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  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include "defs.h"
32 #include <stdarg.h>
33 #include <sys/param.h>
34 #include <fcntl.h>
35 #include <signal.h>
36 #include <sys/resource.h>
37 #include <sys/wait.h>
38 #include <sys/stat.h>
39 #include <pwd.h>
40 #include <grp.h>
41 #include <dirent.h>
42 #include <sys/utsname.h>
43 #ifdef HAVE_PRCTL
44 # include <sys/prctl.h>
45 #endif
46 #include <asm/unistd.h>
47 
48 #include "scno.h"
49 #include "ptrace.h"
50 #include "printsiginfo.h"
51 
52 /* In some libc, these aren't declared. Do it ourself: */
53 extern char **environ;
54 extern int optind;
55 extern char *optarg;
56 
57 #ifdef USE_LIBUNWIND
58 /* if this is true do the stack trace for every system call */
59 bool stack_trace_enabled = false;
60 #endif
61 
62 #define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
63 
64 /* Glue for systems without a MMU that cannot provide fork() */
65 #if !defined(HAVE_FORK)
66 # undef NOMMU_SYSTEM
67 # define NOMMU_SYSTEM 1
68 #endif
69 #if NOMMU_SYSTEM
70 # define fork() vfork()
71 #endif
72 
73 const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
74 
75 cflag_t cflag = CFLAG_NONE;
76 unsigned int followfork = 0;
77 unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC
78 				 | PTRACE_O_TRACEEXIT;
79 unsigned int xflag = 0;
80 bool debug_flag = 0;
81 bool Tflag = 0;
82 bool iflag = 0;
83 bool count_wallclock = 0;
84 unsigned int qflag = 0;
85 static unsigned int tflag = 0;
86 static bool rflag = 0;
87 static bool print_pid_pfx = 0;
88 
89 /* -I n */
90 enum {
91     INTR_NOT_SET        = 0,
92     INTR_ANYWHERE       = 1, /* don't block/ignore any signals */
93     INTR_WHILE_WAIT     = 2, /* block fatal signals while decoding syscall. default */
94     INTR_NEVER          = 3, /* block fatal signals. default if '-o FILE PROG' */
95     INTR_BLOCK_TSTP_TOO = 4, /* block fatal signals and SIGTSTP (^Z) */
96     NUM_INTR_OPTS
97 };
98 static int opt_intr;
99 /* We play with signal mask only if this mode is active: */
100 #define interactive (opt_intr == INTR_WHILE_WAIT)
101 
102 /*
103  * daemonized_tracer supports -D option.
104  * With this option, strace forks twice.
105  * Unlike normal case, with -D *grandparent* process exec's,
106  * becoming a traced process. Child exits (this prevents traced process
107  * from having children it doesn't expect to have), and grandchild
108  * attaches to grandparent similarly to strace -p PID.
109  * This allows for more transparent interaction in cases
110  * when process and its parent are communicating via signals,
111  * wait() etc. Without -D, strace process gets lodged in between,
112  * disrupting parent<->child link.
113  */
114 static bool daemonized_tracer = 0;
115 
116 #if USE_SEIZE
117 static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP;
118 # define use_seize (post_attach_sigstop == 0)
119 #else
120 # define post_attach_sigstop TCB_IGNORE_ONE_SIGSTOP
121 # define use_seize 0
122 #endif
123 
124 /* Sometimes we want to print only succeeding syscalls. */
125 bool not_failing_only = 0;
126 
127 /* Show path associated with fd arguments */
128 unsigned int show_fd_path = 0;
129 
130 static bool detach_on_execve = 0;
131 
132 static int exit_code;
133 static int strace_child = 0;
134 static int strace_tracer_pid = 0;
135 
136 static char *username = NULL;
137 static uid_t run_uid;
138 static gid_t run_gid;
139 
140 unsigned int max_strlen = DEFAULT_STRLEN;
141 static int acolumn = DEFAULT_ACOLUMN;
142 static char *acolumn_spaces;
143 
144 static char *outfname = NULL;
145 /* If -ff, points to stderr. Else, it's our common output log */
146 static FILE *shared_log;
147 
148 struct tcb *printing_tcp = NULL;
149 static struct tcb *current_tcp;
150 
151 static struct tcb **tcbtab;
152 static unsigned int nprocs, tcbtabsize;
153 static const char *progname;
154 
155 unsigned os_release; /* generated from uname()'s u.release */
156 
157 static void detach(struct tcb *tcp);
158 static void cleanup(void);
159 static void interrupt(int sig);
160 static sigset_t empty_set, blocked_set;
161 
162 #ifdef HAVE_SIG_ATOMIC_T
163 static volatile sig_atomic_t interrupted;
164 #else
165 static volatile int interrupted;
166 #endif
167 
168 #ifndef HAVE_STRERROR
169 
170 #if !HAVE_DECL_SYS_ERRLIST
171 extern int sys_nerr;
172 extern char *sys_errlist[];
173 #endif
174 
175 const char *
strerror(int err_no)176 strerror(int err_no)
177 {
178 	static char buf[sizeof("Unknown error %d") + sizeof(int)*3];
179 
180 	if (err_no < 1 || err_no >= sys_nerr) {
181 		sprintf(buf, "Unknown error %d", err_no);
182 		return buf;
183 	}
184 	return sys_errlist[err_no];
185 }
186 
187 #endif /* HAVE_STERRROR */
188 
189 static void
print_version(void)190 print_version(void)
191 {
192 	printf("%s -- version %s\n"
193 	       "Copyright (C) %s The strace developers <%s>.\n"
194 	       "This is free software; see the source for copying conditions.  There is NO\n"
195 	       "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
196 	       PACKAGE_NAME, PACKAGE_VERSION, "1991-2017", PACKAGE_URL);
197 }
198 
199 static void
usage(void)200 usage(void)
201 {
202 	printf("\
203 usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
204               [-a column] [-o file] [-s strsize] [-P path]...\n\
205               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
206    or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
207               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
208 \n\
209 Output format:\n\
210   -a column      alignment COLUMN for printing syscall results (default %d)\n\
211   -i             print instruction pointer at time of syscall\n\
212 "
213 #ifdef USE_LIBUNWIND
214 "\
215   -k             obtain stack trace between each syscall (experimental)\n\
216 "
217 #endif
218 "\
219   -o file        send trace output to FILE instead of stderr\n\
220   -q             suppress messages about attaching, detaching, etc.\n\
221   -r             print relative timestamp\n\
222   -s strsize     limit length of print strings to STRSIZE chars (default %d)\n\
223   -t             print absolute timestamp\n\
224   -tt            print absolute timestamp with usecs\n\
225   -T             print time spent in each syscall\n\
226   -x             print non-ascii strings in hex\n\
227   -xx            print all strings in hex\n\
228   -y             print paths associated with file descriptor arguments\n\
229   -yy            print protocol specific information associated with socket file descriptors\n\
230 \n\
231 Statistics:\n\
232   -c             count time, calls, and errors for each syscall and report summary\n\
233   -C             like -c but also print regular output\n\
234   -O overhead    set overhead for tracing syscalls to OVERHEAD usecs\n\
235   -S sortby      sort syscall counts by: time, calls, name, nothing (default %s)\n\
236   -w             summarise syscall latency (default is system time)\n\
237 \n\
238 Filtering:\n\
239   -e expr        a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
240      options:    trace, abbrev, verbose, raw, signal, read, write, fault\n\
241   -P path        trace accesses to path\n\
242 \n\
243 Tracing:\n\
244   -b execve      detach on execve syscall\n\
245   -D             run tracer process as a detached grandchild, not as parent\n\
246   -f             follow forks\n\
247   -ff            follow forks with output into separate files\n\
248   -I interruptible\n\
249      1:          no signals are blocked\n\
250      2:          fatal signals are blocked while decoding syscall (default)\n\
251      3:          fatal signals are always blocked (default if '-o FILE PROG')\n\
252      4:          fatal signals and SIGTSTP (^Z) are always blocked\n\
253                  (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
254 \n\
255 Startup:\n\
256   -E var         remove var from the environment for command\n\
257   -E var=val     put var=val in the environment for command\n\
258   -p pid         trace process with process id PID, may be repeated\n\
259   -u username    run command as username handling setuid and/or setgid\n\
260 \n\
261 Miscellaneous:\n\
262   -d             enable debug output to stderr\n\
263   -v             verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
264   -h             print help message\n\
265   -V             print version\n\
266 "
267 /* ancient, no one should use it
268 -F -- attempt to follow vforks (deprecated, use -f)\n\
269  */
270 /* this is broken, so don't document it
271 -z -- print only succeeding syscalls\n\
272  */
273 , DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
274 	exit(0);
275 }
276 
277 static void ATTRIBUTE_NORETURN
die(void)278 die(void)
279 {
280 	if (strace_tracer_pid == getpid()) {
281 		cflag = 0;
282 		cleanup();
283 	}
284 	exit(1);
285 }
286 
verror_msg(int err_no,const char * fmt,va_list p)287 static void verror_msg(int err_no, const char *fmt, va_list p)
288 {
289 	char *msg;
290 
291 	fflush(NULL);
292 
293 	/* We want to print entire message with single fprintf to ensure
294 	 * message integrity if stderr is shared with other programs.
295 	 * Thus we use vasprintf + single fprintf.
296 	 */
297 	msg = NULL;
298 	if (vasprintf(&msg, fmt, p) >= 0) {
299 		if (err_no)
300 			fprintf(stderr, "%s: %s: %s\n", progname, msg, strerror(err_no));
301 		else
302 			fprintf(stderr, "%s: %s\n", progname, msg);
303 		free(msg);
304 	} else {
305 		/* malloc in vasprintf failed, try it without malloc */
306 		fprintf(stderr, "%s: ", progname);
307 		vfprintf(stderr, fmt, p);
308 		if (err_no)
309 			fprintf(stderr, ": %s\n", strerror(err_no));
310 		else
311 			putc('\n', stderr);
312 	}
313 	/* We don't switch stderr to buffered, thus fprintf(stderr)
314 	 * always flushes its output and this is not necessary: */
315 	/* fflush(stderr); */
316 }
317 
error_msg(const char * fmt,...)318 void error_msg(const char *fmt, ...)
319 {
320 	va_list p;
321 	va_start(p, fmt);
322 	verror_msg(0, fmt, p);
323 	va_end(p);
324 }
325 
error_msg_and_die(const char * fmt,...)326 void error_msg_and_die(const char *fmt, ...)
327 {
328 	va_list p;
329 	va_start(p, fmt);
330 	verror_msg(0, fmt, p);
331 	die();
332 }
333 
error_msg_and_help(const char * fmt,...)334 void error_msg_and_help(const char *fmt, ...)
335 {
336 	if (fmt != NULL) {
337 		va_list p;
338 		va_start(p, fmt);
339 		verror_msg(0, fmt, p);
340 	}
341 	fprintf(stderr, "Try '%s -h' for more information.\n", progname);
342 	die();
343 }
344 
perror_msg(const char * fmt,...)345 void perror_msg(const char *fmt, ...)
