• 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 <locale.h>
48 #include <sys/utsname.h>
49 #ifdef HAVE_PRCTL
50 # include <sys/prctl.h>
51 #endif
52 #include <asm/unistd.h>
53 
54 #include "largefile_wrappers.h"
55 #include "mmap_cache.h"
56 #include "number_set.h"
57 #include "scno.h"
58 #include "printsiginfo.h"
59 #include "trace_event.h"
60 #include "xstring.h"
61 #include "delay.h"
62 
63 /* In some libc, these aren't declared. Do it ourself: */
64 extern char **environ;
65 extern int optind;
66 extern char *optarg;
67 
68 #ifdef ENABLE_STACKTRACE
69 /* if this is true do the stack trace for every system call */
70 bool stack_trace_enabled;
71 #endif
72 
73 #define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
74 
75 /* Glue for systems without a MMU that cannot provide fork() */
76 #if !defined(HAVE_FORK)
77 # undef NOMMU_SYSTEM
78 # define NOMMU_SYSTEM 1
79 #endif
80 #if NOMMU_SYSTEM
81 # define fork() vfork()
82 #endif
83 
84 const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
85 
86 cflag_t cflag = CFLAG_NONE;
87 unsigned int followfork;
88 unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC
89 				 | PTRACE_O_TRACEEXIT;
90 unsigned int xflag;
91 bool debug_flag;
92 bool Tflag;
93 bool iflag;
94 bool count_wallclock;
95 unsigned int qflag;
96 static unsigned int tflag;
97 static bool rflag;
98 static bool print_pid_pfx;
99 
100 /* -I n */
101 enum {
102 	INTR_NOT_SET        = 0,
103 	INTR_ANYWHERE       = 1, /* don't block/ignore any signals */
104 	INTR_WHILE_WAIT     = 2, /* block fatal signals while decoding syscall. default */
105 	INTR_NEVER          = 3, /* block fatal signals. default if '-o FILE PROG' */
106 	INTR_BLOCK_TSTP_TOO = 4, /* block fatal signals and SIGTSTP (^Z) */
107 	NUM_INTR_OPTS
108 };
109 static int opt_intr;
110 /* We play with signal mask only if this mode is active: */
111 #define interactive (opt_intr == INTR_WHILE_WAIT)
112 
113 /*
114  * daemonized_tracer supports -D option.
115  * With this option, strace forks twice.
116  * Unlike normal case, with -D *grandparent* process exec's,
117  * becoming a traced process. Child exits (this prevents traced process
118  * from having children it doesn't expect to have), and grandchild
119  * attaches to grandparent similarly to strace -p PID.
120  * This allows for more transparent interaction in cases
121  * when process and its parent are communicating via signals,
122  * wait() etc. Without -D, strace process gets lodged in between,
123  * disrupting parent<->child link.
124  */
125 static bool daemonized_tracer;
126 
127 static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP;
128 #define use_seize (post_attach_sigstop == 0)
129 
130 /* Sometimes we want to print only succeeding syscalls. */
131 bool not_failing_only;
132 
133 /* Show path associated with fd arguments */
134 unsigned int show_fd_path;
135 
136 static bool detach_on_execve;
137 
138 static int exit_code;
139 static int strace_child;
140 static int strace_tracer_pid;
141 
142 static const char *username;
143 static uid_t run_uid;
144 static gid_t run_gid;
145 
146 unsigned int max_strlen = DEFAULT_STRLEN;
147 static int acolumn = DEFAULT_ACOLUMN;
148 static char *acolumn_spaces;
149 
150 /* Default output style for xlat entities */
151 enum xlat_style xlat_verbosity = XLAT_STYLE_ABBREV;
152 
153 static const char *outfname;
154 /* If -ff, points to stderr. Else, it's our common output log */
155 static FILE *shared_log;
156 static bool open_append;
157 
158 struct tcb *printing_tcp;
159 static struct tcb *current_tcp;
160 
161 struct tcb_wait_data {
162 	enum trace_event te; /**< Event passed to dispatch_event() */
163 	int status;          /**< status, returned by wait4() */
164 	siginfo_t si;        /**< siginfo, returned by PTRACE_GETSIGINFO */
165 };
166 
167 static struct tcb **tcbtab;
168 static unsigned int nprocs;
169 static size_t tcbtabsize;
170 
171 #ifndef HAVE_PROGRAM_INVOCATION_NAME
172 char *program_invocation_name;
173 #endif
174 
175 unsigned os_release; /* generated from uname()'s u.release */
176 
177 static void detach(struct tcb *tcp);
178 static void cleanup(void);
179 static void interrupt(int sig);
180 
181 #ifdef HAVE_SIG_ATOMIC_T
182 static volatile sig_atomic_t interrupted, restart_failed;
183 #else
184 static volatile int interrupted, restart_failed;
185 #endif
186 
187 static sigset_t timer_set;
188 static void timer_sighandler(int);
189 
190 #ifndef HAVE_STRERROR
191 
192 #if !HAVE_DECL_SYS_ERRLIST
193 extern int sys_nerr;
194 extern char *sys_errlist[];
195 #endif
196 
197 const char *
strerror(int err_no)198 strerror(int err_no)
199 {
200 	static char buf[sizeof("Unknown error %d") + sizeof(int)*3];
201 
202 	if (err_no < 1 || err_no >= sys_nerr) {
203 		xsprintf(buf, "Unknown error %d", err_no);
204 		return buf;
205 	}
206 	return sys_errlist[err_no];
207 }
208 
209 #endif /* HAVE_STERRROR */
210 
211 static void
print_version(void)212 print_version(void)
213 {
214 	static const char features[] =
215 #ifdef ENABLE_STACKTRACE
216 		" stack-trace=" USE_UNWINDER
217 #endif
218 #ifdef USE_DEMANGLE
219 		" stack-demangle"
220 #endif
221 #if SUPPORTED_PERSONALITIES > 1
222 # if defined HAVE_M32_MPERS
223 		" m32-mpers"
224 # else
225 		" no-m32-mpers"
226 # endif
227 #endif /* SUPPORTED_PERSONALITIES > 1 */
228 #if SUPPORTED_PERSONALITIES > 2
229 # if defined HAVE_MX32_MPERS
230 		" mx32-mpers"
231 # else
232 		" no-mx32-mpers"
233 # endif
234 #endif /* SUPPORTED_PERSONALITIES > 2 */
235 		"";
236 
237 	printf("%s -- version %s\n"
238 	       "Copyright (c) 1991-%s The strace developers <%s>.\n"
239 	       "This is free software; see the source for copying conditions.  There is NO\n"
240 	       "warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n",
241 	       PACKAGE_NAME, PACKAGE_VERSION, COPYRIGHT_YEAR, PACKAGE_URL);
242 	printf("\nOptional features enabled:%s\n",
243 	       features[0] ? features : " (none)");
244 }
245 
246 static void
usage(void)247 usage(void)
248 {
249 	printf("\
250 usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
251               [-a column] [-o file] [-s strsize] [-P path]...\n\
252               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
253    or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
254               -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
255 \n\
256 Output format:\n\
257   -a column      alignment COLUMN for printing syscall results (default %d)\n\
258   -i             print instruction pointer at time of syscall\n\
259 "
260 #ifdef ENABLE_STACKTRACE
261 "\
262   -k             obtain stack trace between each syscall\n\
263 "
264 #endif
265 "\
266   -o file        send trace output to FILE instead of stderr\n\
267   -q             suppress messages about attaching, detaching, etc.\n\
268   -r             print relative timestamp\n\
269   -s strsize     limit length of print strings to STRSIZE chars (default %d)\n\
270   -t             print absolute timestamp\n\
271   -tt            print absolute timestamp with usecs\n\
272   -T             print time spent in each syscall\n\
273   -x             print non-ascii strings in hex\n\
274   -xx            print all strings in hex\n\
275   -y             print paths associated with file descriptor arguments\n\
276   -yy            print protocol specific information associated with socket file descriptors\n\
277 \n\
278 Statistics:\n\
279   -c             count time, calls, and errors for each syscall and report summary\n\
280   -C             like -c but also print regular output\n\
281   -O overhead    set overhead for tracing syscalls to OVERHEAD usecs\n\
282   -S sortby      sort syscall counts by: time, calls, name, nothing (default %s)\n\
283   -w             summarise syscall latency (default is system time)\n\
284 \n\
285 Filtering:\n\
286   -e expr        a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
287      options:    trace, abbrev, verbose, raw, signal, read, write, fault, inject, kvm"
288 "\n\
289   -P path        trace accesses to path\n\
290 \n\
291 Tracing:\n\
292   -b execve      detach on execve syscall\n\
293   -D             run tracer process as a detached grandchild, not as parent\n\
294   -f             follow forks\n\
295   -ff            follow forks with output into separate files\n\
296   -I interruptible\n\
297      1:          no signals are blocked\n\
298      2:          fatal signals are blocked while decoding syscall (default)\n\
299      3:          fatal signals are always blocked (default if '-o FILE PROG')\n\
300      4:          fatal signals and SIGTSTP (^Z) are always blocked\n\
301                  (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
302 \n\
303 Startup:\n\
304   -E var         remove var from the environment for command\n\
305   -E var=val     put var=val in the environment for command\n\
306   -p pid         trace process with process id PID, may be repeated\n\
307   -u username    run command as username handling setuid and/or setgid\n\
308 \n\
309 Miscellaneous:\n\
310   -d             enable debug output to stderr\n\
311   -v             verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
312   -h             print help message\n\
313   -V             print version\n\
314 "
315 /* ancient, no one should use it
316 -F -- attempt to follow vforks (deprecated, use -f)\n\
317  */
318 /* this is broken, so don't document it
319 -z -- print only succeeding syscalls\n\
320  */
321 , DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
322 	exit(0);
323 }
324 
325 void ATTRIBUTE_NORETURN
die(void)326 die(void)
327 {
328 	if (strace_tracer_pid == getpid()) {
329 		cflag = 0;
330 		cleanup();
331 		exit(1);
332 	}
333 
334 	_exit(1);
335 }
336 
337 static void
error_opt_arg(int opt,const char * arg)338 error_opt_arg(int opt, const char *arg)
339 {
340 	error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
341 }
342 
343 static const char *ptrace_attach_cmd;
344 
345 static int
ptrace_attach_or_seize(int pid)346 ptrace_attach_or_seize(int pid)
347 {
348 	int r;
349 	if (!use_seize)
350 		return ptrace_attach_cmd = "PTRACE_ATTACH",
351 		       ptrace(PTRACE_ATTACH, pid, 0L, 0L);
352 	r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
353 	if (r)
354 		return ptrace_attach_cmd = "PTRACE_SEIZE", r;
355 	r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
356 	return ptrace_attach_cmd = "PTRACE_INTERRUPT", r;
357 }
358 
359 /*
360  * Used when we want to unblock stopped traced process.
361  * Should be only used with PTRACE_CONT, PTRACE_DETACH and PTRACE_SYSCALL.
362  * Returns 0 on success or if error was ESRCH
363  * (presumably process was killed while we talk to it).
364  * Otherwise prints error message and returns -1.
365  */
366 static int
ptrace_restart(const unsigned int op,struct tcb * const tcp,unsigned int sig)367 ptrace_restart(const unsigned int op, struct tcb *const tcp, unsigned int sig)
368 {
369 	int err;
370 	const char *msg;
371 
372 	errno = 0;
373 	ptrace(op, tcp->pid, 0L, (unsigned long) sig);
374 	err = errno;
375 	if (!err)
376 		return 0;
377 
378 	switch (op) {
379 		case PTRACE_CONT:
380 			msg = "CONT";
381 			break;
382 		case PTRACE_DETACH:
383 			msg = "DETACH";
384 			break;
385 		case PTRACE_LISTEN:
386 			msg = "LISTEN";
387 			break;
388 		default:
389 			msg = "SYSCALL";
390 	}
391 
392 	/*
393 	 * Why curcol != 0? Otherwise sometimes we get this:
394 	 *
395 	 * 10252 kill(10253, SIGKILL)              = 0
396 	 *  <ptrace(SYSCALL,10252):No such process>10253 ...next decode...
