attach to straced process
1103 			 * (grandparent). Grandchild SIGKILLs us after it attached.
1104 			 * Grandparent's wait() is unblocked by our death,
1105 			 * it proceeds to exec the straced program.
1106 			 */
1107 			pause();
1108 			_exit(0); /* paranoia */
1109 		}
1110 		/* grandchild */
1111 		/* We will be the tracer process. Remember our new pid: */
1112 		strace_tracer_pid = getpid();
1113 	}
1114 
1115 	for (tcbi = 0; tcbi < tcbtabsize; tcbi++) {
1116 		tcp = tcbtab[tcbi];
1117 
1118 		if (!tcp->pid)
1119 			continue;
1120 
1121 		/* Is this a process we should attach to, but not yet attached? */
1122 		if (tcp->flags & TCB_ATTACHED)
1123 			continue; /* no, we already attached it */
1124 
1125 		if (tcp->pid == parent_pid || tcp->pid == strace_tracer_pid) {
1126 			errno = EPERM;
1127 			perror_msg("attach: pid %d", tcp->pid);
1128 			droptcb(tcp);
1129 			continue;
1130 		}
1131 
1132 		attach_tcb(tcp);
1133 
1134 		if (interactive) {
1135 			sigprocmask(SIG_SETMASK, &start_set, NULL);
1136 			if (interrupted)
1137 				goto ret;
1138 			sigprocmask(SIG_SETMASK, &blocked_set, NULL);
1139 		}
1140 	} /* for each tcbtab[] */
1141 
1142 	if (daemonized_tracer) {
1143 		/*
1144 		 * Make parent go away.
1145 		 * Also makes grandparent's wait() unblock.
1146 		 */
1147 		kill(parent_pid, SIGKILL);
1148 		strace_child = 0;
1149 	}
1150 
1151  ret:
1152 	if (interactive)
1153 		sigprocmask(SIG_SETMASK, &start_set, NULL);
1154 }
1155 
1156 /* Stack-o-phobic exec helper, in the hope to work around
1157  * NOMMU + "daemonized tracer" difficulty.
1158  */
1159 struct exec_params {
1160 	int fd_to_close;
1161 	uid_t run_euid;
1162 	gid_t run_egid;
1163 	char **argv;
1164 	char *pathname;
1165 	struct sigaction child_sa;
1166 };
1167 static struct exec_params params_for_tracee;
1168 
1169 static void ATTRIBUTE_NOINLINE ATTRIBUTE_NORETURN
exec_or_die(void)1170 exec_or_die(void)
1171 {
1172 	struct exec_params *params = &params_for_tracee;
1173 
1174 	if (params->fd_to_close >= 0)
1175 		close(params->fd_to_close);
1176 	if (!daemonized_tracer && !use_seize) {
1177 		if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) {
1178 			perror_msg_and_die("ptrace(PTRACE_TRACEME, ...)");
1179 		}
1180 	}
1181 
1182 	if (username != NULL) {
1183 		/*
1184 		 * It is important to set groups before we
1185 		 * lose privileges on setuid.
1186 		 */
1187 		if (initgroups(username, run_gid) < 0) {
1188 			perror_msg_and_die("initgroups");
1189 		}
1190 		if (setregid(run_gid, params->run_egid) < 0) {
1191 			perror_msg_and_die("setregid");
1192 		}
1193 		if (setreuid(run_uid, params->run_euid) < 0) {
1194 			perror_msg_and_die("setreuid");
1195 		}
1196 	} else if (geteuid() != 0)
1197 		if (setreuid(run_uid, run_uid) < 0) {
1198 			perror_msg_and_die("setreuid");
1199 		}
1200 
1201 	if (!daemonized_tracer) {
1202 		/*
1203 		 * Induce a ptrace stop. Tracer (our parent)
1204 		 * will resume us with PTRACE_SYSCALL and display
1205 		 * the immediately following execve syscall.
1206 		 * Can't do this on NOMMU systems, we are after
1207 		 * vfork: parent is blocked, stopping would deadlock.
1208 		 */
1209 		if (!NOMMU_SYSTEM)
1210 			kill(getpid(), SIGSTOP);
1211 	} else {
1212 		alarm(3);
1213 		/* we depend on SIGCHLD set to SIG_DFL by init code */
1214 		/* if it happens to be SIG_IGN'ed, wait won't block */
1215 		wait(NULL);
1216 		alarm(0);
1217 	}
1218 
1219 	if (params_for_tracee.child_sa.sa_handler != SIG_DFL)
1220 		sigaction(SIGCHLD, &params_for_tracee.child_sa, NULL);
1221 
1222 	execv(params->pathname, params->argv);
1223 	perror_msg_and_die("exec");
1224 }
1225 
1226 /*
1227  * Open a dummy descriptor for use as a placeholder.
1228  * The descriptor is O_RDONLY with FD_CLOEXEC flag set.
1229  * A read attempt from such descriptor ends with EOF,
1230  * a write attempt is rejected with EBADF.
1231  */
1232 static int
open_dummy_desc(void)1233 open_dummy_desc(void)
1234 {
1235 	int fds[2];
1236 
1237 	if (pipe(fds))
1238 		perror_func_msg_and_die("pipe");
1239 	close(fds[1]);
1240 	set_cloexec_flag(fds[0]);
1241 	return fds[0];
1242 }
1243 
1244 /* placeholder fds status for stdin and stdout */
1245 static bool fd_is_placeholder[2];
1246 
1247 /*
1248  * Ensure that all standard file descriptors are open by opening placeholder
1249  * file descriptors for those standard file descriptors that are not open.
1250  *
1251  * The information which descriptors have been made open is saved
1252  * in fd_is_placeholder for later use.
1253  */
1254 static void
ensure_standard_fds_opened(void)1255 ensure_standard_fds_opened(void)
1256 {
1257 	int fd;
1258 
1259 	while ((fd = open_dummy_desc()) <= 2) {
1260 		if (fd == 2)
1261 			break;
1262 		fd_is_placeholder[fd] = true;
1263 	}
1264 
1265 	if (fd > 2)
1266 		close(fd);
1267 }
1268 
1269 /*
1270  * Redirect stdin and stdout unless they have been opened earlier
1271  * by ensure_standard_fds_opened as placeholders.
1272  */
1273 static void
redirect_standard_fds(void)1274 redirect_standard_fds(void)
1275 {
1276 	int i;
1277 
1278 	/*
1279 	 * It might be a good idea to redirect stderr as well,
1280 	 * but we sometimes need to print error messages.
1281 	 */
1282 	for (i = 0; i <= 1; ++i) {
1283 		if (!fd_is_placeholder[i]) {
1284 			close(i);
1285 			open_dummy_desc();
1286 		}
1287 	}
1288 }
1289 
1290 static void
startup_child(char ** argv)1291 startup_child(char **argv)
1292 {
1293 	struct_stat statbuf;
1294 	const char *filename;
1295 	size_t filename_len;
1296 	char pathname[PATH_MAX];
1297 	int pid;
1298 	struct tcb *tcp;
1299 
1300 	filename = argv[0];
1301 	filename_len = strlen(filename);
1302 
1303 	if (filename_len > sizeof(pathname) - 1) {
1304 		errno = ENAMETOOLONG;
1305 		perror_msg_and_die("exec");
1306 	}
1307 	if (strchr(filename, '/')) {
1308 		strcpy(pathname, filename);
1309 	}
1310 #ifdef USE_DEBUGGING_EXEC
1311 	/*
1312 	 * Debuggers customarily check the current directory
1313 	 * first regardless of the path but doing that gives
1314 	 * security geeks a panic attack.
