strace.c revision 6162a3f34fc8d15095a3b71dc6e4cbdfe1b14ac1
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 <sys/resource.h> 36#include <sys/wait.h> 37#include <sys/stat.h> 38#include <pwd.h> 39#include <grp.h> 40#include <dirent.h> 41#include <sys/utsname.h> 42#ifdef HAVE_PRCTL 43# include <sys/prctl.h> 44#endif 45#if defined(IA64) 46# include <asm/ptrace_offsets.h> 47#endif 48/* In some libc, these aren't declared. Do it ourself: */ 49extern char **environ; 50extern int optind; 51extern char *optarg; 52 53 54#if defined __NR_tkill 55# define my_tkill(tid, sig) syscall(__NR_tkill, (tid), (sig)) 56#else 57 /* kill() may choose arbitrarily the target task of the process group 58 while we later wait on a that specific TID. PID process waits become 59 TID task specific waits for a process under ptrace(2). */ 60# warning "tkill(2) not available, risk of strace hangs!" 61# define my_tkill(tid, sig) kill((tid), (sig)) 62#endif 63 64/* Glue for systems without a MMU that cannot provide fork() */ 65#if !defined(HAVE_FORK) 66# undef NOMMU_SYSTEM 67# define NOMMU_SYSTEM 1 68#endif 69#if NOMMU_SYSTEM 70# define fork() vfork() 71#endif 72 73cflag_t cflag = CFLAG_NONE; 74unsigned int followfork = 0; 75unsigned int ptrace_setoptions = 0; 76unsigned int xflag = 0; 77bool need_fork_exec_workarounds = 0; 78bool debug_flag = 0; 79bool Tflag = 0; 80unsigned int qflag = 0; 81/* Which WSTOPSIG(status) value marks syscall traps? */ 82static unsigned int syscall_trap_sig = SIGTRAP; 83static unsigned int tflag = 0; 84static bool iflag = 0; 85static bool rflag = 0; 86static bool print_pid_pfx = 0; 87 88/* -I n */ 89enum { 90 INTR_NOT_SET = 0, 91 INTR_ANYWHERE = 1, /* don't block/ignore any signals */ 92 INTR_WHILE_WAIT = 2, /* block fatal signals while decoding syscall. default */ 93 INTR_NEVER = 3, /* block fatal signals. default if '-o FILE PROG' */ 94 INTR_BLOCK_TSTP_TOO = 4, /* block fatal signals and SIGTSTP (^Z) */ 95 NUM_INTR_OPTS 96}; 97static int opt_intr; 98/* We play with signal mask only if this mode is active: */ 99#define interactive (opt_intr == INTR_WHILE_WAIT) 100 101/* 102 * daemonized_tracer supports -D option. 103 * With this option, strace forks twice. 104 * Unlike normal case, with -D *grandparent* process exec's, 105 * becoming a traced process. Child exits (this prevents traced process 106 * from having children it doesn't expect to have), and grandchild 107 * attaches to grandparent similarly to strace -p PID. 108 * This allows for more transparent interaction in cases 109 * when process and its parent are communicating via signals, 110 * wait() etc. Without -D, strace process gets lodged in between, 111 * disrupting parent<->child link. 112 */ 113static bool daemonized_tracer = 0; 114 115#if USE_SEIZE 116static int post_attach_sigstop = TCB_IGNORE_ONE_SIGSTOP; 117# define use_seize (post_attach_sigstop == 0) 118#else 119# define post_attach_sigstop TCB_IGNORE_ONE_SIGSTOP 120# define use_seize 0 121#endif 122 123/* Sometimes we want to print only succeeding syscalls. */ 124bool not_failing_only = 0; 125 126/* Show path associated with fd arguments */ 127bool show_fd_path = 0; 128 129static bool detach_on_execve = 0; 130/* Are we "strace PROG" and need to skip detach on first execve? */ 131static bool skip_one_b_execve = 0; 132/* Are we "strace PROG" and need to hide everything until execve? */ 133bool hide_log_until_execve = 0; 134 135static int exit_code = 0; 136static int strace_child = 0; 137static int strace_tracer_pid = 0; 138 139static char *username = NULL; 140static uid_t run_uid; 141static gid_t run_gid; 142 143unsigned int max_strlen = DEFAULT_STRLEN; 144static int acolumn = DEFAULT_ACOLUMN; 145static char *acolumn_spaces; 146 147static char *outfname = NULL; 148/* If -ff, points to stderr. Else, it's our common output log */ 149static FILE *shared_log; 150 151struct tcb *printing_tcp = NULL; 152static struct tcb *current_tcp; 153 154static struct tcb **tcbtab; 155static unsigned int nprocs, tcbtabsize; 156static const char *progname; 157 158unsigned os_release; /* generated from uname()'s u.release */ 159 160static void detach(struct tcb *tcp); 161static void cleanup(void); 162static void interrupt(int sig); 163static sigset_t empty_set, blocked_set; 164 165#ifdef HAVE_SIG_ATOMIC_T 166static volatile sig_atomic_t interrupted; 167#else 168static volatile int interrupted; 169#endif 170 171#ifndef HAVE_STRERROR 172 173#if !HAVE_DECL_SYS_ERRLIST 174extern int sys_nerr; 175extern char *sys_errlist[]; 176#endif 177 178const char * 179strerror(int err_no) 180{ 181 static char buf[sizeof("Unknown error %d") + sizeof(int)*3]; 182 183 if (err_no < 1 || err_no >= sys_nerr) { 184 sprintf(buf, "Unknown error %d", err_no); 185 return buf; 186 } 187 return sys_errlist[err_no]; 188} 189 190#endif /* HAVE_STERRROR */ 191 192static void 193usage(FILE *ofp, int exitval) 194{ 195 fprintf(ofp, "\ 196usage: strace [-CdffhiqrtttTvVxxy] [-I n] [-e expr]...\n\ 197 [-a column] [-o file] [-s strsize] [-P path]...\n\ 198 -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\ 199 or: strace -c[df] [-I n] [-e expr]... [-O overhead] [-S sortby]\n\ 200 -p pid... / [-D] [-E var=val]... [-u username] PROG [ARGS]\n\ 201-c -- count time, calls, and errors for each syscall and report summary\n\ 202-C -- like -c but also print regular output\n\ 203-d -- enable debug output to stderr\n\ 204-D -- run tracer process as a detached grandchild, not as parent\n\ 205-f -- follow forks, -ff -- with output into separate files\n\ 206-i -- print instruction pointer at time of syscall\n\ 207-q -- suppress messages about attaching, detaching, etc.