397 	 *
398 	 * 10252 died after we retrieved syscall exit data,
399 	 * but before we tried to restart it. Log looks ugly.
400 	 */
401 	if (current_tcp && current_tcp->curcol != 0) {
402 		tprintf(" <ptrace(%s):%s>\n", msg, strerror(err));
403 		line_ended();
404 	}
405 	if (err == ESRCH)
406 		return 0;
407 	errno = err;
408 	perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%u)", msg, tcp->pid, sig);
409 	return -1;
410 }
411 
412 static void
set_cloexec_flag(int fd)413 set_cloexec_flag(int fd)
414 {
415 	int flags, newflags;
416 
417 	flags = fcntl(fd, F_GETFD);
418 	if (flags < 0) {
419 		/* Can happen only if fd is bad.
420 		 * Should never happen: if it does, we have a bug
421 		 * in the caller. Therefore we just abort
422 		 * instead of propagating the error.
423 		 */
424 		perror_msg_and_die("fcntl(%d, F_GETFD)", fd);
425 	}
426 
427 	newflags = flags | FD_CLOEXEC;
428 	if (flags == newflags)
429 		return;
430 
431 	if (fcntl(fd, F_SETFD, newflags)) /* never fails */
432 		perror_msg_and_die("fcntl(%d, F_SETFD, %#x)", fd, newflags);
433 }
434 
435 static void
kill_save_errno(pid_t pid,int sig)436 kill_save_errno(pid_t pid, int sig)
437 {
438 	int saved_errno = errno;
439 
440 	(void) kill(pid, sig);
441 	errno = saved_errno;
442 }
443 
444 /*
445  * When strace is setuid executable, we have to swap uids
446  * before and after filesystem and process management operations.
447  */
448 static void
swap_uid(void)449 swap_uid(void)
450 {
451 	int euid = geteuid(), uid = getuid();
452 
453 	if (euid != uid && setreuid(euid, uid) < 0) {
454 		perror_msg_and_die("setreuid");
455 	}
456 }
457 
458 static FILE *
strace_fopen(const char * path)459 strace_fopen(const char *path)
460 {
461 	FILE *fp;
462 
463 	swap_uid();
464 	fp = fopen_stream(path, open_append ? "a" : "w");
465 	if (!fp)
466 		perror_msg_and_die("Can't fopen '%s'", path);
467 	swap_uid();
468 	set_cloexec_flag(fileno(fp));
469 	return fp;
470 }
471 
472 static int popen_pid;
473 
474 #ifndef _PATH_BSHELL
475 # define _PATH_BSHELL "/bin/sh"
476 #endif
477 
478 /*
479  * We cannot use standard popen(3) here because we have to distinguish
480  * popen child process from other processes we trace, and standard popen(3)
481  * does not export its child's pid.
482  */
483 static FILE *
strace_popen(const char * command)484 strace_popen(const char *command)
485 {
486 	FILE *fp;
487 	int pid;
488 	int fds[2];
489 
490 	swap_uid();
491 	if (pipe(fds) < 0)
492 		perror_msg_and_die("pipe");
493 
494 	set_cloexec_flag(fds[1]); /* never fails */
495 
496 	pid = vfork();
497 	if (pid < 0)
498 		perror_msg_and_die("vfork");
499 
500 	if (pid == 0) {
501 		/* child */
502 		close(fds[1]);
503 		if (fds[0] != 0) {
504 			if (dup2(fds[0], 0))
505 				perror_msg_and_die("dup2");
506 			close(fds[0]);
507 		}
508 		execl(_PATH_BSHELL, "sh", "-c", command, NULL);
509 		perror_msg_and_die("Can't execute '%s'", _PATH_BSHELL);
510 	}
511 
512 	/* parent */
513 	popen_pid = pid;
514 	close(fds[0]);
515 	swap_uid();
516 	fp = fdopen(fds[1], "w");
517 	if (!fp)
518 		perror_msg_and_die("fdopen");
519 	return fp;
520 }
521 
522 static void
outf_perror(const struct tcb * const tcp)523 outf_perror(const struct tcb * const tcp)
524 {
525 	if (tcp->outf == stderr)
526 		return;
527 
528 	/* This is ugly, but we don't store separate file names */
529 	if (followfork >= 2)
530 		perror_msg("%s.%u", outfname, tcp->pid);
531 	else
532 		perror_msg("%s", outfname);
533 }
534 
535 ATTRIBUTE_FORMAT((printf, 1, 0))
536 static void
tvprintf(const char * const fmt,va_list args)537 tvprintf(const char *const fmt, va_list args)
538 {
539 	if (current_tcp) {
540 		int n = vfprintf(current_tcp->outf, fmt, args);
541 		if (n < 0) {
542 			/* very unlikely due to vfprintf buffering */
543 			outf_perror(current_tcp);
544 		} else
545 			current_tcp->curcol += n;
546 	}
547 }
548 
549 void
tprintf(const char * fmt,...)550 tprintf(const char *fmt, ...)
551 {
552 	va_list args;
553 	va_start(args, fmt);
554 	tvprintf(fmt, args);
555 	va_end(args);
556 }
557 
558 #ifndef HAVE_FPUTS_UNLOCKED
559 # define fputs_unlocked fputs
560 #endif
561 
562 void
tprints(const char * str)563 tprints(const char *str)
564 {
565 	if (current_tcp) {
566 		int n = fputs_unlocked(str, current_tcp->outf);
567 		if (n >= 0) {
568 			current_tcp->curcol += strlen(str);
569 			return;
570 		}
571 		/* very unlikely due to fputs_unlocked buffering */
572 		outf_perror(current_tcp);
573 	}
574 }
575 
576 void
tprints_comment(const char * const str)577 tprints_comment(const char *const str)
578 {
579 	if (str && *str)
580 		tprintf(" /* %s */", str);
581 }
582 
583 void
tprintf_comment(const char * fmt,...)584 tprintf_comment(const char *fmt, ...)
585 {
586 	if (!fmt || !*fmt)
587 		return;
588 
589 	va_list args;
590 	va_start(args, fmt);
591 	tprints(" /* ");
592 	tvprintf(fmt, args);
593 	tprints(" */");
594 	va_end(args);
595 }
596 
597 static void
flush_tcp_output(const struct tcb * const tcp)598 flush_tcp_output(const struct tcb *const tcp)
599 {
600 	if (fflush(tcp->outf))
601 		outf_perror(tcp);
602 }
603 
604 void
line_ended(void)605 line_ended(void)
606 {
607 	if (current_tcp) {
608 		current_tcp->curcol = 0;
609 		flush_tcp_output(current_tcp);
610 	}
611 	if (printing_tcp) {
612 		printing_tcp->curcol = 0;
613 		printing_tcp = NULL;
614 	}
615 }
616 
617 void
set_current_tcp(const struct tcb * tcp)618 set_current_tcp(const struct tcb *tcp)
619 {
620 	current_tcp = (struct tcb *) tcp;
621 
622 	/* Sync current_personality and stuff */
623 	if (current_tcp)
624 		set_personality(current_tcp->currpers);
625 }
626 
627 void
printleader(struct tcb * tcp)628 printleader(struct tcb *tcp)
629 {
630 	/* If -ff, "previous tcb we printed" is always the same as current,
631 	 * because we have per-tcb output files.
632 	 */
633 	if (followfork >= 2)
634 		printing_tcp = tcp;
635 
636 	if (printing_tcp) {
637 		set_current_tcp(printing_tcp);
638 		if (printing_tcp->curcol != 0 && (followfork < 2 || printing_tcp == tcp)) {
639 			/*
640 			 * case 1: we have a shared log (i.e. not -ff), and last line
641 			 * wasn't finished (same or different tcb, doesn't matter).
642 			 * case 2: split log, we are the same tcb, but our last line
643 			 * didn't finish ("SIGKILL nuked us after syscall entry" etc).
644 			 */
645 			tprints(" <unfinished ...>\n");
646 			printing_tcp->curcol = 0;
647 		}
648 	}
649 
650 	printing_tcp = tcp;
651 	set_current_tcp(tcp);
652 	current_tcp->curcol = 0;
653 
654 	if (print_pid_pfx)
655 		tprintf("%-5d ", tcp->pid);
656 	else if (nprocs > 1 && !outfname)
657 		tprintf("[pid %5u] ", tcp->pid);
658 
659 	if (tflag) {
660 		struct timespec ts;
661 		clock_gettime(CLOCK_REALTIME, &ts);
662 
663 		if (tflag > 2) {
664 			tprintf("%lld.%06ld ",
665 				(long long) ts.tv_sec, (long) ts.tv_nsec / 1000);
666 		} else {
667 			time_t local = ts.tv_sec;
668 			char str[MAX(sizeof("HH:MM:SS"), sizeof(ts.tv_sec) * 3)];
669 			struct tm *tm = localtime(&local);
670 
671 			if (tm)
672 				strftime(str, sizeof(str), "%T", tm);
673 			else
674 				xsprintf(str, "%lld", (long long) local);
675 			if (tflag > 1)
676 				tprintf("%s.%06ld ",
677 					str, (long) ts.tv_nsec / 1000);
678 			else
679 				tprintf("%s ", str);
680 		}
681 	}
682 
683 	if (rflag) {
684 		struct timespec ts;
685 		clock_gettime(CLOCK_MONOTONIC, &ts);
686 
687 		static struct timespec ots;
688 		if (ots.tv_sec == 0)
689 			ots = ts;
690 
691 		struct timespec dts;
692 		ts_sub(&dts, &ts, &ots);
693 		ots = ts;
694 
695 		tprintf("%s%6ld.%06ld%s ",
696 			tflag ? "(+" : "",
697 			(long) dts.tv_sec, (long) dts.tv_nsec / 1000,
698 			tflag ? ")" : "");
699 	}
700 
701 	if (iflag)
702 		print_pc(tcp);
703 }
704 
705 void
tabto(void)706 tabto(void)
707 {
708 	if (current_tcp->curcol < acolumn)
709 		tprints(acolumn_spaces + current_tcp->curcol);
710 }
711 
712 /* Should be only called directly *after successful attach* to a tracee.
713  * Otherwise, "strace -oFILE -ff -p<nonexistant_pid>"
714  * may create bogus empty FILE.<nonexistant_pid>, and then die.
715  */
716 static void
after_successful_attach(struct tcb * tcp,const unsigned int flags)717 after_successful_attach(struct tcb *tcp, const unsigned int flags)
718 {
719 	tcp->flags |= TCB_ATTACHED | TCB_STARTUP | flags;
720 	tcp->outf = shared_log; /* if not -ff mode, the same file is for all */
721 	if (followfork >= 2) {
722 		char name[PATH_MAX];
723 		xsprintf(name, "%s.%u", outfname, tcp->pid);
724 		tcp->outf = strace_fopen(name);
725 	}
726 
727 #ifdef ENABLE_STACKTRACE
728 	if (stack_trace_enabled)
729 		unwind_tcb_init(tcp);
730 #endif
731 }
732 
733 static void
expand_tcbtab(void)734 expand_tcbtab(void)
735 {
736 	/* Allocate some (more) TCBs (and expand the table).
737 	   We don't want to relocate the TCBs because our
738 	   callers have pointers and it would be a pain.