1315 	 */
1316 	else if (stat_file(filename, &statbuf) == 0)
1317 		strcpy(pathname, filename);
1318 #endif /* USE_DEBUGGING_EXEC */
1319 	else {
1320 		const char *path;
1321 		size_t m, n, len;
1322 
1323 		for (path = getenv("PATH"); path && *path; path += m) {
1324 			const char *colon = strchr(path, ':');
1325 			if (colon) {
1326 				n = colon - path;
1327 				m = n + 1;
1328 			} else
1329 				m = n = strlen(path);
1330 			if (n == 0) {
1331 				if (!getcwd(pathname, PATH_MAX))
1332 					continue;
1333 				len = strlen(pathname);
1334 			} else if (n > sizeof(pathname) - 1)
1335 				continue;
1336 			else {
1337 				strncpy(pathname, path, n);
1338 				len = n;
1339 			}
1340 			if (len && pathname[len - 1] != '/')
1341 				pathname[len++] = '/';
1342 			if (filename_len + len > sizeof(pathname) - 1)
1343 				continue;
1344 			strcpy(pathname + len, filename);
1345 			if (stat_file(pathname, &statbuf) == 0 &&
1346 			    /* Accept only regular files
1347 			       with some execute bits set.
1348 			       XXX not perfect, might still fail */
1349 			    S_ISREG(statbuf.st_mode) &&
1350 			    (statbuf.st_mode & 0111))
1351 				break;
1352 		}
1353 		if (!path || !*path)
1354 			pathname[0] = '\0';
1355 	}
1356 	if (stat_file(pathname, &statbuf) < 0) {
1357 		perror_msg_and_die("Can't stat '%s'", filename);
1358 	}
1359 
1360 	params_for_tracee.fd_to_close = (shared_log != stderr) ? fileno(shared_log) : -1;
1361 	params_for_tracee.run_euid = (statbuf.st_mode & S_ISUID) ? statbuf.st_uid : run_uid;
1362 	params_for_tracee.run_egid = (statbuf.st_mode & S_ISGID) ? statbuf.st_gid : run_gid;
1363 	params_for_tracee.argv = argv;
1364 	/*
1365 	 * On NOMMU, can be safely freed only after execve in tracee.
1366 	 * It's hard to know when that happens, so we just leak it.
1367 	 */
1368 	params_for_tracee.pathname = NOMMU_SYSTEM ? xstrdup(pathname) : pathname;
1369 
1370 #if defined HAVE_PRCTL && defined PR_SET_PTRACER && defined PR_SET_PTRACER_ANY
1371 	if (daemonized_tracer)
1372 		prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY);
1373 #endif
1374 
1375 	pid = fork();
1376 	if (pid < 0)
1377 		perror_func_msg_and_die("fork");
1378 
1379 	if ((pid != 0 && daemonized_tracer)
1380 	 || (pid == 0 && !daemonized_tracer)
1381 	) {
1382 		/* We are to become the tracee. Two cases:
1383 		 * -D: we are parent
1384 		 * not -D: we are child
1385 		 */
1386 		exec_or_die();
1387 	}
1388 
1389 	/* We are the tracer */
1390 
1391 	if (!daemonized_tracer) {
1392 		strace_child = pid;
1393 		if (!use_seize) {
1394 			/* child did PTRACE_TRACEME, nothing to do in parent */
1395 		} else {
1396 			if (!NOMMU_SYSTEM) {
1397 				/* Wait until child stopped itself */
1398 				int status;
1399 				while (waitpid(pid, &status, WSTOPPED) < 0) {
1400 					if (errno == EINTR)
1401 						continue;
1402 					perror_msg_and_die("waitpid");
1403 				}
1404 				if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP) {
1405 					kill_save_errno(pid, SIGKILL);
1406 					perror_msg_and_die("Unexpected wait status %#x",
1407 							   status);
1408 				}
1409 			}
1410 			/* Else: NOMMU case, we have no way to sync.
1411 			 * Just attach to it as soon as possible.
1412 			 * This means that we may miss a few first syscalls...
1413 			 */
1414 
1415 			if (ptrace_attach_or_seize(pid)) {
1416 				kill_save_errno(pid, SIGKILL);
1417 				perror_msg_and_die("attach: ptrace(%s, %d)",
1418 						   ptrace_attach_cmd, pid);
1419 			}
1420 			if (!NOMMU_SYSTEM)
1421 				kill(pid, SIGCONT);
1422 		}
1423 		tcp = alloctcb(pid);
1424 		tcp->flags |= TCB_ATTACHED | TCB_STARTUP
1425 			    | TCB_SKIP_DETACH_ON_FIRST_EXEC
1426 			    | (NOMMU_SYSTEM ? 0 : (TCB_HIDE_LOG | post_attach_sigstop));
1427 		newoutf(tcp);
1428 	} else {
1429 		/* With -D, we are *child* here, the tracee is our parent. */
1430 		strace_child = strace_tracer_pid;
1431 		strace_tracer_pid = getpid();
1432 		tcp = alloctcb(strace_child);
1433 		tcp->flags |= TCB_SKIP_DETACH_ON_FIRST_EXEC | TCB_HIDE_LOG;
1434 		/* attaching will be done later, by startup_attach */
1435 		/* note: we don't do newoutf(tcp) here either! */
1436 
1437 		/* NOMMU BUG! -D mode is active, we (child) return,
1438 		 * and we will scribble over parent's stack!
1439 		 * When parent later unpauses, it segfaults.
1440 		 *
1441 		 * We work around it
1442 		 * (1) by declaring exec_or_die() NORETURN,
1443 		 * hopefully compiler will just jump to it
1444 		 * instead of call (won't push anything to stack),
1445 		 * (2) by trying very hard in exec_or_die()
1446 		 * to not use any stack,
1447 		 * (3) having a really big (PATH_MAX) stack object
1448 		 * in this function, which creates a "buffer" between
1449 		 * child's and parent's stack pointers.
1450 		 * This may save us if (1) and (2) failed
1451 		 * and compiler decided to use stack in exec_or_die() anyway
1452 		 * (happens on i386 because of stack parameter passing).
1453 		 *
1454 		 * A cleaner solution is to use makecontext + setcontext
1455 		 * to create a genuine separate stack and execute on it.
1456 		 */
1457 	}
1458 	/*
1459 	 * A case where straced process is part of a pipe:
1460 	 * { sleep 1; yes | head -n99999; } | strace -o/dev/null sh -c 'exec <&-; sleep 9'
1461 	 * If strace won't close its fd#0, closing it in tracee is not enough:
1462 	 * the pipe is still open, it has a reader. Thus, "head" will not get its
1463 	 * SIGPIPE at once, on the first write.
1464 	 *
1465 	 * Preventing it by redirecting strace's stdin/out.
1466 	 * (Don't leave fds 0 and 1 closed, this is bad practice: future opens
1467 	 * will reuse them, unexpectedly making a newly opened object "stdin").