\n\ 208-r -- print relative timestamp, -t -- absolute timestamp, -tt -- with usecs\n\ 209-T -- print time spent in each syscall\n\ 210-v -- verbose mode: print unabbreviated argv, stat, termios, etc. args\n\ 211-x -- print non-ascii strings in hex, -xx -- print all strings in hex\n\ 212-y -- print paths associated with file descriptor arguments\n\ 213-h -- print help message, -V -- print version\n\ 214-a column -- alignment COLUMN for printing syscall results (default %d)\n\ 215-b execve -- detach on this syscall\n\ 216-e expr -- a qualifying expression: option=[!]all or option=[!]val1[,val2]...\n\ 217 options: trace, abbrev, verbose, raw, signal, read, write\n\ 218-I interruptible --\n\ 219 1: no signals are blocked\n\ 220 2: fatal signals are blocked while decoding syscall (default)\n\ 221 3: fatal signals are always blocked (default if '-o FILE PROG')\n\ 222 4: fatal signals and SIGTSTP (^Z) are always blocked\n\ 223 (useful to make 'strace -o FILE PROG' not stop on ^Z)\n\ 224-o file -- send trace output to FILE instead of stderr\n\ 225-O overhead -- set overhead for tracing syscalls to OVERHEAD usecs\n\ 226-p pid -- trace process with process id PID, may be repeated\n\ 227-s strsize -- limit length of print strings to STRSIZE chars (default %d)\n\ 228-S sortby -- sort syscall counts by: time, calls, name, nothing (default %s)\n\ 229-u username -- run command as username handling setuid and/or setgid\n\ 230-E var=val -- put var=val in the environment for command\n\ 231-E var -- remove var from the environment for command\n\ 232-P path -- trace accesses to path\n\ 233" 234/* ancient, no one should use it 235-F -- attempt to follow vforks (deprecated, use -f)\n\ 236 */ 237/* this is broken, so don't document it 238-z -- print only succeeding syscalls\n\ 239 */ 240, DEFAULT_ACOLUMN, DEFAULT_STRLEN, DEFAULT_SORTBY); 241 exit(exitval); 242} 243 244static void die(void) __attribute__ ((noreturn)); 245static void die(void) 246{ 247 if (strace_tracer_pid == getpid()) { 248 cflag = 0; 249 cleanup(); 250 } 251 exit(1); 252} 253 254static void verror_msg(int err_no, const char *fmt, va_list p) 255{ 256 char *msg; 257 258 fflush(NULL); 259 260 /* We want to print entire message with single fprintf to ensure 261 * message integrity if stderr is shared with other programs. 262 * Thus we use vasprintf + single fprintf. 263 */ 264 msg = NULL; 265 if (vasprintf(&msg, fmt, p) >= 0) { 266 if (err_no) 267 fprintf(stderr, "%s: %s: %s\n", progname, msg, strerror(err_no)); 268 else 269 fprintf(stderr, "%s: %s\n", progname, msg); 270 free(msg); 271 } else { 272 /* malloc in vasprintf failed, try it without malloc */ 273 fprintf(stderr, "%s: ", progname); 274 vfprintf(stderr, fmt, p); 275 if (err_no) 276 fprintf(stderr, ": %s\n", strerror(err_no)); 277 else 278 putc('\n', stderr); 279 } 280 /* We don't switch stderr to buffered, thus fprintf(stderr) 281 * always flushes its output and this is not necessary: */ 282 /* fflush(stderr); */ 283} 284 285void error_msg(const char *fmt, ...) 286{ 287 va_list p; 288 va_start(p, fmt); 289 verror_msg(0, fmt, p); 290 va_end(p); 291} 292 293void error_msg_and_die(const char *fmt, ...) 294{ 295 va_list p; 296 va_start(p, fmt); 297 verror_msg(0, fmt, p); 298 die(); 299} 300 301void perror_msg(const char *fmt, ...) 302{ 303 va_list p; 304 va_start(p, fmt); 305 verror_msg(errno, fmt, p); 306 va_end(p); 307} 308 309void perror_msg_and_die(const char *fmt, ...) 310{ 311 va_list p; 312 va_start(p, fmt); 313 verror_msg(errno, fmt, p); 314 die(); 315} 316 317void die_out_of_memory(void) 318{ 319 static bool recursed = 0; 320 if (recursed) 321 exit(1); 322 recursed = 1; 323 error_msg_and_die("Out of memory"); 324} 325 326static void 327error_opt_arg(int opt, const char *arg) 328{ 329 error_msg_and_die("Invalid -%c argument: '%s'", opt, arg); 330} 331 332#if USE_SEIZE 333static int 334ptrace_attach_or_seize(int pid) 335{ 336 int r; 337 if (!use_seize) 338 return ptrace(PTRACE_ATTACH, pid, 0, 0); 339 r = ptrace(PTRACE_SEIZE, pid, 0, 0); 340 if (r) 341 return r; 342 r = ptrace(PTRACE_INTERRUPT, pid, 0, 0); 343 return r; 344} 345#else 346# define ptrace_attach_or_seize(pid) ptrace(PTRACE_ATTACH, (pid), 0, 0) 347#endif 348 349/* 350 * Used when we want to unblock stopped traced process. 351 * Should be only used with PTRACE_CONT, PTRACE_DETACH and PTRACE_SYSCALL. 352 * Returns 0 on success or if error was ESRCH 353 * (presumably process was killed while we talk to it). 354 * Otherwise prints error message and returns -1. 355 */ 356static int 357ptrace_restart(int op, struct tcb *tcp, int sig) 358{ 359 int err; 360 const char *msg; 361 362 errno = 0; 363 ptrace(op, tcp->pid, (void *) 0, (long) sig); 364 err = errno; 365 if (!err) 366 return 0; 367 368 msg = "SYSCALL"; 369 if (op == PTRACE_CONT) 370 msg = "CONT"; 371 if (op == PTRACE_DETACH) 372 msg = "DETACH"; 373#ifdef PTRACE_LISTEN 374 if (op == PTRACE_LISTEN) 375 msg = "LISTEN"; 376#endif 377 /* 378 * Why curcol != 0? Otherwise sometimes we get this: 379 * 380 * 10252 kill(10253, SIGKILL) = 0 381 * <ptrace(SYSCALL,10252):No such process>10253 ...next decode... 382 * 383 * 10252 died after we retrieved syscall exit data, 384 * but before we tried to restart it. Log looks ugly. 385 */ 386 if (current_tcp && current_tcp->curcol != 0) { 387 tprintf(" <ptrace(%s):%s>\n", msg, strerror(err)); 388 line_ended(); 389 } 390 if (err == ESRCH) 391 return 0; 392 errno = err; 393 perror_msg("ptrace(PTRACE_%s,pid:%d,sig:%d)", msg, tcp->pid, sig); 394 return -1; 395} 396 397static void 398set_cloexec_flag(int fd) 399{ 400 int flags, newflags; 401 402 flags = fcntl(fd, F_GETFD); 403 if (flags < 0) { 404 /* Can happen only if fd is bad. 405 * Should never happen: if it does, we have a bug 406 * in the caller. Therefore we just abort 407 * instead of propagating the error. 