739 	   So tcbtab is a table of pointers.  Since we never
740 	   free the TCBs, we allocate a single chunk of many.  */
741 	size_t old_tcbtabsize;
742 	struct tcb *newtcbs;
743 	struct tcb **tcb_ptr;
744 
745 	old_tcbtabsize = tcbtabsize;
746 
747 	tcbtab = xgrowarray(tcbtab, &tcbtabsize, sizeof(tcbtab[0]));
748 	newtcbs = xcalloc(tcbtabsize - old_tcbtabsize, sizeof(newtcbs[0]));
749 
750 	for (tcb_ptr = tcbtab + old_tcbtabsize;
751 	    tcb_ptr < tcbtab + tcbtabsize; tcb_ptr++, newtcbs++)
752 		*tcb_ptr = newtcbs;
753 }
754 
755 static struct tcb *
alloctcb(int pid)756 alloctcb(int pid)
757 {
758 	unsigned int i;
759 	struct tcb *tcp;
760 
761 	if (nprocs == tcbtabsize)
762 		expand_tcbtab();
763 
764 	for (i = 0; i < tcbtabsize; i++) {
765 		tcp = tcbtab[i];
766 		if (!tcp->pid) {
767 			memset(tcp, 0, sizeof(*tcp));
768 			tcp->pid = pid;
769 #if SUPPORTED_PERSONALITIES > 1
770 			tcp->currpers = current_personality;
771 #endif
772 			nprocs++;
773 			debug_msg("new tcb for pid %d, active tcbs:%d",
774 				  tcp->pid, nprocs);
775 			return tcp;
776 		}
777 	}
778 	error_msg_and_die("bug in alloctcb");
779 }
780 
781 void *
get_tcb_priv_data(const struct tcb * tcp)782 get_tcb_priv_data(const struct tcb *tcp)
783 {
784 	return tcp->_priv_data;
785 }
786 
787 int
set_tcb_priv_data(struct tcb * tcp,void * const priv_data,void (* const free_priv_data)(void *))788 set_tcb_priv_data(struct tcb *tcp, void *const priv_data,
789 		  void (*const free_priv_data)(void *))
790 {
791 	if (tcp->_priv_data)
792 		return -1;
793 
794 	tcp->_free_priv_data = free_priv_data;
795 	tcp->_priv_data = priv_data;
796 
797 	return 0;
798 }
799 
800 void
free_tcb_priv_data(struct tcb * tcp)801 free_tcb_priv_data(struct tcb *tcp)
802 {
803 	if (tcp->_priv_data) {
804 		if (tcp->_free_priv_data) {
805 			tcp->_free_priv_data(tcp->_priv_data);
806 			tcp->_free_priv_data = NULL;
807 		}
808 		tcp->_priv_data = NULL;
809 	}
810 }
811 
812 static void
droptcb(struct tcb * tcp)813 droptcb(struct tcb *tcp)
814 {
815 	if (tcp->pid == 0)
816 		return;
817 
818 	int p;
819 	for (p = 0; p < SUPPORTED_PERSONALITIES; ++p)
820 		free(tcp->inject_vec[p]);
821 
822 	free_tcb_priv_data(tcp);
823 
824 #ifdef ENABLE_STACKTRACE
825 	if (stack_trace_enabled)
826 		unwind_tcb_fin(tcp);
827 #endif
828 
829 #ifdef HAVE_LINUX_KVM_H
830 	kvm_vcpu_info_free(tcp);
831 #endif
832 
833 	if (tcp->mmap_cache)
834 		tcp->mmap_cache->free_fn(tcp, __func__);
835 
836 	nprocs--;
837 	debug_msg("dropped tcb for pid %d, %d remain", tcp->pid, nprocs);
838 
839 	if (tcp->outf) {
840 		if (followfork >= 2) {
841 			if (tcp->curcol != 0)
842 				fprintf(tcp->outf, " <detached ...>\n");
843 			fclose(tcp->outf);
844 		} else {
845 			if (printing_tcp == tcp && tcp->curcol != 0)
846 				fprintf(tcp->outf, " <detached ...>\n");
847 			flush_tcp_output(tcp);
848 		}
849 	}
850 
851 	if (current_tcp == tcp)
852 		set_current_tcp(NULL);
853 	if (printing_tcp == tcp)
854 		printing_tcp = NULL;
855 
856 	memset(tcp, 0, sizeof(*tcp));
857 }
858 
859 /* Detach traced process.
860  * Never call DETACH twice on the same process as both unattached and
861  * attached-unstopped processes give the same ESRCH.  For unattached process we
862  * would SIGSTOP it and wait for its SIGSTOP notification forever.
863  */
864 static void
detach(struct tcb * tcp)865 detach(struct tcb *tcp)
866 {
867 	int error;
868 	int status;
869 
870 	/*
871 	 * Linux wrongly insists the child be stopped
872 	 * before detaching.  Arghh.  We go through hoops
873 	 * to make a clean break of things.
874 	 */
875 
876 	if (!(tcp->flags & TCB_ATTACHED))
877 		goto drop;
878 
879 	/* We attached but possibly didn't see the expected SIGSTOP.
880 	 * We must catch exactly one as otherwise the detached process
881 	 * would be left stopped (process state T).
882 	 */
883 	if (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)
884 		goto wait_loop;
885 
886 	error = ptrace(PTRACE_DETACH, tcp->pid, 0, 0);
887 	if (!error) {
888 		/* On a clear day, you can see forever. */
889 		goto drop;
890 	}
891 	if (errno != ESRCH) {
892 		/* Shouldn't happen. */
893 		perror_func_msg("ptrace(PTRACE_DETACH,%u)", tcp->pid);
894 		goto drop;
895 	}
896 	/* ESRCH: process is either not stopped or doesn't exist. */
897 	if (my_tkill(tcp->pid, 0) < 0) {
898 		if (errno != ESRCH)
899 			/* Shouldn't happen. */
900 			perror_func_msg("tkill(%u,0)", tcp->pid);
901 		/* else: process doesn't exist. */
902 		goto drop;
903 	}
904 	/* Process is not stopped, need to stop it. */
905 	if (use_seize) {
906 		/*
907 		 * With SEIZE, tracee can be in group-stop already.
908 		 * In this state sending it another SIGSTOP does nothing.
909 		 * Need to use INTERRUPT.
910 		 * Testcase: trying to ^C a "strace -p <stopped_process>".
911 		 */
912 		error = ptrace(PTRACE_INTERRUPT, tcp->pid, 0, 0);
913 		if (!error)
914 			goto wait_loop;
915 		if (errno != ESRCH)
916 			perror_func_msg("ptrace(PTRACE_INTERRUPT,%u)", tcp->pid);
917 	} else {
918 		error = my_tkill(tcp->pid, SIGSTOP);
919 		if (!error)
920 			goto wait_loop;
921 		if (errno != ESRCH)
922 			perror_func_msg("tkill(%u,SIGSTOP)", tcp->pid);
923 	}
924 	/* Either process doesn't exist, or some weird error. */
925 	goto drop;
926 
927  wait_loop:
928 	/* We end up here in three cases:
929 	 * 1. We sent PTRACE_INTERRUPT (use_seize case)
930 	 * 2. We sent SIGSTOP (!use_seize)
931 	 * 3. Attach SIGSTOP was already pending (TCB_IGNORE_ONE_SIGSTOP set)
932 	 */
933 	for (;;) {
934 		unsigned int sig;
935 		if (waitpid(tcp->pid, &status, __WALL) < 0) {
936 			if (errno == EINTR)
937 				continue;
938 			/*
939 			 * if (errno == ECHILD) break;
940 			 * ^^^  WRONG! We expect this PID to exist,
941 			 * and want to emit a message otherwise:
942 			 */
943 			perror_func_msg("waitpid(%u)", tcp->pid);
944 			break;
945 		}
946 		if (!WIFSTOPPED(status)) {
947 			/*
948 			 * Tracee exited or was killed by signal.
949 			 * We shouldn't normally reach this place:
950 			 * we don't want to consume exit status.
951 			 * Consider "strace -p PID" being ^C-ed:
952 			 * we want merely to detach from PID.
953 			 *
954 			 * However, we _can_ end up here if tracee
955 			 * was SIGKILLed.
956 			 */
957 			break;
958 		}
959 		sig = WSTOPSIG(status);
960 		debug_msg("detach wait: event:%d sig:%d",
961 			  (unsigned) status >> 16, sig);
962 		if (use_seize) {
963 			unsigned event = (unsigned)status >> 16;
964 			if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) {
965 				/*
966 				 * sig == SIGTRAP: PTRACE_INTERRUPT stop.
967 				 * sig == other: process was already stopped
968 				 * with this stopping sig (see tests/detach-stopped).
969 				 * Looks like re-injecting this sig is not necessary
970 				 * in DETACH for the tracee to remain stopped.
971 				 */
972 				sig = 0;
973 			}
974 			/*
975 			 * PTRACE_INTERRUPT is not guaranteed to produce
976 			 * the above event if other ptrace-stop is pending.
977 			 * See tests/detach-sleeping testcase:
978 			 * strace got SIGINT while tracee is sleeping.
979 			 * We sent PTRACE_INTERRUPT.
980 			 * We see syscall exit, not PTRACE_INTERRUPT stop.
981 			 * We won't get PTRACE_INTERRUPT stop
982 			 * if we would CONT now. Need to DETACH.
983 			 */
984 			if (sig == syscall_trap_sig)
985 				sig = 0;
986 			/* else: not sure in which case we can be here.
987 			 * Signal stop? Inject it while detaching.
988 			 */
989 			ptrace_restart(PTRACE_DETACH, tcp, sig);
990 			break;
991 		}
992 		/* Note: this check has to be after use_seize check */
993 		/* (else, in use_seize case SIGSTOP will be mistreated) */
994 		if (sig == SIGSTOP) {
995 			/* Detach, suppressing SIGSTOP */
996 			ptrace_restart(PTRACE_DETACH, tcp, 0);
997 			break;
998 		}
999 		if (sig == syscall_trap_sig)
1000 			sig = 0;
1001 		/* Can't detach just yet, may need to wait for SIGSTOP */
1002 		error = ptrace_restart(PTRACE_CONT, tcp, sig);
1003 		if (error < 0) {
1004 			/* Should not happen.
1005 			 * Note: ptrace_restart returns 0 on ESRCH, so it's not it.
1006 			 * ptrace_restart already emitted error message.
1007 			 */
1008 			break;
1009 		}
1010 	}
1011 
1012  drop:
1013 	if (!qflag && (tcp->flags & TCB_ATTACHED))
1014 		error_msg("Process %u detached", tcp->pid);
1015 
1016 	droptcb(tcp);
1017 }
1018 
1019 static void
process_opt_p_list(char * opt)1020 process_opt_p_list(char *opt)
1021 {
1022 	while (*opt) {
1023 		/*
1024 		 * We accept -p PID,PID; -p "`pidof PROG`"; -p "`pgrep PROG`".