346 {
347 	va_list p;
348 	va_start(p, fmt);
349 	verror_msg(errno, fmt, p);
350 	va_end(p);
351 }
352 
perror_msg_and_die(const char * fmt,...)353 void perror_msg_and_die(const char *fmt, ...)
354 {
355 	va_list p;
356 	va_start(p, fmt);
357 	verror_msg(errno, fmt, p);
358 	die();
359 }
360 
361 static void
error_opt_arg(int opt,const char * arg)362 error_opt_arg(int opt, const char *arg)
363 {
364 	error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
365 }
366 
367 static const char *ptrace_attach_cmd;
368 
369 static int
ptrace_attach_or_seize(int pid)370 ptrace_attach_or_seize(int pid)
371 {
372 #if USE_SEIZE
373 	int r;
374 	if (!use_seize)
375 		return ptrace_attach_cmd = "PTRACE_ATTACH",
376 		       ptrace(PTRACE_ATTACH, pid, 0L, 0L);
377 	r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
378 	if (r)
379 		return ptrace_attach_cmd = "PTRACE_SEIZE", r;
380 	r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
381 	return ptrace_attach_cmd = "PTRACE_INTERRUPT", r;
382 #else
383 		return ptrace_attach_cmd = "PTRACE_ATTACH",
384 		       ptrace(PTRACE_ATTACH, pid, 0L, 0L);
385 #endif
386 }
387 
388 /*
389  * Used when we want to unblock stopped traced process.
390  * Should be only used with PTRACE_CONT, PTRACE_DETACH and PTRACE_SYSCALL.
391  * Returns 0 on success or if error was ESRCH
392  * (presumably process was killed while we talk to it).
393  * Otherwise prints error message and returns -1.
394  */
395 static int
ptrace_restart(const unsigned int op,struct tcb * const tcp,unsigned int sig)396 ptrace_restart(const unsigned int op, struct tcb *const tcp, unsigned int sig)
397 {
398 	int err;
399 	const char *msg;
400 
401 	errno = 0;
402 	ptrace(op, tcp->pid, 0L, (unsigned long) sig);
403 	err = errno;
404 	if (!err)
405 		return 0;
406 
407 	switch (op) {
408 		case PTRACE_CONT:
409 			msg = "CONT";
410 			break;
411 		case PTRACE_DETACH:
412 			msg = "DETACH";
413 			break;
414 		case PTRACE_LISTEN:
415 			msg = "LISTEN";
416 			break;
417 		default:
418 			msg = "SYSCALL";
419 	}
420 
421 	/*
422 	 * Why curcol != 0? Otherwise sometimes we get this:
423 	 *
424 	 * 10252 kill(10253, SIGKILL)              = 0
425 	 *  <ptrace(SYSCALL,10252):No such process>10253 ...next decode...
426 	 *
427 	 * 10252 died after we retrieved syscall exit data,
428 	 * but before we tried to restart it. Log looks ugly.
429 	 */
430 	if (current_tcp && current_tcp->curcol != 0) {
431 		tprintf(" <ptrace(%s):%s>\n", msg, strerror(err));
432 		line_ended();
433 	}
434 	if (err == ESRCH)
435 		return 0;
436 	errno = err;
437 	perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%u)", msg, tcp->pid, sig);
438 	return -1;
439 }
440 
441 static void
set_cloexec_flag(int fd)442 set_cloexec_flag(int fd)
443 {
444 	int flags, newflags;
445 
446 	flags = fcntl(fd, F_GETFD);
447 	if (flags < 0) {
448 		/* Can happen only if fd is bad.
449 		 * Should never happen: if it does, we have a bug
450 		 * in the caller. Therefore we just abort
451 		 * instead of propagating the error.
452 		 */
453 		perror_msg_and_die("fcntl(%d, F_GETFD)", fd);
454 	}
455 
456 	newflags = flags | FD_CLOEXEC;
457 	if (flags == newflags)
458 		return;
459 
460 	fcntl(fd, F_SETFD, newflags); /* never fails */
461 }
462 
463 static void
kill_save_errno(pid_t pid,int sig)464 kill_save_errno(pid_t pid, int sig)
465 {
466 	int saved_errno = errno;
467 
468 	(void) kill(pid, sig);
469 	errno = saved_errno;
470 }
471 
472 /*
473  * When strace is setuid executable, we have to swap uids
474  * before and after filesystem and process management operations.
475  */
476 static void
swap_uid(void)477 swap_uid(void)
478 {
479 	int euid = geteuid(), uid = getuid();
480 
481 	if (euid != uid && setreuid(euid, uid) < 0) {
482 		perror_msg_and_die("setreuid");
483 	}
484 }
485 
486 #ifdef _LARGEFILE64_SOURCE
487 # ifdef HAVE_FOPEN64
488 #  define fopen_for_output fopen64
489 # else
490 #  define fopen_for_output fopen
491 # endif
492 # define struct_stat struct stat64
493 # define stat_file stat64
494 # define struct_dirent struct dirent64
495 # define read_dir readdir64
496 # define struct_rlimit struct rlimit64
497 # define set_rlimit setrlimit64
498 #else
499 # define fopen_for_output fopen
500 # define struct_stat struct stat
501 # define stat_file stat
502 # define struct_dirent struct dirent
503 # define read_dir readdir
504 # define struct_rlimit struct rlimit
505 # define set_rlimit setrlimit
506 #endif
507 
508 static FILE *
strace_fopen(const char * path)509 strace_fopen(const char *path)
510 {
511 	FILE *fp;
512 
513 	swap_uid();
514 	fp = fopen_for_output(path, "w");
515 	if (!fp)
516 		perror_msg_and_die("Can't fopen '%s'", path);
517 	swap_uid();
518 	set_cloexec_flag(fileno(fp));
519 	return fp;
520 }
521 
522 static int popen_pid = 0;
523 
524 #ifndef _PATH_BSHELL
525 # define _PATH_BSHELL "/bin/sh"
526 #endif
527 
528 /*
529  * We cannot use standard popen(3) here because we have to distinguish
530  * popen child process from other processes we trace, and standard popen(3)
531  * does not export its child's pid.
532  */
533 static FILE *
strace_popen(const char * command)534 strace_popen(const char *command)
535 {
536 	FILE *fp;
537 	int pid;
538 	int fds[2];
539 
540 	swap_uid();
541 	if (pipe(fds) < 0)
542 		perror_msg_and_die("pipe");
543 
544 	set_cloexec_flag(fds[1]); /* never fails */
545 
546 	pid = vfork();
547 	if (pid < 0)
548 		perror_msg_and_die("vfork");
549 
550 	if (pid == 0) {
551 		/* child */
552 		close(fds[1]);
553 		if (fds[0] != 0) {
554 			if (dup2(fds[0], 0))
555 				perror_msg_and_die("dup2");
556 			close(fds[0]);
557 		}
558 		execl(_PATH_BSHELL, "sh", "-c", command, NULL);
559 		perror_msg_and_die("Can't execute '%s'", _PATH_BSHELL);
560 	}
561 
562 	/* parent */
563 	popen_pid = pid;
564 	close(fds[0]);
565 	swap_uid();
566 	fp = fdopen(fds[1], "w");
567 	if (!fp)
568 		die_out_of_memory();
569 	return fp;
570 }
571 
572 void
tprintf(const char * fmt,...)573 tprintf(const char *fmt, ...)
574 {
575 	va_list args;
576 
577 	va_start(args, fmt);
578 	if (current_tcp) {
579 		int n = vfprintf(current_tcp->outf, fmt, args);
580 		if (n < 0) {
581 			if (current_tcp->outf != stderr)
582 				perror_msg("%s", outfname);
583 		} else
584 			current_tcp->curcol += n;
585 	}
586 	va_end(args);
587 }
588 
589 #ifndef HAVE_FPUTS_UNLOCKED
590 # define fputs_unlocked fputs
591 #endif
592 
593 void
tprints(const char * str)594 tprints(const char *str)
595 {
596 	if (current_tcp) {
597 		int n = fputs_unlocked(str, current_tcp->outf);
598 		if (n >= 0) {
599 			current_tcp->curcol += strlen(str);
600 			return;
601 		}
602 		if (current_tcp->outf != stderr)
603 			perror_msg("%s", outfname);
604 	}
605 }
606 
607 void
line_ended(void)608 line_ended(void)
609 {
610 	if (current_tcp) {
611 		current_tcp->curcol = 0;
612 		fflush(current_tcp->outf);
613 	}
614 	if (printing_tcp) {
615 		printing_tcp->curcol = 0;
616 		printing_tcp = NULL;
617 	}
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 		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 	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 		}
666 		else if (tflag > 2) {
667 			tprintf("%ld.%06ld ",
668 				(long) tv.tv_sec, (long) tv.tv_usec);
669 		}
670 		else {
671 			time_t local = tv.tv_sec;
672 			strftime(str, sizeof(str), "%T", localtime(&local));
673 			if (tflag > 1)
674 				tprintf("%s.%06ld ", str, (long) tv.tv_usec);
675 			else
676 				tprintf("%s ", str);
677 		}
678 	}
679 	if (iflag)
680 		print_pc(tcp);
681 }
682 
683 void
tabto(void)684 tabto(void)
685 {
686 	if (current_tcp->curcol < acolumn)
687 		tprints(acolumn_spaces + current_tcp->curcol);
688 }
689 
690 /* Should be only called directly *after successful attach* to a tracee.
691  * Otherwise, "strace -oFILE -ff -p<nonexistant_pid>"
692  * may create bogus empty FILE.<nonexistant_pid>, and then die.
693  */
694 static void
newoutf(struct tcb * tcp)695 newoutf(struct tcb *tcp)
696 {
697 	tcp->outf = shared_log; /* if not -ff mode, the same file is for all */
698 	if (followfork >= 2) {
699 		char name[520 + sizeof(int) * 3];
700 		sprintf(name, "%.512s.%u", outfname, tcp->pid);
701 		tcp->outf = strace_fopen(name);
702 	}
703 }
704 
705 static void
expand_tcbtab(void)706 expand_tcbtab(void)
707 {
708 	/* Allocate some (more) TCBs (and expand the table).
709 	   We don't want to relocate the TCBs because our
710 	   callers have pointers and it would be a pain.