408 */ 409 perror_msg_and_die("fcntl(%d, F_GETFD)", fd); 410 } 411 412 newflags = flags | FD_CLOEXEC; 413 if (flags == newflags) 414 return; 415 416 fcntl(fd, F_SETFD, newflags); /* never fails */ 417} 418 419static void kill_save_errno(pid_t pid, int sig) 420{ 421 int saved_errno = errno; 422 423 (void) kill(pid, sig); 424 errno = saved_errno; 425} 426 427/* 428 * When strace is setuid executable, we have to swap uids 429 * before and after filesystem and process management operations. 430 */ 431static void 432swap_uid(void) 433{ 434 int euid = geteuid(), uid = getuid(); 435 436 if (euid != uid && setreuid(euid, uid) < 0) { 437 perror_msg_and_die("setreuid"); 438 } 439} 440 441#if _LFS64_LARGEFILE 442# define fopen_for_output fopen64 443# define struct_stat struct stat64 444# define stat_file stat64 445# define struct_dirent struct dirent64 446# define read_dir readdir64 447# define struct_rlimit struct rlimit64 448# define set_rlimit setrlimit64 449#else 450# define fopen_for_output fopen 451# define struct_stat struct stat 452# define stat_file stat 453# define struct_dirent struct dirent 454# define read_dir readdir 455# define struct_rlimit struct rlimit 456# define set_rlimit setrlimit 457#endif 458 459static FILE * 460strace_fopen(const char *path) 461{ 462 FILE *fp; 463 464 swap_uid(); 465 fp = fopen_for_output(path, "w"); 466 if (!fp) 467 perror_msg_and_die("Can't fopen '%s'", path); 468 swap_uid(); 469 set_cloexec_flag(fileno(fp)); 470 return fp; 471} 472 473static int popen_pid = 0; 474 475#ifndef _PATH_BSHELL 476# define _PATH_BSHELL "/bin/sh" 477#endif 478 479/* 480 * We cannot use standard popen(3) here because we have to distinguish 481 * popen child process from other processes we trace, and standard popen(3) 482 * does not export its child's pid. 483 */ 484static FILE * 485strace_popen(const char *command) 486{ 487 FILE *fp; 488 int pid; 489 int fds[2]; 490 491 swap_uid(); 492 if (pipe(fds) < 0) 493 perror_msg_and_die("pipe"); 494 495 set_cloexec_flag(fds[1]); /* never fails */ 496 497 pid = vfork(); 498 if (pid < 0) 499 perror_msg_and_die("vfork"); 500 501 if (pid == 0) { 502 /* child */ 503 close(fds[1]); 504 if (fds[0] != 0) { 505 if (dup2(fds[0], 0)) 506 perror_msg_and_die("dup2"); 507 close(fds[0]); 508 } 509 execl(_PATH_BSHELL, "sh", "-c", command, NULL); 510 perror_msg_and_die("Can't execute '%s'", _PATH_BSHELL); 511 } 512 513 /* parent */ 514 popen_pid = pid; 515 close(fds[0]); 516 swap_uid(); 517 fp = fdopen(fds[1], "w"); 518 if (!fp) 519 die_out_of_memory(); 520 return fp; 521} 522 523void 524tprintf(const char *fmt, ...) 525{ 526 va_list args; 527 528 va_start(args, fmt); 529 if (current_tcp) { 530 int n = strace_vfprintf(current_tcp->outf, fmt, args); 531 if (n < 0) { 532 if (current_tcp->outf != stderr) 533 perror_msg("%s", outfname); 534 } else 535 current_tcp->curcol += n; 536 } 537 va_end(args); 538} 539 540void 541tprints(const char *str) 542{ 543 if (current_tcp) { 544 int n = fputs_unlocked(str, current_tcp->outf); 545 if (n >= 0) { 546 current_tcp->curcol += strlen(str); 547 return; 548 } 549 if (current_tcp->outf != stderr) 550 perror_msg("%s", outfname); 551 } 552} 553 554void 555line_ended(void) 556{ 557 if (current_tcp) { 558 current_tcp->curcol = 0; 559 fflush(current_tcp->outf); 560 } 561 if (printing_tcp) { 562 printing_tcp->curcol = 0; 563 printing_tcp = NULL; 564 } 565} 566 567void 568printleader(struct tcb *tcp) 569{ 570 /* If -ff, "previous tcb we printed" is always the same as current, 571 * because we have per-tcb output files. 572 */ 573 if (followfork >= 2) 574 printing_tcp = tcp; 575 576 if (printing_tcp) { 577 current_tcp = printing_tcp; 578 if (printing_tcp->curcol != 0 && (followfork < 2 || printing_tcp == tcp)) { 579 /* 580 * case 1: we have a shared log (i.e. not -ff), and last line 581 * wasn't finished (same or different tcb, doesn't matter). 582 * case 2: split log, we are the same tcb, but our last line 583 * didn't finish ("SIGKILL nuked us after syscall entry" etc). 584 */ 585 tprints(" <unfinished ...>\n"); 586 printing_tcp->curcol = 0; 587 } 588 } 589 590 printing_tcp = tcp; 591 current_tcp = tcp; 592 current_tcp->curcol = 0; 593 594 if (print_pid_pfx) 595 tprintf("%-5d ", tcp->pid); 596 else if (nprocs > 1 && !outfname) 597 tprintf("[pid %5u] ", tcp->pid); 598 599 if (tflag) { 600 char str[sizeof("HH:MM:SS")]; 601 struct timeval tv, dtv; 602 static struct timeval otv; 603 604 gettimeofday(&tv, NULL); 605 if (rflag) { 606 if (otv.tv_sec == 0) 607 otv = tv; 608 tv_sub(&dtv, &tv, &otv); 609 tprintf("%6ld.%06ld ", 610 (long) dtv.tv_sec, (long) dtv.tv_usec); 611 otv = tv; 612 } 613 else if (tflag > 2) { 614 tprintf("%ld.%06ld ", 615 (long) tv.tv_sec, (long) tv.tv_usec); 616 } 617 else { 618 time_t local = tv.tv_sec; 619 strftime(str, sizeof(str), "%T", localtime(&local)); 620 if (tflag > 1) 621 tprintf("%s.%06ld ", str, (long) tv.tv_usec); 622 else 623 tprintf("%s ", str); 624 } 625 } 626 if (iflag) 627 print_pc(tcp); 628} 629 630void 631tabto(void) 632{ 633 if (current_tcp->curcol < acolumn) 634 tprints(acolumn_spaces + current_tcp->curcol); 635} 636 637/* Should be only called directly *after successful attach* to a tracee. 638 * Otherwise, "strace -oFILE -ff -p<nonexistant_pid>" 639 * may create bogus empty FILE.<nonexistant_pid>, and then die. 640 */ 641static void 642newoutf(struct tcb *tcp) 643{ 644 tcp->outf = shared_log; /* if not -ff mode, the same file is for all */ 645 if (followfork >= 2) { 646 char name[520 + sizeof(int) * 3]; 647 sprintf(name, "%.512s.