1468 	 */
1469 	redirect_standard_fds();
1470 }
1471 
1472 #if USE_SEIZE
1473 static void
test_ptrace_seize(void)1474 test_ptrace_seize(void)
1475 {
1476 	int pid;
1477 
1478 	/* Need fork for test. NOMMU has no forks */
1479 	if (NOMMU_SYSTEM) {
1480 		post_attach_sigstop = 0; /* this sets use_seize to 1 */
1481 		return;
1482 	}
1483 
1484 	pid = fork();
1485 	if (pid < 0)
1486 		perror_func_msg_and_die("fork");
1487 
1488 	if (pid == 0) {
1489 		pause();
1490 		_exit(0);
1491 	}
1492 
1493 	/* PTRACE_SEIZE, unlike ATTACH, doesn't force tracee to trap.  After
1494 	 * attaching tracee continues to run unless a trap condition occurs.
1495 	 * PTRACE_SEIZE doesn't affect signal or group stop state.
1496 	 */
1497 	if (ptrace(PTRACE_SEIZE, pid, 0, 0) == 0) {
1498 		post_attach_sigstop = 0; /* this sets use_seize to 1 */
1499 	} else {
1500 		debug_msg("PTRACE_SEIZE doesn't work");
1501 	}
1502 
1503 	kill(pid, SIGKILL);
1504 
1505 	while (1) {
1506 		int status, tracee_pid;
1507 
1508 		errno = 0;
1509 		tracee_pid = waitpid(pid, &status, 0);
1510 		if (tracee_pid <= 0) {
1511 			if (errno == EINTR)
1512 				continue;
1513 			perror_func_msg_and_die("unexpected wait result %d",
1514 						tracee_pid);
1515 		}
1516 		if (WIFSIGNALED(status))
1517 			return;
1518 
1519 		error_func_msg_and_die("unexpected wait status %#x", status);
1520 	}
1521 }
1522 #else /* !USE_SEIZE */
1523 # define test_ptrace_seize() ((void)0)
1524 #endif
1525 
1526 static unsigned
get_os_release(void)1527 get_os_release(void)
1528 {
1529 	unsigned rel;
1530 	const char *p;
1531 	struct utsname u;
1532 	if (uname(&u) < 0)
1533 		perror_msg_and_die("uname");
1534 	/* u.release has this form: "3.2.9[-some-garbage]" */
1535 	rel = 0;
1536 	p = u.release;
1537 	for (;;) {
1538 		if (!(*p >= '0' && *p <= '9'))
1539 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1540 		/* Note: this open-codes KERNEL_VERSION(): */
1541 		rel = (rel << 8) | atoi(p);
1542 		if (rel >= KERNEL_VERSION(1, 0, 0))
1543 			break;
1544 		while (*p >= '0' && *p <= '9')
1545 			p++;
1546 		if (*p != '.') {
1547 			if (rel >= KERNEL_VERSION(0, 1, 0)) {
1548 				/* "X.Y-something" means "X.Y.0" */
1549 				rel <<= 8;
1550 				break;
1551 			}
1552 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1553 		}
1554 		p++;
1555 	}
1556 	return rel;
1557 }
1558 
1559 static void
set_sighandler(int signo,void (* sighandler)(int),struct sigaction * oldact)1560 set_sighandler(int signo, void (*sighandler)(int), struct sigaction *oldact)
1561 {
1562 	/* if signal handler is a function, add the signal to blocked_set */
1563 	if (sighandler != SIG_IGN && sighandler != SIG_DFL)
1564 		sigaddset(&blocked_set, signo);
1565 
1566 	const struct sigaction sa = { .sa_handler = sighandler };
1567 	sigaction(signo, &sa, oldact);
1568 }
1569 
1570 /*
1571  * Initialization part of main() was eating much stack (~0.5k),
1572  * which was unused after init.
1573  * We can reuse it if we move init code into a separate function.
1574  *
1575  * Don't want main() to inline us and defeat the reason
1576  * we have a separate function.
1577  */
1578 static void ATTRIBUTE_NOINLINE
init(int argc,char * argv[])1579 init(int argc, char *argv[])
1580 {
1581 	int c, i;
1582 	int optF = 0;
1583 
1584 	if (!program_invocation_name || !*program_invocation_name) {
1585 		static char name[] = "strace";
1586 		program_invocation_name =
1587 			(argc > 0 && argv[0] && *argv[0]) ? argv[0] : name;
1588 	}
1589 
1590 	strace_tracer_pid = getpid();
1591 
1592 	os_release = get_os_release();
1593 
1594 	shared_log = stderr;
1595 	set_sortby(DEFAULT_SORTBY);
1596 	set_personality(DEFAULT_PERSONALITY);
1597 	qualify("trace=all");
1598 	qualify("abbrev=all");
1599 	qualify("verbose=all");
1600 #if DEFAULT_QUAL_FLAGS != (QUAL_TRACE | QUAL_ABBREV | QUAL_VERBOSE)
1601 # error Bug in DEFAULT_QUAL_FLAGS
1602 #endif
1603 	qualify("signal=all");
1604 	while ((c = getopt(argc, argv, "+"
1605 #ifdef USE_LIBUNWIND
1606 	    "k"
1607 #endif
1608 	    "a:b:cCdDe:E:fFhiI:o:O:p:P:qrs:S:tTu:vVwxyz")) != EOF) {
1609 		switch (c) {
1610 		case 'a':
1611 			acolumn = string_to_uint(optarg);
1612 			if (acolumn < 0)
1613 				error_opt_arg(c, optarg);
1614 			break;
1615 		case 'b':
1616 			if (strcmp(optarg, "execve") != 0)
1617 				error_msg_and_die("Syscall '%s' for -b isn't supported",
1618 					optarg);
1619 			detach_on_execve = 1;
1620 			break;
1621 		case 'c':
1622 			if (cflag == CFLAG_BOTH) {
1623 				error_msg_and_help("-c and -C are mutually exclusive");
1624 			}
1625 			cflag = CFLAG_ONLY_STATS;
1626 			break;
1627 		case 'C':
1628 			if (cflag == CFLAG_ONLY_STATS) {
1629 				error_msg_and_help("-c and -C are mutually exclusive");
1630 			}
1631 			cflag = CFLAG_BOTH;
1632 			break;
1633 		case 'd':
1634 			debug_flag = 1;
1635 			break;
1636 		case 'D':
1637 			daemonized_tracer = 1;
1638 			break;
1639 		case 'e':
1640 			qualify(optarg);
1641 			break;
1642 		case 'E':
1643 			if (putenv(optarg) < 0)
1644 				perror_msg_and_die("putenv");
1645 			break;
1646 		case 'f':
1647 			followfork++;
1648 			break;
1649 		case 'F':
1650 			optF = 1;
1651 			break;
1652 		case 'h':
1653 			usage();
1654 			break;
1655 		case 'i':
1656 			iflag = 1;
1657 			break;
1658 		case 'I':
1659 			opt_intr = string_to_uint_upto(optarg, NUM_INTR_OPTS - 1);
1660 			if (opt_intr <= 0)
1661 				error_opt_arg(c, optarg);
1662 			break;
1663 #ifdef USE_LIBUNWIND
1664 		case 'k':
1665 			stack_trace_enabled = true;
1666 			break;
1667 #endif
1668 		case 'o':
1669 			outfname = optarg;
1670 			break;
1671 		case 'O':
1672 			i = string_to_uint(optarg);
1673 			if (i < 0)
1674 				error_opt_arg(c, optarg);
1675 			set_overhead(i);
1676 			break;
1677 		case 'p':
1678 			process_opt_p_list(optarg);
1679 			break;
1680 		case 'P':
1681 			pathtrace_select(optarg);
