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
47 #include "ptrace.h"
48 #include "printsiginfo.h"
49
50 /* In some libc, these aren't declared. Do it ourself: */
51 extern char **environ;
52 extern int optind;
53 extern char *optarg;
54
55 #ifdef USE_LIBUNWIND
56 /* if this is true do the stack trace for every system call */
57 bool stack_trace_enabled = false;
58 #endif
59
60 #if defined __NR_tkill
61 # define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig))
62 #else
63 /* kill() may choose arbitrarily the target task of the process group
64 while we later wait on a that specific TID. PID process waits become
65 TID task specific waits for a process under ptrace(2). */
66 # warning "tkill(2) not available, risk of strace hangs!"
67 # define my_tkill(tid, sig) kill((tid), (sig))
68 #endif
69
70 /* Glue for systems without a MMU that cannot provide fork() */
71 #if !defined(HAVE_FORK)
72 # undef NOMMU_SYSTEM
73 # define NOMMU_SYSTEM 1
74 #endif
75 #if NOMMU_SYSTEM
76 # define fork() vfork()
77 #endif
78
79 const unsigned int syscall_trap_sig = SIGTRAP | 0x80;
80
81 cflag_t cflag = CFLAG_NONE;
82 unsigned int followfork = 0;
83 unsigned int ptrace_setoptions = PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC;
84 unsigned int xflag = 0;
85 bool debug_flag = 0;
86 bool Tflag = 0;
87 bool iflag = 0;
88 bool count_wallclock = 0;
89 unsigned int qflag = 0;
90 static unsigned int tflag = 0;
91 static bool rflag = 0;
92 static bool print_pid_pfx = 0;
93
94 /* -I n */
95 enum {
96 INTR_NOT_SET = 0,
97 INTR_ANYWHERE = 1, /* don't block/ignore any signals */
98 INTR_WHILE_WAIT = 2, /* block fatal signals while decoding syscall. default */
99 INTR_NEVER = 3, /* block fatal signals. default if '-o FILE PROG' */
100 INTR_BLOCK_TSTP_TOO = 4, /* block fatal signals and SIGTSTP (^Z) */
101 NUM_INTR_OPTS
102 };
103 static int opt_intr;
104 /* We play with signal mask only if this mode is active: */
105 #define interactive (opt_intr == INTR_WHILE_WAIT)
106
107 /*
108 * daemonized_tracer supports -D option.
109 * With this option, strace forks twice.
110 * Unlike normal case, with -D *grandparent* process exec's,
111 * becoming a traced process. Child exits (this prevents traced process
112 * from having children it doesn't expect to have), and grandchild
113 * attaches to grandparent similarly to strace -p PID.
114 * This allows for more transparent interaction in cases
115 * when process and its parent are communicating via signals,
116 * wait() etc. Without -D, strace process gets lodged in between,
117 * disrupting parent<->child link.
118 */
119 static bool daemonized_tracer = 0;
120
121 #if USE_SEIZE
122 static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP;
123 # define use_seize (post_attach_sigstop == 0)
124 #else
125 # define post_attach_sigstop TCB_IGNORE_ONE_SIGSTOP
126 # define use_seize 0
127 #endif
128
129 /* Sometimes we want to print only succeeding syscalls. */
130 bool not_failing_only = 0;
131
132 /* Show path associated with fd arguments */
133 unsigned int show_fd_path = 0;
134
135 static bool detach_on_execve = 0;
136 /* Are we "strace PROG" and need to skip detach on first execve? */
137 static bool skip_one_b_execve = 0;
138 /* Are we "strace PROG" and need to hide everything until execve? */
139 bool hide_log_until_execve = 0;
140
141 static int exit_code = 0;
142 static int strace_child = 0;
143 static int strace_tracer_pid = 0;
144
145 static char *username = NULL;
146 static uid_t run_uid;
147 static gid_t run_gid;
148
149 unsigned int max_strlen = DEFAULT_STRLEN;
150 static int acolumn = DEFAULT_ACOLUMN;
151 static char *acolumn_spaces;
152
153 static char *outfname = NULL;
154 /* If -ff, points to stderr. Else, it's our common output log */
155 static FILE *shared_log;
156
157 struct tcb *printing_tcp = NULL;
158 static struct tcb *current_tcp;
159
160 static struct tcb **tcbtab;
161 static unsigned int nprocs, tcbtabsize;
162 static const char *progname;
163
164 unsigned os_release; /* generated from uname()'s u.release */
165
166 static void detach(struct tcb *tcp);
167 static void cleanup(void);
168 static void interrupt(int sig);
169 static sigset_t empty_set, blocked_set;
170
171 #ifdef HAVE_SIG_ATOMIC_T
172 static volatile sig_atomic_t interrupted;
173 #else
174 static volatile int interrupted;
175 #endif
176
177 #ifndef HAVE_STRERROR
178
179 #if !HAVE_DECL_SYS_ERRLIST
180 extern int sys_nerr;
181 extern char *sys_errlist[];
182 #endif
183
184 const char *
strerror(int err_no)185 strerror(int err_no)
186 {
187 static char buf[sizeof("Unknown error %d") + sizeof(int)*3];
188
189 if (err_no < 1 || err_no >= sys_nerr) {
190 sprintf(buf, "Unknown error %d", err_no);
191 return buf;
192 }
193 return sys_errlist[err_no];
194 }
195
196 #endif /* HAVE_STERRROR */
197
198 static void
usage()199 usage()
200 {
201 printf("\
202 usage: strace [-CdffhiqrtttTvVwxxy] [-I n] [-e expr]...\n\
203 [-a column] [-o file] [-s strsize] [-P path]...\n\
204 -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
205 or: strace -c[dfw] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\
206 -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\
207 \n\
208 Output format:\n\
209 -a column alignment COLUMN for printing syscall results (default %d)\n\
210 -i print instruction pointer at time of syscall\n\
211 -o file send trace output to FILE instead of stderr\n\
212 -q suppress messages about attaching, detaching, etc.\n\
213 -r print relative timestamp\n\
214 -s strsize limit length of print strings to STRSIZE chars (default %d)\n\
215 -t print absolute timestamp\n\
216 -tt print absolute timestamp with usecs\n\
217 -T print time spent in each syscall\n\
218 -x print non-ascii strings in hex\n\
219 -xx print all strings in hex\n\
220 -y print paths associated with file descriptor arguments\n\
221 -yy print ip:port pairs associated with socket file descriptors\n\
222 \n\
223 Statistics:\n\
224 -c count time, calls, and errors for each syscall and report summary\n\
225 -C like -c but also print regular output\n\
226 -O overhead set overhead for tracing syscalls to OVERHEAD usecs\n\
227 -S sortby sort syscall counts by: time, calls, name, nothing (default %s)\n\
228 -w summarise syscall latency (default is system time)\n\
229 \n\
230 Filtering:\n\
231 -e expr a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\
232 options: trace, abbrev, verbose, raw, signal, read, write\n\
233 -P path trace accesses to path\n\
234 \n\
235 Tracing:\n\
236 -b execve detach on execve syscall\n\
237 -D run tracer process as a detached grandchild, not as parent\n\
238 -f follow forks\n\
239 -ff follow forks with output into separate files\n\
240 -I interruptible\n\
241 1: no signals are blocked\n\
242 2: fatal signals are blocked while decoding syscall (default)\n\
243 3: fatal signals are always blocked (default if '-o FILE PROG')\n\
244 4: fatal signals and SIGTSTP (^Z) are always blocked\n\
245 (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\
246 \n\
247 Startup:\n\
248 -E var remove var from the environment for command\n\
249 -E var=val put var=val in the environment for command\n\
250 -p pid trace process with process id PID, may be repeated\n\
251 -u username run command as username handling setuid and/or setgid\n\
252 \n\
253 Miscellaneous:\n\
254 -d enable debug output to stderr\n\
255 -v verbose mode: print unabbreviated argv, stat, termios, etc. args\n\
256 -h print help message\n\
257 -V print version\n\
258 "
259 #ifdef USE_LIBUNWIND
260 " -k obtain stack trace between each syscall (experimental)\n\
261 "
262 #endif
263 /* ancient, no one should use it
264 -F -- attempt to follow vforks (deprecated, use -f)\n\
265 */
266 /* this is broken, so don't document it
267 -z -- print only succeeding syscalls\n\
268 */
269 , DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY);
270 exit(0);
271 }
272
273 static void ATTRIBUTE_NORETURN
die(void)274 die(void)
275 {
276 if (strace_tracer_pid == getpid()) {
277 cflag = 0;
278 cleanup();
279 }
280 exit(1);
281 }
282
verror_msg(int err_no,const char * fmt,va_list p)283 static void verror_msg(int err_no, const char *fmt, va_list p)
284 {
285 char *msg;
286
287 fflush(NULL);
288
289 /* We want to print entire message with single fprintf to ensure
290 * message integrity if stderr is shared with other programs.
291 * Thus we use vasprintf + single fprintf.
292 */
293 msg = NULL;
294 if (vasprintf(&msg, fmt, p) >= 0) {
295 if (err_no)
296 fprintf(stderr, "%s: %s: %s\n", progname, msg, strerror(err_no));
297 else
298 fprintf(stderr, "%s: %s\n", progname, msg);
299 free(msg);
300 } else {
301 /* malloc in vasprintf failed, try it without malloc */
302 fprintf(stderr, "%s: ", progname);
303 vfprintf(stderr, fmt, p);
304 if (err_no)
305 fprintf(stderr, ": %s\n", strerror(err_no));
306 else
307 putc('\n', stderr);
308 }
309 /* We don't switch stderr to buffered, thus fprintf(stderr)
310 * always flushes its output and this is not necessary: */
311 /* fflush(stderr); */
312 }
313
error_msg(const char * fmt,...)314 void error_msg(const char *fmt, ...)
315 {
316 va_list p;
317 va_start(p, fmt);
318 verror_msg(0, fmt, p);
319 va_end(p);
320 }
321
error_msg_and_die(const char * fmt,...)322 void error_msg_and_die(const char *fmt, ...)