%u", outfname, tcp->pid); 648 tcp->outf = strace_fopen(name); 649 } 650} 651 652static void 653expand_tcbtab(void) 654{ 655 /* Allocate some more TCBs and expand the table. 656 We don't want to relocate the TCBs because our 657 callers have pointers and it would be a pain. 658 So tcbtab is a table of pointers. Since we never 659 free the TCBs, we allocate a single chunk of many. */ 660 int i = tcbtabsize; 661 struct tcb *newtcbs = calloc(tcbtabsize, sizeof(newtcbs[0])); 662 struct tcb **newtab = realloc(tcbtab, tcbtabsize * 2 * sizeof(tcbtab[0])); 663 if (!newtab || !newtcbs) 664 die_out_of_memory(); 665 tcbtabsize *= 2; 666 tcbtab = newtab; 667 while (i < tcbtabsize) 668 tcbtab[i++] = newtcbs++; 669} 670 671static struct tcb * 672alloctcb(int pid) 673{ 674 int i; 675 struct tcb *tcp; 676 677 if (nprocs == tcbtabsize) 678 expand_tcbtab(); 679 680 for (i = 0; i < tcbtabsize; i++) { 681 tcp = tcbtab[i]; 682 if (!tcp->pid) { 683 memset(tcp, 0, sizeof(*tcp)); 684 tcp->pid = pid; 685#if SUPPORTED_PERSONALITIES > 1 686 tcp->currpers = current_personality; 687#endif 688 nprocs++; 689 if (debug_flag) 690 fprintf(stderr, "new tcb for pid %d, active tcbs:%d\n", tcp->pid, nprocs); 691 return tcp; 692 } 693 } 694 error_msg_and_die("bug in alloctcb"); 695} 696 697static void 698droptcb(struct tcb *tcp) 699{ 700 if (tcp->pid == 0) 701 return; 702 703 nprocs--; 704 if (debug_flag) 705 fprintf(stderr, "dropped tcb for pid %d, %d remain\n", tcp->pid, nprocs); 706 707 if (tcp->outf) { 708 if (followfork >= 2) { 709 if (tcp->curcol != 0) 710 fprintf(tcp->outf, " <detached ...>\n"); 711 fclose(tcp->outf); 712 } else { 713 if (printing_tcp == tcp && tcp->curcol != 0) 714 fprintf(tcp->outf, " <detached ...>\n"); 715 fflush(tcp->outf); 716 } 717 } 718 719 if (current_tcp == tcp) 720 current_tcp = NULL; 721 if (printing_tcp == tcp) 722 printing_tcp = NULL; 723 724 memset(tcp, 0, sizeof(*tcp)); 725} 726 727/* Detach traced process. 728 * Never call DETACH twice on the same process as both unattached and 729 * attached-unstopped processes give the same ESRCH. For unattached process we 730 * would SIGSTOP it and wait for its SIGSTOP notification forever. 731 */ 732static void 733detach(struct tcb *tcp) 734{ 735 int error; 736 int status; 737 738 if (tcp->flags & TCB_BPTSET) 739 clearbpt(tcp); 740 741 /* 742 * Linux wrongly insists the child be stopped 743 * before detaching. Arghh. We go through hoops 744 * to make a clean break of things. 745 */ 746#if defined(SPARC) 747# undef PTRACE_DETACH 748# define PTRACE_DETACH PTRACE_SUNDETACH 749#endif 750 751 if (!(tcp->flags & TCB_ATTACHED)) 752 goto drop; 753 754 /* We attached but possibly didn't see the expected SIGSTOP. 755 * We must catch exactly one as otherwise the detached process 756 * would be left stopped (process state T). 757 */ 758 if (tcp->flags & TCB_IGNORE_ONE_SIGSTOP) 759 goto wait_loop; 760 761 error = ptrace(PTRACE_DETACH, tcp->pid, 0, 0); 762 if (!error) { 763 /* On a clear day, you can see forever. */ 764 goto drop; 765 } 766 if (errno != ESRCH) { 767 /* Shouldn't happen. */ 768 perror_msg("detach: ptrace(PTRACE_DETACH,%u)", tcp->pid); 769 goto drop; 770 } 771 /* ESRCH: process is either not stopped or doesn't exist. */ 772 if (my_tkill(tcp->pid, 0) < 0) { 773 if (errno != ESRCH) 774 /* Shouldn't happen. */ 775 perror_msg("detach: tkill(%u,0)", tcp->pid); 776 /* else: process doesn't exist. */ 777 goto drop; 778 } 779 /* Process is not stopped, need to stop it. */ 780 if (use_seize) { 781 /* 782 * With SEIZE, tracee can be in group-stop already. 783 * In this state sending it another SIGSTOP does nothing. 784 * Need to use INTERRUPT. 785 * Testcase: trying to ^C a "strace -p <stopped_process>". 786 */ 787 error = ptrace(PTRACE_INTERRUPT, tcp->pid, 0, 0); 788 if (!error) 789 goto wait_loop; 790 if (errno != ESRCH) 791 perror_msg("detach: ptrace(PTRACE_INTERRUPT,%u)", tcp->pid); 792 } 793 else { 794 error = my_tkill(tcp->pid, SIGSTOP); 795 if (!error) 796 goto wait_loop; 797 if (errno != ESRCH) 798 perror_msg("detach: tkill(%u,SIGSTOP)", tcp->pid); 799 } 800 /* Either process doesn't exist, or some weird error. */ 801 goto drop; 802 803 wait_loop: 804 /* We end up here in three cases: 805 * 1. We sent PTRACE_INTERRUPT (use_seize case) 806 * 2. We sent SIGSTOP (!use_seize) 807 * 3. Attach SIGSTOP was already pending (TCB_IGNORE_ONE_SIGSTOP set) 808 */ 809 for (;;) { 810 int sig; 811 if (waitpid(tcp->pid, &status, __WALL) < 0) { 812 if (errno == EINTR) 813 continue; 814 /* 815 * if (errno == ECHILD) break; 816 * ^^^ WRONG! We expect this PID to exist, 817 * and want to emit a message otherwise: 818 */ 819 perror_msg("detach: waitpid(%u)", tcp->pid); 820 break; 821 } 822 if (!WIFSTOPPED(status)) { 823 /* 824 * Tracee exited or was killed by signal. 825 * We shouldn't normally reach this place: 826 * we don't want to consume exit status. 827 * Consider "strace -p PID" being ^C-ed: 828 * we want merely to detach from PID. 829 * 830 * However, we _can_ end up here if tracee 831 * was SIGKILLed. 832 */ 833 break; 834 } 835 sig = WSTOPSIG(status); 836 if (debug_flag) 837 fprintf(stderr, "detach wait: event:%d sig:%d\n", 838 (unsigned)status >> 16, sig); 839 if (use_seize) { 840 unsigned event = (unsigned)status >> 16; 841 if (event == PTRACE_EVENT_STOP /*&& sig == SIGTRAP*/) { 842 /* 843 * sig == SIGTRAP: PTRACE_INTERRUPT stop. 844 * sig == other: process was already stopped 845 * with this stopping sig (see tests/detach-stopped). 