1682 			break;
1683 		case 'q':
1684 			qflag++;
1685 			break;
1686 		case 'r':
1687 			rflag = 1;
1688 			break;
1689 		case 's':
1690 			i = string_to_uint(optarg);
1691 			if (i < 0 || (unsigned int) i > -1U / 4)
1692 				error_opt_arg(c, optarg);
1693 			max_strlen = i;
1694 			break;
1695 		case 'S':
1696 			set_sortby(optarg);
1697 			break;
1698 		case 't':
1699 			tflag++;
1700 			break;
1701 		case 'T':
1702 			Tflag = 1;
1703 			break;
1704 		case 'u':
1705 			username = optarg;
1706 			break;
1707 		case 'v':
1708 			qualify("abbrev=none");
1709 			break;
1710 		case 'V':
1711 			print_version();
1712 			exit(0);
1713 			break;
1714 		case 'w':
1715 			count_wallclock = 1;
1716 			break;
1717 		case 'x':
1718 			xflag++;
1719 			break;
1720 		case 'y':
1721 			show_fd_path++;
1722 			break;
1723 		case 'z':
1724 			not_failing_only = 1;
1725 			break;
1726 		default:
1727 			error_msg_and_help(NULL);
1728 			break;
1729 		}
1730 	}
1731 
1732 	argv += optind;
1733 	argc -= optind;
1734 
1735 	if (argc < 0 || (!nprocs && !argc)) {
1736 		error_msg_and_help("must have PROG [ARGS] or -p PID");
1737 	}
1738 
1739 	if (!argc && daemonized_tracer) {
1740 		error_msg_and_help("PROG [ARGS] must be specified with -D");
1741 	}
1742 
1743 	if (optF) {
1744 		if (followfork) {
1745 			error_msg("deprecated option -F ignored");
1746 		} else {
1747 			error_msg("option -F is deprecated, "
1748 				  "please use -f instead");
1749 			followfork = optF;
1750 		}
1751 	}
1752 
1753 	if (followfork >= 2 && cflag) {
1754 		error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
1755 	}
1756 
1757 	if (count_wallclock && !cflag) {
1758 		error_msg_and_help("-w must be given with (-c or -C)");
1759 	}
1760 
1761 	if (cflag == CFLAG_ONLY_STATS) {
1762 		if (iflag)
1763 			error_msg("-%c has no effect with -c", 'i');
1764 #ifdef USE_LIBUNWIND
1765 		if (stack_trace_enabled)
1766 			error_msg("-%c has no effect with -c", 'k');
1767 #endif
1768 		if (rflag)
1769 			error_msg("-%c has no effect with -c", 'r');
1770 		if (tflag)
1771 			error_msg("-%c has no effect with -c", 't');
1772 		if (Tflag)
1773 			error_msg("-%c has no effect with -c", 'T');
1774 		if (show_fd_path)
1775 			error_msg("-%c has no effect with -c", 'y');
1776 	}
1777 
1778 	if (rflag) {
1779 		if (tflag > 1)
1780 			error_msg("-tt has no effect with -r");
1781 		tflag = 1;
1782 	}
1783 
1784 	acolumn_spaces = xmalloc(acolumn + 1);
1785 	memset(acolumn_spaces, ' ', acolumn);
1786 	acolumn_spaces[acolumn] = '\0';
1787 
1788 	sigprocmask(SIG_SETMASK, NULL, &start_set);
1789 	memcpy(&blocked_set, &start_set, sizeof(blocked_set));
1790 
1791 	set_sighandler(SIGCHLD, SIG_DFL, &params_for_tracee.child_sa);
1792 
1793 #ifdef USE_LIBUNWIND
1794 	if (stack_trace_enabled) {
1795 		unsigned int tcbi;
1796 
1797 		unwind_init();
1798 		for (tcbi = 0; tcbi < tcbtabsize; ++tcbi) {
1799 			unwind_tcb_init(tcbtab[tcbi]);
1800 		}
1801 	}
1802 #endif
1803 
1804 	/* See if they want to run as another user. */
1805 	if (username != NULL) {
1806 		struct passwd *pent;
1807 
1808 		if (getuid() != 0 || geteuid() != 0) {
1809 			error_msg_and_die("You must be root to use the -u option");
1810 		}
1811 		pent = getpwnam(username);
1812 		if (pent == NULL) {
1813 			error_msg_and_die("Cannot find user '%s'", username);
1814 		}
1815 		run_uid = pent->pw_uid;
1816 		run_gid = pent->pw_gid;
1817 	} else {
1818 		run_uid = getuid();
1819 		run_gid = getgid();
1820 	}
1821 
1822 	if (followfork)
1823 		ptrace_setoptions |= PTRACE_O_TRACECLONE |
1824 				     PTRACE_O_TRACEFORK |
1825 				     PTRACE_O_TRACEVFORK;
1826 	debug_msg("ptrace_setoptions = %#x", ptrace_setoptions);
1827 	test_ptrace_seize();
1828 
1829 	/*
1830 	 * Is something weird with our stdin and/or stdout -
1831 	 * for example, may they be not open? In this case,
1832 	 * ensure that none of the future opens uses them.
1833 	 *
1834 	 * This was seen in the wild when /proc/sys/kernel/core_pattern
1835 	 * was set to "|/bin/strace -o/tmp/LOG PROG":
1836 	 * kernel runs coredump helper with fd#0 open but fd#1 closed (!),
1837 	 * therefore LOG gets opened to fd#1, and fd#1 is closed by
1838 	 * "don't hold up stdin/out open" code soon after.
1839 	 */
1840 	ensure_standard_fds_opened();
1841 
1842 	/* Check if they want to redirect the output. */
1843 	if (outfname) {
1844 		/* See if they want to pipe the output. */
1845 		if (outfname[0] == '|' || outfname[0] == '!') {
1846 			/*
1847 			 * We can't do the <outfname>.PID funny business
1848 			 * when using popen, so prohibit it.
1849 			 */
1850 			if (followfork >= 2)
1851 				error_msg_and_help("piping the output and -ff "
1852 						   "are mutually exclusive");
1853 			shared_log = strace_popen(outfname + 1);
1854 		} else if (followfork < 2) {
1855 			shared_log = strace_fopen(outfname);
1856 		} else if (strlen(outfname) >= PATH_MAX - sizeof(int) * 3) {
1857 			errno = ENAMETOOLONG;
1858 			perror_msg_and_die("%s", outfname);
1859 		}
1860 	} else {
1861 		/* -ff without -o FILE is the same as single -f */
1862 		if (followfork >= 2)
1863 			followfork = 1;
1864 	}
1865 
1866 	if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
1867 		setvbuf(shared_log, NULL, _IOLBF, 0);
1868 	}
1869 
1870 	/*
1871 	 * argv[0]	-pPID	-oFILE	Default interactive setting
1872 	 * yes		*	0	INTR_WHILE_WAIT
1873 	 * no		1	0	INTR_WHILE_WAIT
1874 	 * yes		*	1	INTR_NEVER
1875 	 * no		1	1	INTR_WHILE_WAIT
1876 	 */
1877 
1878 	if (outfname && argc) {
1879 		if (!opt_intr)
1880 			opt_intr = INTR_NEVER;
1881 		if (!qflag)
1882 			qflag = 1;
1883 	}
1884 	if (!opt_intr)
1885 		opt_intr = INTR_WHILE_WAIT;
1886 
1887 	/*
1888 	 * startup_child() must be called before the signal handlers get
1889 	 * installed below as they are inherited into the spawned process.