323 {
324 va_list p;
325 va_start(p, fmt);
326 verror_msg(0, fmt, p);
327 die();
328 }
329
error_msg_and_help(const char * fmt,...)330 void error_msg_and_help(const char *fmt, ...)
331 {
332 if (fmt != NULL) {
333 va_list p;
334 va_start(p, fmt);
335 verror_msg(0, fmt, p);
336 }
337 fprintf(stderr, "Try '%s -h' for more information.\n", progname);
338 die();
339 }
340
perror_msg(const char * fmt,...)341 void perror_msg(const char *fmt, ...)
342 {
343 va_list p;
344 va_start(p, fmt);
345 verror_msg(errno, fmt, p);
346 va_end(p);
347 }
348
perror_msg_and_die(const char * fmt,...)349 void perror_msg_and_die(const char *fmt, ...)
350 {
351 va_list p;
352 va_start(p, fmt);
353 verror_msg(errno, fmt, p);
354 die();
355 }
356
357 static void
error_opt_arg(int opt,const char * arg)358 error_opt_arg(int opt, const char *arg)
359 {
360 error_msg_and_help("invalid -%c argument: '%s'", opt, arg);
361 }
362
363 #if USE_SEIZE
364 static int
ptrace_attach_or_seize(int pid)365 ptrace_attach_or_seize(int pid)
366 {
367 int r;
368 if (!use_seize)
369 return ptrace(PTRACE_ATTACH, pid, 0L, 0L);
370 r = ptrace(PTRACE_SEIZE, pid, 0L, (unsigned long) ptrace_setoptions);
371 if (r)
372 return r;
373 r = ptrace(PTRACE_INTERRUPT, pid, 0L, 0L);
374 return r;
375 }
376 #else
377 # define ptrace_attach_or_seize(pid) ptrace(PTRACE_ATTACH, (pid), 0, 0)
378 #endif
379
380 /*
381 * Used when we want to unblock stopped traced process.
382 * Should be only used with PTRACE_CONT, PTRACE_DETACH and PTRACE_SYSCALL.
383 * Returns 0 on success or if error was ESRCH
384 * (presumably process was killed while we talk to it).
385 * Otherwise prints error message and returns -1.
386 */
387 static int
ptrace_restart(int op,struct tcb * tcp,int sig)388 ptrace_restart(int op, struct tcb *tcp, int sig)
389 {
390 int err;
391 const char *msg;
392
393 errno = 0;
394 ptrace(op, tcp->pid, (void *) 0, (long) sig);
395 err = errno;
396 if (!err)
397 return 0;
398
399 msg = "SYSCALL";
400 if (op == PTRACE_CONT)
401 msg = "CONT";
402 if (op == PTRACE_DETACH)
403 msg = "DETACH";
404 #ifdef PTRACE_LISTEN
405 if (op == PTRACE_LISTEN)
406 msg = "LISTEN";
407 #endif
408 /*
409 * Why curcol != 0? Otherwise sometimes we get this:
410 *
411 * 10252 kill(10253, SIGKILL) = 0
412 * <ptrace(SYSCALL,10252):No such process>10253 ...next decode...
413 *
414 * 10252 died after we retrieved syscall exit data,
415 * but before we tried to restart it. Log looks ugly.
416 */
417 if (current_tcp && current_tcp->curcol != 0) {
418 tprintf(" <ptrace(%s):%s>\n", msg, strerror(err));
419 line_ended();
420 }
421 if (err == ESRCH)
422 return 0;
423 errno = err;
424 perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%d)", msg, tcp->pid, sig);
425 return -1;
426 }
427
428 static void
set_cloexec_flag(int fd)429 set_cloexec_flag(int fd)
430 {
431 int flags, newflags;
432
433 flags = fcntl(fd, F_GETFD);
434 if (flags < 0) {
435 /* Can happen only if fd is bad.
436 * Should never happen: if it does, we have a bug
437 * in the caller. Therefore we just abort
438 * instead of propagating the error.
439 */
440 perror_msg_and_die("fcntl(%d, F_GETFD)", fd);
441 }
442
443 newflags = flags | FD_CLOEXEC;
444 if (flags == newflags)
445 return;
446
447 fcntl(fd, F_SETFD, newflags); /* never fails */
448 }
449
kill_save_errno(pid_t pid,int sig)450 static void kill_save_errno(pid_t pid, int sig)
451 {
452 int saved_errno = errno;
453
454 (void) kill(pid, sig);
455 errno = saved_errno;
456 }
457
458 /*
459 * When strace is setuid executable, we have to swap uids
460 * before and after filesystem and process management operations.
461 */
462 static void
swap_uid(void)463 swap_uid(void)
464 {
465 int euid = geteuid(), uid = getuid();
466
467 if (euid != uid && setreuid(euid, uid) < 0) {
468 perror_msg_and_die("setreuid");
469 }
470 }
471
472 #ifdef _LARGEFILE64_SOURCE
473 # ifdef HAVE_FOPEN64
474 # define fopen_for_output fopen64
475 # else
476 # define fopen_for_output fopen
477 # endif
478 # define struct_stat struct stat64
479 # define stat_file stat64
480 # define struct_dirent struct dirent64
481 # define read_dir readdir64
482 # define struct_rlimit struct rlimit64
483 # define set_rlimit setrlimit64
484 #else
485 # define fopen_for_output fopen
486 # define struct_stat struct stat
487 # define stat_file stat
488 # define struct_dirent struct dirent
489 # define read_dir readdir
490 # define struct_rlimit struct rlimit
491 # define set_rlimit setrlimit
492 #endif
493
494 static FILE *
strace_fopen(const char * path)495 strace_fopen(const char *path)
496 {
497 FILE *fp;
498
499 swap_uid();
500 fp = fopen_for_output(path, "w");
501 if (!fp)
502 perror_msg_and_die("Can't fopen '%s'", path);
503 swap_uid();
504 set_cloexec_flag(fileno(fp));
505 return fp;
506 }
507
508 static int popen_pid = 0;
509
510 #ifndef _PATH_BSHELL
511 # define _PATH_BSHELL "/bin/sh"
512 #endif
513
514 /*
515 * We cannot use standard popen(3) here because we have to distinguish
516 * popen child process from other processes we trace, and standard popen(3)
517 * does not export its child's pid.
518 */
519 static FILE *
strace_popen(const char * command)520 strace_popen(const char *command)
521 {
522 FILE *fp;
523 int pid;
524 int fds[2];
525
526 swap_uid();
527 if (pipe(fds) < 0)
528 perror_msg_and_die("pipe");
529
530 set_cloexec_flag(fds[1]); /* never fails */
531
532 pid = vfork();
533 if (pid < 0)
534 perror_msg_and_die("vfork");
535
536 if (pid == 0) {
537 /* child */
538 close(fds[1]);
539 if (fds[0] != 0) {
540 if (dup2(fds[0], 0))
541 perror_msg_and_die("dup2");
542 close(fds[0]);
543 }
544 execl(_PATH_BSHELL, "sh", "-c", command, NULL);
545 perror_msg_and_die("Can't execute '%s'", _PATH_BSHELL);
546 }
547
548 /* parent */
549 popen_pid = pid;
550 close(fds[0]);
551 swap_uid();
552 fp = fdopen(fds[1], "w");
553 if (!fp)
554 die_out_of_memory();
555 return fp;
556 }
557
558 void
tprintf(const char * fmt,...)559 tprintf(const char *fmt, ...)
560 {
561 va_list args;
562
563 va_start(args, fmt);
564 if (current_tcp) {
565 int n = strace_vfprintf(current_tcp->outf, fmt, args);
566 if (n < 0) {
567 if (current_tcp->outf != stderr)
568 perror_msg("%s", outfname);
569 } else
570 current_tcp->curcol += n;
571 }
572 va_end(args);
573 }
574
575 #ifndef HAVE_FPUTS_UNLOCKED
576 # define fputs_unlocked fputs
577 #endif
578
579 void
tprints(const char * str)580 tprints(const char *str)
581 {
582 if (current_tcp) {
583 int n = fputs_unlocked(str, current_tcp->outf);
584 if (n >= 0) {
585 current_tcp->curcol += strlen(str);
586 return;
587 }
588 if (current_tcp->outf != stderr)
589 perror_msg("%s", outfname);
590 }
591 }
592
593 void
line_ended(void)594 line_ended(void)
595 {
596 if (current_tcp) {
597 current_tcp->curcol = 0;
598 fflush(current_tcp->outf);
599 }
600 if (printing_tcp) {
601 printing_tcp->curcol = 0;
602 printing_tcp = NULL;
603 }
604 }
605
606 void
printleader(struct tcb * tcp)607 printleader(struct tcb *tcp)
608 {
609 /* If -ff, "previous tcb we printed" is always the same as current,
610 * because we have per-tcb output files.
611 */
612 if (followfork >= 2)
613 printing_tcp = tcp;
614
615 if (printing_tcp) {
616 current_tcp = printing_tcp;
617 if (printing_tcp->curcol != 0 && (followfork < 2 || printing_tcp == tcp)) {
618 /*
619 * case 1: we have a shared log (i.e. not -ff), and last line
620 * wasn't finished (same or different tcb, doesn't matter).
621 * case 2: split log, we are the same tcb, but our last line
622 * didn't finish ("SIGKILL nuked us after syscall entry" etc).