846 * Looks like re-injecting this sig is not necessary 847 * in DETACH for the tracee to remain stopped. 848 */ 849 sig = 0; 850 } 851 /* 852 * PTRACE_INTERRUPT is not guaranteed to produce 853 * the above event if other ptrace-stop is pending. 854 * See tests/detach-sleeping testcase: 855 * strace got SIGINT while tracee is sleeping. 856 * We sent PTRACE_INTERRUPT. 857 * We see syscall exit, not PTRACE_INTERRUPT stop. 858 * We won't get PTRACE_INTERRUPT stop 859 * if we would CONT now. Need to DETACH. 860 */ 861 if (sig == syscall_trap_sig) 862 sig = 0; 863 /* else: not sure in which case we can be here. 864 * Signal stop? Inject it while detaching. 865 */ 866 ptrace_restart(PTRACE_DETACH, tcp, sig); 867 break; 868 } 869 /* Note: this check has to be after use_seize check */ 870 /* (else, in use_seize case SIGSTOP will be mistreated) */ 871 if (sig == SIGSTOP) { 872 /* Detach, suppressing SIGSTOP */ 873 ptrace_restart(PTRACE_DETACH, tcp, 0); 874 break; 875 } 876 if (sig == syscall_trap_sig) 877 sig = 0; 878 /* Can't detach just yet, may need to wait for SIGSTOP */ 879 error = ptrace_restart(PTRACE_CONT, tcp, sig); 880 if (error < 0) { 881 /* Should not happen. 882 * Note: ptrace_restart returns 0 on ESRCH, so it's not it. 883 * ptrace_restart already emitted error message. 884 */ 885 break; 886 } 887 } 888 889 drop: 890 if (!qflag && (tcp->flags & TCB_ATTACHED)) 891 fprintf(stderr, "Process %u detached\n", tcp->pid); 892 893 droptcb(tcp); 894} 895 896static void 897process_opt_p_list(char *opt) 898{ 899 while (*opt) { 900 /* 901 * We accept -p PID,PID; -p "`pidof PROG`"; -p "`pgrep PROG`". 902 * pidof uses space as delim, pgrep uses newline. :( 903 */ 904 int pid; 905 char *delim = opt + strcspn(opt, ", \n\t"); 906 char c = *delim; 907 908 *delim = '\0'; 909 pid = string_to_uint(opt); 910 if (pid <= 0) { 911 error_msg_and_die("Invalid process id: '%s'", opt); 912 } 913 if (pid == strace_tracer_pid) { 914 error_msg_and_die("I'm sorry, I can't let you do that, Dave."); 915 } 916 *delim = c; 917 alloctcb(pid); 918 if (c == '\0') 919 break; 920 opt = delim + 1; 921 } 922} 923 924static void 925startup_attach(void) 926{ 927 int tcbi; 928 struct tcb *tcp; 929 930 /* 931 * Block user interruptions as we would leave the traced 932 * process stopped (process state T) if we would terminate in 933 * between PTRACE_ATTACH and wait4() on SIGSTOP. 934 * We rely on cleanup() from this point on. 935 */ 936 if (interactive) 937 sigprocmask(SIG_BLOCK, &blocked_set, NULL); 938 939 if (daemonized_tracer) { 940 pid_t pid = fork(); 941 if (pid < 0) { 942 perror_msg_and_die("fork"); 943 } 944 if (pid) { /* parent */ 945 /* 946 * Wait for grandchild to attach to straced process 947 * (grandparent). Grandchild SIGKILLs us after it attached. 948 * Grandparent's wait() is unblocked by our death, 949 * it proceeds to exec the straced program. 950 */ 951 pause(); 952 _exit(0); /* paranoia */ 953 } 954 /* grandchild */ 955 /* We will be the tracer process. Remember our new pid: */ 956 strace_tracer_pid = getpid(); 957 } 958 959 for (tcbi = 0; tcbi < tcbtabsize; tcbi++) { 960 tcp = tcbtab[tcbi]; 961 962 if (!tcp->pid) 963 continue; 964 965 /* Is this a process we should attach to, but not yet attached? */ 966 if (tcp->flags & TCB_ATTACHED) 967 continue; /* no, we already attached it */ 968 969 if (followfork && !daemonized_tracer) { 970 char procdir[sizeof("/proc/%d/task") + sizeof(int) * 3]; 971 DIR *dir; 972 973 sprintf(procdir, "/proc/%d/task", tcp->pid); 974 dir = opendir(procdir); 975 if (dir != NULL) { 976 unsigned int ntid = 0, nerr = 0; 977 struct_dirent *de; 978 979 while ((de = read_dir(dir)) != NULL) { 980 struct tcb *cur_tcp; 981 int tid; 982 983 if (de->d_fileno == 0) 984 continue; 985 /* we trust /proc filesystem */ 986 tid = atoi(de->d_name); 987 if (tid <= 0) 988 continue; 989 ++ntid; 990 if (ptrace_attach_or_seize(tid) < 0) { 991 ++nerr; 992 if (debug_flag) 993 fprintf(stderr, "attach to pid %d failed\n", tid); 994 continue; 995 } 996 if (debug_flag) 997 fprintf(stderr, "attach to pid %d succeeded\n", tid); 998 cur_tcp = tcp; 999 if (tid != tcp->pid) 1000 cur_tcp = alloctcb(tid); 1001 cur_tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop; 1002 newoutf(cur_tcp); 1003 } 1004 closedir(dir); 1005 if (interactive) { 1006 sigprocmask(SIG_SETMASK, &empty_set, NULL); 1007 if (interrupted) 1008 goto ret; 1009 sigprocmask(SIG_BLOCK, &blocked_set, NULL); 1010 } 1011 ntid -= nerr; 1012 if (ntid == 0) { 1013 perror_msg("attach: ptrace(PTRACE_ATTACH, ...)"); 1014 droptcb(tcp); 1015 continue; 1016 } 1017 if (!qflag) { 1018 fprintf(stderr, ntid > 1 1019? "Process %u attached with %u threads\n" 1020: "Process %u attached\n", 1021 tcp->pid, ntid); 1022 } 1023 if (!(tcp->flags & TCB_ATTACHED)) { 1024 /* -p PID, we failed to attach to PID itself 1025 * but did attach to some of its sibling threads. 1026 * Drop PID's tcp. 1027 */ 1028 droptcb(tcp); 1029 } 1030 continue; 1031 } /* if (opendir worked) */ 1032 } /* if (-f) */ 1033 if (ptrace_attach_or_seize(tcp->pid) < 0) { 1034 perror_msg("attach: ptrace(PTRACE_ATTACH, ...)"); 1035 droptcb(tcp); 1036 continue; 1037 } 1038 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop; 1039 newoutf(tcp); 1040 if (debug_flag) 1041 fprintf(stderr, "attach to pid %d (main) succeeded\n", tcp->pid); 1042 1043 if (daemonized_tracer) { 1044 /* 1045 * Make parent go away. 1046 * Also makes grandparent's wait() unblock. 1047 */ 1048 kill(getppid(), SIGKILL); 1049 } 1050 1051 if (!qflag) 1052 fprintf(stderr, 1053 "Process %u attached\n", 1054 tcp->pid); 1055 } /* for each tcbtab[] */ 1056 1057 ret: 1058 if (interactive) 1059 sigprocmask(SIG_SETMASK, &empty_set, NULL); 1060} 1061 1062/* Stack-o-phobic exec helper, in the hope to work around 1063 * NOMMU + "daemonized tracer" difficulty. 