711 	   So tcbtab is a table of pointers.  Since we never
712 	   free the TCBs, we allocate a single chunk of many.  */
713 	unsigned int new_tcbtabsize, alloc_tcbtabsize;
714 	struct tcb *newtcbs;
715 
716 	if (tcbtabsize) {
717 		alloc_tcbtabsize = tcbtabsize;
718 		new_tcbtabsize = tcbtabsize * 2;
719 	} else {
720 		new_tcbtabsize = alloc_tcbtabsize = 1;
721 	}
722 
723 	newtcbs = xcalloc(alloc_tcbtabsize, sizeof(newtcbs[0]));
724 	tcbtab = xreallocarray(tcbtab, new_tcbtabsize, sizeof(tcbtab[0]));
725 	while (tcbtabsize < new_tcbtabsize)
726 		tcbtab[tcbtabsize++] = newtcbs++;
727 }
728 
729 static struct tcb *
alloctcb(int pid)730 alloctcb(int pid)
731 {
732 	unsigned int i;
733 	struct tcb *tcp;
734 
735 	if (nprocs == tcbtabsize)
736 		expand_tcbtab();
737 
738 	for (i = 0; i < tcbtabsize; i++) {
739 		tcp = tcbtab[i];
740 		if (!tcp->pid) {
741 			memset(tcp, 0, sizeof(*tcp));
742 			tcp->pid = pid;
743 #if SUPPORTED_PERSONALITIES > 1
744 			tcp->currpers = current_personality;
745 #endif
746 
747 #ifdef USE_LIBUNWIND
748 			if (stack_trace_enabled)
749 				unwind_tcb_init(tcp);
750 #endif
751 
752 			nprocs++;
753 			if (debug_flag)
754 				error_msg("new tcb for pid %d, active tcbs:%d",
755 					  tcp->pid, nprocs);
756 			return tcp;
757 		}
758 	}
759 	error_msg_and_die("bug in alloctcb");
760 }
761 
762 void *
get_tcb_priv_data(const struct tcb * tcp)763 get_tcb_priv_data(const struct tcb *tcp)
764 {
765 	return tcp->_priv_data;
766 }
767 
768 int
set_tcb_priv_data(struct tcb * tcp,void * const priv_data,void (* const free_priv_data)(void *))769 set_tcb_priv_data(struct tcb *tcp, void *const priv_data,
770 		  void (*const free_priv_data)(void *))
771 {
772 	if (tcp->_priv_data)
773 		return -1;
774 
775 	tcp->_free_priv_data = free_priv_data;
776 	tcp->_priv_data = priv_data;
777 
778 	return 0;
779 }
780 
781 void
free_tcb_priv_data(struct tcb * tcp)782 free_tcb_priv_data(struct tcb *tcp)
783 {
784 	if (tcp->_priv_data) {
785 		if (tcp->_free_priv_data) {
786 			tcp->_free_priv_data(tcp->_priv_data);
787 			tcp->_free_priv_data = NULL;
788 		}
789 		tcp->_priv_data = NULL;
790 	}
791 }
792 
793 static void
droptcb(struct tcb * tcp)794 droptcb(struct tcb *tcp)
795 {
796 	if (tcp->pid == 0)
797 		return;
798 
799 	int p;
800 	for (p = 0; p < SUPPORTED_PERSONALITIES; ++p)
801 		free(tcp->inject_vec[p]);
802 
803 	free_tcb_priv_data(tcp);
804 
805 #ifdef USE_LIBUNWIND
806 	if (stack_trace_enabled) {
807 		unwind_tcb_fin(tcp);
808 	}
809 #endif
810 
811 	nprocs--;
812 	if (debug_flag)
813 		error_msg("dropped tcb for pid %d, %d remain",
814 			  tcp->pid, nprocs);
815 
816 	if (tcp->outf) {
817 		if (followfork >= 2) {
818 			if (tcp->curcol != 0)
819 				fprintf(tcp->outf, " <detached ...>\n");
820 			fclose(tcp->outf);
821 		} else {
822 			if (printing_tcp == tcp && tcp->curcol != 0)
823 				fprintf(tcp->outf, " <detached ...>\n");
824 			fflush(tcp->outf);
825 		}
826 	}
827 
828 	if (current_tcp == tcp)
829 		current_tcp = NULL;
830 	if (printing_tcp == tcp)
831 		printing_tcp = NULL;
832 
833 	memset(tcp, 0, sizeof(*tcp));
834 }
835 
836 /* Detach traced process.
837  * Never call DETACH twice on the same process as both unattached and
838  * attached-unstopped processes give the same ESRCH.  For unattached process we
839  * would SIGSTOP it and wait for its SIGSTOP notification forever.
840  */
841 static void
detach(struct tcb * tcp)842 detach(struct tcb *tcp)
843 {
844 	int error;
845 	int status;
846 
847 	/*
848 	 * Linux wrongly insists the child be stopped
849 	 * before detaching.  Arghh.  We go through hoops
850 	 * to make a clean break of things.
851 	 */
852 
853 	if (!(tcp->flags & TCB_ATTACHED))
854 		goto drop;
855 
856 	/* We attached but possibly didn't see the expected SIGSTOP.
857 	 * We must catch exactly one as otherwise the detached process
858 	 * would be left stopped (process state T).
859 	 */
860 	if (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)
861 		goto wait_loop;
862 
863 	error = ptrace(PTRACE_DETACH, tcp->pid, 0, 0);
864 	if (!error) {
865 		/* On a clear day, you can see forever. */
866 		goto drop;
867 	}
868 	if (errno != ESRCH) {
869 		/* Shouldn't happen. */
870 		perror_msg("detach: ptrace(PTRACE_DETACH,%u)", tcp->pid);
871 		goto drop;
872 	}
873 	/* ESRCH: process is either not stopped or doesn't exist. */
874 	if (my_tkill(tcp->pid, 0) < 0) {
875 		if (errno != ESRCH)
876 			/* Shouldn't happen. */
877 			perror_msg("detach: tkill(%u,0)", tcp->pid);
878 		/* else: process doesn't exist. */
879 		goto drop;
880 	}
881 	/* Process is not stopped, need to stop it. */
882 	if (use_seize) {
883 		/*
884 		 * With SEIZE, tracee can be in group-stop already.
885 		 * In this state sending it another SIGSTOP does nothing.
886 		 * Need to use INTERRUPT.
887 		 * Testcase: trying to ^C a "strace -p <stopped_process>".
888 		 */
889 		error = ptrace(PTRACE_INTERRUPT, tcp->pid, 0, 0);
890 		if (!error)
891 			goto wait_loop;
892 		if (errno != ESRCH)
893 			perror_msg("detach: ptrace(PTRACE_INTERRUPT,%u)", tcp->pid);
894 	}
895 	else {
896 		error = my_tkill(tcp->pid, SIGSTOP);
897 		if (!error)
898 			goto wait_loop;
899 		if (errno != ESRCH)
900 			perror_msg("detach: tkill(%u,SIGSTOP)", tcp->pid);
901 	}
902 	/* Either process doesn't exist, or some weird error. */
903 	goto drop;
904 
905  wait_loop:
906 	/* We end up here in three cases:
907 	 * 1. We sent PTRACE_INTERRUPT (use_seize case)
908 	 * 2. We sent SIGSTOP (!use_seize)
909 	 * 3. Attach SIGSTOP was already pending (TCB_IGNORE_ONE_SIGSTOP set)
910 	 */
911 	for (;;) {
912 		unsigned int sig;
913 		if (waitpid(tcp->pid, &status, __WALL) < 0) {
914 			if (errno == EINTR)
915 				continue;
916 			/*
917 			 * if (errno == ECHILD) break;
918 			 * ^^^  WRONG! We expect this PID to exist,
919 			 * and want to emit a message otherwise:
920 			 */
921 			perror_msg("detach: waitpid(%u)", tcp->pid);
922 			break;
923 		}
924 		if (!WIFSTOPPED(status)) {
925 			/*
926 			 * Tracee exited or was killed by signal.
927 			 * We shouldn't normally reach this place:
928 			 * we don't want to consume exit status.
929 			 * Consider "strace -p PID" being ^C-ed:
930 			 * we want merely to detach from PID.
931 			 *
932 			 * However, we _can_ end up here if tracee
933 			 * was SIGKILLed.
934 			 */
935 			break;
936 		}
937 		sig = WSTOPSIG(status);
938 		if (debug_flag)
939 			error_msg("detach wait: event:%d sig:%d",
940 				  (unsigned)status >> 16, sig);
941 		if (use_seize) {
942 			unsigned event = (unsigned)status >> 16;
943 			if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) {
944 				/*
945 				 * sig == SIGTRAP: PTRACE_INTERRUPT stop.
946 				 * sig == other: process was already stopped
947 				 * with this stopping sig (see tests/detach-stopped).
948 				 * Looks like re-injecting this sig is not necessary
949 				 * in DETACH for the tracee to remain stopped.
950 				 */
951 				sig = 0;
952 			}
953 			/*
954 			 * PTRACE_INTERRUPT is not guaranteed to produce
955 			 * the above event if other ptrace-stop is pending.
956 			 * See tests/detach-sleeping testcase:
957 			 * strace got SIGINT while tracee is sleeping.
958 			 * We sent PTRACE_INTERRUPT.
959 			 * We see syscall exit, not PTRACE_INTERRUPT stop.
960 			 * We won't get PTRACE_INTERRUPT stop
961 			 * if we would CONT now. Need to DETACH.
962 			 */
963 			if (sig == syscall_trap_sig)
964 				sig = 0;
965 			/* else: not sure in which case we can be here.
966 			 * Signal stop? Inject it while detaching.
967 			 */
968 			ptrace_restart(PTRACE_DETACH, tcp, sig);
969 			break;
970 		}
971 		/* Note: this check has to be after use_seize check */
972 		/* (else, in use_seize case SIGSTOP will be mistreated) */
973 		if (sig == SIGSTOP) {
974 			/* Detach, suppressing SIGSTOP */
975 			ptrace_restart(PTRACE_DETACH, tcp, 0);
976 			break;
977 		}
978 		if (sig == syscall_trap_sig)
979 			sig = 0;
980 		/* Can't detach just yet, may need to wait for SIGSTOP */
981 		error = ptrace_restart(PTRACE_CONT, tcp, sig);
982 		if (error < 0) {
983 			/* Should not happen.
984 			 * Note: ptrace_restart returns 0 on ESRCH, so it's not it.
985 			 * ptrace_restart already emitted error message.
986 			 */
987 			break;
988 		}
989 	}
990 
991  drop:
992 	if (!qflag && (tcp->flags & TCB_ATTACHED))
993 		error_msg("Process %u detached", tcp->pid);
994 
995 	droptcb(tcp);
996 }
997 
998 static void
process_opt_p_list(char * opt)999 process_opt_p_list(char *opt)
1000 {
1001 	while (*opt) {
1002 		/*
1003 		 * We accept -p PID,PID; -p "`pidof PROG`"; -p "`pgrep PROG`".