623 */
624 tprints(" <unfinished ...>\n");
625 printing_tcp->curcol = 0;
626 }
627 }
628
629 printing_tcp = tcp;
630 current_tcp = tcp;
631 current_tcp->curcol = 0;
632
633 if (print_pid_pfx)
634 tprintf("%-5d ", tcp->pid);
635 else if (nprocs > 1 && !outfname)
636 tprintf("[pid %5u] ", tcp->pid);
637
638 if (tflag) {
639 char str[sizeof("HH:MM:SS")];
640 struct timeval tv, dtv;
641 static struct timeval otv;
642
643 gettimeofday(&tv, NULL);
644 if (rflag) {
645 if (otv.tv_sec == 0)
646 otv = tv;
647 tv_sub(&dtv, &tv, &otv);
648 tprintf("%6ld.%06ld ",
649 (long) dtv.tv_sec, (long) dtv.tv_usec);
650 otv = tv;
651 }
652 else if (tflag > 2) {
653 tprintf("%ld.%06ld ",
654 (long) tv.tv_sec, (long) tv.tv_usec);
655 }
656 else {
657 time_t local = tv.tv_sec;
658 strftime(str, sizeof(str), "%T", localtime(&local));
659 if (tflag > 1)
660 tprintf("%s.%06ld ", str, (long) tv.tv_usec);
661 else
662 tprintf("%s ", str);
663 }
664 }
665 if (iflag)
666 print_pc(tcp);
667 }
668
669 void
tabto(void)670 tabto(void)
671 {
672 if (current_tcp->curcol < acolumn)
673 tprints(acolumn_spaces + current_tcp->curcol);
674 }
675
676 /* Should be only called directly *after successful attach* to a tracee.
677 * Otherwise, "strace -oFILE -ff -p<nonexistant_pid>"
678 * may create bogus empty FILE.<nonexistant_pid>, and then die.
679 */
680 static void
newoutf(struct tcb * tcp)681 newoutf(struct tcb *tcp)
682 {
683 tcp->outf = shared_log; /* if not -ff mode, the same file is for all */
684 if (followfork >= 2) {
685 char name[520 + sizeof(int) * 3];
686 sprintf(name, "%.512s.%u", outfname, tcp->pid);
687 tcp->outf = strace_fopen(name);
688 }
689 }
690
691 static void
expand_tcbtab(void)692 expand_tcbtab(void)
693 {
694 /* Allocate some more TCBs and expand the table.
695 We don't want to relocate the TCBs because our
696 callers have pointers and it would be a pain.
697 So tcbtab is a table of pointers. Since we never
698 free the TCBs, we allocate a single chunk of many. */
699 unsigned int i = tcbtabsize;
700 struct tcb *newtcbs = xcalloc(tcbtabsize, sizeof(newtcbs[0]));
701 struct tcb **newtab = xreallocarray(tcbtab, tcbtabsize * 2,
702 sizeof(tcbtab[0]));
703 tcbtabsize *= 2;
704 tcbtab = newtab;
705 while (i < tcbtabsize)
706 tcbtab[i++] = newtcbs++;
707 }
708
709 static struct tcb *
alloctcb(int pid)710 alloctcb(int pid)
711 {
712 unsigned int i;
713 struct tcb *tcp;
714
715 if (nprocs == tcbtabsize)
716 expand_tcbtab();
717
718 for (i = 0; i < tcbtabsize; i++) {
719 tcp = tcbtab[i];
720 if (!tcp->pid) {
721 memset(tcp, 0, sizeof(*tcp));
722 tcp->pid = pid;
723 #if SUPPORTED_PERSONALITIES > 1
724 tcp->currpers = current_personality;
725 #endif
726
727 #ifdef USE_LIBUNWIND
728 if (stack_trace_enabled)
729 unwind_tcb_init(tcp);
730 #endif
731
732 nprocs++;
733 if (debug_flag)
734 error_msg("new tcb for pid %d, active tcbs:%d",
735 tcp->pid, nprocs);
736 return tcp;
737 }
738 }
739 error_msg_and_die("bug in alloctcb");
740 }
741
742 static void
droptcb(struct tcb * tcp)743 droptcb(struct tcb *tcp)
744 {
745 if (tcp->pid == 0)
746 return;
747
748 #ifdef USE_LIBUNWIND
749 if (stack_trace_enabled) {
750 unwind_tcb_fin(tcp);
751 }
752 #endif
753
754 nprocs--;
755 if (debug_flag)
756 error_msg("dropped tcb for pid %d, %d remain",
757 tcp->pid, nprocs);
758
759 if (tcp->outf) {
760 if (followfork >= 2) {
761 if (tcp->curcol != 0)
762 fprintf(tcp->outf, " <detached ...>\n");
763 fclose(tcp->outf);
764 } else {
765 if (printing_tcp == tcp && tcp->curcol != 0)
766 fprintf(tcp->outf, " <detached ...>\n");
767 fflush(tcp->outf);
768 }
769 }
770
771 if (current_tcp == tcp)
772 current_tcp = NULL;
773 if (printing_tcp == tcp)
774 printing_tcp = NULL;
775
776 memset(tcp, 0, sizeof(*tcp));
777 }
778
779 /* Detach traced process.
780 * Never call DETACH twice on the same process as both unattached and
781 * attached-unstopped processes give the same ESRCH. For unattached process we
782 * would SIGSTOP it and wait for its SIGSTOP notification forever.
783 */
784 static void
detach(struct tcb * tcp)785 detach(struct tcb *tcp)
786 {
787 int error;
788 int status;
789
790 /*
791 * Linux wrongly insists the child be stopped
792 * before detaching. Arghh. We go through hoops
793 * to make a clean break of things.
794 */
795 #if defined(SPARC)
796 # undef PTRACE_DETACH
797 # define PTRACE_DETACH PTRACE_SUNDETACH
798 #endif
799
800 if (!(tcp->flags & TCB_ATTACHED))
801 goto drop;
802
803 /* We attached but possibly didn't see the expected SIGSTOP.
804 * We must catch exactly one as otherwise the detached process
805 * would be left stopped (process state T).
806 */
807 if (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)
808 goto wait_loop;
809
810 error = ptrace(PTRACE_DETACH, tcp->pid, 0, 0);
811 if (!error) {
812 /* On a clear day, you can see forever. */
813 goto drop;
814 }
815 if (errno != ESRCH) {
816 /* Shouldn't happen. */
817 perror_msg("detach: ptrace(PTRACE_DETACH,%u)", tcp->pid);
818 goto drop;
819 }
820 /* ESRCH: process is either not stopped or doesn't exist. */
821 if (my_tkill(tcp->pid, 0) < 0) {
822 if (errno != ESRCH)
823 /* Shouldn't happen. */
824 perror_msg("detach: tkill(%u,0)", tcp->pid);
825 /* else: process doesn't exist. */
826 goto drop;
827 }
828 /* Process is not stopped, need to stop it. */
829 if (use_seize) {
830 /*
831 * With SEIZE, tracee can be in group-stop already.
832 * In this state sending it another SIGSTOP does nothing.
833 * Need to use INTERRUPT.
834 * Testcase: trying to ^C a "strace -p <stopped_process>".
835 */
836 error = ptrace(PTRACE_INTERRUPT, tcp->pid, 0, 0);
837 if (!error)
838 goto wait_loop;
839 if (errno != ESRCH)
840 perror_msg("detach: ptrace(PTRACE_INTERRUPT,%u)", tcp->pid);
841 }
842 else {
843 error = my_tkill(tcp->pid, SIGSTOP);
844 if (!error)
845 goto wait_loop;
846 if (errno != ESRCH)
847 perror_msg("detach: tkill(%u,SIGSTOP)", tcp->pid);
848 }
849 /* Either process doesn't exist, or some weird error. */
850 goto drop;
851
852 wait_loop:
853 /* We end up here in three cases:
854 * 1. We sent PTRACE_INTERRUPT (use_seize case)
855 * 2. We sent SIGSTOP (!use_seize)
856 * 3. Attach SIGSTOP was already pending (TCB_IGNORE_ONE_SIGSTOP set)
857 */
858 for (;;) {
859 unsigned int sig;
860 if (waitpid(tcp->pid, &status, __WALL) < 0) {
861 if (errno == EINTR)
862 continue;
863 /*
864 * if (errno == ECHILD) break;
865 * ^^^ WRONG! We expect this PID to exist,
866 * and want to emit a message otherwise:
867 */
868 perror_msg("detach: waitpid(%u)", tcp->pid);
869 break;
870 }
871 if (!WIFSTOPPED(status)) {
872 /*
873 * Tracee exited or was killed by signal.
874 * We shouldn't normally reach this place:
875 * we don't want to consume exit status.
876 * Consider "strace -p PID" being ^C-ed:
877 * we want merely to detach from PID.
878 *
879 * However, we _can_ end up here if tracee
880 * was SIGKILLed.
881 */
882 break;
883 }
884 sig = WSTOPSIG(status);
885 if (debug_flag)
886 error_msg("detach wait: event:%d sig:%d",
887 (unsigned)status >> 16, sig);
888 if (use_seize) {
889 unsigned event = (unsigned)status >> 16;
890 if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) {
891 /*
892 * sig == SIGTRAP: PTRACE_INTERRUPT stop.
893 * sig == other: process was already stopped
894 * with this stopping sig (see tests/detach-stopped).
895 * Looks like re-injecting this sig is not necessary
896 * in DETACH for the tracee to remain stopped.
897 */
898 sig = 0;
899 }
900 /*
901 * PTRACE_INTERRUPT is not guaranteed to produce
902 * the above event if other ptrace-stop is pending.
903 * See tests/detach-sleeping testcase:
904 * strace got SIGINT while tracee is sleeping.
905 * We sent PTRACE_INTERRUPT.
906 * We see syscall exit, not PTRACE_INTERRUPT stop.
907 * We won't get PTRACE_INTERRUPT stop
908 * if we would CONT now. Need to DETACH.
909 */
910 if (sig == syscall_trap_sig)
911 sig = 0;
912 /* else: not sure in which case we can be here.
913 * Signal stop? Inject it while detaching.
914 */
915 ptrace_restart(PTRACE_DETACH, tcp, sig);
916 break;
917 }
918 /* Note: this check has to be after use_seize check */
919 /* (else, in use_seize case SIGSTOP will be mistreated) */
920 if (sig == SIGSTOP) {
921 /* Detach, suppressing SIGSTOP */
922 ptrace_restart(PTRACE_DETACH, tcp, 0);
923 break;
924 }
925 if (sig == syscall_trap_sig)
926 sig = 0;
927 /* Can't detach just yet, may need to wait for SIGSTOP */
928 error = ptrace_restart(PTRACE_CONT, tcp, sig);
929 if (error < 0) {
930 /* Should not happen.