1025 		 * pidof uses space as delim, pgrep uses newline. :(
1026 		 */
1027 		int pid;
1028 		char *delim = opt + strcspn(opt, "\n\t ,");
1029 		char c = *delim;
1030 
1031 		*delim = '\0';
1032 		pid = string_to_uint(opt);
1033 		if (pid <= 0) {
1034 			error_msg_and_die("Invalid process id: '%s'", opt);
1035 		}
1036 		if (pid == strace_tracer_pid) {
1037 			error_msg_and_die("I'm sorry, I can't let you do that, Dave.");
1038 		}
1039 		*delim = c;
1040 		alloctcb(pid);
1041 		if (c == '\0')
1042 			break;
1043 		opt = delim + 1;
1044 	}
1045 }
1046 
1047 static void
attach_tcb(struct tcb * const tcp)1048 attach_tcb(struct tcb *const tcp)
1049 {
1050 	if (ptrace_attach_or_seize(tcp->pid) < 0) {
1051 		perror_msg("attach: ptrace(%s, %d)",
1052 			   ptrace_attach_cmd, tcp->pid);
1053 		droptcb(tcp);
1054 		return;
1055 	}
1056 
1057 	after_successful_attach(tcp, TCB_GRABBED | post_attach_sigstop);
1058 	debug_msg("attach to pid %d (main) succeeded", tcp->pid);
1059 
1060 	static const char task_path[] = "/proc/%d/task";
1061 	char procdir[sizeof(task_path) + sizeof(int) * 3];
1062 	DIR *dir;
1063 	unsigned int ntid = 0, nerr = 0;
1064 
1065 	if (followfork && tcp->pid != strace_child &&
1066 	    xsprintf(procdir, task_path, tcp->pid) > 0 &&
1067 	    (dir = opendir(procdir)) != NULL) {
1068 		struct_dirent *de;
1069 
1070 		while ((de = read_dir(dir)) != NULL) {
1071 			if (de->d_fileno == 0)
1072 				continue;
1073 
1074 			int tid = string_to_uint(de->d_name);
1075 			if (tid <= 0 || tid == tcp->pid)
1076 				continue;
1077 
1078 			++ntid;
1079 			if (ptrace_attach_or_seize(tid) < 0) {
1080 				++nerr;
1081 				debug_perror_msg("attach: ptrace(%s, %d)",
1082 						 ptrace_attach_cmd, tid);
1083 				continue;
1084 			}
1085 
1086 			after_successful_attach(alloctcb(tid),
1087 						TCB_GRABBED | post_attach_sigstop);
1088 			debug_msg("attach to pid %d succeeded", tid);
1089 		}
1090 
1091 		closedir(dir);
1092 	}
1093 
1094 	if (!qflag) {
1095 		if (ntid > nerr)
1096 			error_msg("Process %u attached"
1097 				  " with %u threads",
1098 				  tcp->pid, ntid - nerr + 1);
1099 		else
1100 			error_msg("Process %u attached",
1101 				  tcp->pid);
1102 	}
1103 }
1104 
1105 static void
startup_attach(void)1106 startup_attach(void)
1107 {
1108 	pid_t parent_pid = strace_tracer_pid;
1109 	unsigned int tcbi;
1110 	struct tcb *tcp;
1111 
1112 	if (daemonized_tracer) {
1113 		pid_t pid = fork();
1114 		if (pid < 0)
1115 			perror_func_msg_and_die("fork");
1116 
1117 		if (pid) { /* parent */
1118 			/*
1119 			 * Wait for grandchild to attach to straced process
1120 			 * (grandparent). Grandchild SIGKILLs us after it attached.
1121 			 * Grandparent's wait() is unblocked by our death,
1122 			 * it proceeds to exec the straced program.
1123 			 */
1124 			pause();
1125 			_exit(0); /* paranoia */
1126 		}
1127 		/* grandchild */
1128 		/* We will be the tracer process. Remember our new pid: */
1129 		strace_tracer_pid = getpid();
1130 	}
1131 
1132 	for (tcbi = 0; tcbi < tcbtabsize; tcbi++) {
1133 		tcp = tcbtab[tcbi];
1134 
1135 		if (!tcp->pid)
1136 			continue;
1137 
1138 		/* Is this a process we should attach to, but not yet attached? */
1139 		if (tcp->flags & TCB_ATTACHED)
1140 			continue; /* no, we already attached it */
1141 
1142 		if (tcp->pid == parent_pid || tcp->pid == strace_tracer_pid) {
1143 			errno = EPERM;
1144 			perror_msg("attach: pid %d", tcp->pid);
1145 			droptcb(tcp);
1146 			continue;
1147 		}
1148 
1149 		attach_tcb(tcp);
1150 
1151 		if (interrupted)
1152 			return;
1153 	} /* for each tcbtab[] */
1154 
1155 	if (daemonized_tracer) {
1156 		/*
1157 		 * Make parent go away.
1158 		 * Also makes grandparent's wait() unblock.
1159 		 */
1160 		kill(parent_pid, SIGKILL);
1161 		strace_child = 0;
1162 	}
1163 }
1164 
1165 /* Stack-o-phobic exec helper, in the hope to work around
1166  * NOMMU + "daemonized tracer" difficulty.
1167  */
1168 struct exec_params {
1169 	int fd_to_close;
1170 	uid_t run_euid;
1171 	gid_t run_egid;
1172 	char **argv;
1173 	char *pathname;
1174 	struct sigaction child_sa;
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 	} else if (geteuid() != 0)
1206 		if (setreuid(run_uid, run_uid) < 0) {
1207 			perror_msg_and_die("setreuid");
1208 		}
1209 
1210 	if (!daemonized_tracer) {
1211 		/*
1212 		 * Induce a ptrace stop. Tracer (our parent)
1213 		 * will resume us with PTRACE_SYSCALL and display
1214 		 * the immediately following execve syscall.
1215 		 * Can't do this on NOMMU systems, we are after
1216 		 * vfork: parent is blocked, stopping would deadlock.
1217 		 */
1218 		if (!NOMMU_SYSTEM)
1219 			kill(getpid(), SIGSTOP);
1220 	} else {
1221 		alarm(3);
1222 		/* we depend on SIGCHLD set to SIG_DFL by init code */
1223 		/* if it happens to be SIG_IGN'ed, wait won't block */
1224 		wait(NULL);
1225 		alarm(0);
1226 	}
1227 
1228 	if (params_for_tracee.child_sa.sa_handler != SIG_DFL)
1229 		sigaction(SIGCHLD, &params_for_tracee.child_sa, NULL);
1230 
1231 	execv(params->pathname, params->argv);
1232 	perror_msg_and_die("exec");
1233 }
1234 
1235 /*
1236  * Open a dummy descriptor for use as a placeholder.
1237  * The descriptor is O_RDONLY with FD_CLOEXEC flag set.
1238  * A read attempt from such descriptor ends with EOF,
1239  * a write attempt is rejected with EBADF.
1240  */
1241 static int
open_dummy_desc(void)1242 open_dummy_desc(void)
1243 {
1244 	int fds[2];
1245 
1246 	if (pipe(fds))
1247 		perror_func_msg_and_die("pipe");
1248 	close(fds[1]);
1249 	set_cloexec_flag(fds[0]);
1250 	return fds[0];
1251 }
1252 
1253 /* placeholder fds status for stdin and stdout */
1254 static bool fd_is_placeholder[2];
1255 
1256 /*
1257  * Ensure that all standard file descriptors are open by opening placeholder
1258  * file descriptors for those standard file descriptors that are not open.
1259  *
1260  * The information which descriptors have been made open is saved
1261  * in fd_is_placeholder for later use.
1262  */
1263 static void
ensure_standard_fds_opened(void)1264 ensure_standard_fds_opened(void)
1265 {
1266 	int fd;
1267 
1268 	while ((fd = open_dummy_desc()) <= 2) {
1269 		if (fd == 2)
1270 			break;
1271 		fd_is_placeholder[fd] = true;
1272 	}
1273 
1274 	if (fd > 2)
1275 		close(fd);
1276 }
1277 
1278 /*
1279  * Redirect stdin and stdout unless they have been opened earlier
1280  * by ensure_standard_fds_opened as placeholders.
1281  */
1282 static void
redirect_standard_fds(void)1283 redirect_standard_fds(void)
1284 {
1285 	int i;
1286 
1287 	/*
1288 	 * It might be a good idea to redirect stderr as well,
1289 	 * but we sometimes need to print error messages.
1290 	 */
1291 	for (i = 0; i <= 1; ++i) {
1292 		if (!fd_is_placeholder[i]) {
1293 			close(i);
1294 			open_dummy_desc();
1295 		}
1296 	}
1297 }
1298 
1299 static void
startup_child(char ** argv)1300 startup_child(char **argv)
1301 {
1302 	struct_stat statbuf;
1303 	const char *filename;
1304 	size_t filename_len;
1305 	char pathname[PATH_MAX];
1306 	int pid;
1307 	struct tcb *tcp;
1308 
1309 	filename = argv[0];
1310 	filename_len = strlen(filename);
1311 
1312 	if (filename_len > sizeof(pathname) - 1) {
1313 		errno = ENAMETOOLONG;
1314 		perror_msg_and_die("exec");
1315 	}
1316 	if (strchr(filename, '/')) {
1317 		strcpy(pathname, filename);
1318 	}
1319 #ifdef USE_DEBUGGING_EXEC
1320 	/*
1321 	 * Debuggers customarily check the current directory
1322 	 * first regardless of the path but doing that gives
1323 	 * security geeks a panic attack.
1324 	 */
1325 	else if (stat_file(filename, &statbuf) == 0)
1326 		strcpy(pathname, filename);
1327 #endif /* USE_DEBUGGING_EXEC */
1328 	else {
1329 		const char *path;
1330 		size_t m, n, len;
1331 
1332 		for (path = getenv("PATH"); path && *path; path += m) {
1333 			const char *colon = strchr(path, ':');
1334 			if (colon) {
1335 				n = colon - path;
1336 				m = n + 1;
1337 			} else
1338 				m = n = strlen(path);
1339 			if (n == 0) {
1340 				if (!getcwd(pathname, PATH_MAX))
1341 					continue;
1342 				len = strlen(pathname);
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_func_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 		after_successful_attach(tcp, TCB_SKIP_DETACH_ON_FIRST_EXEC
1434 					     | (NOMMU_SYSTEM ? 0
1435 						: (TCB_HIDE_LOG
1436 						   | post_attach_sigstop)));
1437 	} else {
1438 		/* With -D, we are *child* here, the tracee is our parent. */
1439 		strace_child = strace_tracer_pid;
1440 		strace_tracer_pid = getpid();
1441 		tcp = alloctcb(strace_child);
1442 		tcp->flags |= TCB_SKIP_DETACH_ON_FIRST_EXEC | TCB_HIDE_LOG;
1443 		/*
1444 		 * Attaching will be done later, by startup_attach.
1445 		 * Note: we don't do after_successful_attach() here either!
1446 		 */
1447 
1448 		/* NOMMU BUG! -D mode is active, we (child) return,
1449 		 * and we will scribble over parent's stack!
1450 		 * When parent later unpauses, it segfaults.
1451 		 *
1452 		 * We work around it
1453 		 * (1) by declaring exec_or_die() NORETURN,
1454 		 * hopefully compiler will just jump to it
1455 		 * instead of call (won't push anything to stack),
1456 		 * (2) by trying very hard in exec_or_die()
1457 		 * to not use any stack,
1458 		 * (3) having a really big (PATH_MAX) stack object
1459 		 * in this function, which creates a "buffer" between
1460 		 * child's and parent's stack pointers.
1461 		 * This may save us if (1) and (2) failed
1462 		 * and compiler decided to use stack in exec_or_die() anyway
1463 		 * (happens on i386 because of stack parameter passing).
1464 		 *
1465 		 * A cleaner solution is to use makecontext + setcontext
1466 		 * to create a genuine separate stack and execute on it.
1467 		 */
1468 	}
1469 	/*
1470 	 * A case where straced process is part of a pipe:
1471 	 * { sleep 1; yes | head -n99999; } | strace -o/dev/null sh -c 'exec <&-; sleep 9'
1472 	 * If strace won't close its fd#0, closing it in tracee is not enough:
1473 	 * the pipe is still open, it has a reader. Thus, "head" will not get its
1474 	 * SIGPIPE at once, on the first write.
1475 	 *
1476 	 * Preventing it by redirecting strace's stdin/out.