1890 	 * Also we do not need to be protected by them as during interruption
1891 	 * in the startup_child() mode we kill the spawned process anyway.
1892 	 */
1893 	if (argc) {
1894 		startup_child(argv);
1895 	}
1896 
1897 	set_sighandler(SIGTTOU, SIG_IGN, NULL);
1898 	set_sighandler(SIGTTIN, SIG_IGN, NULL);
1899 	if (opt_intr != INTR_ANYWHERE) {
1900 		if (opt_intr == INTR_BLOCK_TSTP_TOO)
1901 			set_sighandler(SIGTSTP, SIG_IGN, NULL);
1902 		/*
1903 		 * In interactive mode (if no -o OUTFILE, or -p PID is used),
1904 		 * fatal signals are blocked while syscall stop is processed,
1905 		 * and acted on in between, when waiting for new syscall stops.
1906 		 * In non-interactive mode, signals are ignored.
1907 		 */
1908 		set_sighandler(SIGHUP, interactive ? interrupt : SIG_IGN, NULL);
1909 		set_sighandler(SIGINT, interactive ? interrupt : SIG_IGN, NULL);
1910 		set_sighandler(SIGQUIT, interactive ? interrupt : SIG_IGN, NULL);
1911 		set_sighandler(SIGPIPE, interactive ? interrupt : SIG_IGN, NULL);
1912 		set_sighandler(SIGTERM, interactive ? interrupt : SIG_IGN, NULL);
1913 	}
1914 
1915 	if (nprocs != 0 || daemonized_tracer)
1916 		startup_attach();
1917 
1918 	/* Do we want pids printed in our -o OUTFILE?
1919 	 * -ff: no (every pid has its own file); or
1920 	 * -f: yes (there can be more pids in the future); or
1921 	 * -p PID1,PID2: yes (there are already more than one pid)
1922 	 */
1923 	print_pid_pfx = (outfname && followfork < 2 && (followfork == 1 || nprocs > 1));
1924 }
1925 
1926 static struct tcb *
pid2tcb(int pid)1927 pid2tcb(int pid)
1928 {
1929 	unsigned int i;
1930 
1931 	if (pid <= 0)
1932 		return NULL;
1933 
1934 	for (i = 0; i < tcbtabsize; i++) {
1935 		struct tcb *tcp = tcbtab[i];
1936 		if (tcp->pid == pid)
1937 			return tcp;
1938 	}
1939 
1940 	return NULL;
1941 }
1942 
1943 static void
cleanup(void)1944 cleanup(void)
1945 {
1946 	unsigned int i;
1947 	struct tcb *tcp;
1948 	int fatal_sig;
1949 
1950 	/* 'interrupted' is a volatile object, fetch it only once */
1951 	fatal_sig = interrupted;
1952 	if (!fatal_sig)
1953 		fatal_sig = SIGTERM;
1954 
1955 	for (i = 0; i < tcbtabsize; i++) {
1956 		tcp = tcbtab[i];
1957 		if (!tcp->pid)
1958 			continue;
1959 		debug_func_msg("looking at pid %u", tcp->pid);
1960 		if (tcp->pid == strace_child) {
1961 			kill(tcp->pid, SIGCONT);
1962 			kill(tcp->pid, fatal_sig);
1963 		}
1964 		detach(tcp);
1965 	}
1966 	if (cflag)
1967 		call_summary(shared_log);
1968 }
1969 
1970 static void
interrupt(int sig)1971 interrupt(int sig)
1972 {
1973 	interrupted = sig;
1974 }
1975 
1976 static void
print_debug_info(const int pid,int status)1977 print_debug_info(const int pid, int status)
1978 {
1979 	const unsigned int event = (unsigned int) status >> 16;
1980 	char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16];
1981 	char evbuf[sizeof(",EVENT_VFORK_DONE (%u)") + sizeof(int)*3 /*paranoia:*/ + 16];
1982 
1983 	strcpy(buf, "???");
1984 	if (WIFSIGNALED(status))
1985 #ifdef WCOREDUMP
1986 		xsprintf(buf, "WIFSIGNALED,%ssig=%s",
1987 				WCOREDUMP(status) ? "core," : "",
1988 				signame(WTERMSIG(status)));
1989 #else
1990 		xsprintf(buf, "WIFSIGNALED,sig=%s",
1991 				signame(WTERMSIG(status)));
1992 #endif
1993 	if (WIFEXITED(status))
1994 		xsprintf(buf, "WIFEXITED,exitcode=%u", WEXITSTATUS(status));
1995 	if (WIFSTOPPED(status))
1996 		xsprintf(buf, "WIFSTOPPED,sig=%s", signame(WSTOPSIG(status)));
1997 	evbuf[0] = '\0';
1998 	if (event != 0) {
1999 		static const char *const event_names[] = {
2000 			[PTRACE_EVENT_CLONE] = "CLONE",
2001 			[PTRACE_EVENT_FORK]  = "FORK",
2002 			[PTRACE_EVENT_VFORK] = "VFORK",
2003 			[PTRACE_EVENT_VFORK_DONE] = "VFORK_DONE",
2004 			[PTRACE_EVENT_EXEC]  = "EXEC",
2005 			[PTRACE_EVENT_EXIT]  = "EXIT",
2006 			/* [PTRACE_EVENT_STOP (=128)] would make biggish array */
2007 		};
2008 		const char *e = "??";
2009 		if (event < ARRAY_SIZE(event_names))
2010 			e = event_names[event];
2011 		else if (event == PTRACE_EVENT_STOP)
2012 			e = "STOP";
2013 		xsprintf(evbuf, ",EVENT_%s (%u)", e, event);
2014 	}
2015 	error_msg("[wait(0x%06x) = %u] %s%s", status, pid, buf, evbuf);
2016 }
2017 
2018 static struct tcb *
maybe_allocate_tcb(const int pid,int status)2019 maybe_allocate_tcb(const int pid, int status)
2020 {
2021 	if (!WIFSTOPPED(status)) {
2022 		if (detach_on_execve && pid == strace_child) {
2023 			/* example: strace -bexecve sh -c 'exec true' */
2024 			strace_child = 0;
2025 			return NULL;
2026 		}
2027 		/*
2028 		 * This can happen if we inherited an unknown child.