931 * Note: ptrace_restart returns 0 on ESRCH, so it's not it.
932 * ptrace_restart already emitted error message.
933 */
934 break;
935 }
936 }
937
938 drop:
939 if (!qflag && (tcp->flags & TCB_ATTACHED))
940 error_msg("Process %u detached", tcp->pid);
941
942 droptcb(tcp);
943 }
944
945 static void
process_opt_p_list(char * opt)946 process_opt_p_list(char *opt)
947 {
948 while (*opt) {
949 /*
950 * We accept -p PID,PID; -p "`pidof PROG`"; -p "`pgrep PROG`".
951 * pidof uses space as delim, pgrep uses newline. :(
952 */
953 int pid;
954 char *delim = opt + strcspn(opt, ", \n\t");
955 char c = *delim;
956
957 *delim = '\0';
958 pid = string_to_uint(opt);
959 if (pid <= 0) {
960 error_msg_and_die("Invalid process id: '%s'", opt);
961 }
962 if (pid == strace_tracer_pid) {
963 error_msg_and_die("I'm sorry, I can't let you do that, Dave.");
964 }
965 *delim = c;
966 alloctcb(pid);
967 if (c == '\0')
968 break;
969 opt = delim + 1;
970 }
971 }
972
973 static void
startup_attach(void)974 startup_attach(void)
975 {
976 unsigned int tcbi;
977 struct tcb *tcp;
978
979 /*
980 * Block user interruptions as we would leave the traced
981 * process stopped (process state T) if we would terminate in
982 * between PTRACE_ATTACH and wait4() on SIGSTOP.
983 * We rely on cleanup() from this point on.
984 */
985 if (interactive)
986 sigprocmask(SIG_BLOCK, &blocked_set, NULL);
987
988 if (daemonized_tracer) {
989 pid_t pid = fork();
990 if (pid < 0) {
991 perror_msg_and_die("fork");
992 }
993 if (pid) { /* parent */
994 /*
995 * Wait for grandchild to attach to straced process
996 * (grandparent). Grandchild SIGKILLs us after it attached.
997 * Grandparent's wait() is unblocked by our death,
998 * it proceeds to exec the straced program.
999 */
1000 pause();
1001 _exit(0); /* paranoia */
1002 }
1003 /* grandchild */
1004 /* We will be the tracer process. Remember our new pid: */
1005 strace_tracer_pid = getpid();
1006 }
1007
1008 for (tcbi = 0; tcbi < tcbtabsize; tcbi++) {
1009 tcp = tcbtab[tcbi];
1010
1011 if (!tcp->pid)
1012 continue;
1013
1014 /* Is this a process we should attach to, but not yet attached? */
1015 if (tcp->flags & TCB_ATTACHED)
1016 continue; /* no, we already attached it */
1017
1018 if (followfork && !daemonized_tracer) {
1019 char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3];
1020 DIR *dir;
1021
1022 sprintf(procdir, "/proc/%d/task", tcp->pid);
1023 dir = opendir(procdir);
1024 if (dir != NULL) {
1025 unsigned int ntid = 0, nerr = 0;
1026 struct_dirent *de;
1027
1028 while ((de = read_dir(dir)) != NULL) {
1029 struct tcb *cur_tcp;
1030 int tid;
1031
1032 if (de->d_fileno == 0)
1033 continue;
1034 /* we trust /proc filesystem */
1035 tid = atoi(de->d_name);
1036 if (tid <= 0)
1037 continue;
1038 ++ntid;
1039 if (ptrace_attach_or_seize(tid) < 0) {
1040 ++nerr;
1041 if (debug_flag)
1042 error_msg("attach to pid %d failed", tid);
1043 continue;
1044 }
1045 if (debug_flag)
1046 error_msg("attach to pid %d succeeded", tid);
1047 cur_tcp = tcp;
1048 if (tid != tcp->pid)
1049 cur_tcp = alloctcb(tid);
1050 cur_tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
1051 newoutf(cur_tcp);
1052 }
1053 closedir(dir);
1054 if (interactive) {
1055 sigprocmask(SIG_SETMASK, &empty_set, NULL);
1056 if (interrupted)
1057 goto ret;
1058 sigprocmask(SIG_BLOCK, &blocked_set, NULL);
1059 }
1060 ntid -= nerr;
1061 if (ntid == 0) {
1062 perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
1063 droptcb(tcp);
1064 continue;
1065 }
1066 if (!qflag) {
1067 if (ntid > 1)
1068 error_msg("Process %u attached"
1069 " with %u threads",
1070 tcp->pid, ntid);
1071 else
1072 error_msg("Process %u attached",
1073 tcp->pid);
1074 }
1075 if (!(tcp->flags & TCB_ATTACHED)) {
1076 /* -p PID, we failed to attach to PID itself
1077 * but did attach to some of its sibling threads.
1078 * Drop PID's tcp.
1079 */
1080 droptcb(tcp);
1081 }
1082 continue;
1083 } /* if (opendir worked) */
1084 } /* if (-f) */
1085 if (ptrace_attach_or_seize(tcp->pid) < 0) {
1086 perror_msg("attach: ptrace(PTRACE_ATTACH, ...)");
1087 droptcb(tcp);
1088 continue;
1089 }
1090 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
1091 newoutf(tcp);
1092 if (debug_flag)
1093 error_msg("attach to pid %d (main) succeeded", tcp->pid);
1094
1095 if (daemonized_tracer) {
1096 /*
1097 * Make parent go away.
1098 * Also makes grandparent's wait() unblock.
1099 */
1100 kill(getppid(), SIGKILL);
1101 }
1102
1103 if (!qflag)
1104 error_msg("Process %u attached", tcp->pid);
1105 } /* for each tcbtab[] */
1106
1107 ret:
1108 if (interactive)
1109 sigprocmask(SIG_SETMASK, &empty_set, NULL);
1110 }
1111
1112 /* Stack-o-phobic exec helper, in the hope to work around
1113 * NOMMU + "daemonized tracer" difficulty.
1114 */
1115 struct exec_params {
1116 int fd_to_close;
1117 uid_t run_euid;
1118 gid_t run_egid;
1119 char **argv;
1120 char *pathname;
1121 };
1122 static struct exec_params params_for_tracee;
1123
1124 static void ATTRIBUTE_NOINLINE ATTRIBUTE_NORETURN
exec_or_die(void)1125 exec_or_die(void)
1126 {
1127 struct exec_params *params = ¶ms_for_tracee;
1128
1129 if (params->fd_to_close >= 0)
1130 close(params->fd_to_close);
1131 if (!daemonized_tracer && !use_seize) {
1132 if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) {
1133 perror_msg_and_die("ptrace(PTRACE_TRACEME, ...)");
1134 }
1135 }
1136
1137 if (username != NULL) {
1138 /*
1139 * It is important to set groups before we
1140 * lose privileges on setuid.
1141 */
1142 if (initgroups(username, run_gid) < 0) {
1143 perror_msg_and_die("initgroups");
1144 }
1145 if (setregid(run_gid, params->run_egid) < 0) {
1146 perror_msg_and_die("setregid");
1147 }
1148 if (setreuid(run_uid, params->run_euid) < 0) {
1149 perror_msg_and_die("setreuid");
1150 }
1151 }
1152 else if (geteuid() != 0)
1153 if (setreuid(run_uid, run_uid) < 0) {
1154 perror_msg_and_die("setreuid");
1155 }
1156
1157 if (!daemonized_tracer) {
1158 /*
1159 * Induce a ptrace stop. Tracer (our parent)
1160 * will resume us with PTRACE_SYSCALL and display
1161 * the immediately following execve syscall.
1162 * Can't do this on NOMMU systems, we are after
1163 * vfork: parent is blocked, stopping would deadlock.
1164 */
1165 if (!NOMMU_SYSTEM)
1166 kill(getpid(), SIGSTOP);
1167 } else {
1168 alarm(3);
1169 /* we depend on SIGCHLD set to SIG_DFL by init code */
1170 /* if it happens to be SIG_IGN'ed, wait won't block */
1171 wait(NULL);
1172 alarm(0);
1173 }
1174
1175 execv(params->pathname, params->argv);
1176 perror_msg_and_die("exec");
1177 }
1178
1179 static void
startup_child(char ** argv)1180 startup_child(char **argv)
1181 {
1182 struct_stat statbuf;
1183 const char *filename;
1184 size_t filename_len;
1185 char pathname[PATH_MAX];
1186 int pid;
1187 struct tcb *tcp;
1188
1189 filename = argv[0];
1190 filename_len = strlen(filename);
1191
1192 if (filename_len > sizeof(pathname) - 1) {
1193 errno = ENAMETOOLONG;
1194 perror_msg_and_die("exec");
1195 }
1196 if (strchr(filename, '/')) {
1197 strcpy(pathname, filename);
1198 }
1199 #ifdef USE_DEBUGGING_EXEC
1200 /*
1201 * Debuggers customarily check the current directory
1202 * first regardless of the path but doing that gives
1203 * security geeks a panic attack.
1204 */
1205 else if (stat_file(filename, &statbuf) == 0)
1206 strcpy(pathname, filename);
1207 #endif /* USE_DEBUGGING_EXEC */
1208 else {
1209 const char *path;
1210 size_t m, n, len;
1211
1212 for (path = getenv("PATH"); path && *path; path += m) {
1213 const char *colon = strchr(path, ':');
1214 if (colon) {
1215 n = colon - path;
1216 m = n + 1;
1217 }
1218 else
1219 m = n = strlen(path);
1220 if (n == 0) {
1221 if (!getcwd(pathname, PATH_MAX))
1222 continue;
1223 len = strlen(pathname);
1224 }
1225 else if (n > sizeof pathname - 1)
1226 continue;
1227 else {
1228 strncpy(pathname, path, n);
1229 len = n;
1230 }
1231 if (len && pathname[len - 1] != '/')
1232 pathname[len++] = '/';
1233 if (filename_len + len > sizeof(pathname) - 1)
1234 continue;
1235 strcpy(pathname + len, filename);
1236 if (stat_file(pathname, &statbuf) == 0 &&
1237 /* Accept only regular files
1238 with some execute bits set.