1004 		 * pidof uses space as delim, pgrep uses newline. :(
1005 		 */
1006 		int pid;
1007 		char *delim = opt + strcspn(opt, ", \n\t");
1008 		char c = *delim;
1009 
1010 		*delim = '\0';
1011 		pid = string_to_uint(opt);
1012 		if (pid <= 0) {
1013 			error_msg_and_die("Invalid process id: '%s'", opt);
1014 		}
1015 		if (pid == strace_tracer_pid) {
1016 			error_msg_and_die("I'm sorry, I can't let you do that, Dave.");
1017 		}
1018 		*delim = c;
1019 		alloctcb(pid);
1020 		if (c == '\0')
1021 			break;
1022 		opt = delim + 1;
1023 	}
1024 }
1025 
1026 static void
attach_tcb(struct tcb * const tcp)1027 attach_tcb(struct tcb *const tcp)
1028 {
1029 	if (ptrace_attach_or_seize(tcp->pid) < 0) {
1030 		perror_msg("attach: ptrace(%s, %d)",
1031 			   ptrace_attach_cmd, tcp->pid);
1032 		droptcb(tcp);
1033 		return;
1034 	}
1035 
1036 	tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
1037 	newoutf(tcp);
1038 	if (debug_flag)
1039 		error_msg("attach to pid %d (main) succeeded", tcp->pid);
1040 
1041 	char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
1042 	DIR *dir;
1043 	unsigned int ntid = 0, nerr = 0;
1044 
1045 	if (followfork && tcp->pid != strace_child &&
1046 	    sprintf(procdir, "/proc/%d/task", tcp->pid) > 0 &&
1047 	    (dir = opendir(procdir)) != NULL) {
1048 		struct_dirent *de;
1049 
1050 		while ((de = read_dir(dir)) != NULL) {
1051 			if (de->d_fileno == 0)
1052 				continue;
1053 
1054 			int tid = string_to_uint(de->d_name);
1055 			if (tid <= 0 || tid == tcp->pid)
1056 				continue;
1057 
1058 			++ntid;
1059 			if (ptrace_attach_or_seize(tid) < 0) {
1060 				++nerr;
1061 				if (debug_flag)
1062 					perror_msg("attach: ptrace(%s, %d)",
1063 						   ptrace_attach_cmd, tid);
1064 				continue;
1065 			}
1066 			if (debug_flag)
1067 				error_msg("attach to pid %d succeeded", tid);
1068 
1069 			struct tcb *tid_tcp = alloctcb(tid);
1070 			tid_tcp->flags |= TCB_ATTACHED | TCB_STARTUP |
1071 					  post_attach_sigstop;
1072 			newoutf(tid_tcp);
1073 		}
1074 
1075 		closedir(dir);
1076 	}
1077 
1078 	if (!qflag) {
1079 		if (ntid > nerr)
1080 			error_msg("Process %u attached"
1081 				  " with %u threads",
1082 				  tcp->pid, ntid - nerr + 1);
1083 		else
1084 			error_msg("Process %u attached",
1085 				  tcp->pid);
1086 	}
1087 }
1088 
1089 static void
startup_attach(void)1090 startup_attach(void)
1091 {
1092 	pid_t parent_pid = strace_tracer_pid;
1093 	unsigned int tcbi;
1094 	struct tcb *tcp;
1095 
1096 	/*
1097 	 * Block user interruptions as we would leave the traced
1098 	 * process stopped (process state T) if we would terminate in
1099 	 * between PTRACE_ATTACH and wait4() on SIGSTOP.
1100 	 * We rely on cleanup() from this point on.
1101 	 */
1102 	if (interactive)
1103 		sigprocmask(SIG_BLOCK, &blocked_set, NULL);
1104 
1105 	if (daemonized_tracer) {
1106 		pid_t pid = fork();
1107 		if (pid < 0) {
1108 			perror_msg_and_die("fork");
1109 		}
1110 		if (pid) { /* parent */
1111 			/*
1112 			 * Wait for grandchild to attach to straced process
1113 			 * (grandparent). Grandchild SIGKILLs us after it attached.
1114 			 * Grandparent's wait() is unblocked by our death,
1115 			 * it proceeds to exec the straced program.
1116 			 */
1117 			pause();
1118 			_exit(0); /* paranoia */
1119 		}
1120 		/* grandchild */
1121 		/* We will be the tracer process. Remember our new pid: */
1122 		strace_tracer_pid = getpid();
1123 	}
1124 
1125 	for (tcbi = 0; tcbi < tcbtabsize; tcbi++) {
1126 		tcp = tcbtab[tcbi];
1127 
1128 		if (!tcp->pid)
1129 			continue;
1130 
1131 		/* Is this a process we should attach to, but not yet attached? */
1132 		if (tcp->flags & TCB_ATTACHED)
1133 			continue; /* no, we already attached it */
1134 
1135 		if (tcp->pid == parent_pid || tcp->pid == strace_tracer_pid) {
1136 			errno = EPERM;
1137 			perror_msg("attach: pid %d", tcp->pid);
1138 			droptcb(tcp);
1139 			continue;
1140 		}
1141 
1142 		attach_tcb(tcp);
1143 
1144 		if (interactive) {
1145 			sigprocmask(SIG_SETMASK, &empty_set, NULL);
1146 			if (interrupted)
1147 				goto ret;
1148 			sigprocmask(SIG_BLOCK, &blocked_set, NULL);
1149 		}
1150 	} /* for each tcbtab[] */
1151 
1152 	if (daemonized_tracer) {
1153 		/*
1154 		 * Make parent go away.
1155 		 * Also makes grandparent's wait() unblock.
1156 		 */
1157 		kill(parent_pid, SIGKILL);
1158 		strace_child = 0;
1159 	}
1160 
1161  ret:
1162 	if (interactive)
1163 		sigprocmask(SIG_SETMASK, &empty_set, NULL);
1164 }
1165 
1166 /* Stack-o-phobic exec helper, in the hope to work around
1167  * NOMMU + "daemonized tracer" difficulty.
1168  */
1169 struct exec_params {
1170 	int fd_to_close;
1171 	uid_t run_euid;
1172 	gid_t run_egid;
1173 	char **argv;
1174 	char *pathname;
1175 };
1176 static struct exec_params params_for_tracee;
1177 
1178 static void ATTRIBUTE_NOINLINE ATTRIBUTE_NORETURN
exec_or_die(void)1179 exec_or_die(void)
1180 {
1181 	struct exec_params *params = &params_for_tracee;
1182 
1183 	if (params->fd_to_close >= 0)
1184 		close(params->fd_to_close);
1185 	if (!daemonized_tracer && !use_seize) {
1186 		if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) {
1187 			perror_msg_and_die("ptrace(PTRACE_TRACEME, ...)");
1188 		}
1189 	}
1190 
1191 	if (username != NULL) {
1192 		/*
1193 		 * It is important to set groups before we
1194 		 * lose privileges on setuid.
1195 		 */
1196 		if (initgroups(username, run_gid) < 0) {
1197 			perror_msg_and_die("initgroups");
1198 		}
1199 		if (setregid(run_gid, params->run_egid) < 0) {
1200 			perror_msg_and_die("setregid");
1201 		}
1202 		if (setreuid(run_uid, params->run_euid) < 0) {
1203 			perror_msg_and_die("setreuid");
1204 		}
1205 	}
1206 	else if (geteuid() != 0)
1207 		if (setreuid(run_uid, run_uid) < 0) {
1208 			perror_msg_and_die("setreuid");
1209 		}
1210 
1211 	if (!daemonized_tracer) {
1212 		/*
1213 		 * Induce a ptrace stop. Tracer (our parent)
1214 		 * will resume us with PTRACE_SYSCALL and display
1215 		 * the immediately following execve syscall.
1216 		 * Can't do this on NOMMU systems, we are after
1217 		 * vfork: parent is blocked, stopping would deadlock.
1218 		 */
1219 		if (!NOMMU_SYSTEM)
1220 			kill(getpid(), SIGSTOP);
1221 	} else {
1222 		alarm(3);
1223 		/* we depend on SIGCHLD set to SIG_DFL by init code */
1224 		/* if it happens to be SIG_IGN'ed, wait won't block */
1225 		wait(NULL);
1226 		alarm(0);
1227 	}
1228 
1229 	execv(params->pathname, params->argv);
1230 	perror_msg_and_die("exec");
1231 }
1232 
1233 /*
1234  * Open a dummy descriptor for use as a placeholder.
1235  * The descriptor is O_RDONLY with FD_CLOEXEC flag set.
1236  * A read attempt from such descriptor ends with EOF,
1237  * a write attempt is rejected with EBADF.
1238  */
1239 static int
open_dummy_desc(void)1240 open_dummy_desc(void)
1241 {
1242 	int fds[2];
1243 
1244 	if (pipe(fds))
1245 		perror_msg_and_die("pipe");
1246 	close(fds[1]);
1247 	set_cloexec_flag(fds[0]);
1248 	return fds[0];
1249 }
1250 
1251 /* placeholder fds status for stdin and stdout */
1252 static bool fd_is_placeholder[2];
1253 
1254 /*
1255  * Ensure that all standard file descriptors are open by opening placeholder
1256  * file descriptors for those standard file descriptors that are not open.
1257  *
1258  * The information which descriptors have been made open is saved
1259  * in fd_is_placeholder for later use.
1260  */
1261 static void
ensure_standard_fds_opened(void)1262 ensure_standard_fds_opened(void)
1263 {
1264 	int fd;
1265 
1266 	while ((fd = open_dummy_desc()) <= 2) {
1267 		if (fd == 2)
1268 			break;
1269 		fd_is_placeholder[fd] = true;
1270 	}
1271 
1272 	if (fd > 2)
1273 		close(fd);
1274 }
1275 
1276 /*
1277  * Redirect stdin and stdout unless they have been opened earlier
1278  * by ensure_standard_fds_opened as placeholders.
1279  */
1280 static void
redirect_standard_fds(void)1281 redirect_standard_fds(void)
1282 {
1283 	int i;
1284 
1285 	/*
1286 	 * It might be a good idea to redirect stderr as well,
1287 	 * but we sometimes need to print error messages.
1288 	 */
1289 	for (i = 0; i <= 1; ++i) {
1290 		if (!fd_is_placeholder[i]) {
1291 			close(i);
1292 			open_dummy_desc();
1293 		}
1294 	}
1295 }
1296 
1297 static void
startup_child(char ** argv)1298 startup_child(char **argv)
1299 {
1300 	struct_stat statbuf;
1301 	const char *filename;
1302 	size_t filename_len;
1303 	char pathname[PATH_MAX];
1304 	int pid;
1305 	struct tcb *tcp;
1306 
1307 	filename = argv[0];
1308 	filename_len = strlen(filename);
1309 
1310 	if (filename_len > sizeof(pathname) - 1) {
1311 		errno = ENAMETOOLONG;
1312 		perror_msg_and_die("exec");
1313 	}
1314 	if (strchr(filename, '/')) {
1315 		strcpy(pathname, filename);
1316 	}
1317 #ifdef USE_DEBUGGING_EXEC
1318 	/*
1319 	 * Debuggers customarily check the current directory
1320 	 * first regardless of the path but doing that gives
1321 	 * security geeks a panic attack.
1322 	 */
1323 	else if (stat_file(filename, &statbuf) == 0)
1324 		strcpy(pathname, filename);
1325 #endif /* USE_DEBUGGING_EXEC */
1326 	else {
1327 		const char *path;
1328 		size_t m, n, len;
1329 
1330 		for (path = getenv("PATH"); path && *path; path += m) {
1331 			const char *colon = strchr(path, ':');
1332 			if (colon) {
1333 				n = colon - path;
1334 				m = n + 1;
1335 			}
1336 			else
1337 				m = n = strlen(path);
1338 			if (n == 0) {
1339 				if (!getcwd(pathname, PATH_MAX))
1340 					continue;
1341 				len = strlen(pathname);
1342 			}
1343 			else if (n > sizeof pathname - 1)
1344 				continue;
1345 			else {
1346 				strncpy(pathname, path, n);
1347 				len = n;
1348 			}
1349 			if (len && pathname[len - 1] != '/')
1350 				pathname[len++] = '/';
1351 			if (filename_len + len > sizeof(pathname) - 1)
1352 				continue;
1353 			strcpy(pathname + len, filename);
1354 			if (stat_file(pathname, &statbuf) == 0 &&
1355 			    /* Accept only regular files
1356 			       with some execute bits set.