2029 		 * Example: (sleep 1 & exec strace true)
2030 		 */
2031 		error_msg("Exit of unknown pid %u ignored", pid);
2032 		return NULL;
2033 	}
2034 	if (followfork) {
2035 		/* We assume it's a fork/vfork/clone child */
2036 		struct tcb *tcp = alloctcb(pid);
2037 		tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
2038 		newoutf(tcp);
2039 		if (!qflag)
2040 			error_msg("Process %d attached", pid);
2041 		return tcp;
2042 	} else {
2043 		/*
2044 		 * This can happen if a clone call misused CLONE_PTRACE itself.
2045 		 *
2046 		 * There used to be a dance around possible re-injection of
2047 		 * WSTOPSIG(status), but it was later removed as the only
2048 		 * observable stop here is the initial ptrace-stop.
2049 		 */
2050 		ptrace(PTRACE_DETACH, pid, NULL, 0L);
2051 		error_msg("Detached unknown pid %d", pid);
2052 		return NULL;
2053 	}
2054 }
2055 
2056 static struct tcb *
maybe_switch_tcbs(struct tcb * tcp,const int pid)2057 maybe_switch_tcbs(struct tcb *tcp, const int pid)
2058 {
2059 	FILE *fp;
2060 	struct tcb *execve_thread;
2061 	long old_pid = 0;
2062 
2063 	if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, &old_pid) < 0)
2064 		return tcp;
2065 	/* Avoid truncation in pid2tcb() param passing */
2066 	if (old_pid <= 0 || old_pid == pid)
2067 		return tcp;
2068 	if ((unsigned long) old_pid > UINT_MAX)
2069 		return tcp;
2070 	execve_thread = pid2tcb(old_pid);
2071 	/* It should be !NULL, but I feel paranoid */
2072 	if (!execve_thread)
2073 		return tcp;
2074 
2075 	if (execve_thread->curcol != 0) {
2076 		/*
2077 		 * One case we are here is -ff:
2078 		 * try "strace -oLOG -ff test/threaded_execve"
2079 		 */
2080 		fprintf(execve_thread->outf, " <pid changed to %d ...>\n", pid);
2081 		/*execve_thread->curcol = 0; - no need, see code below */
2082 	}
2083 	/* Swap output FILEs (needed for -ff) */
2084 	fp = execve_thread->outf;
2085 	execve_thread->outf = tcp->outf;
2086 	tcp->outf = fp;
2087 	/* And their column positions */
2088 	execve_thread->curcol = tcp->curcol;
2089 	tcp->curcol = 0;
2090 	/* Drop leader, but close execve'd thread outfile (if -ff) */
2091 	droptcb(tcp);
2092 	/* Switch to the thread, reusing leader's outfile and pid */
2093 	tcp = execve_thread;
2094 	tcp->pid = pid;
2095 	if (cflag != CFLAG_ONLY_STATS) {
2096 		printleader(tcp);
2097 		tprintf("+++ superseded by execve in pid %lu +++\n", old_pid);
2098 		line_ended();
2099 		tcp->flags |= TCB_REPRINT;
2100 	}
2101 
2102 	return tcp;
2103 }
2104 
2105 static void
print_signalled(struct tcb * tcp,const int pid,int status)2106 print_signalled(struct tcb *tcp, const int pid, int status)
2107 {
2108 	if (pid == strace_child) {
2109 		exit_code = 0x100 | WTERMSIG(status);
2110 		strace_child = 0;
2111 	}
2112 
2113 	if (cflag != CFLAG_ONLY_STATS
2114 	    && is_number_in_set(WTERMSIG(status), signal_set)) {
2115 		printleader(tcp);
2116 #ifdef WCOREDUMP
2117 		tprintf("+++ killed by %s %s+++\n",
2118 			signame(WTERMSIG(status)),
2119 			WCOREDUMP(status) ? "(core dumped) " : "");
2120 #else
2121 		tprintf("+++ killed by %s +++\n",
2122 			signame(WTERMSIG(status)));
2123 #endif
2124 		line_ended();
2125 	}
2126 }
2127 
2128 static void
print_exited(struct tcb * tcp,const int pid,int status)2129 print_exited(struct tcb *tcp, const int pid, int status)
2130 {
2131 	if (pid == strace_child) {
2132 		exit_code = WEXITSTATUS(status);
2133 		strace_child = 0;
2134 	}
2135 
2136 	if (cflag != CFLAG_ONLY_STATS &&
2137 	    qflag < 2) {
2138 		printleader(tcp);
2139 		tprintf("+++ exited with %d +++\n", WEXITSTATUS(status));
2140 		line_ended();
2141 	}
2142 }
2143 
2144 static void
print_stopped(struct tcb * tcp,const siginfo_t * si,const unsigned int sig)2145 print_stopped(struct tcb *tcp, const siginfo_t *si, const unsigned int sig)
2146 {
2147 	if (cflag != CFLAG_ONLY_STATS
2148 	    && !hide_log(tcp)
2149 	    && is_number_in_set(sig, signal_set)) {
2150 		printleader(tcp);
2151 		if (si) {
2152 			tprintf("--- %s ", signame(sig));
2153 			printsiginfo(si);
2154 			tprints(" ---\n");
2155 		} else
2156 			tprintf("--- stopped by %s ---\n", signame(sig));
2157 		line_ended();
2158 	}
2159 }
2160 
2161 static void
startup_tcb(struct tcb * tcp)2162 startup_tcb(struct tcb *tcp)
2163 {
2164 	debug_msg("pid %d has TCB_STARTUP, initializing it", tcp->pid);
2165 
2166 	tcp->flags &= ~TCB_STARTUP;
2167 
2168 	if (!use_seize) {
2169 		debug_msg("setting opts 0x%x on pid %d",
2170 			  ptrace_setoptions, tcp->pid);
2171 		if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) {
2172 			if (errno != ESRCH) {
2173 				/* Should never happen, really */
2174 				perror_msg_and_die("PTRACE_SETOPTIONS");
2175 			}
2176 		}
2177 	}
2178 
2179 	if ((tcp->flags & TCB_GRABBED) && (get_scno(tcp) == 1))
2180 		tcp->s_prev_ent = tcp->s_ent;
2181 }
2182 
2183 static void
print_event_exit(struct tcb * tcp)2184 print_event_exit(struct tcb *tcp)
2185 {
2186 	if (entering(tcp) || filtered(tcp) || hide_log(tcp)
2187 	    || cflag == CFLAG_ONLY_STATS) {
2188 		return;
2189 	}
2190 
2191 	if (followfork < 2 && printing_tcp && printing_tcp != tcp
2192 	    && printing_tcp->curcol != 0) {
2193 		set_current_tcp(printing_tcp);
2194 		tprints(" <unfinished ...>\n");
2195 		flush_tcp_output(printing_tcp);
2196 		printing_tcp->curcol = 0;
2197 		set_current_tcp(tcp);
2198 	}
2199 
2200 	if ((followfork < 2 && printing_tcp != tcp)
2201 	    || (tcp->flags & TCB_REPRINT)) {
2202 		tcp->flags &= ~TCB_REPRINT;
2203 		printleader(tcp);
2204 		tprintf("<... %s resumed>", tcp->s_ent->sys_name);
2205 	}
2206 
2207 	if (!(tcp->sys_func_rval & RVAL_DECODED)) {
2208 		/*
2209 		 * The decoder has probably decided to print something
2210 		 * on exiting syscall which is not going to happen.