1239 XXX not perfect, might still fail */
1240 S_ISREG(statbuf.st_mode) &&
1241 (statbuf.st_mode & 0111))
1242 break;
1243 }
1244 if (!path || !*path)
1245 pathname[0] = '\0';
1246 }
1247 if (stat_file(pathname, &statbuf) < 0) {
1248 perror_msg_and_die("Can't stat '%s'", filename);
1249 }
1250
1251 params_for_tracee.fd_to_close = (shared_log != stderr) ? fileno(shared_log) : -1;
1252 params_for_tracee.run_euid = (statbuf.st_mode & S_ISUID) ? statbuf.st_uid : run_uid;
1253 params_for_tracee.run_egid = (statbuf.st_mode & S_ISGID) ? statbuf.st_gid : run_gid;
1254 params_for_tracee.argv = argv;
1255 /*
1256 * On NOMMU, can be safely freed only after execve in tracee.
1257 * It's hard to know when that happens, so we just leak it.
1258 */
1259 params_for_tracee.pathname = NOMMU_SYSTEM ? xstrdup(pathname) : pathname;
1260
1261 #if defined HAVE_PRCTL && defined PR_SET_PTRACER && defined PR_SET_PTRACER_ANY
1262 if (daemonized_tracer)
1263 prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY);
1264 #endif
1265
1266 pid = fork();
1267 if (pid < 0) {
1268 perror_msg_and_die("fork");
1269 }
1270 if ((pid != 0 && daemonized_tracer)
1271 || (pid == 0 && !daemonized_tracer)
1272 ) {
1273 /* We are to become the tracee. Two cases:
1274 * -D: we are parent
1275 * not -D: we are child
1276 */
1277 exec_or_die();
1278 }
1279
1280 /* We are the tracer */
1281
1282 if (!daemonized_tracer) {
1283 strace_child = pid;
1284 if (!use_seize) {
1285 /* child did PTRACE_TRACEME, nothing to do in parent */
1286 } else {
1287 if (!NOMMU_SYSTEM) {
1288 /* Wait until child stopped itself */
1289 int status;
1290 while (waitpid(pid, &status, WSTOPPED) < 0) {
1291 if (errno == EINTR)
1292 continue;
1293 perror_msg_and_die("waitpid");
1294 }
1295 if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP) {
1296 kill_save_errno(pid, SIGKILL);
1297 perror_msg_and_die("Unexpected wait status %x", status);
1298 }
1299 }
1300 /* Else: NOMMU case, we have no way to sync.
1301 * Just attach to it as soon as possible.
1302 * This means that we may miss a few first syscalls...
1303 */
1304
1305 if (ptrace_attach_or_seize(pid)) {
1306 kill_save_errno(pid, SIGKILL);
1307 perror_msg_and_die("Can't attach to %d", pid);
1308 }
1309 if (!NOMMU_SYSTEM)
1310 kill(pid, SIGCONT);
1311 }
1312 tcp = alloctcb(pid);
1313 if (!NOMMU_SYSTEM)
1314 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
1315 else
1316 tcp->flags |= TCB_ATTACHED | TCB_STARTUP;
1317 newoutf(tcp);
1318 }
1319 else {
1320 /* With -D, we are *child* here, IOW: different pid. Fetch it: */
1321 strace_tracer_pid = getpid();
1322 /* The tracee is our parent: */
1323 pid = getppid();
1324 alloctcb(pid);
1325 /* attaching will be done later, by startup_attach */
1326 /* note: we don't do newoutf(tcp) here either! */
1327
1328 /* NOMMU BUG! -D mode is active, we (child) return,
1329 * and we will scribble over parent's stack!
1330 * When parent later unpauses, it segfaults.
1331 *
1332 * We work around it
1333 * (1) by declaring exec_or_die() NORETURN,
1334 * hopefully compiler will just jump to it
1335 * instead of call (won't push anything to stack),
1336 * (2) by trying very hard in exec_or_die()
1337 * to not use any stack,
1338 * (3) having a really big (PATH_MAX) stack object
1339 * in this function, which creates a "buffer" between
1340 * child's and parent's stack pointers.
1341 * This may save us if (1) and (2) failed
1342 * and compiler decided to use stack in exec_or_die() anyway
1343 * (happens on i386 because of stack parameter passing).
1344 *
1345 * A cleaner solution is to use makecontext + setcontext
1346 * to create a genuine separate stack and execute on it.
1347 */
1348 }
1349 }
1350
1351 #if USE_SEIZE
1352 static void
test_ptrace_seize(void)1353 test_ptrace_seize(void)
1354 {
1355 int pid;
1356
1357 /* Need fork for test. NOMMU has no forks */
1358 if (NOMMU_SYSTEM) {
1359 post_attach_sigstop = 0; /* this sets use_seize to 1 */
1360 return;
1361 }
1362
1363 pid = fork();
1364 if (pid < 0)
1365 perror_msg_and_die("fork");
1366
1367 if (pid == 0) {
1368 pause();
1369 _exit(0);
1370 }
1371
1372 /* PTRACE_SEIZE, unlike ATTACH, doesn't force tracee to trap. After
1373 * attaching tracee continues to run unless a trap condition occurs.
1374 * PTRACE_SEIZE doesn't affect signal or group stop state.
1375 */
1376 if (ptrace(PTRACE_SEIZE, pid, 0, 0) == 0) {
1377 post_attach_sigstop = 0; /* this sets use_seize to 1 */
1378 } else if (debug_flag) {
1379 error_msg("PTRACE_SEIZE doesn't work");
1380 }
1381
1382 kill(pid, SIGKILL);
1383
1384 while (1) {
1385 int status, tracee_pid;
1386
1387 errno = 0;
1388 tracee_pid = waitpid(pid, &status, 0);
1389 if (tracee_pid <= 0) {
1390 if (errno == EINTR)
1391 continue;
1392 perror_msg_and_die("%s: unexpected wait result %d",
1393 __func__, tracee_pid);
1394 }
1395 if (WIFSIGNALED(status)) {
1396 return;
1397 }
1398 error_msg_and_die("%s: unexpected wait status %x",
1399 __func__, status);
1400 }
1401 }
1402 #else /* !USE_SEIZE */
1403 # define test_ptrace_seize() ((void)0)
1404 #endif
1405
1406 static unsigned
get_os_release(void)1407 get_os_release(void)
1408 {
1409 unsigned rel;
1410 const char *p;
1411 struct utsname u;
1412 if (uname(&u) < 0)
1413 perror_msg_and_die("uname");
1414 /* u.release has this form: "3.2.9[-some-garbage]" */
1415 rel = 0;
1416 p = u.release;
1417 for (;;) {
1418 if (!(*p >= '0' && *p <= '9'))
1419 error_msg_and_die("Bad OS release string: '%s'", u.release);
1420 /* Note: this open-codes KERNEL_VERSION(): */
1421 rel = (rel << 8) | atoi(p);
1422 if (rel >= KERNEL_VERSION(1,0,0))
1423 break;
1424 while (*p >= '0' && *p <= '9')
1425 p++;
1426 if (*p != '.') {
1427 if (rel >= KERNEL_VERSION(0,1,0)) {
1428 /* "X.Y-something" means "X.Y.0" */
1429 rel <<= 8;
1430 break;
1431 }
1432 error_msg_and_die("Bad OS release string: '%s'", u.release);
1433 }
1434 p++;
1435 }
1436 return rel;
1437 }
1438
1439 /*
1440 * Initialization part of main() was eating much stack (~0.5k),
1441 * which was unused after init.
1442 * We can reuse it if we move init code into a separate function.
1443 *
1444 * Don't want main() to inline us and defeat the reason
1445 * we have a separate function.