1477 	 * (Don't leave fds 0 and 1 closed, this is bad practice: future opens
1478 	 * will reuse them, unexpectedly making a newly opened object "stdin").
1479 	 */
1480 	redirect_standard_fds();
1481 }
1482 
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_func_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 {
1510 		debug_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_func_msg_and_die("unexpected wait result %d",
1524 						tracee_pid);
1525 		}
1526 		if (WIFSIGNALED(status))
1527 			return;
1528 
1529 		error_func_msg_and_die("unexpected wait status %#x", status);
1530 	}
1531 }
1532 
1533 static unsigned
get_os_release(void)1534 get_os_release(void)
1535 {
1536 	unsigned rel;
1537 	const char *p;
1538 	struct utsname u;
1539 	if (uname(&u) < 0)
1540 		perror_msg_and_die("uname");
1541 	/* u.release has this form: "3.2.9[-some-garbage]" */
1542 	rel = 0;
1543 	p = u.release;
1544 	for (;;) {
1545 		if (!(*p >= '0' && *p <= '9'))
1546 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1547 		/* Note: this open-codes KERNEL_VERSION(): */
1548 		rel = (rel << 8) | atoi(p);
1549 		if (rel >= KERNEL_VERSION(1, 0, 0))
1550 			break;
1551 		while (*p >= '0' && *p <= '9')
1552 			p++;
1553 		if (*p != '.') {
1554 			if (rel >= KERNEL_VERSION(0, 1, 0)) {
1555 				/* "X.Y-something" means "X.Y.0" */
1556 				rel <<= 8;
1557 				break;
1558 			}
1559 			error_msg_and_die("Bad OS release string: '%s'", u.release);
1560 		}
1561 		p++;
1562 	}
1563 	return rel;
1564 }
1565 
1566 static void
set_sighandler(int signo,void (* sighandler)(int),struct sigaction * oldact)1567 set_sighandler(int signo, void (*sighandler)(int), struct sigaction *oldact)
1568 {
1569 	const struct sigaction sa = { .sa_handler = sighandler };
1570 	sigaction(signo, &sa, oldact);
1571 }
1572 
1573 /*
1574  * Initialization part of main() was eating much stack (~0.5k),
1575  * which was unused after init.
1576  * We can reuse it if we move init code into a separate function.
1577  *
1578  * Don't want main() to inline us and defeat the reason
1579  * we have a separate function.
1580  */
1581 static void ATTRIBUTE_NOINLINE
init(int argc,char * argv[])1582 init(int argc, char *argv[])
1583 {
1584 	int c, i;
1585 	int optF = 0;
1586 
1587 	if (!program_invocation_name || !*program_invocation_name) {
1588 		static char name[] = "strace";
1589 		program_invocation_name =
1590 			(argc > 0 && argv[0] && *argv[0]) ? argv[0] : name;
1591 	}
1592 
1593 	strace_tracer_pid = getpid();
1594 
1595 	os_release = get_os_release();
1596 
1597 	shared_log = stderr;
1598 	set_sortby(DEFAULT_SORTBY);
1599 	set_personality(DEFAULT_PERSONALITY);
1600 	qualify("trace=all");
1601 	qualify("abbrev=all");
1602 	qualify("verbose=all");
1603 #if DEFAULT_QUAL_FLAGS != (QUAL_TRACE | QUAL_ABBREV | QUAL_VERBOSE)
1604 # error Bug in DEFAULT_QUAL_FLAGS
1605 #endif
1606 	qualify("signal=all");
1607 	while ((c = getopt(argc, argv, "+"
1608 #ifdef ENABLE_STACKTRACE
1609 	    "k"
1610 #endif
1611 	    "a:Ab:cCdDe:E:fFhiI:o:O:p:P:qrs:S:tTu:vVwxX:yz")) != EOF) {
1612 		switch (c) {
1613 		case 'a':
1614 			acolumn = string_to_uint(optarg);
1615 			if (acolumn < 0)
1616 				error_opt_arg(c, optarg);
1617 			break;
1618 		case 'A':
1619 			open_append = true;
1620 			break;
1621 		case 'b':
1622 			if (strcmp(optarg, "execve") != 0)
1623 				error_msg_and_die("Syscall '%s' for -b isn't supported",
1624 					optarg);
1625 			detach_on_execve = 1;
1626 			break;
1627 		case 'c':
1628 			if (cflag == CFLAG_BOTH) {
1629 				error_msg_and_help("-c and -C are mutually exclusive");
1630 			}
1631 			cflag = CFLAG_ONLY_STATS;
1632 			break;
1633 		case 'C':
1634 			if (cflag == CFLAG_ONLY_STATS) {
1635 				error_msg_and_help("-c and -C are mutually exclusive");
1636 			}
1637 			cflag = CFLAG_BOTH;
1638 			break;
1639 		case 'd':
1640 			debug_flag = 1;
1641 			break;
1642 		case 'D':
1643 			daemonized_tracer = 1;
1644 			break;
1645 		case 'e':
1646 			qualify(optarg);
1647 			break;
1648 		case 'E':
1649 			if (putenv(optarg) < 0)
1650 				perror_msg_and_die("putenv");
1651 			break;
1652 		case 'f':
1653 			followfork++;
1654 			break;
1655 		case 'F':
1656 			optF = 1;
1657 			break;
1658 		case 'h':
1659 			usage();
1660 			break;
1661 		case 'i':
1662 			iflag = 1;
1663 			break;
1664 		case 'I':
1665 			opt_intr = string_to_uint_upto(optarg, NUM_INTR_OPTS - 1);
1666 			if (opt_intr <= 0)
1667 				error_opt_arg(c, optarg);
1668 			break;
1669 #ifdef ENABLE_STACKTRACE
1670 		case 'k':
1671 			stack_trace_enabled = true;
1672 			break;
1673 #endif
1674 		case 'o':
1675 			outfname = optarg;
1676 			break;
1677 		case 'O':
1678 			i = string_to_uint(optarg);
1679 			if (i < 0)
1680 				error_opt_arg(c, optarg);
1681 			set_overhead(i);
1682 			break;
1683 		case 'p':
1684 			process_opt_p_list(optarg);
1685 			break;
1686 		case 'P':
1687 			pathtrace_select(optarg);
1688 			break;
1689 		case 'q':
1690 			qflag++;
1691 			break;
1692 		case 'r':
1693 			rflag = 1;
1694 			break;
1695 		case 's':
1696 			i = string_to_uint(optarg);
1697 			if (i < 0 || (unsigned int) i > -1U / 4)
1698 				error_opt_arg(c, optarg);
1699 			max_strlen = i;
1700 			break;
1701 		case 'S':
1702 			set_sortby(optarg);
1703 			break;
1704 		case 't':
1705 			tflag++;
1706 			break;
1707 		case 'T':
1708 			Tflag = 1;
1709 			break;
1710 		case 'u':
1711 			username = optarg;
1712 			break;
1713 		case 'v':
1714 			qualify("abbrev=none");
1715 			break;
1716 		case 'V':
1717 			print_version();
1718 			exit(0);
1719 			break;
1720 		case 'w':
1721 			count_wallclock = 1;
1722 			break;
1723 		case 'x':
1724 			xflag++;
1725 			break;
1726 		case 'X':
1727 			if (!strcmp(optarg, "raw"))
1728 				xlat_verbosity = XLAT_STYLE_RAW;
1729 			else if (!strcmp(optarg, "abbrev"))
1730 				xlat_verbosity = XLAT_STYLE_ABBREV;
1731 			else if (!strcmp(optarg, "verbose"))
1732 				xlat_verbosity = XLAT_STYLE_VERBOSE;
1733 			else
1734 				error_opt_arg(c, optarg);
1735 			break;
1736 		case 'y':
1737 			show_fd_path++;
1738 			break;
1739 		case 'z':
1740 			not_failing_only = 1;
1741 			break;
1742 		default:
1743 			error_msg_and_help(NULL);
1744 			break;
1745 		}
1746 	}
1747 
1748 	argv += optind;
1749 	argc -= optind;
1750 
1751 	if (argc < 0 || (!nprocs && !argc)) {
1752 		error_msg_and_help("must have PROG [ARGS] or -p PID");
1753 	}
1754 
1755 	if (!argc && daemonized_tracer) {
1756 		error_msg_and_help("PROG [ARGS] must be specified with -D");
1757 	}
1758 
1759 	if (optF) {
1760 		if (followfork) {
1761 			error_msg("deprecated option -F ignored");
1762 		} else {
1763 			error_msg("option -F is deprecated, "
1764 				  "please use -f instead");
1765 			followfork = optF;
1766 		}
1767 	}
1768 
1769 	if (followfork >= 2 && cflag) {
1770 		error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
1771 	}
1772 
1773 	if (count_wallclock && !cflag) {
1774 		error_msg_and_help("-w must be given with (-c or -C)");
1775 	}
1776 
1777 	if (cflag == CFLAG_ONLY_STATS) {
1778 		if (iflag)
1779 			error_msg("-%c has no effect with -c", 'i');
1780 #ifdef ENABLE_STACKTRACE
1781 		if (stack_trace_enabled)
1782 			error_msg("-%c has no effect with -c", 'k');
1783 #endif
1784 		if (rflag)
1785 			error_msg("-%c has no effect with -c", 'r');
1786 		if (tflag)
1787 			error_msg("-%c has no effect with -c", 't');
1788 		if (Tflag)
1789 			error_msg("-%c has no effect with -c", 'T');
1790 		if (show_fd_path)
1791 			error_msg("-%c has no effect with -c", 'y');
1792 	}
1793 
1794 	acolumn_spaces = xmalloc(acolumn + 1);
1795 	memset(acolumn_spaces, ' ', acolumn);
1796 	acolumn_spaces[acolumn] = '\0';
1797 
1798 	set_sighandler(SIGCHLD, SIG_DFL, &params_for_tracee.child_sa);
1799 
1800 #ifdef ENABLE_STACKTRACE
1801 	if (stack_trace_enabled)
1802 		unwind_init();
1803 #endif
1804 
1805 	/* See if they want to run as another user. */
1806 	if (username != NULL) {
1807 		struct passwd *pent;
1808 
1809 		if (getuid() != 0 || geteuid() != 0) {
1810 			error_msg_and_die("You must be root to use the -u option");
1811 		}
1812 		pent = getpwnam(username);
1813 		if (pent == NULL) {
1814 			error_msg_and_die("Cannot find user '%s'", username);
1815 		}
1816 		run_uid = pent->pw_uid;
1817 		run_gid = pent->pw_gid;
1818 	} else {
1819 		run_uid = getuid();
1820 		run_gid = getgid();
1821 	}
1822 
1823 	if (followfork)
1824 		ptrace_setoptions |= PTRACE_O_TRACECLONE |
1825 				     PTRACE_O_TRACEFORK |
1826 				     PTRACE_O_TRACEVFORK;
1827 	debug_msg("ptrace_setoptions = %#x", ptrace_setoptions);
1828 	test_ptrace_seize();
1829 
1830 	/*
1831 	 * Is something weird with our stdin and/or stdout -
1832 	 * for example, may they be not open? In this case,
1833 	 * ensure that none of the future opens uses them.
1834 	 *
1835 	 * This was seen in the wild when /proc/sys/kernel/core_pattern
1836 	 * was set to "|/bin/strace -o/tmp/LOG PROG":
1837 	 * kernel runs coredump helper with fd#0 open but fd#1 closed (!),
1838 	 * therefore LOG gets opened to fd#1, and fd#1 is closed by
1839 	 * "don't hold up stdin/out open" code soon after.