1357 			       XXX not perfect, might still fail */
1358 			    S_ISREG(statbuf.st_mode) &&
1359 			    (statbuf.st_mode & 0111))
1360 				break;
1361 		}
1362 		if (!path || !*path)
1363 			pathname[0] = '\0';
1364 	}
1365 	if (stat_file(pathname, &statbuf) < 0) {
1366 		perror_msg_and_die("Can't stat '%s'", filename);
1367 	}
1368 
1369 	params_for_tracee.fd_to_close = (shared_log != stderr) ? fileno(shared_log) : -1;
1370 	params_for_tracee.run_euid = (statbuf.st_mode & S_ISUID) ? statbuf.st_uid : run_uid;
1371 	params_for_tracee.run_egid = (statbuf.st_mode & S_ISGID) ? statbuf.st_gid : run_gid;
1372 	params_for_tracee.argv = argv;
1373 	/*
1374 	 * On NOMMU, can be safely freed only after execve in tracee.
1375 	 * It's hard to know when that happens, so we just leak it.
1376 	 */
1377 	params_for_tracee.pathname = NOMMU_SYSTEM ? xstrdup(pathname) : pathname;
1378 
1379 #if defined HAVE_PRCTL && defined PR_SET_PTRACER && defined PR_SET_PTRACER_ANY
1380 	if (daemonized_tracer)
1381 		prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY);
1382 #endif
1383 
1384 	pid = fork();
1385 	if (pid < 0) {
1386 		perror_msg_and_die("fork");
1387 	}
1388 	if ((pid != 0 && daemonized_tracer)
1389 	 || (pid == 0 && !daemonized_tracer)
1390 	) {
1391 		/* We are to become the tracee. Two cases:
1392 		 * -D: we are parent
1393 		 * not -D: we are child
1394 		 */
1395 		exec_or_die();
1396 	}
1397 
1398 	/* We are the tracer */
1399 
1400 	if (!daemonized_tracer) {
1401 		strace_child = pid;
1402 		if (!use_seize) {
1403 			/* child did PTRACE_TRACEME, nothing to do in parent */
1404 		} else {
1405 			if (!NOMMU_SYSTEM) {
1406 				/* Wait until child stopped itself */
1407 				int status;
1408 				while (waitpid(pid, &status, WSTOPPED) < 0) {
1409 					if (errno == EINTR)
1410 						continue;
1411 					perror_msg_and_die("waitpid");
1412 				}
1413 				if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP) {
1414 					kill_save_errno(pid, SIGKILL);
1415 					perror_msg_and_die("Unexpected wait status %#x",
1416 							   status);
1417 				}
1418 			}
1419 			/* Else: NOMMU case, we have no way to sync.
1420 			 * Just attach to it as soon as possible.
1421 			 * This means that we may miss a few first syscalls...
1422 			 */
1423 
1424 			if (ptrace_attach_or_seize(pid)) {
1425 				kill_save_errno(pid, SIGKILL);
1426 				perror_msg_and_die("attach: ptrace(%s, %d)",
1427 						   ptrace_attach_cmd, pid);
1428 			}
1429 			if (!NOMMU_SYSTEM)
1430 				kill(pid, SIGCONT);
1431 		}
1432 		tcp = alloctcb(pid);
1433 		tcp->flags |= TCB_ATTACHED | TCB_STARTUP
1434 			    | TCB_SKIP_DETACH_ON_FIRST_EXEC
1435 			    | (NOMMU_SYSTEM ? 0 : (TCB_HIDE_LOG | post_attach_sigstop));
1436 		newoutf(tcp);
1437 	}
1438 	else {
1439 		/* With -D, we are *child* here, the tracee is our parent. */
1440 		strace_child = strace_tracer_pid;
1441 		strace_tracer_pid = getpid();
1442 		tcp = alloctcb(strace_child);
1443 		tcp->flags |= TCB_SKIP_DETACH_ON_FIRST_EXEC | TCB_HIDE_LOG;
1444 		/* attaching will be done later, by startup_attach */
1445 		/* note: we don't do newoutf(tcp) here either! */
1446 
1447 		/* NOMMU BUG! -D mode is active, we (child) return,
1448 		 * and we will scribble over parent's stack!
1449 		 * When parent later unpauses, it segfaults.
1450 		 *
1451 		 * We work around it
1452 		 * (1) by declaring exec_or_die() NORETURN,
1453 		 * hopefully compiler will just jump to it
1454 		 * instead of call (won't push anything to stack),
1455 		 * (2) by trying very hard in exec_or_die()
1456 		 * to not use any stack,
1457 		 * (3) having a really big (PATH_MAX) stack object
1458 		 * in this function, which creates a "buffer" between
1459 		 * child's and parent's stack pointers.
1460 		 * This may save us if (1) and (2) failed
1461 		 * and compiler decided to use stack in exec_or_die() anyway
1462 		 * (happens on i386 because of stack parameter passing).
1463 		 *
1464 		 * A cleaner solution is to use makecontext + setcontext
1465 		 * to create a genuine separate stack and execute on it.
1466 		 */
1467 	}
1468 	/*
1469 	 * A case where straced process is part of a pipe:
1470 	 * { sleep 1; yes | head -n99999; } | strace -o/dev/null sh -c 'exec <&-; sleep 9'
1471 	 * If strace won't close its fd#0, closing it in tracee is not enough:
1472 	 * the pipe is still open, it has a reader. Thus, "head" will not get its
1473 	 * SIGPIPE at once, on the first write.
1474 	 *
1475 	 * Preventing it by redirecting strace's stdin/out.
1476 	 * (Don't leave fds 0 and 1 closed, this is bad practice: future opens
1477 	 * will reuse them, unexpectedly making a newly opened object "stdin").
1478 	 */
1479 	redirect_standard_fds();
1480 }
1481 
1482 #if USE_SEIZE
1483 static void
test_ptrace_seize(void)1484 test_ptrace_seize(void)
1485 {
1486 	int pid;
1487 
1488 	/* Need fork for test. NOMMU has no forks */
1489 	if (NOMMU_SYSTEM) {
1490 		post_attach_sigstop = 0; /* this sets use_seize to 1 */
1491 		return;
1492 	}
1493 
1494 	pid = fork();
1495 	if (pid < 0)
1496 		perror_msg_and_die("fork");
1497 
1498 	if (pid == 0) {
1499 		pause();
1500 		_exit(0);
1501 	}
1502 
1503 	/* PTRACE_SEIZE, unlike ATTACH, doesn't force tracee to trap.  After
1504 	 * attaching tracee continues to run unless a trap condition occurs.
1505 	 * PTRACE_SEIZE doesn't affect signal or group stop state.
1506 	 */
1507 	if (ptrace(PTRACE_SEIZE, pid, 0, 0) == 0) {
1508 		post_attach_sigstop = 0; /* this sets use_seize to 1 */
1509 	} else if (debug_flag) {
1510 		error_msg("PTRACE_SEIZE doesn't work");
1511 	}
1512 
1513 	kill(pid, SIGKILL);
1514 
1515 	while (1) {
1516 		int status, tracee_pid;
1517 
1518 		errno = 0;
1519 		tracee_pid = waitpid(pid, &status, 0);
1520 		if (tracee_pid <= 0) {
1521 			if (errno == EINTR)
1522 				continue;
1523 			perror_msg_and_die("%s: unexpected wait result %d",
1524 					 __func__, tracee_pid);
1525 		}
1526 		if (WIFSIGNALED(status)) {
1527 			return;
1528 		}
1529 		error_msg_and_die("%s: unexpected wait status %#x",
1530 				  __func__, status);
1531 	}
1532 }
1533 #else /* !USE_SEIZE */
1534 # define test_ptrace_seize() ((void)0)
1535 #endif
1536 
1537 static unsigned
get_os_release(void)1538 get_os_release(void)
1539 {
1540 	unsigned rel;
1541 	const char *p;
1542 	struct utsname u;
1543 	if (uname(&u) < 0)
1544 		perror_msg_and_die("uname");
1545 	/* u.release has this form: "3.2.9[-some-garbage]" */
1546 	rel = 0;
1547 	p = u.release;
1548 	for (;;) {
1549 		if (!(*p >= '0' && *p <= '9'))
1550 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1551 		/* Note: this open-codes KERNEL_VERSION(): */
1552 		rel = (rel << 8) | atoi(p);
1553 		if (rel >= KERNEL_VERSION(1,0,0))
1554 			break;
1555 		while (*p >= '0' && *p <= '9')
1556 			p++;
1557 		if (*p != '.') {
1558 			if (rel >= KERNEL_VERSION(0,1,0)) {
1559 				/* "X.Y-something" means "X.Y.0" */
1560 				rel <<= 8;
1561 				break;
1562 			}
1563 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1564 		}
1565 		p++;
1566 	}
1567 	return rel;
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 	struct sigaction sa;
1584 
1585 	progname = argv[0] ? argv[0] : "strace";
1586 
1587 	/* Make sure SIGCHLD has the default action so that waitpid
1588 	   definitely works without losing track of children.  The user
1589 	   should not have given us a bogus state to inherit, but he might
1590 	   have.  Arguably we should detect SIG_IGN here and pass it on
1591 	   to children, but probably noone really needs that.  */
1592 	signal(SIGCHLD, SIG_DFL);
1593 
1594 	strace_tracer_pid = getpid();
1595 
1596 	os_release = get_os_release();
1597 
1598 	shared_log = stderr;
1599 	set_sortby(DEFAULT_SORTBY);
1600 	set_personality(DEFAULT_PERSONALITY);
1601 	qualify("trace=all");
1602 	qualify("abbrev=all");
1603 	qualify("verbose=all");
1604 #if DEFAULT_QUAL_FLAGS != (QUAL_TRACE | QUAL_ABBREV | QUAL_VERBOSE)
1605 # error Bug in DEFAULT_QUAL_FLAGS
1606 #endif
1607 	qualify("signal=all");
1608 	while ((c = getopt(argc, argv,
1609 		"+b:cCdfFhiqrtTvVwxyz"
1610 #ifdef USE_LIBUNWIND
1611 		"k"
1612 #endif
1613 		"D"
1614 		"a:e:o:O:p:s:S:u:E:P:I:")) != EOF) {
1615 		switch (c) {
1616 		case 'b':
1617 			if (strcmp(optarg, "execve") != 0)
1618 				error_msg_and_die("Syscall '%s' for -b isn't supported",
1619 					optarg);
1620 			detach_on_execve = 1;
1621 			break;
1622 		case 'c':