1446 */
1447 static void ATTRIBUTE_NOINLINE
init(int argc,char * argv[])1448 init(int argc, char *argv[])
1449 {
1450 struct tcb *tcp;
1451 int c, i;
1452 int optF = 0;
1453 unsigned int tcbi;
1454 struct sigaction sa;
1455
1456 progname = argv[0] ? argv[0] : "strace";
1457
1458 /* Make sure SIGCHLD has the default action so that waitpid
1459 definitely works without losing track of children. The user
1460 should not have given us a bogus state to inherit, but he might
1461 have. Arguably we should detect SIG_IGN here and pass it on
1462 to children, but probably noone really needs that. */
1463 signal(SIGCHLD, SIG_DFL);
1464
1465 strace_tracer_pid = getpid();
1466
1467 os_release = get_os_release();
1468
1469 /* Allocate the initial tcbtab. */
1470 tcbtabsize = argc; /* Surely enough for all -p args. */
1471 tcbtab = xcalloc(tcbtabsize, sizeof(tcbtab[0]));
1472 tcp = xcalloc(tcbtabsize, sizeof(*tcp));
1473 for (tcbi = 0; tcbi < tcbtabsize; tcbi++)
1474 tcbtab[tcbi] = tcp++;
1475
1476 shared_log = stderr;
1477 set_sortby(DEFAULT_SORTBY);
1478 set_personality(DEFAULT_PERSONALITY);
1479 qualify("trace=all");
1480 qualify("abbrev=all");
1481 qualify("verbose=all");
1482 #if DEFAULT_QUAL_FLAGS != (QUAL_TRACE | QUAL_ABBREV | QUAL_VERBOSE)
1483 # error Bug in DEFAULT_QUAL_FLAGS
1484 #endif
1485 qualify("signal=all");
1486 while ((c = getopt(argc, argv,
1487 "+b:cCdfFhiqrtTvVwxyz"
1488 #ifdef USE_LIBUNWIND
1489 "k"
1490 #endif
1491 "D"
1492 "a:e:o:O:p:s:S:u:E:P:I:")) != EOF) {
1493 switch (c) {
1494 case 'b':
1495 if (strcmp(optarg, "execve") != 0)
1496 error_msg_and_die("Syscall '%s' for -b isn't supported",
1497 optarg);
1498 detach_on_execve = 1;
1499 break;
1500 case 'c':
1501 if (cflag == CFLAG_BOTH) {
1502 error_msg_and_help("-c and -C are mutually exclusive");
1503 }
1504 cflag = CFLAG_ONLY_STATS;
1505 break;
1506 case 'C':
1507 if (cflag == CFLAG_ONLY_STATS) {
1508 error_msg_and_help("-c and -C are mutually exclusive");
1509 }
1510 cflag = CFLAG_BOTH;
1511 break;
1512 case 'd':
1513 debug_flag = 1;
1514 break;
1515 case 'D':
1516 daemonized_tracer = 1;
1517 break;
1518 case 'F':
1519 optF = 1;
1520 break;
1521 case 'f':
1522 followfork++;
1523 break;
1524 case 'h':
1525 usage();
1526 break;
1527 case 'i':
1528 iflag = 1;
1529 break;
1530 case 'q':
1531 qflag++;
1532 break;
1533 case 'r':
1534 rflag = 1;
1535 /* fall through to tflag++ */
1536 case 't':
1537 tflag++;
1538 break;
1539 case 'T':
1540 Tflag = 1;
1541 break;
1542 case 'w':
1543 count_wallclock = 1;
1544 break;
1545 case 'x':
1546 xflag++;
1547 break;
1548 case 'y':
1549 show_fd_path++;
1550 break;
1551 case 'v':
1552 qualify("abbrev=none");
1553 break;
1554 case 'V':
1555 printf("%s -- version %s\n", PACKAGE_NAME, VERSION);
1556 exit(0);
1557 break;
1558 case 'z':
1559 not_failing_only = 1;
1560 break;
1561 case 'a':
1562 acolumn = string_to_uint(optarg);
1563 if (acolumn < 0)
1564 error_opt_arg(c, optarg);
1565 break;
1566 case 'e':
1567 qualify(optarg);
1568 break;
1569 case 'o':
1570 outfname = xstrdup(optarg);
1571 break;
1572 case 'O':
1573 i = string_to_uint(optarg);
1574 if (i < 0)
1575 error_opt_arg(c, optarg);
1576 set_overhead(i);
1577 break;
1578 case 'p':
1579 process_opt_p_list(optarg);
1580 break;
1581 case 'P':
1582 pathtrace_select(optarg);
1583 break;
1584 case 's':
1585 i = string_to_uint(optarg);
1586 if (i < 0)
1587 error_opt_arg(c, optarg);
1588 max_strlen = i;
1589 break;
1590 case 'S':
1591 set_sortby(optarg);
1592 break;
1593 case 'u':
1594 username = xstrdup(optarg);
1595 break;
1596 #ifdef USE_LIBUNWIND
1597 case 'k':
1598 stack_trace_enabled = true;
1599 break;
1600 #endif
1601 case 'E':
1602 if (putenv(optarg) < 0)
1603 die_out_of_memory();
1604 break;
1605 case 'I':
1606 opt_intr = string_to_uint(optarg);
1607 if (opt_intr <= 0 || opt_intr >= NUM_INTR_OPTS)
1608 error_opt_arg(c, optarg);
1609 break;
1610 default:
1611 error_msg_and_help(NULL);
1612 break;
1613 }
1614 }
1615 argv += optind;
1616 /* argc -= optind; - no need, argc is not used below */
1617
1618 acolumn_spaces = xmalloc(acolumn + 1);
1619 memset(acolumn_spaces, ' ', acolumn);
1620 acolumn_spaces[acolumn] = '\0';
1621
1622 /* Must have PROG [ARGS], or -p PID. Not both. */
1623 if (!argv[0] == !nprocs) {
1624 error_msg_and_help("must have PROG [ARGS] or -p PID");
1625 }
1626
1627 if (nprocs != 0 && daemonized_tracer) {
1628 error_msg_and_help("-D and -p are mutually exclusive");
1629 }
1630
1631 if (!followfork)
1632 followfork = optF;
1633
1634 if (followfork >= 2 && cflag) {
1635 error_msg_and_help("(-c or -C) and -ff are mutually exclusive");
1636 }
1637
1638 if (count_wallclock && !cflag) {
1639 error_msg_and_help("-w must be given with (-c or -C)");
1640 }
1641
1642 if (cflag == CFLAG_ONLY_STATS) {
1643 if (iflag)
1644 error_msg("-%c has no effect with -c", 'i');
1645 #ifdef USE_LIBUNWIND
1646 if (stack_trace_enabled)
1647 error_msg("-%c has no effect with -c", 'k');
1648 #endif
1649 if (rflag)
1650 error_msg("-%c has no effect with -c", 'r');
1651 if (tflag)
1652 error_msg("-%c has no effect with -c", 't');
1653 if (Tflag)
1654 error_msg("-%c has no effect with -c", 'T');
1655 if (show_fd_path)
1656 error_msg("-%c has no effect with -c", 'y');
1657 }
1658
1659 #ifdef USE_LIBUNWIND
1660 if (stack_trace_enabled)
1661 unwind_init();
1662 #endif
1663
1664 /* See if they want to run as another user. */
1665 if (username != NULL) {
1666 struct passwd *pent;
1667
1668 if (getuid() != 0 || geteuid() != 0) {
1669 error_msg_and_die("You must be root to use the -u option");
1670 }
1671 pent = getpwnam(username);
1672 if (pent == NULL) {
1673 error_msg_and_die("Cannot find user '%s'", username);
1674 }
1675 run_uid = pent->pw_uid;
1676 run_gid = pent->pw_gid;
1677 }
1678 else {
1679 run_uid = getuid();
1680 run_gid = getgid();
1681 }
1682
1683 if (followfork)
1684 ptrace_setoptions |= PTRACE_O_TRACECLONE |
1685 PTRACE_O_TRACEFORK |
1686 PTRACE_O_TRACEVFORK;
1687 if (debug_flag)
1688 error_msg("ptrace_setoptions = %#x", ptrace_setoptions);
1689 test_ptrace_seize();
1690
1691 /* Check if they want to redirect the output. */
1692 if (outfname) {
1693 /* See if they want to pipe the output. */
1694 if (outfname[0] == '|' || outfname[0] == '!') {
1695 /*
1696 * We can't do the <outfname>.PID funny business
1697 * when using popen, so prohibit it.
1698 */
1699 if (followfork >= 2)
1700 error_msg_and_help("piping the output and -ff are mutually exclusive");
1701 shared_log = strace_popen(outfname + 1);
1702 }
1703 else if (followfork < 2)
1704 shared_log = strace_fopen(outfname);
1705 } else {
1706 /* -ff without -o FILE is the same as single -f */
1707 if (followfork >= 2)
1708 followfork = 1;
1709 }
1710
1711 if (!outfname || outfname[0] == '|' || outfname[0] == '!') {
1712 char *buf = xmalloc(BUFSIZ);
1713 setvbuf(shared_log, buf, _IOLBF, BUFSIZ);
1714 }
1715 if (outfname && argv[0]) {
1716 if (!opt_intr)
1717 opt_intr = INTR_NEVER;
1718 if (!qflag)
1719 qflag = 1;
1720 }
1721 if (!opt_intr)
1722 opt_intr = INTR_WHILE_WAIT;
1723
1724 /* argv[0] -pPID -oFILE Default interactive setting
1725 * yes 0 0 INTR_WHILE_WAIT
1726 * no 1 0 INTR_WHILE_WAIT
1727 * yes 0 1 INTR_NEVER
1728 * no 1 1 INTR_WHILE_WAIT
1729 */
1730
1731 sigemptyset(&empty_set);
1732 sigemptyset(&blocked_set);
1733
1734 /* startup_child() must be called before the signal handlers get
1735 * installed below as they are inherited into the spawned process.
1736 * Also we do not need to be protected by them as during interruption
1737 * in the startup_child() mode we kill the spawned process anyway.
1738 */
1739 if (argv[0]) {
1740 if (!NOMMU_SYSTEM || daemonized_tracer)
1741 hide_log_until_execve = 1;
1742 skip_one_b_execve = 1;
1743 startup_child(argv);
1744 }
1745
1746 sa.sa_handler = SIG_IGN;
1747 sigemptyset(&sa.sa_mask);
1748 sa.sa_flags = 0;
1749 sigaction(SIGTTOU, &sa, NULL); /* SIG_IGN */
1750 sigaction(SIGTTIN, &sa, NULL); /* SIG_IGN */
1751 if (opt_intr != INTR_ANYWHERE) {
1752 if (opt_intr == INTR_BLOCK_TSTP_TOO)
1753 sigaction(SIGTSTP, &sa, NULL); /* SIG_IGN */
1754 /*
1755 * In interactive mode (if no -o OUTFILE, or -p PID is used),
1756 * fatal signals are blocked while syscall stop is processed,
1757 * and acted on in between, when waiting for new syscall stops.
1758 * In non-interactive mode, signals are ignored.
1759 */
1760 if (opt_intr == INTR_WHILE_WAIT) {
1761 sigaddset(&blocked_set, SIGHUP);
1762 sigaddset(&blocked_set, SIGINT);
1763 sigaddset(&blocked_set, SIGQUIT);
1764 sigaddset(&blocked_set, SIGPIPE);
1765 sigaddset(&blocked_set, SIGTERM);
1766 sa.sa_handler = interrupt;
1767 }
1768 /* SIG_IGN, or set handler for these */
1769 sigaction(SIGHUP, &sa, NULL);
1770 sigaction(SIGINT, &sa, NULL);
1771 sigaction(SIGQUIT, &sa, NULL);
1772 sigaction(SIGPIPE, &sa, NULL);
1773 sigaction(SIGTERM, &sa, NULL);
1774 }
1775 if (nprocs != 0 || daemonized_tracer)
1776 startup_attach();
1777
1778 /* Do we want pids printed in our -o OUTFILE?