2211 		 */
2212 		tprints(" <unfinished ...>");
2213 	}
2214 
2215 	printing_tcp = tcp;
2216 	tprints(") ");
2217 	tabto();
2218 	tprints("= ?\n");
2219 	line_ended();
2220 }
2221 
2222 static enum trace_event
next_event(int * pstatus,siginfo_t * si)2223 next_event(int *pstatus, siginfo_t *si)
2224 {
2225 	int pid;
2226 	int wait_errno;
2227 	int status;
2228 	struct tcb *tcp;
2229 	struct rusage ru;
2230 
2231 	if (interrupted)
2232 		return TE_BREAK;
2233 
2234 	/*
2235 	 * Used to exit simply when nprocs hits zero, but in this testcase:
2236 	 *  int main(void) { _exit(!!fork()); }
2237 	 * under strace -f, parent sometimes (rarely) manages
2238 	 * to exit before we see the first stop of the child,
2239 	 * and we are losing track of it:
2240 	 *  19923 clone(...) = 19924
2241 	 *  19923 exit_group(1)     = ?
2242 	 *  19923 +++ exited with 1 +++
2243 	 * Exiting only when wait() returns ECHILD works better.
2244 	 */
2245 	if (popen_pid != 0) {
2246 		/* However, if -o|logger is in use, we can't do that.
2247 		 * Can work around that by double-forking the logger,
2248 		 * but that loses the ability to wait for its completion
2249 		 * on exit. Oh well...
2250 		 */
2251 		if (nprocs == 0)
2252 			return TE_BREAK;
2253 	}
2254 
2255 	if (interactive)
2256 		sigprocmask(SIG_SETMASK, &start_set, NULL);
2257 	pid = wait4(-1, pstatus, __WALL, (cflag ? &ru : NULL));
2258 	wait_errno = errno;
2259 	if (interactive)
2260 		sigprocmask(SIG_SETMASK, &blocked_set, NULL);
2261 
2262 	if (pid < 0) {
2263 		if (wait_errno == EINTR)
2264 			return TE_NEXT;
2265 		if (nprocs == 0 && wait_errno == ECHILD)
2266 			return TE_BREAK;
2267 		/*
2268 		 * If nprocs > 0, ECHILD is not expected,
2269 		 * treat it as any other error here:
2270 		 */
2271 		errno = wait_errno;
2272 		perror_msg_and_die("wait4(__WALL)");
2273 	}
2274 
2275 	status = *pstatus;
2276 
2277 	if (pid == popen_pid) {
2278 		if (!WIFSTOPPED(status))
2279 			popen_pid = 0;
2280 		return TE_NEXT;
2281 	}
2282 
2283 	if (debug_flag)
2284 		print_debug_info(pid, status);
2285 
2286 	/* Look up 'pid' in our table. */
2287 	tcp = pid2tcb(pid);
2288 
2289 	if (!tcp) {
2290 		tcp = maybe_allocate_tcb(pid, status);
2291 		if (!tcp)
2292 			return TE_NEXT;
2293 	}
2294 
2295 	clear_regs(tcp);
2296 
2297 	/* Set current output file */
2298 	set_current_tcp(tcp);
2299 
2300 	if (cflag) {
2301 		tv_sub(&tcp->dtime, &ru.ru_stime, &tcp->stime);
2302 		tcp->stime = ru.ru_stime;
2303 	}
2304 
2305 	if (WIFSIGNALED(status))
2306 		return TE_SIGNALLED;
2307 
2308 	if (WIFEXITED(status))
2309 		return TE_EXITED;
2310 
2311 	/*
2312 	 * As WCONTINUED flag has not been specified to wait4,
2313 	 * it cannot be WIFCONTINUED(status), so the only case
2314 	 * that remains is WIFSTOPPED(status).
2315 	 */
2316 
2317 	/* Is this the very first time we see this tracee stopped? */
2318 	if (tcp->flags & TCB_STARTUP)
2319 		startup_tcb(tcp);
2320 
2321 	const unsigned int sig = WSTOPSIG(status);
2322 	const unsigned int event = (unsigned int) status >> 16;
2323 
2324 	switch (event) {
2325 	case 0:
2326 		/*
2327 		 * Is this post-attach SIGSTOP?
2328 		 * Interestingly, the process may stop
2329 		 * with STOPSIG equal to some other signal
2330 		 * than SIGSTOP if we happened to attach
2331 		 * just before the process takes a signal.
2332 		 */
2333 		if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) {
2334 			debug_func_msg("ignored SIGSTOP on pid %d", tcp->pid);
2335 			tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP;
2336 			return TE_RESTART;
2337 		} else if (sig == syscall_trap_sig) {
2338 			return TE_SYSCALL_STOP;
2339 		} else {
2340 			*si = (siginfo_t) {};
2341 			/*
2342 			 * True if tracee is stopped by signal
2343 			 * (as opposed to "tracee received signal").
2344 			 * TODO: shouldn't we check for errno == EINVAL too?
2345 			 * We can get ESRCH instead, you know...
2346 			 */
2347 			bool stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, si) < 0;
2348 			return stopped ? TE_GROUP_STOP : TE_SIGNAL_DELIVERY_STOP;
2349 		}
2350 		break;
2351 #if USE_SEIZE
2352 	case PTRACE_EVENT_STOP:
2353 		/*
2354 		 * PTRACE_INTERRUPT-stop or group-stop.
2355 		 * PTRACE_INTERRUPT-stop has sig == SIGTRAP here.
2356 		 */
2357 		switch (sig) {
2358 		case SIGSTOP:
2359 		case SIGTSTP:
2360 		case SIGTTIN:
2361 		case SIGTTOU:
2362 			return TE_GROUP_STOP;
2363 		}
2364 		return TE_RESTART;
2365 #endif
2366 	case PTRACE_EVENT_EXEC:
2367 		return TE_STOP_BEFORE_EXECVE;
2368 	case PTRACE_EVENT_EXIT:
2369 		return TE_STOP_BEFORE_EXIT;
2370 	default:
2371 		return TE_RESTART;
2372 	}
2373 }
2374 
2375 static int
trace_syscall(struct tcb * tcp,unsigned int * sig)2376 trace_syscall(struct tcb *tcp, unsigned int *sig)
2377 {
2378 	if (entering(tcp)) {
2379 		int res = syscall_entering_decode(tcp);
2380 		switch (res) {
2381 		case 0:
2382 			return 0;
2383 		case 1:
2384 			res = syscall_entering_trace(tcp, sig);
2385 		}
2386 		syscall_entering_finish(tcp, res);
2387 		return res;
2388 	} else {
2389 		struct timeval tv = {};
2390 		int res = syscall_exiting_decode(tcp, &tv);
2391 		if (res != 0) {
2392 			res = syscall_exiting_trace(tcp, tv, res);
2393 		}
2394 		syscall_exiting_finish(tcp);
2395 		return res;
2396 	}
2397 }
2398 
2399 /* Returns true iff the main trace loop has to continue. */
2400 static bool
dispatch_event(enum trace_event ret,int * pstatus,siginfo_t * si)2401 dispatch_event(enum trace_event ret, int *pstatus, siginfo_t *si)
2402 {
2403 	unsigned int restart_op = PTRACE_SYSCALL;
2404 	unsigned int restart_sig = 0;
2405 
2406 	switch (ret) {
2407 	case TE_BREAK:
2408 		return false;
2409 
2410 	case TE_NEXT:
2411 		return true;
2412 
2413 	case TE_RESTART:
2414 		break;
2415 
2416 	case TE_SYSCALL_STOP:
2417 		if (trace_syscall(current_tcp, &restart_sig) < 0) {
2418 			/*
2419 			 * ptrace() failed in trace_syscall().