1623 			if (cflag == CFLAG_BOTH) {
1624 				error_msg_and_help("-c and -C are mutually exclusive");
1625 			}
1626 			cflag = CFLAG_ONLY_STATS;
1627 			break;
1628 		case 'C':
1629 			if (cflag == CFLAG_ONLY_STATS) {
1630 				error_msg_and_help("-c and -C are mutually exclusive");
1631 			}
1632 			cflag = CFLAG_BOTH;
1633 			break;
1634 		case 'd':
1635 			debug_flag = 1;
1636 			break;
1637 		case 'D':
1638 			daemonized_tracer = 1;
1639 			break;
1640 		case 'F':
1641 			optF = 1;
1642 			break;
1643 		case 'f':
1644 			followfork++;
1645 			break;
1646 		case 'h':
1647 			usage();
1648 			break;
1649 		case 'i':
1650 			iflag = 1;
1651 			break;
1652 		case 'q':
1653 			qflag++;
1654 			break;
1655 		case 'r':
1656 			rflag = 1;
1657 			break;
1658 		case 't':
1659 			tflag++;
1660 			break;
1661 		case 'T':
1662 			Tflag = 1;
1663 			break;
1664 		case 'w':
1665 			count_wallclock = 1;
1666 			break;
1667 		case 'x':
1668 			xflag++;
1669 			break;
1670 		case 'y':
1671 			show_fd_path++;
1672 			break;
1673 		case 'v':
1674 			qualify("abbrev=none");
1675 			break;
1676 		case 'V':
1677 			print_version();
1678 			exit(0);
1679 			break;
1680 		case 'z':
1681 			not_failing_only = 1;
1682 			break;
1683 		case 'a':
1684 			acolumn = string_to_uint(optarg);
1685 			if (acolumn < 0)
1686 				error_opt_arg(c, optarg);
1687 			break;
1688 		case 'e':
1689 			qualify(optarg);
1690 			break;
1691 		case 'o':
1692 			outfname = xstrdup(optarg);
1693 			break;
1694 		case 'O':
1695 			i = string_to_uint(optarg);
1696 			if (i < 0)
1697 				error_opt_arg(c, optarg);
1698 			set_overhead(i);
1699 			break;
1700 		case 'p':
1701 			process_opt_p_list(optarg);
1702 			break;
1703 		case 'P':
1704 			pathtrace_select(optarg);
1705 			break;
1706 		case 's':
1707 			i = string_to_uint(optarg);
1708 			if (i < 0)
1709 				error_opt_arg(c, optarg);
1710 			max_strlen = i;
1711 			break;
1712 		case 'S':
1713 			set_sortby(optarg);
1714 			break;
1715 		case 'u':
1716 			username = xstrdup(optarg);
1717 			break;
1718 #ifdef USE_LIBUNWIND
1719 		case 'k':
1720 			stack_trace_enabled = true;
1721 			break;
1722 #endif
1723 		case 'E':
1724 			if (putenv(optarg) < 0)
1725 				die_out_of_memory();
1726 			break;
1727 		case 'I':
1728 			opt_intr = string_to_uint_upto(optarg, NUM_INTR_OPTS - 1);
1729 			if (opt_intr <= 0)
1730 				error_opt_arg(c, optarg);
1731 			break;
1732 		default:
1733 			error_msg_and_help(NULL);
1734 			break;
1735 		}
1736 	}
1737 	argv += optind;
1738 	/* argc -= optind; - no need, argc is not used below */
1739 
1740 	acolumn_spaces = xmalloc(acolumn + 1);
1741 	memset(acolumn_spaces, ' ', acolumn);
1742 	acolumn_spaces[acolumn] = '\0';
1743 
1744 	if (!argv[0] && !nprocs) {
1745 		error_msg_and_help("must have PROG [ARGS] or -p PID");
1746 	}
1747 
1748 	if (!argv[0] && daemonized_tracer) {
1749 		error_msg_and_help("PROG [ARGS] must be specified with -D");
1750 	}
1751 
1752 	if (!followfork)
1753 		followfork = optF;
1754 
1755 	if (followfork >= 2 && cflag) {
1756 		error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
1757 	}
1758 
1759 	if (count_wallclock && !cflag) {
1760 		error_msg_and_help("-w must be given with (-c or -C)");
1761 	}
1762 
1763 	if (cflag == CFLAG_ONLY_STATS) {
1764 		if (iflag)
1765 			error_msg("-%c has no effect with -c", 'i');
1766 #ifdef USE_LIBUNWIND
1767 		if (stack_trace_enabled)
1768 			error_msg("-%c has no effect with -c", 'k');
1769 #endif
1770 		if (rflag)
1771 			error_msg("-%c has no effect with -c", 'r');
1772 		if (tflag)
1773 			error_msg("-%c has no effect with -c", 't');
1774 		if (Tflag)
1775 			error_msg("-%c has no effect with -c", 'T');
1776 		if (show_fd_path)
1777 			error_msg("-%c has no effect with -c", 'y');
1778 	}
1779 
1780 	if (rflag) {
1781 		if (tflag > 1)
1782 			error_msg("-tt has no effect with -r");
1783 		tflag = 1;
1784 	}
1785 
1786 #ifdef USE_LIBUNWIND
1787 	if (stack_trace_enabled) {
1788 		unsigned int tcbi;
1789 
1790 		unwind_init();
1791 		for (tcbi = 0; tcbi < tcbtabsize; ++tcbi) {
1792 			unwind_tcb_init(tcbtab[tcbi]);
1793 		}
1794 	}
1795 #endif
1796 
1797 	/* See if they want to run as another user. */
1798 	if (username != NULL) {
1799 		struct passwd *pent;
1800 
1801 		if (getuid() != 0 || geteuid() != 0) {
1802 			error_msg_and_die("You must be root to use the -u option");
1803 		}
1804 		pent = getpwnam(username);
1805 		if (pent == NULL) {
1806 			error_msg_and_die("Cannot find user '%s'", username);
1807 		}
1808 		run_uid = pent->pw_uid;
1809 		run_gid = pent->pw_gid;
1810 	}
1811 	else {
1812 		run_uid = getuid();
1813 		run_gid = getgid();
1814 	}
1815 
1816 	if (followfork)
1817 		ptrace_setoptions |= PTRACE_O_TRACECLONE |
1818 				     PTRACE_O_TRACEFORK |
1819 				     PTRACE_O_TRACEVFORK;
1820 	if (debug_flag)
1821 		error_msg("ptrace_setoptions = %#x", ptrace_setoptions);
1822 	test_ptrace_seize();
1823 
1824 	/*
1825 	 * Is something weird with our stdin and/or stdout -
1826 	 * for example, may they be not open? In this case,
1827 	 * ensure that none of the future opens uses them.
1828 	 *
1829 	 * This was seen in the wild when /proc/sys/kernel/core_pattern
1830 	 * was set to "|/bin/strace -o/tmp/LOG PROG":
1831 	 * kernel runs coredump helper with fd#0 open but fd#1 closed (!),
1832 	 * therefore LOG gets opened to fd#1, and fd#1 is closed by
1833 	 * "don't hold up stdin/out open" code soon after.
1834 	 */
1835 	ensure_standard_fds_opened();
1836 
1837 	/* Check if they want to redirect the output. */
1838 	if (outfname) {
1839 		/* See if they want to pipe the output. */
1840 		if (outfname[0] == '|' || outfname[0] == '!') {
1841 			/*
1842 			 * We can't do the <outfname>.PID funny business
1843 			 * when using popen, so prohibit it.
1844 			 */
1845 			if (followfork >= 2)
1846 				error_msg_and_help("piping the output and -ff are mutually exclusive");
1847 			shared_log = strace_popen(outfname + 1);
1848 		}
1849 		else if (followfork < 2)
1850 			shared_log = strace_fopen(outfname);
1851 	} else {
1852 		/* -ff without -o FILE is the same as single -f */
1853 		if (followfork >= 2)
1854 			followfork = 1;
1855 	}
1856 
1857 	if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
1858 		setvbuf(shared_log, NULL, _IOLBF, 0);
1859 	}
1860 	if (outfname && argv[0]) {
1861 		if (!opt_intr)
1862 			opt_intr = INTR_NEVER;
1863 		if (!qflag)
1864 			qflag = 1;
1865 	}
1866 	if (!opt_intr)
1867 		opt_intr = INTR_WHILE_WAIT;
1868 
1869 	/* argv[0]	-pPID	-oFILE	Default interactive setting
1870 	 * yes		*	0	INTR_WHILE_WAIT
1871 	 * no		1	0	INTR_WHILE_WAIT
1872 	 * yes		*	1	INTR_NEVER
1873 	 * no		1	1	INTR_WHILE_WAIT
1874 	 */
1875 
1876 	sigemptyset(&empty_set);
1877 	sigemptyset(&blocked_set);
1878 
1879 	/* startup_child() must be called before the signal handlers get
1880 	 * installed below as they are inherited into the spawned process.
1881 	 * Also we do not need to be protected by them as during interruption
1882 	 * in the startup_child() mode we kill the spawned process anyway.
1883 	 */
1884 	if (argv[0]) {
1885 		startup_child(argv);
1886 	}
1887 
1888 	sa.sa_handler = SIG_IGN;
1889 	sigemptyset(&sa.sa_mask);
1890 	sa.sa_flags = 0;
1891 	sigaction(SIGTTOU, &sa, NULL); /* SIG_IGN */
1892 	sigaction(SIGTTIN, &sa, NULL); /* SIG_IGN */
1893 	if (opt_intr != INTR_ANYWHERE) {
1894 		if (opt_intr == INTR_BLOCK_TSTP_TOO)
1895 			sigaction(SIGTSTP, &sa, NULL); /* SIG_IGN */
1896 		/*
1897 		 * In interactive mode (if no -o OUTFILE, or -p PID is used),
1898 		 * fatal signals are blocked while syscall stop is processed,
1899 		 * and acted on in between, when waiting for new syscall stops.
1900 		 * In non-interactive mode, signals are ignored.
1901 		 */
1902 		if (opt_intr == INTR_WHILE_WAIT) {
1903 			sigaddset(&blocked_set, SIGHUP);
1904 			sigaddset(&blocked_set, SIGINT);
1905 			sigaddset(&blocked_set, SIGQUIT);
1906 			sigaddset(&blocked_set, SIGPIPE);
1907 			sigaddset(&blocked_set, SIGTERM);
1908 			sa.sa_handler = interrupt;
1909 		}
1910 		/* SIG_IGN, or set handler for these */
1911 		sigaction(SIGHUP, &sa, NULL);
1912 		sigaction(SIGINT, &sa, NULL);
1913 		sigaction(SIGQUIT, &sa, NULL);
1914 		sigaction(SIGPIPE, &sa, NULL);
1915 		sigaction(SIGTERM, &sa, NULL);
1916 	}
1917 	if (nprocs != 0 || daemonized_tracer)
1918 		startup_attach();
1919 
1920 	/* Do we want pids printed in our -o OUTFILE?