1064 */ 1065struct exec_params { 1066 int fd_to_close; 1067 uid_t run_euid; 1068 gid_t run_egid; 1069 char **argv; 1070 char *pathname; 1071}; 1072static struct exec_params params_for_tracee; 1073static void __attribute__ ((noinline, noreturn)) 1074exec_or_die(void) 1075{ 1076 struct exec_params *params = ¶ms_for_tracee; 1077 1078 if (params->fd_to_close >= 0) 1079 close(params->fd_to_close); 1080 if (!daemonized_tracer && !use_seize) { 1081 if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) { 1082 perror_msg_and_die("ptrace(PTRACE_TRACEME, ...)"); 1083 } 1084 } 1085 1086 if (username != NULL) { 1087 /* 1088 * It is important to set groups before we 1089 * lose privileges on setuid. 1090 */ 1091 if (initgroups(username, run_gid) < 0) { 1092 perror_msg_and_die("initgroups"); 1093 } 1094 if (setregid(run_gid, params->run_egid) < 0) { 1095 perror_msg_and_die("setregid"); 1096 } 1097 if (setreuid(run_uid, params->run_euid) < 0) { 1098 perror_msg_and_die("setreuid"); 1099 } 1100 } 1101 else if (geteuid() != 0) 1102 if (setreuid(run_uid, run_uid) < 0) { 1103 perror_msg_and_die("setreuid"); 1104 } 1105 1106 if (!daemonized_tracer) { 1107 /* 1108 * Induce a ptrace stop. Tracer (our parent) 1109 * will resume us with PTRACE_SYSCALL and display 1110 * the immediately following execve syscall. 1111 * Can't do this on NOMMU systems, we are after 1112 * vfork: parent is blocked, stopping would deadlock. 1113 */ 1114 if (!NOMMU_SYSTEM) 1115 kill(getpid(), SIGSTOP); 1116 } else { 1117 alarm(3); 1118 /* we depend on SIGCHLD set to SIG_DFL by init code */ 1119 /* if it happens to be SIG_IGN'ed, wait won't block */ 1120 wait(NULL); 1121 alarm(0); 1122 } 1123 1124 execv(params->pathname, params->argv); 1125 perror_msg_and_die("exec"); 1126} 1127 1128static void 1129startup_child(char **argv) 1130{ 1131 struct_stat statbuf; 1132 const char *filename; 1133 char pathname[MAXPATHLEN]; 1134 int pid; 1135 struct tcb *tcp; 1136 1137 filename = argv[0]; 1138 if (strchr(filename, '/')) { 1139 if (strlen(filename) > sizeof pathname - 1) { 1140 errno = ENAMETOOLONG; 1141 perror_msg_and_die("exec"); 1142 } 1143 strcpy(pathname, filename); 1144 } 1145#ifdef USE_DEBUGGING_EXEC 1146 /* 1147 * Debuggers customarily check the current directory 1148 * first regardless of the path but doing that gives 1149 * security geeks a panic attack. 1150 */ 1151 else if (stat_file(filename, &statbuf) == 0) 1152 strcpy(pathname, filename); 1153#endif /* USE_DEBUGGING_EXEC */ 1154 else { 1155 const char *path; 1156 int m, n, len; 1157 1158 for (path = getenv("PATH"); path && *path; path += m) { 1159 const char *colon = strchr(path, ':'); 1160 if (colon) { 1161 n = colon - path; 1162 m = n + 1; 1163 } 1164 else 1165 m = n = strlen(path); 1166 if (n == 0) { 1167 if (!getcwd(pathname, MAXPATHLEN)) 1168 continue; 1169 len = strlen(pathname); 1170 } 1171 else if (n > sizeof pathname - 1) 1172 continue; 1173 else { 1174 strncpy(pathname, path, n); 1175 len = n; 1176 } 1177 if (len && pathname[len - 1] != '/') 1178 pathname[len++] = '/'; 1179 strcpy(pathname + len, filename); 1180 if (stat_file(pathname, &statbuf) == 0 && 1181 /* Accept only regular files 1182 with some execute bits set. 1183 XXX not perfect, might still fail */ 1184 S_ISREG(statbuf.st_mode) && 1185 (statbuf.st_mode & 0111)) 1186 break; 1187 } 1188 } 1189 if (stat_file(pathname, &statbuf) < 0) { 1190 perror_msg_and_die("Can't stat '%s'", filename); 1191 } 1192 1193 params_for_tracee.fd_to_close = (shared_log != stderr) ? fileno(shared_log) : -1; 1194 params_for_tracee.run_euid = (statbuf.st_mode & S_ISUID) ? statbuf.st_uid : run_uid; 1195 params_for_tracee.run_egid = (statbuf.st_mode & S_ISGID) ? statbuf.st_gid : run_gid; 1196 params_for_tracee.argv = argv; 1197 /* 1198 * On NOMMU, can be safely freed only after execve in tracee. 1199 * It's hard to know when that happens, so we just leak it. 1200 */ 1201 params_for_tracee.pathname = NOMMU_SYSTEM ? strdup(pathname) : pathname; 1202 1203#if defined HAVE_PRCTL && defined PR_SET_PTRACER && defined PR_SET_PTRACER_ANY 1204 if (daemonized_tracer) 1205 prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY); 1206#endif 1207 1208 pid = fork(); 1209 if (pid < 0) { 1210 perror_msg_and_die("fork"); 1211 } 1212 if ((pid != 0 && daemonized_tracer) 1213 || (pid == 0 && !daemonized_tracer) 1214 ) { 1215 /* We are to become the tracee. Two cases: 1216 * -D: we are parent 1217 * not -D: we are child 1218 */ 1219 exec_or_die(); 1220 } 1221 1222 /* We are the tracer */ 1223 1224 if (!daemonized_tracer) { 1225 strace_child = pid; 1226 if (!use_seize) { 1227 /* child did PTRACE_TRACEME, nothing to do in parent */ 1228 } else { 1229 if (!NOMMU_SYSTEM) { 1230 /* Wait until child stopped itself */ 1231 int status; 1232 while (waitpid(pid, &status, WSTOPPED) < 0) { 1233 if (errno == EINTR) 1234 continue; 1235 perror_msg_and_die("waitpid"); 1236 } 1237 if (!WIFSTOPPED(status) || WSTOPSIG(status) != SIGSTOP) { 1238 kill_save_errno(pid, SIGKILL); 1239 perror_msg_and_die("Unexpected wait status %x", status); 1240 } 1241 } 1242 /* Else: NOMMU case, we have no way to sync. 1243 * Just attach to it as soon as possible. 1244 * This means that we may miss a few first syscalls... 