1840 	 */
1841 	ensure_standard_fds_opened();
1842 
1843 	/* Check if they want to redirect the output. */
1844 	if (outfname) {
1845 		/* See if they want to pipe the output. */
1846 		if (outfname[0] == '|' || outfname[0] == '!') {
1847 			/*
1848 			 * We can't do the <outfname>.PID funny business
1849 			 * when using popen, so prohibit it.
1850 			 */
1851 			if (followfork >= 2)
1852 				error_msg_and_help("piping the output and -ff "
1853 						   "are mutually exclusive");
1854 			shared_log = strace_popen(outfname + 1);
1855 		} else if (followfork < 2) {
1856 			shared_log = strace_fopen(outfname);
1857 		} else if (strlen(outfname) >= PATH_MAX - sizeof(int) * 3) {
1858 			errno = ENAMETOOLONG;
1859 			perror_msg_and_die("%s", outfname);
1860 		}
1861 	} else {
1862 		/* -ff without -o FILE is the same as single -f */
1863 		if (followfork >= 2)
1864 			followfork = 1;
1865 	}
1866 
1867 	if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
1868 		setvbuf(shared_log, NULL, _IOLBF, 0);
1869 	}
1870 
1871 	/*
1872 	 * argv[0]	-pPID	-oFILE	Default interactive setting
1873 	 * yes		*	0	INTR_WHILE_WAIT
1874 	 * no		1	0	INTR_WHILE_WAIT
1875 	 * yes		*	1	INTR_NEVER
1876 	 * no		1	1	INTR_WHILE_WAIT
1877 	 */
1878 
1879 	if (outfname && argc) {
1880 		if (!opt_intr)
1881 			opt_intr = INTR_NEVER;
1882 		if (!qflag)
1883 			qflag = 1;
1884 	}
1885 	if (!opt_intr)
1886 		opt_intr = INTR_WHILE_WAIT;
1887 
1888 	/*
1889 	 * startup_child() must be called before the signal handlers get
1890 	 * installed below as they are inherited into the spawned process.
1891 	 * Also we do not need to be protected by them as during interruption
1892 	 * in the startup_child() mode we kill the spawned process anyway.
1893 	 */
1894 	if (argc) {
1895 		startup_child(argv);
1896 	}
1897 
1898 	set_sighandler(SIGTTOU, SIG_IGN, NULL);
1899 	set_sighandler(SIGTTIN, SIG_IGN, NULL);
1900 	if (opt_intr != INTR_ANYWHERE) {
1901 		if (opt_intr == INTR_BLOCK_TSTP_TOO)
1902 			set_sighandler(SIGTSTP, SIG_IGN, NULL);
1903 		/*
1904 		 * In interactive mode (if no -o OUTFILE, or -p PID is used),
1905 		 * fatal signals are handled asynchronously and acted
1906 		 * when waiting for process state changes.
1907 		 * In non-interactive mode these signals are ignored.
1908 		 */
1909 		set_sighandler(SIGHUP, interactive ? interrupt : SIG_IGN, NULL);
1910 		set_sighandler(SIGINT, interactive ? interrupt : SIG_IGN, NULL);
1911 		set_sighandler(SIGQUIT, interactive ? interrupt : SIG_IGN, NULL);
1912 		set_sighandler(SIGPIPE, interactive ? interrupt : SIG_IGN, NULL);
1913 		set_sighandler(SIGTERM, interactive ? interrupt : SIG_IGN, NULL);
1914 	}
1915 
1916 	sigemptyset(&timer_set);
1917 	sigaddset(&timer_set, SIGALRM);
1918 	sigprocmask(SIG_BLOCK, &timer_set, NULL);
1919 	set_sighandler(SIGALRM, timer_sighandler, NULL);
1920 
1921 	if (nprocs != 0 || daemonized_tracer)
1922 		startup_attach();
1923 
1924 	/* Do we want pids printed in our -o OUTFILE?
1925 	 * -ff: no (every pid has its own file); or
1926 	 * -f: yes (there can be more pids in the future); or
1927 	 * -p PID1,PID2: yes (there are already more than one pid)
1928 	 */
1929 	print_pid_pfx = (outfname && followfork < 2 && (followfork == 1 || nprocs > 1));
1930 }
1931 
1932 static struct tcb *
pid2tcb(const int pid)1933 pid2tcb(const int pid)
1934 {
1935 	if (pid <= 0)
1936 		return NULL;
1937 
1938 #define PID2TCB_CACHE_SIZE 1024U
1939 #define PID2TCB_CACHE_MASK (PID2TCB_CACHE_SIZE - 1)
1940 
1941 	static struct tcb *pid2tcb_cache[PID2TCB_CACHE_SIZE];
1942 	struct tcb **const ptcp = &pid2tcb_cache[pid & PID2TCB_CACHE_MASK];
1943 	struct tcb *tcp = *ptcp;
1944 
1945 	if (tcp && tcp->pid == pid)
1946 		return tcp;
1947 
1948 	for (unsigned int i = 0; i < tcbtabsize; ++i) {
1949 		tcp = tcbtab[i];
1950 		if (tcp->pid == pid)
1951 			return *ptcp = tcp;
1952 	}
1953 
1954 	return NULL;
1955 }
1956 
1957 static void
cleanup(void)1958 cleanup(void)
1959 {
1960 	unsigned int i;
1961 	struct tcb *tcp;
1962 	int fatal_sig;
1963 
1964 	/* 'interrupted' is a volatile object, fetch it only once */
1965 	fatal_sig = interrupted;
1966 	if (!fatal_sig)
1967 		fatal_sig = SIGTERM;
1968 
1969 	for (i = 0; i < tcbtabsize; i++) {
1970 		tcp = tcbtab[i];
1971 		if (!tcp->pid)
1972 			continue;
1973 		debug_func_msg("looking at pid %u", tcp->pid);
1974 		if (tcp->pid == strace_child) {
1975 			kill(tcp->pid, SIGCONT);
1976 			kill(tcp->pid, fatal_sig);
1977 		}
1978 		detach(tcp);
1979 	}
1980 	if (cflag)
1981 		call_summary(shared_log);
1982 }
1983 
1984 static void
interrupt(int sig)1985 interrupt(int sig)
1986 {
1987 	interrupted = sig;
1988 }
1989 
1990 static void
print_debug_info(const int pid,int status)1991 print_debug_info(const int pid, int status)
1992 {
1993 	const unsigned int event = (unsigned int) status >> 16;
1994 	char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16];
1995 	char evbuf[sizeof(",EVENT_VFORK_DONE (%u)") + sizeof(int)*3 /*paranoia:*/ + 16];
1996 
1997 	strcpy(buf, "???");
1998 	if (WIFSIGNALED(status))
1999 #ifdef WCOREDUMP
2000 		xsprintf(buf, "WIFSIGNALED,%ssig=%s",
2001 				WCOREDUMP(status) ? "core," : "",
2002 				signame(WTERMSIG(status)));
2003 #else
2004 		xsprintf(buf, "WIFSIGNALED,sig=%s",
2005 				signame(WTERMSIG(status)));
2006 #endif
2007 	if (WIFEXITED(status))
2008 		xsprintf(buf, "WIFEXITED,exitcode=%u", WEXITSTATUS(status));
2009 	if (WIFSTOPPED(status))
2010 		xsprintf(buf, "WIFSTOPPED,sig=%s", signame(WSTOPSIG(status)));
2011 	evbuf[0] = '\0';
2012 	if (event != 0) {
2013 		static const char *const event_names[] = {
2014 			[PTRACE_EVENT_CLONE] = "CLONE",
2015 			[PTRACE_EVENT_FORK]  = "FORK",
2016 			[PTRACE_EVENT_VFORK] = "VFORK",
2017 			[PTRACE_EVENT_VFORK_DONE] = "VFORK_DONE",
2018 			[PTRACE_EVENT_EXEC]  = "EXEC",
2019 			[PTRACE_EVENT_EXIT]  = "EXIT",
2020 			/* [PTRACE_EVENT_STOP (=128)] would make biggish array */
2021 		};
2022 		const char *e = "??";
2023 		if (event < ARRAY_SIZE(event_names))
2024 			e = event_names[event];
2025 		else if (event == PTRACE_EVENT_STOP)
2026 			e = "STOP";
2027 		xsprintf(evbuf, ",EVENT_%s (%u)", e, event);
2028 	}
2029 	error_msg("[wait(0x%06x) = %u] %s%s", status, pid, buf, evbuf);
2030 }
2031 
2032 static struct tcb *
maybe_allocate_tcb(const int pid,int status)2033 maybe_allocate_tcb(const int pid, int status)
2034 {
2035 	if (!WIFSTOPPED(status)) {
2036 		if (detach_on_execve && pid == strace_child) {
2037 			/* example: strace -bexecve sh -c 'exec true' */
2038 			strace_child = 0;
2039 			return NULL;
2040 		}
2041 		/*
2042 		 * This can happen if we inherited an unknown child.
2043 		 * Example: (sleep 1 & exec strace true)
2044 		 */
2045 		error_msg("Exit of unknown pid %u ignored", pid);
2046 		return NULL;
2047 	}
2048 	if (followfork) {
2049 		/* We assume it's a fork/vfork/clone child */
2050 		struct tcb *tcp = alloctcb(pid);
2051 		after_successful_attach(tcp, post_attach_sigstop);
2052 		if (!qflag)
2053 			error_msg("Process %d attached", pid);
2054 		return tcp;
2055 	} else {
2056 		/*
2057 		 * This can happen if a clone call misused CLONE_PTRACE itself.
2058 		 *
2059 		 * There used to be a dance around possible re-injection of
2060 		 * WSTOPSIG(status), but it was later removed as the only
2061 		 * observable stop here is the initial ptrace-stop.