1779 * -ff: no (every pid has its own file); or
1780 * -f: yes (there can be more pids in the future); or
1781 * -p PID1,PID2: yes (there are already more than one pid)
1782 */
1783 print_pid_pfx = (outfname && followfork < 2 && (followfork == 1 || nprocs > 1));
1784 }
1785
1786 static struct tcb *
pid2tcb(int pid)1787 pid2tcb(int pid)
1788 {
1789 unsigned int i;
1790
1791 if (pid <= 0)
1792 return NULL;
1793
1794 for (i = 0; i < tcbtabsize; i++) {
1795 struct tcb *tcp = tcbtab[i];
1796 if (tcp->pid == pid)
1797 return tcp;
1798 }
1799
1800 return NULL;
1801 }
1802
1803 static void
cleanup(void)1804 cleanup(void)
1805 {
1806 unsigned int i;
1807 struct tcb *tcp;
1808 int fatal_sig;
1809
1810 /* 'interrupted' is a volatile object, fetch it only once */
1811 fatal_sig = interrupted;
1812 if (!fatal_sig)
1813 fatal_sig = SIGTERM;
1814
1815 for (i = 0; i < tcbtabsize; i++) {
1816 tcp = tcbtab[i];
1817 if (!tcp->pid)
1818 continue;
1819 if (debug_flag)
1820 error_msg("cleanup: looking at pid %u", tcp->pid);
1821 if (tcp->pid == strace_child) {
1822 kill(tcp->pid, SIGCONT);
1823 kill(tcp->pid, fatal_sig);
1824 }
1825 detach(tcp);
1826 }
1827 if (cflag)
1828 call_summary(shared_log);
1829 }
1830
1831 static void
interrupt(int sig)1832 interrupt(int sig)
1833 {
1834 interrupted = sig;
1835 }
1836
1837 static void
print_debug_info(const int pid,int status)1838 print_debug_info(const int pid, int status)
1839 {
1840 const unsigned int event = (unsigned int) status >> 16;
1841 char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16];
1842 char evbuf[sizeof(",EVENT_VFORK_DONE (%u)") + sizeof(int)*3 /*paranoia:*/ + 16];
1843
1844 strcpy(buf, "???");
1845 if (WIFSIGNALED(status))
1846 #ifdef WCOREDUMP
1847 sprintf(buf, "WIFSIGNALED,%ssig=%s",
1848 WCOREDUMP(status) ? "core," : "",
1849 signame(WTERMSIG(status)));
1850 #else
1851 sprintf(buf, "WIFSIGNALED,sig=%s",
1852 signame(WTERMSIG(status)));
1853 #endif
1854 if (WIFEXITED(status))
1855 sprintf(buf, "WIFEXITED,exitcode=%u", WEXITSTATUS(status));
1856 if (WIFSTOPPED(status))
1857 sprintf(buf, "WIFSTOPPED,sig=%s", signame(WSTOPSIG(status)));
1858 #ifdef WIFCONTINUED
1859 /* Should never be seen */
1860 if (WIFCONTINUED(status))
1861 strcpy(buf, "WIFCONTINUED");
1862 #endif
1863 evbuf[0] = '\0';
1864 if (event != 0) {
1865 static const char *const event_names[] = {
1866 [PTRACE_EVENT_CLONE] = "CLONE",
1867 [PTRACE_EVENT_FORK] = "FORK",
1868 [PTRACE_EVENT_VFORK] = "VFORK",
1869 [PTRACE_EVENT_VFORK_DONE] = "VFORK_DONE",
1870 [PTRACE_EVENT_EXEC] = "EXEC",
1871 [PTRACE_EVENT_EXIT] = "EXIT",
1872 /* [PTRACE_EVENT_STOP (=128)] would make biggish array */
1873 };
1874 const char *e = "??";
1875 if (event < ARRAY_SIZE(event_names))
1876 e = event_names[event];
1877 else if (event == PTRACE_EVENT_STOP)
1878 e = "STOP";
1879 sprintf(evbuf, ",EVENT_%s (%u)", e, event);
1880 }
1881 error_msg("[wait(0x%06x) = %u] %s%s", status, pid, buf, evbuf);
1882 }
1883
1884 static struct tcb *
maybe_allocate_tcb(const int pid,int status)1885 maybe_allocate_tcb(const int pid, int status)
1886 {
1887 if (!WIFSTOPPED(status)) {
1888 if (detach_on_execve && pid == strace_child) {
1889 /* example: strace -bexecve sh -c 'exec true' */
1890 strace_child = 0;
1891 return NULL;
1892 }
1893 /*
1894 * This can happen if we inherited an unknown child.
1895 * Example: (sleep 1 & exec strace true)
1896 */
1897 error_msg("Exit of unknown pid %u ignored", pid);
1898 return NULL;
1899 }
1900 if (followfork) {
1901 /* We assume it's a fork/vfork/clone child */
1902 struct tcb *tcp = alloctcb(pid);
1903 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop;
1904 newoutf(tcp);
1905 if (!qflag)
1906 error_msg("Process %d attached", pid);
1907 return tcp;
1908 } else {
1909 /* This can happen if a clone call used
1910 * CLONE_PTRACE itself.
1911 */
1912 ptrace(PTRACE_CONT, pid, (char *) 0, 0);
1913 error_msg("Stop of unknown pid %u seen, PTRACE_CONTed it", pid);
1914 return NULL;
1915 }
1916 }
1917
1918 static struct tcb *
maybe_switch_tcbs(struct tcb * tcp,const int pid)1919 maybe_switch_tcbs(struct tcb *tcp, const int pid)
1920 {
1921 FILE *fp;
1922 struct tcb *execve_thread;
1923 long old_pid = 0;
1924
1925 if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, (long) &old_pid) < 0)
1926 return tcp;
1927 /* Avoid truncation in pid2tcb() param passing */
1928 if (old_pid <= 0 || old_pid == pid)
1929 return tcp;
1930 if ((unsigned long) old_pid > UINT_MAX)
1931 return tcp;
1932 execve_thread = pid2tcb(old_pid);
1933 /* It should be !NULL, but I feel paranoid */
1934 if (!execve_thread)
1935 return tcp;
1936
1937 if (execve_thread->curcol != 0) {
1938 /*
1939 * One case we are here is -ff:
1940 * try "strace -oLOG -ff test/threaded_execve"
1941 */
1942 fprintf(execve_thread->outf, " <pid changed to %d ...>\n", pid);
1943 /*execve_thread->curcol = 0; - no need, see code below */
1944 }
1945 /* Swap output FILEs (needed for -ff) */
1946 fp = execve_thread->outf;
1947 execve_thread->outf = tcp->outf;
1948 tcp->outf = fp;
1949 /* And their column positions */
1950 execve_thread->curcol = tcp->curcol;
1951 tcp->curcol = 0;
1952 /* Drop leader, but close execve'd thread outfile (if -ff) */
1953 droptcb(tcp);
1954 /* Switch to the thread, reusing leader's outfile and pid */
1955 tcp = execve_thread;
1956 tcp->pid = pid;
1957 if (cflag != CFLAG_ONLY_STATS) {
1958 printleader(tcp);
1959 tprintf("+++ superseded by execve in pid %lu +++\n", old_pid);
1960 line_ended();
1961 tcp->flags |= TCB_REPRINT;
1962 }
1963
1964 return tcp;
1965 }
1966
1967 static void
print_signalled(struct tcb * tcp,const int pid,int status)1968 print_signalled(struct tcb *tcp, const int pid, int status)
1969 {
1970 if (pid == strace_child) {
1971 exit_code = 0x100 | WTERMSIG(status);
1972 strace_child = 0;
1973 }
1974
1975 if (cflag != CFLAG_ONLY_STATS
1976 && (qual_flags[WTERMSIG(status)] & QUAL_SIGNAL)
1977 ) {
1978 printleader(tcp);
1979 #ifdef WCOREDUMP
1980 tprintf("+++ killed by %s %s+++\n",
1981 signame(WTERMSIG(status)),
1982 WCOREDUMP(status) ? "(core dumped) " : "");
1983 #else
1984 tprintf("+++ killed by %s +++\n",
1985 signame(WTERMSIG(status)));
1986 #endif
1987 line_ended();
1988 }
1989 }
1990
1991 static void
print_exited(struct tcb * tcp,const int pid,int status)1992 print_exited(struct tcb *tcp, const int pid, int status)
1993 {
1994 if (pid == strace_child) {
1995 exit_code = WEXITSTATUS(status);
1996 strace_child = 0;
1997 }
1998
1999 if (cflag != CFLAG_ONLY_STATS &&
2000 qflag < 2) {
2001 printleader(tcp);
2002 tprintf("+++ exited with %d +++\n", WEXITSTATUS(status));
2003 line_ended();
2004 }
2005 }
2006
2007 static void
print_stopped(struct tcb * tcp,const siginfo_t * si,const unsigned int sig)2008 print_stopped(struct tcb *tcp, const siginfo_t *si, const unsigned int sig)
2009 {
2010 if (cflag != CFLAG_ONLY_STATS
2011 && !hide_log_until_execve
2012 && (qual_flags[sig] & QUAL_SIGNAL)
2013 ) {
2014 printleader(tcp);
2015 if (si) {
2016 tprintf("--- %s ", signame(sig));
2017 printsiginfo(si, verbose(tcp));
2018 tprints(" ---\n");
2019 } else
2020 tprintf("--- stopped by %s ---\n", signame(sig));
2021 line_ended();
2022 }
2023 }
2024
2025 static void
startup_tcb(struct tcb * tcp)2026 startup_tcb(struct tcb *tcp)
2027 {
2028 if (debug_flag)
2029 error_msg("pid %d has TCB_STARTUP, initializing it", tcp->pid);
2030
2031 tcp->flags &= ~TCB_STARTUP;
2032
2033 if (!use_seize) {
2034 if (debug_flag)
2035 error_msg("setting opts 0x%x on pid %d",
2036 ptrace_setoptions, tcp->pid);
2037 if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) {
2038 if (errno != ESRCH) {
2039 /* Should never happen, really */
2040 perror_msg_and_die("PTRACE_SETOPTIONS");
2041 }
2042 }
2043 }
2044 }
2045
2046 /* Returns true iff the main trace loop has to continue. */
2047 static bool
trace(void)2048 trace(void)
2049 {
2050 int pid;
2051 int wait_errno;
2052 int status;
2053 bool stopped;
2054 unsigned int sig;
2055 unsigned int event;
2056 struct tcb *tcp;
2057 struct rusage ru;
2058
2059 if (interrupted)
2060 return false;
2061
2062 /*
2063 * Used to exit simply when nprocs hits zero, but in this testcase:
2064 * int main() { _exit(!!fork()); }
2065 * under strace -f, parent sometimes (rarely) manages
2066 * to exit before we see the first stop of the child,
2067 * and we are losing track of it:
2068 * 19923 clone(...) = 19924
2069 * 19923 exit_group(1) = ?