1921 	 * -ff: no (every pid has its own file); or
1922 	 * -f: yes (there can be more pids in the future); or
1923 	 * -p PID1,PID2: yes (there are already more than one pid)
1924 	 */
1925 	print_pid_pfx = (outfname && followfork < 2 && (followfork == 1 || nprocs > 1));
1926 }
1927 
1928 static struct tcb *
pid2tcb(int pid)1929 pid2tcb(int pid)
1930 {
1931 	unsigned int i;
1932 
1933 	if (pid <= 0)
1934 		return NULL;
1935 
1936 	for (i = 0; i < tcbtabsize; i++) {
1937 		struct tcb *tcp = tcbtab[i];
1938 		if (tcp->pid == pid)
1939 			return tcp;
1940 	}
1941 
1942 	return NULL;
1943 }
1944 
1945 static void
cleanup(void)1946 cleanup(void)
1947 {
1948 	unsigned int i;
1949 	struct tcb *tcp;
1950 	int fatal_sig;
1951 
1952 	/* 'interrupted' is a volatile object, fetch it only once */
1953 	fatal_sig = interrupted;
1954 	if (!fatal_sig)
1955 		fatal_sig = SIGTERM;
1956 
1957 	for (i = 0; i < tcbtabsize; i++) {
1958 		tcp = tcbtab[i];
1959 		if (!tcp->pid)
1960 			continue;
1961 		if (debug_flag)
1962 			error_msg("cleanup: looking at pid %u", tcp->pid);
1963 		if (tcp->pid == strace_child) {
1964 			kill(tcp->pid, SIGCONT);
1965 			kill(tcp->pid, fatal_sig);
1966 		}
1967 		detach(tcp);
1968 	}
1969 	if (cflag)
1970 		call_summary(shared_log);
1971 }
1972 
1973 static void
interrupt(int sig)1974 interrupt(int sig)
1975 {
1976 	interrupted = sig;
1977 }
1978 
1979 static void
print_debug_info(const int pid,int status)1980 print_debug_info(const int pid, int status)
1981 {
1982 	const unsigned int event = (unsigned int) status >> 16;
1983 	char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16];
1984 	char evbuf[sizeof(",EVENT_VFORK_DONE (%u)") + sizeof(int)*3 /*paranoia:*/ + 16];
1985 
1986 	strcpy(buf, "???");
1987 	if (WIFSIGNALED(status))
1988 #ifdef WCOREDUMP
1989 		sprintf(buf, "WIFSIGNALED,%ssig=%s",
1990 				WCOREDUMP(status) ? "core," : "",
1991 				signame(WTERMSIG(status)));
1992 #else
1993 		sprintf(buf, "WIFSIGNALED,sig=%s",
1994 				signame(WTERMSIG(status)));
1995 #endif
1996 	if (WIFEXITED(status))
1997 		sprintf(buf, "WIFEXITED,exitcode=%u", WEXITSTATUS(status));
1998 	if (WIFSTOPPED(status))
1999 		sprintf(buf, "WIFSTOPPED,sig=%s", signame(WSTOPSIG(status)));
2000 #ifdef WIFCONTINUED
2001 	/* Should never be seen */
2002 	if (WIFCONTINUED(status))
2003 		strcpy(buf, "WIFCONTINUED");
2004 #endif
2005 	evbuf[0] = '\0';
2006 	if (event != 0) {
2007 		static const char *const event_names[] = {
2008 			[PTRACE_EVENT_CLONE] = "CLONE",
2009 			[PTRACE_EVENT_FORK]  = "FORK",
2010 			[PTRACE_EVENT_VFORK] = "VFORK",
2011 			[PTRACE_EVENT_VFORK_DONE] = "VFORK_DONE",
2012 			[PTRACE_EVENT_EXEC]  = "EXEC",
2013 			[PTRACE_EVENT_EXIT]  = "EXIT",
2014 			/* [PTRACE_EVENT_STOP (=128)] would make biggish array */
2015 		};
2016 		const char *e = "??";
2017 		if (event < ARRAY_SIZE(event_names))
2018 			e = event_names[event];
2019 		else if (event == PTRACE_EVENT_STOP)
2020 			e = "STOP";
2021 		sprintf(evbuf, ",EVENT_%s (%u)", e, event);
2022 	}
2023 	error_msg("[wait(0x%06x) = %u] %s%s", status, pid, buf, evbuf);
2024 }
2025 
2026 static struct tcb *
maybe_allocate_tcb(const int pid,int status)2027 maybe_allocate_tcb(const int pid, int status)
2028 {
2029 	if (!WIFSTOPPED(status)) {
2030 		if (detach_on_execve && pid == strace_child) {
2031 			/* example: strace -bexecve sh -c 'exec true' */
2032 			strace_child = 0;
2033 			return NULL;
2034 		}
2035 		/*
2036 		 * This can happen if we inherited an unknown child.
2037 		 * Example: (sleep 1 & exec strace true)
2038 		 */
2039 		error_msg("Exit of unknown pid %u ignored", pid);
2040 		return NULL;
2041 	}
2042 	if (followfork) {
2043 		/* We assume it's a fork/vfork/clone child */
2044 		struct tcb *tcp = alloctcb(pid);
2045 		tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
2046 		newoutf(tcp);
2047 		if (!qflag)
2048 			error_msg("Process %d attached", pid);
2049 		return tcp;
2050 	} else {
2051 		/* This can happen if a clone call used
2052 		 * CLONE_PTRACE itself.
2053 		 */
2054 		ptrace(PTRACE_CONT, pid, NULL, 0);
2055 		error_msg("Stop of unknown pid %u seen, PTRACE_CONTed it", pid);
2056 		return NULL;
2057 	}
2058 }
2059 
2060 static struct tcb *
maybe_switch_tcbs(struct tcb * tcp,const int pid)2061 maybe_switch_tcbs(struct tcb *tcp, const int pid)
2062 {
2063 	FILE *fp;
2064 	struct tcb *execve_thread;
2065 	long old_pid = 0;
2066 
2067 	if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, &old_pid) < 0)
2068 		return tcp;
2069 	/* Avoid truncation in pid2tcb() param passing */
2070 	if (old_pid <= 0 || old_pid == pid)
2071 		return tcp;
2072 	if ((unsigned long) old_pid > UINT_MAX)
2073 		return tcp;
2074 	execve_thread = pid2tcb(old_pid);
2075 	/* It should be !NULL, but I feel paranoid */
2076 	if (!execve_thread)
2077 		return tcp;
2078 
2079 	if (execve_thread->curcol != 0) {
2080 		/*
2081 		 * One case we are here is -ff:
2082 		 * try "strace -oLOG -ff test/threaded_execve"
2083 		 */
2084 		fprintf(execve_thread->outf, " <pid changed to %d ...>\n", pid);
2085 		/*execve_thread->curcol = 0; - no need, see code below */
2086 	}
2087 	/* Swap output FILEs (needed for -ff) */
2088 	fp = execve_thread->outf;
2089 	execve_thread->outf = tcp->outf;
2090 	tcp->outf = fp;
2091 	/* And their column positions */
2092 	execve_thread->curcol = tcp->curcol;
2093 	tcp->curcol = 0;
2094 	/* Drop leader, but close execve'd thread outfile (if -ff) */
2095 	droptcb(tcp);
2096 	/* Switch to the thread, reusing leader's outfile and pid */
2097 	tcp = execve_thread;
2098 	tcp->pid = pid;
2099 	if (cflag != CFLAG_ONLY_STATS) {
2100 		printleader(tcp);
2101 		tprintf("+++ superseded by execve in pid %lu +++\n", old_pid);
2102 		line_ended();
2103 		tcp->flags |= TCB_REPRINT;
2104 	}
2105 
2106 	return tcp;
2107 }
2108 
2109 static void
print_signalled(struct tcb * tcp,const int pid,int status)2110 print_signalled(struct tcb *tcp, const int pid, int status)
2111 {
2112 	if (pid == strace_child) {
2113 		exit_code = 0x100 | WTERMSIG(status);
2114 		strace_child = 0;
2115 	}
2116 
2117 	if (cflag != CFLAG_ONLY_STATS
2118 	    && is_number_in_set(WTERMSIG(status), &signal_set)) {
2119 		printleader(tcp);
2120 #ifdef WCOREDUMP
2121 		tprintf("+++ killed by %s %s+++\n",
2122 			signame(WTERMSIG(status)),
2123 			WCOREDUMP(status) ? "(core dumped) " : "");
2124 #else
2125 		tprintf("+++ killed by %s +++\n",
2126 			signame(WTERMSIG(status)));
2127 #endif
2128 		line_ended();
2129 	}
2130 }
2131 
2132 static void
print_exited(struct tcb * tcp,const int pid,int status)2133 print_exited(struct tcb *tcp, const int pid, int status)
2134 {
2135 	if (pid == strace_child) {
2136 		exit_code = WEXITSTATUS(status);
2137 		strace_child = 0;
2138 	}
2139 
2140 	if (cflag != CFLAG_ONLY_STATS &&
2141 	    qflag < 2) {
2142 		printleader(tcp);
2143 		tprintf("+++ exited with %d +++\n", WEXITSTATUS(status));
2144 		line_ended();
2145 	}
2146 }
2147 
2148 static void
print_stopped(struct tcb * tcp,const siginfo_t * si,const unsigned int sig)2149 print_stopped(struct tcb *tcp, const siginfo_t *si, const unsigned int sig)
2150 {
2151 	if (cflag != CFLAG_ONLY_STATS
2152 	    && !hide_log(tcp)
2153 	    && is_number_in_set(sig, &signal_set)) {
2154 		printleader(tcp);
2155 		if (si) {
2156 			tprintf("--- %s ", signame(sig));
2157 			printsiginfo(si);
2158 			tprints(" ---\n");
2159 		} else
2160 			tprintf("--- stopped by %s ---\n", signame(sig));
2161 		line_ended();
2162 	}
2163 }
2164 
2165 static void
startup_tcb(struct tcb * tcp)2166 startup_tcb(struct tcb *tcp)
2167 {
2168 	if (debug_flag)
2169 		error_msg("pid %d has TCB_STARTUP, initializing it", tcp->pid);
2170 
2171 	tcp->flags &= ~TCB_STARTUP;
2172 
2173 	if (!use_seize) {
2174 		if (debug_flag)
2175 			error_msg("setting opts 0x%x on pid %d",
2176 				  ptrace_setoptions, tcp->pid);
2177 		if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) {
2178 			if (errno != ESRCH) {
2179 				/* Should never happen, really */
2180 				perror_msg_and_die("PTRACE_SETOPTIONS");
2181 			}
2182 		}
2183 	}
2184 }
2185 
2186 static void
print_event_exit(struct tcb * tcp)2187 print_event_exit(struct tcb *tcp)
2188 {
2189 	if (entering(tcp) || filtered(tcp) || hide_log(tcp)
2190 	    || cflag == CFLAG_ONLY_STATS) {
2191 		return;
2192 	}
2193 
2194 	if (followfork < 2 && printing_tcp && printing_tcp != tcp
2195 	    && printing_tcp->curcol != 0) {
2196 		current_tcp = printing_tcp;
2197 		tprints(" <unfinished ...>\n");
2198 		fflush(printing_tcp->outf);
2199 		printing_tcp->curcol = 0;
2200 		current_tcp = tcp;
2201 	}
2202 
2203 	if ((followfork < 2 && printing_tcp != tcp)
2204 	    || (tcp->flags & TCB_REPRINT)) {
2205 		tcp->flags &= ~TCB_REPRINT;
2206 		printleader(tcp);
2207 		tprintf("<... %s resumed>", tcp->s_ent->sys_name);
2208 	}
2209 
2210 	if (!(tcp->sys_func_rval & RVAL_DECODED)) {
2211 		/*
2212 		 * The decoder has probably decided to print something
2213 		 * on exiting syscall which is not going to happen.