2062 		 */
2063 		ptrace(PTRACE_DETACH, pid, NULL, 0L);
2064 		error_msg("Detached unknown pid %d", pid);
2065 		return NULL;
2066 	}
2067 }
2068 
2069 static struct tcb *
maybe_switch_tcbs(struct tcb * tcp,const int pid)2070 maybe_switch_tcbs(struct tcb *tcp, const int pid)
2071 {
2072 	FILE *fp;
2073 	struct tcb *execve_thread;
2074 	long old_pid = 0;
2075 
2076 	if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, &old_pid) < 0)
2077 		return tcp;
2078 	/* Avoid truncation in pid2tcb() param passing */
2079 	if (old_pid <= 0 || old_pid == pid)
2080 		return tcp;
2081 	if ((unsigned long) old_pid > UINT_MAX)
2082 		return tcp;
2083 	execve_thread = pid2tcb(old_pid);
2084 	/* It should be !NULL, but I feel paranoid */
2085 	if (!execve_thread)
2086 		return tcp;
2087 
2088 	if (execve_thread->curcol != 0) {
2089 		/*
2090 		 * One case we are here is -ff:
2091 		 * try "strace -oLOG -ff test/threaded_execve"
2092 		 */
2093 		fprintf(execve_thread->outf, " <pid changed to %d ...>\n", pid);
2094 		/*execve_thread->curcol = 0; - no need, see code below */
2095 	}
2096 	/* Swap output FILEs (needed for -ff) */
2097 	fp = execve_thread->outf;
2098 	execve_thread->outf = tcp->outf;
2099 	tcp->outf = fp;
2100 	/* And their column positions */
2101 	execve_thread->curcol = tcp->curcol;
2102 	tcp->curcol = 0;
2103 	/* Drop leader, but close execve'd thread outfile (if -ff) */
2104 	droptcb(tcp);
2105 	/* Switch to the thread, reusing leader's outfile and pid */
2106 	tcp = execve_thread;
2107 	tcp->pid = pid;
2108 	if (cflag != CFLAG_ONLY_STATS) {
2109 		printleader(tcp);
2110 		tprintf("+++ superseded by execve in pid %lu +++\n", old_pid);
2111 		line_ended();
2112 		tcp->flags |= TCB_REPRINT;
2113 	}
2114 
2115 	return tcp;
2116 }
2117 
2118 static void
print_signalled(struct tcb * tcp,const int pid,int status)2119 print_signalled(struct tcb *tcp, const int pid, int status)
2120 {
2121 	if (pid == strace_child) {
2122 		exit_code = 0x100 | WTERMSIG(status);
2123 		strace_child = 0;
2124 	}
2125 
2126 	if (cflag != CFLAG_ONLY_STATS
2127 	    && is_number_in_set(WTERMSIG(status), signal_set)) {
2128 		printleader(tcp);
2129 #ifdef WCOREDUMP
2130 		tprintf("+++ killed by %s %s+++\n",
2131 			signame(WTERMSIG(status)),
2132 			WCOREDUMP(status) ? "(core dumped) " : "");
2133 #else
2134 		tprintf("+++ killed by %s +++\n",
2135 			signame(WTERMSIG(status)));
2136 #endif
2137 		line_ended();
2138 	}
2139 }
2140 
2141 static void
print_exited(struct tcb * tcp,const int pid,int status)2142 print_exited(struct tcb *tcp, const int pid, int status)
2143 {
2144 	if (pid == strace_child) {
2145 		exit_code = WEXITSTATUS(status);
2146 		strace_child = 0;
2147 	}
2148 
2149 	if (cflag != CFLAG_ONLY_STATS &&
2150 	    qflag < 2) {
2151 		printleader(tcp);
2152 		tprintf("+++ exited with %d +++\n", WEXITSTATUS(status));
2153 		line_ended();
2154 	}
2155 }
2156 
2157 static void
print_stopped(struct tcb * tcp,const siginfo_t * si,const unsigned int sig)2158 print_stopped(struct tcb *tcp, const siginfo_t *si, const unsigned int sig)
2159 {
2160 	if (cflag != CFLAG_ONLY_STATS
2161 	    && !hide_log(tcp)
2162 	    && is_number_in_set(sig, signal_set)) {
2163 		printleader(tcp);
2164 		if (si) {
2165 			tprintf("--- %s ", signame(sig));
2166 			printsiginfo(si);
2167 			tprints(" ---\n");
2168 		} else
2169 			tprintf("--- stopped by %s ---\n", signame(sig));
2170 		line_ended();
2171 
2172 #ifdef ENABLE_STACKTRACE
2173 		if (stack_trace_enabled)
2174 			unwind_tcb_print(tcp);
2175 #endif
2176 	}
2177 }
2178 
2179 static void
startup_tcb(struct tcb * tcp)2180 startup_tcb(struct tcb *tcp)
2181 {
2182 	debug_msg("pid %d has TCB_STARTUP, initializing it", tcp->pid);
2183 
2184 	tcp->flags &= ~TCB_STARTUP;
2185 
2186 	if (!use_seize) {
2187 		debug_msg("setting opts 0x%x on pid %d",
2188 			  ptrace_setoptions, tcp->pid);
2189 		if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) {
2190 			if (errno != ESRCH) {
2191 				/* Should never happen, really */
2192 				perror_msg_and_die("PTRACE_SETOPTIONS");
2193 			}
2194 		}
2195 	}
2196 
2197 	if ((tcp->flags & TCB_GRABBED) && (get_scno(tcp) == 1))
2198 		tcp->s_prev_ent = tcp->s_ent;
2199 }
2200 
2201 static void
print_event_exit(struct tcb * tcp)2202 print_event_exit(struct tcb *tcp)
2203 {
2204 	if (entering(tcp) || filtered(tcp) || hide_log(tcp)
2205 	    || cflag == CFLAG_ONLY_STATS) {
2206 		return;
2207 	}
2208 
2209 	if (followfork < 2 && printing_tcp && printing_tcp != tcp
2210 	    && printing_tcp->curcol != 0) {
2211 		set_current_tcp(printing_tcp);
2212 		tprints(" <unfinished ...>\n");
2213 		flush_tcp_output(printing_tcp);
2214 		printing_tcp->curcol = 0;
2215 		set_current_tcp(tcp);
2216 	}
2217 
2218 	if ((followfork < 2 && printing_tcp != tcp)
2219 	    || (tcp->flags & TCB_REPRINT)) {
2220 		tcp->flags &= ~TCB_REPRINT;
2221 		printleader(tcp);
2222 		tprintf("<... %s resumed>", tcp->s_ent->sys_name);
2223 	}
2224 
2225 	if (!(tcp->sys_func_rval & RVAL_DECODED)) {
2226 		/*
2227 		 * The decoder has probably decided to print something
2228 		 * on exiting syscall which is not going to happen.
2229 		 */
2230 		tprints(" <unfinished ...>");
2231 	}
2232 
2233 	printing_tcp = tcp;
2234 	tprints(") ");
2235 	tabto();
2236 	tprints("= ?\n");
2237 	line_ended();
2238 }
2239 
2240 static const struct tcb_wait_data *
next_event(void)2241 next_event(void)
2242 {
2243 	static struct tcb_wait_data wait_data;
2244 
2245 	int pid;
2246 	int status;
2247 	struct tcb *tcp;
2248 	struct tcb_wait_data *wd = &wait_data;
2249 	struct rusage ru;
2250 
2251 	if (interrupted)
2252 		return NULL;
2253 
2254 	/*
2255 	 * Used to exit simply when nprocs hits zero, but in this testcase:
2256 	 *  int main(void) { _exit(!!fork()); }
2257 	 * under strace -f, parent sometimes (rarely) manages
2258 	 * to exit before we see the first stop of the child,
2259 	 * and we are losing track of it:
2260 	 *  19923 clone(...) = 19924
2261 	 *  19923 exit_group(1)     = ?
2262 	 *  19923 +++ exited with 1 +++
2263 	 * Exiting only when wait() returns ECHILD works better.
2264 	 */
2265 	if (popen_pid != 0) {
2266 		/* However, if -o|logger is in use, we can't do that.
2267 		 * Can work around that by double-forking the logger,
2268 		 * but that loses the ability to wait for its completion
2269 		 * on exit. Oh well...
2270 		 */
2271 		if (nprocs == 0)
2272 			return NULL;
2273 	}
2274 
2275 	const bool unblock_delay_timer = is_delay_timer_armed();
2276 
2277 	/*
2278 	 * The window of opportunity to handle expirations
2279 	 * of the delay timer opens here.
2280 	 *
2281 	 * Unblock the signal handler for the delay timer
2282 	 * iff the delay timer is already created.
2283 	 */
2284 	if (unblock_delay_timer)
2285 		sigprocmask(SIG_UNBLOCK, &timer_set, NULL);
2286 
2287 	/*
2288 	 * If the delay timer has expired, then its expiration
2289 	 * has been handled already by the signal handler.
2290 	 *
2291 	 * If the delay timer expires during wait4(),
2292 	 * then the system call will be interrupted and
2293 	 * the expiration will be handled by the signal handler.
2294 	 */
2295 	pid = wait4(-1, &status, __WALL, (cflag ? &ru : NULL));
2296 	const int wait_errno = errno;
2297 
2298 	/*
2299 	 * The window of opportunity to handle expirations
2300 	 * of the delay timer closes here.
2301 	 *
2302 	 * Block the signal handler for the delay timer
2303 	 * iff it was unblocked earlier.
2304 	 */
2305 	if (unblock_delay_timer) {
2306 		sigprocmask(SIG_BLOCK, &timer_set, NULL);
2307 
2308 		if (restart_failed)
2309 			return NULL;
2310 	}
2311 
2312 	if (pid < 0) {
2313 		if (wait_errno == EINTR) {
2314 			wd->te = TE_NEXT;
2315 			return wd;
2316 		}
2317 		if (nprocs == 0 && wait_errno == ECHILD)
2318 			return NULL;
2319 		/*
2320 		 * If nprocs > 0, ECHILD is not expected,
2321 		 * treat it as any other error here:
2322 		 */
2323 		errno = wait_errno;
2324 		perror_msg_and_die("wait4(__WALL)");
2325 	}
2326 
2327 	wd->status = status;
2328 
2329 	if (pid == popen_pid) {
2330 		if (!WIFSTOPPED(status))
2331 			popen_pid = 0;
2332 		wd->te = TE_NEXT;
2333 		return wd;
2334 	}
2335 
2336 	if (debug_flag)
2337 		print_debug_info(pid, status);
2338 
2339 	/* Look up 'pid' in our table. */
2340 	tcp = pid2tcb(pid);
2341 
2342 	if (!tcp) {
2343 		tcp = maybe_allocate_tcb(pid, status);
2344 		if (!tcp) {
2345 			wd->te = TE_NEXT;
2346 			return wd;
2347 		}
2348 	}
2349 
2350 	clear_regs(tcp);
2351 
2352 	/* Set current output file */
2353 	set_current_tcp(tcp);
2354 
2355 	if (cflag) {
2356 		struct timespec stime = {
2357 			.tv_sec = ru.ru_stime.tv_sec,
2358 			.tv_nsec = ru.ru_stime.tv_usec * 1000
2359 		};
2360 		ts_sub(&tcp->dtime, &stime, &tcp->stime);
2361 		tcp->stime = stime;
2362 	}
2363 
2364 	if (WIFSIGNALED(status)) {
2365 		wd->te = TE_SIGNALLED;
2366 		return wd;
2367 	}
2368 
2369 	if (WIFEXITED(status)) {
2370 		wd->te = TE_EXITED;
2371 		return wd;
2372 	}
2373 
2374 	/*
2375 	 * As WCONTINUED flag has not been specified to wait4,
2376 	 * it cannot be WIFCONTINUED(status), so the only case
2377 	 * that remains is WIFSTOPPED(status).
2378 	 */
2379 
2380 	/* Is this the very first time we see this tracee stopped? */
2381 	if (tcp->flags & TCB_STARTUP)
2382 		startup_tcb(tcp);
2383 
2384 	const unsigned int sig = WSTOPSIG(status);
2385 	const unsigned int event = (unsigned int) status >> 16;
2386 
2387 	switch (event) {
2388 	case 0:
2389 		/*
2390 		 * Is this post-attach SIGSTOP?
2391 		 * Interestingly, the process may stop
2392 		 * with STOPSIG equal to some other signal
2393 		 * than SIGSTOP if we happened to attach
2394 		 * just before the process takes a signal.
2395 		 */
2396 		if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) {
2397 			debug_func_msg("ignored SIGSTOP on pid %d", tcp->pid);
2398 			tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP;
2399 			wd->te = TE_RESTART;
2400 		} else if (sig == syscall_trap_sig) {
2401 			wd->te = TE_SYSCALL_STOP;
2402 		} else {
2403 			memset(&wd->si, 0, sizeof(wd->si));
2404 			/*
2405 			 * True if tracee is stopped by signal
2406 			 * (as opposed to "tracee received signal").
2407 			 * TODO: shouldn't we check for errno == EINVAL too?
2408 			 * We can get ESRCH instead, you know...
2409 			 */
2410 			bool stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, &wd->si) < 0;
2411 			wd->te = stopped ? TE_GROUP_STOP : TE_SIGNAL_DELIVERY_STOP;
2412 		}
2413 		break;
2414 	case PTRACE_EVENT_STOP:
2415 		/*
2416 		 * PTRACE_INTERRUPT-stop or group-stop.
2417 		 * PTRACE_INTERRUPT-stop has sig == SIGTRAP here.