1245 */ 1246 1247 if (ptrace_attach_or_seize(pid)) { 1248 kill_save_errno(pid, SIGKILL); 1249 perror_msg_and_die("Can't attach to %d", pid); 1250 } 1251 if (!NOMMU_SYSTEM) 1252 kill(pid, SIGCONT); 1253 } 1254 tcp = alloctcb(pid); 1255 if (!NOMMU_SYSTEM) 1256 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop; 1257 else 1258 tcp->flags |= TCB_ATTACHED | TCB_STARTUP; 1259 newoutf(tcp); 1260 } 1261 else { 1262 /* With -D, we are *child* here, IOW: different pid. Fetch it: */ 1263 strace_tracer_pid = getpid(); 1264 /* The tracee is our parent: */ 1265 pid = getppid(); 1266 alloctcb(pid); 1267 /* attaching will be done later, by startup_attach */ 1268 /* note: we don't do newoutf(tcp) here either! */ 1269 1270 /* NOMMU BUG! -D mode is active, we (child) return, 1271 * and we will scribble over parent's stack! 1272 * When parent later unpauses, it segfaults. 1273 * 1274 * We work around it 1275 * (1) by declaring exec_or_die() NORETURN, 1276 * hopefully compiler will just jump to it 1277 * instead of call (won't push anything to stack), 1278 * (2) by trying very hard in exec_or_die() 1279 * to not use any stack, 1280 * (3) having a really big (MAXPATHLEN) stack object 1281 * in this function, which creates a "buffer" between 1282 * child's and parent's stack pointers. 1283 * This may save us if (1) and (2) failed 1284 * and compiler decided to use stack in exec_or_die() anyway 1285 * (happens on i386 because of stack parameter passing). 1286 * 1287 * A cleaner solution is to use makecontext + setcontext 1288 * to create a genuine separate stack and execute on it. 1289 */ 1290 } 1291} 1292 1293/* 1294 * Test whether the kernel support PTRACE_O_TRACECLONE et al options. 1295 * First fork a new child, call ptrace with PTRACE_SETOPTIONS on it, 1296 * and then see which options are supported by the kernel. 1297 */ 1298static int 1299test_ptrace_setoptions_followfork(void) 1300{ 1301 int pid, expected_grandchild = 0, found_grandchild = 0; 1302 const unsigned int test_options = PTRACE_O_TRACECLONE | 1303 PTRACE_O_TRACEFORK | 1304 PTRACE_O_TRACEVFORK; 1305 1306 /* Need fork for test. NOMMU has no forks */ 1307 if (NOMMU_SYSTEM) 1308 goto worked; /* be bold, and pretend that test succeeded */ 1309 1310 pid = fork(); 1311 if (pid < 0) 1312 perror_msg_and_die("fork"); 1313 if (pid == 0) { 1314 pid = getpid(); 1315 if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) 1316 perror_msg_and_die("%s: PTRACE_TRACEME doesn't work", 1317 __func__); 1318 kill_save_errno(pid, SIGSTOP); 1319 if (fork() < 0) 1320 perror_msg_and_die("fork"); 1321 _exit(0); 1322 } 1323 1324 while (1) { 1325 int status, tracee_pid; 1326 1327 errno = 0; 1328 tracee_pid = wait(&status); 1329 if (tracee_pid <= 0) { 1330 if (errno == EINTR) 1331 continue; 1332 if (errno == ECHILD) 1333 break; 1334 kill_save_errno(pid, SIGKILL); 1335 perror_msg_and_die("%s: unexpected wait result %d", 1336 __func__, tracee_pid); 1337 } 1338 if (WIFEXITED(status)) { 1339 if (WEXITSTATUS(status)) { 1340 if (tracee_pid != pid) 1341 kill_save_errno(pid, SIGKILL); 1342 error_msg_and_die("%s: unexpected exit status %u", 1343 __func__, WEXITSTATUS(status)); 1344 } 1345 continue; 1346 } 1347 if (WIFSIGNALED(status)) { 1348 if (tracee_pid != pid) 1349 kill_save_errno(pid, SIGKILL); 1350 error_msg_and_die("%s: unexpected signal %u", 1351 __func__, WTERMSIG(status)); 1352 } 1353 if (!WIFSTOPPED(status)) { 1354 if (tracee_pid != pid) 1355 kill_save_errno(tracee_pid, SIGKILL); 1356 kill_save_errno(pid, SIGKILL); 1357 error_msg_and_die("%s: unexpected wait status %x", 1358 __func__, status); 1359 } 1360 if (tracee_pid != pid) { 1361 found_grandchild = tracee_pid; 1362 if (ptrace(PTRACE_CONT, tracee_pid, 0, 0) < 0) { 1363 kill_save_errno(tracee_pid, SIGKILL); 1364 kill_save_errno(pid, SIGKILL); 1365 perror_msg_and_die("PTRACE_CONT doesn't work"); 1366 } 1367 continue; 1368 } 1369 switch (WSTOPSIG(status)) { 1370 case SIGSTOP: 1371 if (ptrace(PTRACE_SETOPTIONS, pid, 0, test_options) < 0 1372 && errno != EINVAL && errno != EIO) 1373 perror_msg("PTRACE_SETOPTIONS"); 1374 break; 1375 case SIGTRAP: 1376 if (status >> 16 == PTRACE_EVENT_FORK) { 1377 long msg = 0; 1378 1379 if (ptrace(PTRACE_GETEVENTMSG, pid, 1380 NULL, (long) &msg) == 0) 1381 expected_grandchild = msg; 1382 } 1383 break; 1384 } 1385 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0) { 1386 kill_save_errno(pid, SIGKILL); 1387 perror_msg_and_die("PTRACE_SYSCALL doesn't work"); 1388 } 1389 } 1390 if (expected_grandchild && expected_grandchild == found_grandchild) { 1391 worked: 1392 ptrace_setoptions |= test_options; 1393 if (debug_flag) 1394 fprintf(stderr, "ptrace_setoptions = %#x\n", 1395 ptrace_setoptions); 1396 return 0; 1397 } 1398 error_msg("Test for PTRACE_O_TRACECLONE failed, " 1399 "giving up using this feature."); 1400 return 1; 1401} 1402 1403/* 1404 * Test whether the kernel support PTRACE_O_TRACESYSGOOD. 1405 * First fork a new child, call ptrace(PTRACE_SETOPTIONS) on it, 1406 * and then see whether it will stop with (SIGTRAP | 0x80). 