2420 			 * Likely a result of process disappearing mid-flight.
2421 			 * Observed case: exit_group() or SIGKILL terminating
2422 			 * all processes in thread group.
2423 			 * We assume that ptrace error was caused by process death.
2424 			 * We used to detach(current_tcp) here, but since we no
2425 			 * longer implement "detach before death" policy/hack,
2426 			 * we can let this process to report its death to us
2427 			 * normally, via WIFEXITED or WIFSIGNALED wait status.
2428 			 */
2429 			return true;
2430 		}
2431 		break;
2432 
2433 	case TE_SIGNAL_DELIVERY_STOP:
2434 		restart_sig = WSTOPSIG(*pstatus);
2435 		print_stopped(current_tcp, si, restart_sig);
2436 		break;
2437 
2438 	case TE_SIGNALLED:
2439 		print_signalled(current_tcp, current_tcp->pid, *pstatus);
2440 		droptcb(current_tcp);
2441 		return true;
2442 
2443 	case TE_GROUP_STOP:
2444 		restart_sig = WSTOPSIG(*pstatus);
2445 		print_stopped(current_tcp, NULL, restart_sig);
2446 		if (use_seize) {
2447 			/*
2448 			 * This ends ptrace-stop, but does *not* end group-stop.
2449 			 * This makes stopping signals work properly on straced
2450 			 * process (that is, process really stops. It used to
2451 			 * continue to run).
2452 			 */
2453 			restart_op = PTRACE_LISTEN;
2454 			restart_sig = 0;
2455 		}
2456 		break;
2457 
2458 	case TE_EXITED:
2459 		print_exited(current_tcp, current_tcp->pid, *pstatus);
2460 		droptcb(current_tcp);
2461 		return true;
2462 
2463 	case TE_STOP_BEFORE_EXECVE:
2464 		/*
2465 		 * Check that we are inside syscall now (next event after
2466 		 * PTRACE_EVENT_EXEC should be for syscall exiting).  If it is
2467 		 * not the case, we might have a situation when we attach to a
2468 		 * process and the first thing we see is a PTRACE_EVENT_EXEC
2469 		 * and all the following syscall state tracking is screwed up
2470 		 * otherwise.
2471 		 */
2472 		if (entering(current_tcp)) {
2473 			int ret;
2474 
2475 			error_msg("Stray PTRACE_EVENT_EXEC from pid %d"
2476 				  ", trying to recover...",
2477 				  current_tcp->pid);
2478 
2479 			current_tcp->flags |= TCB_RECOVERING;
2480 			ret = trace_syscall(current_tcp, &restart_sig);
2481 			current_tcp->flags &= ~TCB_RECOVERING;
2482 
2483 			if (ret < 0) {
2484 				/* The reason is described in TE_SYSCALL_STOP */
2485 				return true;
2486 			}
2487 		}
2488 
2489 		/*
2490 		 * Under Linux, execve changes pid to thread leader's pid,
2491 		 * and we see this changed pid on EVENT_EXEC and later,
2492 		 * execve sysexit. Leader "disappears" without exit
2493 		 * notification. Let user know that, drop leader's tcb,
2494 		 * and fix up pid in execve thread's tcb.
2495 		 * Effectively, execve thread's tcb replaces leader's tcb.
2496 		 *
2497 		 * BTW, leader is 'stuck undead' (doesn't report WIFEXITED
2498 		 * on exit syscall) in multithreaded programs exactly
2499 		 * in order to handle this case.
2500 		 *
2501 		 * PTRACE_GETEVENTMSG returns old pid starting from Linux 3.0.
2502 		 * On 2.6 and earlier, it can return garbage.
2503 		 */
2504 		if (os_release >= KERNEL_VERSION(3, 0, 0))
2505 			set_current_tcp(maybe_switch_tcbs(current_tcp,
2506 							  current_tcp->pid));
2507 
2508 		if (detach_on_execve) {
2509 			if (current_tcp->flags & TCB_SKIP_DETACH_ON_FIRST_EXEC) {
2510 				current_tcp->flags &= ~TCB_SKIP_DETACH_ON_FIRST_EXEC;
2511 			} else {
2512 				detach(current_tcp); /* do "-b execve" thingy */
2513 				return true;
2514 			}
2515 		}
2516 		break;
2517 
2518 	case TE_STOP_BEFORE_EXIT:
2519 		print_event_exit(current_tcp);
2520 		break;
2521 	}
2522 
2523 	/* We handled quick cases, we are permitted to interrupt now. */
2524 	if (interrupted)
2525 		return false;
2526 
2527 	if (ptrace_restart(restart_op, current_tcp, restart_sig) < 0) {
2528 		/* Note: ptrace_restart emitted error message */
2529 		exit_code = 1;
2530 		return false;
2531 	}
2532 	return true;
2533 }
2534 
2535 #ifdef ENABLE_COVERAGE_GCOV
2536 extern void __gcov_flush(void);
2537 #endif
2538 
2539 static void ATTRIBUTE_NORETURN
terminate(void)2540 terminate(void)
2541 {
2542 	cleanup();
2543 	fflush(NULL);
2544 	if (shared_log != stderr)
2545 		fclose(shared_log);
2546 	if (popen_pid) {
2547 		while (waitpid(popen_pid, NULL, 0) < 0 && errno == EINTR)
2548 			;
2549 	}
2550 	if (exit_code > 0xff) {
2551 		/* Avoid potential core file clobbering.  */
2552 		struct_rlimit rlim = {0, 0};
2553 		set_rlimit(RLIMIT_CORE, &rlim);
2554 
2555 		/* Child was killed by a signal, mimic that.  */
2556 		exit_code &= 0xff;
2557 		signal(exit_code, SIG_DFL);
2558 #ifdef ENABLE_COVERAGE_GCOV
2559 		__gcov_flush();
2560 #endif
2561 		raise(exit_code);
2562 
2563 		/* Unblock the signal.  */
2564 		sigset_t mask;
2565 		sigemptyset(&mask);
2566 		sigaddset(&mask, exit_code);
2567 #ifdef ENABLE_COVERAGE_GCOV
2568 		__gcov_flush();
2569 #endif
2570 		sigprocmask(SIG_UNBLOCK, &mask, NULL);
2571 
2572 		/* Paranoia - what if this signal is not fatal?
2573 		   Exit with 128 + signo then.  */
2574 		exit_code += 128;
2575 	}
2576 	exit(exit_code);
2577 }
2578 
2579 int
main(int argc,char * argv[])2580 main(int argc, char *argv[])
2581 {
2582 	init(argc, argv);
2583 
2584 	exit_code = !nprocs;
2585 
2586 	int status;
2587 	siginfo_t si;
2588 	while (dispatch_event(next_event(&status, &si), &status, &si))
2589 		;
2590 	terminate();
2591 }
2592