2418 		 */
2419 		switch (sig) {
2420 		case SIGSTOP:
2421 		case SIGTSTP:
2422 		case SIGTTIN:
2423 		case SIGTTOU:
2424 			wd->te = TE_GROUP_STOP;
2425 			break;
2426 		default:
2427 			wd->te = TE_RESTART;
2428 		}
2429 		break;
2430 	case PTRACE_EVENT_EXEC:
2431 		wd->te = TE_STOP_BEFORE_EXECVE;
2432 		break;
2433 	case PTRACE_EVENT_EXIT:
2434 		wd->te = TE_STOP_BEFORE_EXIT;
2435 		break;
2436 	default:
2437 		wd->te = TE_RESTART;
2438 	}
2439 
2440 	return wd;
2441 }
2442 
2443 static int
trace_syscall(struct tcb * tcp,unsigned int * sig)2444 trace_syscall(struct tcb *tcp, unsigned int *sig)
2445 {
2446 	if (entering(tcp)) {
2447 		int res = syscall_entering_decode(tcp);
2448 		switch (res) {
2449 		case 0:
2450 			return 0;
2451 		case 1:
2452 			res = syscall_entering_trace(tcp, sig);
2453 		}
2454 		syscall_entering_finish(tcp, res);
2455 		return res;
2456 	} else {
2457 		struct timespec ts = {};
2458 		int res = syscall_exiting_decode(tcp, &ts);
2459 		if (res != 0) {
2460 			res = syscall_exiting_trace(tcp, &ts, res);
2461 		}
2462 		syscall_exiting_finish(tcp);
2463 		return res;
2464 	}
2465 }
2466 
2467 /* Returns true iff the main trace loop has to continue. */
2468 static bool
dispatch_event(const struct tcb_wait_data * wd)2469 dispatch_event(const struct tcb_wait_data *wd)
2470 {
2471 	unsigned int restart_op = PTRACE_SYSCALL;
2472 	unsigned int restart_sig = 0;
2473 	enum trace_event te = wd ? wd->te : TE_BREAK;
2474 	/*
2475 	 * Copy wd->status to a non-const variable to workaround glibc bugs
2476 	 * around union wait fixed by glibc commit glibc-2.24~391
2477 	 */
2478 	int status = wd ? wd->status : 0;
2479 
2480 	switch (te) {
2481 	case TE_BREAK:
2482 		return false;
2483 
2484 	case TE_NEXT:
2485 		return true;
2486 
2487 	case TE_RESTART:
2488 		break;
2489 
2490 	case TE_SYSCALL_STOP:
2491 		if (trace_syscall(current_tcp, &restart_sig) < 0) {
2492 			/*
2493 			 * ptrace() failed in trace_syscall().
2494 			 * Likely a result of process disappearing mid-flight.
2495 			 * Observed case: exit_group() or SIGKILL terminating
2496 			 * all processes in thread group.
2497 			 * We assume that ptrace error was caused by process death.
2498 			 * We used to detach(current_tcp) here, but since we no
2499 			 * longer implement "detach before death" policy/hack,
2500 			 * we can let this process to report its death to us
2501 			 * normally, via WIFEXITED or WIFSIGNALED wait status.
2502 			 */
2503 			return true;
2504 		}
2505 		break;
2506 
2507 	case TE_SIGNAL_DELIVERY_STOP:
2508 		restart_sig = WSTOPSIG(status);
2509 		print_stopped(current_tcp, &wd->si, restart_sig);
2510 		break;
2511 
2512 	case TE_SIGNALLED:
2513 		print_signalled(current_tcp, current_tcp->pid, status);
2514 		droptcb(current_tcp);
2515 		return true;
2516 
2517 	case TE_GROUP_STOP:
2518 		restart_sig = WSTOPSIG(status);
2519 		print_stopped(current_tcp, NULL, restart_sig);
2520 		if (use_seize) {
2521 			/*
2522 			 * This ends ptrace-stop, but does *not* end group-stop.
2523 			 * This makes stopping signals work properly on straced
2524 			 * process (that is, process really stops. It used to
2525 			 * continue to run).
2526 			 */
2527 			restart_op = PTRACE_LISTEN;
2528 			restart_sig = 0;
2529 		}
2530 		break;
2531 
2532 	case TE_EXITED:
2533 		print_exited(current_tcp, current_tcp->pid, status);
2534 		droptcb(current_tcp);
2535 		return true;
2536 
2537 	case TE_STOP_BEFORE_EXECVE:
2538 		/*
2539 		 * Check that we are inside syscall now (next event after
2540 		 * PTRACE_EVENT_EXEC should be for syscall exiting).  If it is
2541 		 * not the case, we might have a situation when we attach to a
2542 		 * process and the first thing we see is a PTRACE_EVENT_EXEC
2543 		 * and all the following syscall state tracking is screwed up
2544 		 * otherwise.
2545 		 */
2546 		if (entering(current_tcp)) {
2547 			int ret;
2548 
2549 			error_msg("Stray PTRACE_EVENT_EXEC from pid %d"
2550 				  ", trying to recover...",
2551 				  current_tcp->pid);
2552 
2553 			current_tcp->flags |= TCB_RECOVERING;
2554 			ret = trace_syscall(current_tcp, &restart_sig);
2555 			current_tcp->flags &= ~TCB_RECOVERING;
2556 
2557 			if (ret < 0) {
2558 				/* The reason is described in TE_SYSCALL_STOP */
2559 				return true;
2560 			}
2561 		}
2562 
2563 		/*
2564 		 * Under Linux, execve changes pid to thread leader's pid,
2565 		 * and we see this changed pid on EVENT_EXEC and later,
2566 		 * execve sysexit. Leader "disappears" without exit
2567 		 * notification. Let user know that, drop leader's tcb,
2568 		 * and fix up pid in execve thread's tcb.
2569 		 * Effectively, execve thread's tcb replaces leader's tcb.
2570 		 *
2571 		 * BTW, leader is 'stuck undead' (doesn't report WIFEXITED
2572 		 * on exit syscall) in multithreaded programs exactly
2573 		 * in order to handle this case.
2574 		 *
2575 		 * PTRACE_GETEVENTMSG returns old pid starting from Linux 3.0.
2576 		 * On 2.6 and earlier, it can return garbage.
2577 		 */
2578 		if (os_release >= KERNEL_VERSION(3, 0, 0))
2579 			set_current_tcp(maybe_switch_tcbs(current_tcp,
2580 							  current_tcp->pid));
2581 
2582 		if (detach_on_execve) {
2583 			if (current_tcp->flags & TCB_SKIP_DETACH_ON_FIRST_EXEC) {
2584 				current_tcp->flags &= ~TCB_SKIP_DETACH_ON_FIRST_EXEC;
2585 			} else {
2586 				detach(current_tcp); /* do "-b execve" thingy */
2587 				return true;
2588 			}
2589 		}
2590 		break;
2591 
2592 	case TE_STOP_BEFORE_EXIT:
2593 		print_event_exit(current_tcp);
2594 		break;
2595 	}
2596 
2597 	/* We handled quick cases, we are permitted to interrupt now. */
2598 	if (interrupted)
2599 		return false;
2600 
2601 	/* If the process is being delayed, do not ptrace_restart just yet */
2602 	if (syscall_delayed(current_tcp))
2603 		return true;
2604 
2605 	if (ptrace_restart(restart_op, current_tcp, restart_sig) < 0) {
2606 		/* Note: ptrace_restart emitted error message */
2607 		exit_code = 1;
2608 		return false;
2609 	}
2610 	return true;
2611 }
2612 
2613 static bool
restart_delayed_tcb(struct tcb * const tcp)2614 restart_delayed_tcb(struct tcb *const tcp)
2615 {
2616 	const struct tcb_wait_data wd = { .te = TE_RESTART };
2617 
2618 	debug_func_msg("pid %d", tcp->pid);
2619 
2620 	tcp->flags &= ~TCB_DELAYED;
2621 
2622 	struct tcb *const prev_tcp = current_tcp;
2623 	current_tcp = tcp;
2624 	bool ret = dispatch_event(&wd);
2625 	current_tcp = prev_tcp;
2626 
2627 	return ret;
2628 }
2629 
2630 static bool
restart_delayed_tcbs(void)2631 restart_delayed_tcbs(void)
2632 {
2633 	struct tcb *tcp_next = NULL;
2634 	struct timespec ts_now;
2635 
2636 	clock_gettime(CLOCK_MONOTONIC, &ts_now);
2637 
2638 	for (size_t i = 0; i < tcbtabsize; i++) {
2639 		struct tcb *tcp = tcbtab[i];
2640 
2641 		if (tcp->pid && syscall_delayed(tcp)) {
2642 			if (ts_cmp(&ts_now, &tcp->delay_expiration_time) > 0) {
2643 				if (!restart_delayed_tcb(tcp))
2644 					return false;
2645 			} else {
2646 				/* Check whether this tcb is the next.  */
2647 				if (!tcp_next ||
2648 				    ts_cmp(&tcp_next->delay_expiration_time,
2649 					   &tcp->delay_expiration_time) > 0) {
2650 					tcp_next = tcp;
2651 				}
2652 			}
2653 		}
2654 	}
2655 
2656 	if (tcp_next)
2657 		arm_delay_timer(tcp_next);
2658 
2659 	return true;
2660 }
2661 
2662 /*
2663  * As this signal handler does a lot of work that is not suitable
2664  * for signal handlers, extra care must be taken to ensure that
2665  * it is enabled only in those places where it's safe.
2666  */
2667 static void
timer_sighandler(int sig)2668 timer_sighandler(int sig)
2669 {
2670 	delay_timer_expired();
2671 
2672 	if (restart_failed)
2673 		return;
2674 
2675 	int saved_errno = errno;
2676 
2677 	if (!restart_delayed_tcbs())
2678 		restart_failed = 1;
2679 
2680 	errno = saved_errno;
2681 }
2682 
2683 #ifdef ENABLE_COVERAGE_GCOV
2684 extern void __gcov_flush(void);
2685 #endif
2686 
2687 static void ATTRIBUTE_NORETURN
terminate(void)2688 terminate(void)
2689 {
2690 	cleanup();
2691 	fflush(NULL);
2692 	if (shared_log != stderr)
2693 		fclose(shared_log);
2694 	if (popen_pid) {
2695 		while (waitpid(popen_pid, NULL, 0) < 0 && errno == EINTR)
2696 			;
2697 	}
2698 	if (exit_code > 0xff) {
2699 		/* Avoid potential core file clobbering.  */
2700 		struct_rlimit rlim = {0, 0};
2701 		set_rlimit(RLIMIT_CORE, &rlim);
2702 
2703 		/* Child was killed by a signal, mimic that.  */
2704 		exit_code &= 0xff;
2705 		signal(exit_code, SIG_DFL);
2706 #ifdef ENABLE_COVERAGE_GCOV
2707 		__gcov_flush();
2708 #endif
2709 		raise(exit_code);
2710 
2711 		/* Unblock the signal.  */
2712 		sigset_t mask;
2713 		sigemptyset(&mask);
2714 		sigaddset(&mask, exit_code);
2715 #ifdef ENABLE_COVERAGE_GCOV
2716 		__gcov_flush();
2717 #endif
2718 		sigprocmask(SIG_UNBLOCK, &mask, NULL);
2719 
2720 		/* Paranoia - what if this signal is not fatal?
2721 		   Exit with 128 + signo then.  */
2722 		exit_code += 128;
2723 	}
2724 	exit(exit_code);
2725 }
2726 
2727 int
main(int argc,char * argv[])2728 main(int argc, char *argv[])
2729 {
2730 	setlocale(LC_ALL, "");
2731 	init(argc, argv);
2732 
2733 	exit_code = !nprocs;
2734 
2735 	while (dispatch_event(next_event()))
2736 		;
2737 	terminate();
2738 }
2739