2070 * 19923 +++ exited with 1 +++
2071 * Exiting only when wait() returns ECHILD works better.
2072 */
2073 if (popen_pid != 0) {
2074 /* However, if -o|logger is in use, we can't do that.
2075 * Can work around that by double-forking the logger,
2076 * but that loses the ability to wait for its completion
2077 * on exit. Oh well...
2078 */
2079 if (nprocs == 0)
2080 return false;
2081 }
2082
2083 if (interactive)
2084 sigprocmask(SIG_SETMASK, &empty_set, NULL);
2085 pid = wait4(-1, &status, __WALL, (cflag ? &ru : NULL));
2086 wait_errno = errno;
2087 if (interactive)
2088 sigprocmask(SIG_BLOCK, &blocked_set, NULL);
2089
2090 if (pid < 0) {
2091 if (wait_errno == EINTR)
2092 return true;
2093 if (nprocs == 0 && wait_errno == ECHILD)
2094 return false;
2095 /*
2096 * If nprocs > 0, ECHILD is not expected,
2097 * treat it as any other error here:
2098 */
2099 errno = wait_errno;
2100 perror_msg_and_die("wait4(__WALL)");
2101 }
2102
2103 if (pid == popen_pid) {
2104 if (!WIFSTOPPED(status))
2105 popen_pid = 0;
2106 return true;
2107 }
2108
2109 if (debug_flag)
2110 print_debug_info(pid, status);
2111
2112 /* Look up 'pid' in our table. */
2113 tcp = pid2tcb(pid);
2114
2115 if (!tcp) {
2116 tcp = maybe_allocate_tcb(pid, status);
2117 if (!tcp)
2118 return true;
2119 }
2120
2121 if (WIFSTOPPED(status))
2122 get_regs(pid);
2123 else
2124 clear_regs();
2125
2126 event = (unsigned int) status >> 16;
2127
2128 if (event == PTRACE_EVENT_EXEC) {
2129 /*
2130 * Under Linux, execve changes pid to thread leader's pid,
2131 * and we see this changed pid on EVENT_EXEC and later,
2132 * execve sysexit. Leader "disappears" without exit
2133 * notification. Let user know that, drop leader's tcb,
2134 * and fix up pid in execve thread's tcb.
2135 * Effectively, execve thread's tcb replaces leader's tcb.
2136 *
2137 * BTW, leader is 'stuck undead' (doesn't report WIFEXITED
2138 * on exit syscall) in multithreaded programs exactly
2139 * in order to handle this case.
2140 *
2141 * PTRACE_GETEVENTMSG returns old pid starting from Linux 3.0.
2142 * On 2.6 and earlier, it can return garbage.
2143 */
2144 if (os_release >= KERNEL_VERSION(3,0,0))
2145 tcp = maybe_switch_tcbs(tcp, pid);
2146
2147 if (detach_on_execve && !skip_one_b_execve) {
2148 detach(tcp); /* do "-b execve" thingy */
2149 return true;
2150 }
2151 skip_one_b_execve = 0;
2152 }
2153
2154 /* Set current output file */
2155 current_tcp = tcp;
2156
2157 if (cflag) {
2158 tv_sub(&tcp->dtime, &ru.ru_stime, &tcp->stime);
2159 tcp->stime = ru.ru_stime;
2160 }
2161
2162 if (WIFSIGNALED(status)) {
2163 print_signalled(tcp, pid, status);
2164 droptcb(tcp);
2165 return true;
2166 }
2167
2168 if (WIFEXITED(status)) {
2169 print_exited(tcp, pid, status);
2170 droptcb(tcp);
2171 return true;
2172 }
2173
2174 if (!WIFSTOPPED(status)) {
2175 /*
2176 * Neither signalled, exited or stopped.
2177 * How could that be?
2178 */
2179 error_msg("pid %u not stopped!", pid);
2180 droptcb(tcp);
2181 return true;
2182 }
2183
2184 /* Is this the very first time we see this tracee stopped? */
2185 if (tcp->flags & TCB_STARTUP) {
2186 startup_tcb(tcp);
2187 if (get_scno(tcp) == 1)
2188 tcp->s_prev_ent = tcp->s_ent;
2189 }
2190
2191 sig = WSTOPSIG(status);
2192
2193 if (event != 0) {
2194 /* Ptrace event */
2195 #if USE_SEIZE
2196 if (event == PTRACE_EVENT_STOP) {
2197 /*
2198 * PTRACE_INTERRUPT-stop or group-stop.
2199 * PTRACE_INTERRUPT-stop has sig == SIGTRAP here.
2200 */
2201 switch (sig) {
2202 case SIGSTOP:
2203 case SIGTSTP:
2204 case SIGTTIN:
2205 case SIGTTOU:
2206 stopped = true;
2207 goto show_stopsig;
2208 }
2209 }
2210 #endif
2211 goto restart_tracee_with_sig_0;
2212 }
2213
2214 /*
2215 * Is this post-attach SIGSTOP?
2216 * Interestingly, the process may stop
2217 * with STOPSIG equal to some other signal
2218 * than SIGSTOP if we happend to attach
2219 * just before the process takes a signal.
2220 */
2221 if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) {
2222 if (debug_flag)
2223 error_msg("ignored SIGSTOP on pid %d", tcp->pid);
2224 tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP;
2225 goto restart_tracee_with_sig_0;
2226 }
2227
2228 if (sig != syscall_trap_sig) {
2229 siginfo_t si = {};
2230
2231 /*
2232 * True if tracee is stopped by signal
2233 * (as opposed to "tracee received signal").
2234 * TODO: shouldn't we check for errno == EINVAL too?
2235 * We can get ESRCH instead, you know...
2236 */
2237 stopped = ptrace(PTRACE_GETSIGINFO, pid, 0, (long) &si) < 0;
2238 #if USE_SEIZE
2239 show_stopsig:
2240 #endif
2241 print_stopped(tcp, stopped ? NULL : &si, sig);
2242
2243 if (!stopped)
2244 /* It's signal-delivery-stop. Inject the signal */
2245 goto restart_tracee;
2246
2247 /* It's group-stop */
2248 if (use_seize) {
2249 /*
2250 * This ends ptrace-stop, but does *not* end group-stop.
2251 * This makes stopping signals work properly on straced process
2252 * (that is, process really stops. It used to continue to run).
2253 */
2254 if (ptrace_restart(PTRACE_LISTEN, tcp, 0) < 0) {
2255 /* Note: ptrace_restart emitted error message */
2256 exit_code = 1;
2257 return false;
2258 }
2259 return true;
2260 }
2261 /* We don't have PTRACE_LISTEN support... */
2262 goto restart_tracee;
2263 }
2264
2265 /* We handled quick cases, we are permitted to interrupt now. */
2266 if (interrupted)
2267 return false;
2268
2269 /*
2270 * This should be syscall entry or exit.
2271 * Handle it.
2272 */
2273 if (trace_syscall(tcp) < 0) {
2274 /*
2275 * ptrace() failed in trace_syscall().
2276 * Likely a result of process disappearing mid-flight.
2277 * Observed case: exit_group() or SIGKILL terminating
2278 * all processes in thread group.
2279 * We assume that ptrace error was caused by process death.
2280 * We used to detach(tcp) here, but since we no longer
2281 * implement "detach before death" policy/hack,
2282 * we can let this process to report its death to us
2283 * normally, via WIFEXITED or WIFSIGNALED wait status.
2284 */
2285 return true;
2286 }
2287
2288 restart_tracee_with_sig_0:
2289 sig = 0;
2290
2291 restart_tracee:
2292 if (ptrace_restart(PTRACE_SYSCALL, tcp, sig) < 0) {
2293 /* Note: ptrace_restart emitted error message */
2294 exit_code = 1;
2295 return false;
2296 }
2297
2298 return true;
2299 }
2300
2301 int
main(int argc,char * argv[])2302 main(int argc, char *argv[])
2303 {
2304 init(argc, argv);
2305
2306 while (trace())
2307 ;
2308
2309 cleanup();
2310 fflush(NULL);
2311 if (shared_log != stderr)
2312 fclose(shared_log);
2313 if (popen_pid) {
2314 while (waitpid(popen_pid, NULL, 0) < 0 && errno == EINTR)
2315 ;
2316 }
2317 if (exit_code > 0xff) {
2318 /* Avoid potential core file clobbering. */
2319 struct_rlimit rlim = {0, 0};
2320 set_rlimit(RLIMIT_CORE, &rlim);
2321
2322 /* Child was killed by a signal, mimic that. */
2323 exit_code &= 0xff;
2324 signal(exit_code, SIG_DFL);
2325 raise(exit_code);
2326 /* Paranoia - what if this signal is not fatal?
2327 Exit with 128 + signo then. */
2328 exit_code += 128;
2329 }
2330
2331 return exit_code;
2332 }
2333