2214 		 */
2215 		tprints(" <unfinished ...>");
2216 	}
2217 	tprints(") ");
2218 	tabto();
2219 	tprints("= ?\n");
2220 	line_ended();
2221 }
2222 
2223 /* Returns true iff the main trace loop has to continue. */
2224 static bool
trace(void)2225 trace(void)
2226 {
2227 	int pid;
2228 	int wait_errno;
2229 	int status;
2230 	bool stopped;
2231 	unsigned int sig;
2232 	unsigned int event;
2233 	struct tcb *tcp;
2234 	struct rusage ru;
2235 
2236 	if (interrupted)
2237 		return false;
2238 
2239 	/*
2240 	 * Used to exit simply when nprocs hits zero, but in this testcase:
2241 	 *  int main() { _exit(!!fork()); }
2242 	 * under strace -f, parent sometimes (rarely) manages
2243 	 * to exit before we see the first stop of the child,
2244 	 * and we are losing track of it:
2245 	 *  19923 clone(...) = 19924
2246 	 *  19923 exit_group(1)     = ?
2247 	 *  19923 +++ exited with 1 +++
2248 	 * Exiting only when wait() returns ECHILD works better.
2249 	 */
2250 	if (popen_pid != 0) {
2251 		/* However, if -o|logger is in use, we can't do that.
2252 		 * Can work around that by double-forking the logger,
2253 		 * but that loses the ability to wait for its completion
2254 		 * on exit. Oh well...
2255 		 */
2256 		if (nprocs == 0)
2257 			return false;
2258 	}
2259 
2260 	if (interactive)
2261 		sigprocmask(SIG_SETMASK, &empty_set, NULL);
2262 	pid = wait4(-1, &status, __WALL, (cflag ? &ru : NULL));
2263 	wait_errno = errno;
2264 	if (interactive)
2265 		sigprocmask(SIG_BLOCK, &blocked_set, NULL);
2266 
2267 	if (pid < 0) {
2268 		if (wait_errno == EINTR)
2269 			return true;
2270 		if (nprocs == 0 && wait_errno == ECHILD)
2271 			return false;
2272 		/*
2273 		 * If nprocs > 0, ECHILD is not expected,
2274 		 * treat it as any other error here:
2275 		 */
2276 		errno = wait_errno;
2277 		perror_msg_and_die("wait4(__WALL)");
2278 	}
2279 
2280 	if (pid == popen_pid) {
2281 		if (!WIFSTOPPED(status))
2282 			popen_pid = 0;
2283 		return true;
2284 	}
2285 
2286 	if (debug_flag)
2287 		print_debug_info(pid, status);
2288 
2289 	/* Look up 'pid' in our table. */
2290 	tcp = pid2tcb(pid);
2291 
2292 	if (!tcp) {
2293 		tcp = maybe_allocate_tcb(pid, status);
2294 		if (!tcp)
2295 			return true;
2296 	}
2297 
2298 	if (WIFSTOPPED(status))
2299 		get_regs(pid);
2300 	else
2301 		clear_regs();
2302 
2303 	event = (unsigned int) status >> 16;
2304 
2305 	if (event == PTRACE_EVENT_EXEC) {
2306 		/*
2307 		 * Under Linux, execve changes pid to thread leader's pid,
2308 		 * and we see this changed pid on EVENT_EXEC and later,
2309 		 * execve sysexit. Leader "disappears" without exit
2310 		 * notification. Let user know that, drop leader's tcb,
2311 		 * and fix up pid in execve thread's tcb.
2312 		 * Effectively, execve thread's tcb replaces leader's tcb.
2313 		 *
2314 		 * BTW, leader is 'stuck undead' (doesn't report WIFEXITED
2315 		 * on exit syscall) in multithreaded programs exactly
2316 		 * in order to handle this case.
2317 		 *
2318 		 * PTRACE_GETEVENTMSG returns old pid starting from Linux 3.0.
2319 		 * On 2.6 and earlier, it can return garbage.
2320 		 */
2321 		if (os_release >= KERNEL_VERSION(3,0,0))
2322 			tcp = maybe_switch_tcbs(tcp, pid);
2323 
2324 		if (detach_on_execve) {
2325 			if (tcp->flags & TCB_SKIP_DETACH_ON_FIRST_EXEC) {
2326 				tcp->flags &= ~TCB_SKIP_DETACH_ON_FIRST_EXEC;
2327 			} else {
2328 				detach(tcp); /* do "-b execve" thingy */
2329 				return true;
2330 			}
2331 		}
2332 	}
2333 
2334 	/* Set current output file */
2335 	current_tcp = tcp;
2336 
2337 	if (cflag) {
2338 		tv_sub(&tcp->dtime, &ru.ru_stime, &tcp->stime);
2339 		tcp->stime = ru.ru_stime;
2340 	}
2341 
2342 	if (WIFSIGNALED(status)) {
2343 		print_signalled(tcp, pid, status);
2344 		droptcb(tcp);
2345 		return true;
2346 	}
2347 
2348 	if (WIFEXITED(status)) {
2349 		print_exited(tcp, pid, status);
2350 		droptcb(tcp);
2351 		return true;
2352 	}
2353 
2354 	if (!WIFSTOPPED(status)) {
2355 		/*
2356 		 * Neither signalled, exited or stopped.
2357 		 * How could that be?
2358 		 */
2359 		error_msg("pid %u not stopped!", pid);
2360 		droptcb(tcp);
2361 		return true;
2362 	}
2363 
2364 	/* Is this the very first time we see this tracee stopped? */
2365 	if (tcp->flags & TCB_STARTUP) {
2366 		startup_tcb(tcp);
2367 		if (get_scno(tcp) == 1)
2368 			tcp->s_prev_ent = tcp->s_ent;
2369 	}
2370 
2371 	sig = WSTOPSIG(status);
2372 
2373 	switch (event) {
2374 		case 0:
2375 			break;
2376 		case PTRACE_EVENT_EXIT:
2377 			print_event_exit(tcp);
2378 			goto restart_tracee_with_sig_0;
2379 #if USE_SEIZE
2380 		case PTRACE_EVENT_STOP:
2381 			/*
2382 			 * PTRACE_INTERRUPT-stop or group-stop.
2383 			 * PTRACE_INTERRUPT-stop has sig == SIGTRAP here.
2384 			 */
2385 			switch (sig) {
2386 				case SIGSTOP:
2387 				case SIGTSTP:
2388 				case SIGTTIN:
2389 				case SIGTTOU:
2390 					stopped = true;
2391 					goto show_stopsig;
2392 			}
2393 			/* fall through */
2394 #endif
2395 		default:
2396 			goto restart_tracee_with_sig_0;
2397 	}
2398 
2399 	/*
2400 	 * Is this post-attach SIGSTOP?
2401 	 * Interestingly, the process may stop
2402 	 * with STOPSIG equal to some other signal
2403 	 * than SIGSTOP if we happend to attach
2404 	 * just before the process takes a signal.
2405 	 */
2406 	if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) {
2407 		if (debug_flag)
2408 			error_msg("ignored SIGSTOP on pid %d", tcp->pid);
2409 		tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP;
2410 		goto restart_tracee_with_sig_0;
2411 	}
2412 
2413 	if (sig != syscall_trap_sig) {
2414 		siginfo_t si = {};
2415 
2416 		/*
2417 		 * True if tracee is stopped by signal
2418 		 * (as opposed to "tracee received signal").
2419 		 * TODO: shouldn't we check for errno == EINVAL too?
2420 		 * We can get ESRCH instead, you know...
2421 		 */
2422 		stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, &si) < 0;
2423 #if USE_SEIZE
2424 show_stopsig:
2425 #endif
2426 		print_stopped(tcp, stopped ? NULL : &si, sig);
2427 
2428 		if (!stopped)
2429 			/* It's signal-delivery-stop. Inject the signal */
2430 			goto restart_tracee;
2431 
2432 		/* It's group-stop */
2433 		if (use_seize) {
2434 			/*
2435 			 * This ends ptrace-stop, but does *not* end group-stop.
2436 			 * This makes stopping signals work properly on straced process
2437 			 * (that is, process really stops. It used to continue to run).
2438 			 */
2439 			if (ptrace_restart(PTRACE_LISTEN, tcp, 0) < 0) {
2440 				/* Note: ptrace_restart emitted error message */
2441 				exit_code = 1;
2442 				return false;
2443 			}
2444 			return true;
2445 		}
2446 		/* We don't have PTRACE_LISTEN support... */
2447 		goto restart_tracee;
2448 	}
2449 
2450 	/* We handled quick cases, we are permitted to interrupt now. */
2451 	if (interrupted)
2452 		return false;
2453 
2454 	/*
2455 	 * This should be syscall entry or exit.
2456 	 * Handle it.
2457 	 */
2458 	sig = 0;
2459 	if (trace_syscall(tcp, &sig) < 0) {
2460 		/*
2461 		 * ptrace() failed in trace_syscall().
2462 		 * Likely a result of process disappearing mid-flight.
2463 		 * Observed case: exit_group() or SIGKILL terminating
2464 		 * all processes in thread group.
2465 		 * We assume that ptrace error was caused by process death.
2466 		 * We used to detach(tcp) here, but since we no longer
2467 		 * implement "detach before death" policy/hack,
2468 		 * we can let this process to report its death to us
2469 		 * normally, via WIFEXITED or WIFSIGNALED wait status.
2470 		 */
2471 		return true;
2472 	}
2473 	goto restart_tracee;
2474 
2475 restart_tracee_with_sig_0:
2476 	sig = 0;
2477 
2478 restart_tracee:
2479 	if (ptrace_restart(PTRACE_SYSCALL, tcp, sig) < 0) {
2480 		/* Note: ptrace_restart emitted error message */
2481 		exit_code = 1;
2482 		return false;
2483 	}
2484 
2485 	return true;
2486 }
2487 
2488 int
main(int argc,char * argv[])2489 main(int argc, char *argv[])
2490 {
2491 	init(argc, argv);
2492 
2493 	exit_code = !nprocs;
2494 
2495 	while (trace())
2496 		;
2497 
2498 	cleanup();
2499 	fflush(NULL);
2500 	if (shared_log != stderr)
2501 		fclose(shared_log);
2502 	if (popen_pid) {
2503 		while (waitpid(popen_pid, NULL, 0) < 0 && errno == EINTR)
2504 			;
2505 	}
2506 	if (exit_code > 0xff) {
2507 		/* Avoid potential core file clobbering.  */
2508 		struct_rlimit rlim = {0, 0};
2509 		set_rlimit(RLIMIT_CORE, &rlim);
2510 
2511 		/* Child was killed by a signal, mimic that.  */
2512 		exit_code &= 0xff;
2513 		signal(exit_code, SIG_DFL);
2514 		raise(exit_code);
2515 		/* Paranoia - what if this signal is not fatal?
2516 		   Exit with 128 + signo then.  */
2517 		exit_code += 128;
2518 	}
2519 
2520 	return exit_code;
2521 }
2522