1407 * 1408 * Use of this option enables correct handling of user-generated SIGTRAPs, 1409 * and SIGTRAPs generated by special instructions such as int3 on x86: 1410 * _start: .globl _start 1411 * int3 1412 * movl $42, %ebx 1413 * movl $1, %eax 1414 * int $0x80 1415 * (compile with: "gcc -nostartfiles -nostdlib -o int3 int3.S") 1416 */ 1417static int 1418test_ptrace_setoptions_for_all(void) 1419{ 1420 const unsigned int test_options = PTRACE_O_TRACESYSGOOD | 1421 PTRACE_O_TRACEEXEC; 1422 int pid; 1423 int it_worked = 0; 1424 1425 /* Need fork for test. NOMMU has no forks */ 1426 if (NOMMU_SYSTEM) 1427 goto worked; /* be bold, and pretend that test succeeded */ 1428 1429 pid = fork(); 1430 if (pid < 0) 1431 perror_msg_and_die("fork"); 1432 1433 if (pid == 0) { 1434 pid = getpid(); 1435 if (ptrace(PTRACE_TRACEME, 0L, 0L, 0L) < 0) 1436 /* Note: exits with exitcode 1 */ 1437 perror_msg_and_die("%s: PTRACE_TRACEME doesn't work", 1438 __func__); 1439 kill(pid, SIGSTOP); 1440 _exit(0); /* parent should see entry into this syscall */ 1441 } 1442 1443 while (1) { 1444 int status, tracee_pid; 1445 1446 errno = 0; 1447 tracee_pid = wait(&status); 1448 if (tracee_pid <= 0) { 1449 if (errno == EINTR) 1450 continue; 1451 kill_save_errno(pid, SIGKILL); 1452 perror_msg_and_die("%s: unexpected wait result %d", 1453 __func__, tracee_pid); 1454 } 1455 if (WIFEXITED(status)) { 1456 if (WEXITSTATUS(status) == 0) 1457 break; 1458 error_msg_and_die("%s: unexpected exit status %u", 1459 __func__, WEXITSTATUS(status)); 1460 } 1461 if (WIFSIGNALED(status)) { 1462 error_msg_and_die("%s: unexpected signal %u", 1463 __func__, WTERMSIG(status)); 1464 } 1465 if (!WIFSTOPPED(status)) { 1466 kill(pid, SIGKILL); 1467 error_msg_and_die("%s: unexpected wait status %x", 1468 __func__, status); 1469 } 1470 if (WSTOPSIG(status) == SIGSTOP) { 1471 /* 1472 * We don't check "options aren't accepted" error. 1473 * If it happens, we'll never get (SIGTRAP | 0x80), 1474 * and thus will decide to not use the option. 1475 * IOW: the outcome of the test will be correct. 1476 */ 1477 if (ptrace(PTRACE_SETOPTIONS, pid, 0L, test_options) < 0 1478 && errno != EINVAL && errno != EIO) 1479 perror_msg("PTRACE_SETOPTIONS"); 1480 } 1481 if (WSTOPSIG(status) == (SIGTRAP | 0x80)) { 1482 it_worked = 1; 1483 } 1484 if (ptrace(PTRACE_SYSCALL, pid, 0L, 0L) < 0) { 1485 kill_save_errno(pid, SIGKILL); 1486 perror_msg_and_die("PTRACE_SYSCALL doesn't work"); 1487 } 1488 } 1489 1490 if (it_worked) { 1491 worked: 1492 syscall_trap_sig = (SIGTRAP | 0x80); 1493 ptrace_setoptions |= test_options; 1494 if (debug_flag) 1495 fprintf(stderr, "ptrace_setoptions = %#x\n", 1496 ptrace_setoptions); 1497 return 0; 1498 } 1499 1500 error_msg("Test for PTRACE_O_TRACESYSGOOD failed, " 1501 "giving up using this feature."); 1502 return 1; 1503} 1504 1505#if USE_SEIZE 1506static void 1507test_ptrace_seize(void) 1508{ 1509 int pid; 1510 1511 /* Need fork for test. NOMMU has no forks */ 1512 if (NOMMU_SYSTEM) { 1513 post_attach_sigstop = 0; /* this sets use_seize to 1 */ 1514 return; 1515 } 1516 1517 pid = fork(); 1518 if (pid < 0) 1519 perror_msg_and_die("fork"); 1520 1521 if (pid == 0) { 1522 pause(); 1523 _exit(0); 1524 } 1525 1526 /* PTRACE_SEIZE, unlike ATTACH, doesn't force tracee to trap. After 1527 * attaching tracee continues to run unless a trap condition occurs. 1528 * PTRACE_SEIZE doesn't affect signal or group stop state. 1529 */ 1530 if (ptrace(PTRACE_SEIZE, pid, 0, 0) == 0) { 1531 post_attach_sigstop = 0; /* this sets use_seize to 1 */ 1532 } else if (debug_flag) { 1533 fprintf(stderr, "PTRACE_SEIZE doesn't work\n"); 1534 } 1535 1536 kill(pid, SIGKILL); 1537 1538 while (1) { 1539 int status, tracee_pid; 1540 1541 errno = 0; 1542 tracee_pid = waitpid(pid, &status, 0); 1543 if (tracee_pid <= 0) { 1544 if (errno == EINTR) 1545 continue; 1546 perror_msg_and_die("%s: unexpected wait result %d", 1547 __func__, tracee_pid); 1548 } 1549 if (WIFSIGNALED(status)) { 1550 return; 1551 } 1552 error_msg_and_die("%s: unexpected wait status %x", 1553 __func__, status); 1554 } 1555} 1556#else /* !USE_SEIZE */ 1557# define test_ptrace_seize() ((void)0) 1558#endif 1559 1560static unsigned 1561get_os_release(void) 1562{ 1563 unsigned rel; 1564 const char *p; 1565 struct utsname u; 1566 if (uname(&u) < 0) 1567 perror_msg_and_die("uname"); 1568 /* u.release has this form: "3.2.9[-some-garbage]" */ 1569 rel = 0; 1570 p = u.release; 1571 for (;;) { 1572 if (!(*p >= '0' && *p <= '9')) 1573 error_msg_and_die("Bad OS release string: '%s'", u.release); 1574 /* Note: this open-codes KERNEL_VERSION(): */ 1575 rel = (rel << 8) | atoi(p); 1576 if (rel >= KERNEL_VERSION(1,0,0)) 1577 break; 1578 while (*p >= '0' && *p <= '9') 1579 p++; 1580 if (*p != '.') { 1581 if (rel >= KERNEL_VERSION(0,1,0)) { 1582 /* "X.Y-something" means "X.Y.0" */ 1583 rel <<= 8; 1584 break; 1585 } 1586 error_msg_and_die("Bad OS release string: '%s'", u.release); 1587 } 1588 p++; 1589 } 1590 return rel; 1591} 1592 1593/* 1594 * Initialization part of main() was eating much stack (~0.5k), 1595 * which was unused after init. 1596 * We can reuse it if we move init code into a separate function. 1597 * 1598 * Don't want main() to inline us and defeat the reason 1599 * we have a separate function. 1600 */ 1601static void __attribute__ ((noinline)) 1602init(int argc, char *argv[]) 1603{ 1604 struct tcb *tcp; 1605 int c, i; 1606 int optF = 0; 1607 struct sigaction sa; 1608 1609 progname = argv[0] ? argv[0] : "strace"; 1610 1611 /* Make sure SIGCHLD has the default action so that waitpid 1612 definitely works without losing track of children. The user 1613 should not have given us a bogus state to inherit, but he might 1614 have. Arguably we should detect SIG_IGN here and pass it on 1615 to children, but probably noone really needs that. */ 1616 signal(SIGCHLD, SIG_DFL); 1617 1618 strace_tracer_pid = getpid(); 1619 1620 os_release = get_os_release(); 1621 1622 /* Allocate the initial tcbtab. */ 1623 tcbtabsize = argc; /* Surely enough for all -p args. */ 1624 tcbtab = calloc(tcbtabsize, sizeof(tcbtab[0])); 1625 if (!tcbtab) 1626 die_out_of_memory(); 1627 tcp = calloc(tcbtabsize, sizeof(*tcp)); 1628 if (!tcp) 1629 die_out_of_memory(); 1630 for (c = 0; c < tcbtabsize; c++) 1631 tcbtab[c] = tcp++; 1632 1633 shared_log = stderr; 1634 set_sortby(DEFAULT_SORTBY); 1635 set_personality(DEFAULT_PERSONALITY); 1636 qualify("trace=all"); 1637 qualify("abbrev=all"); 1638 qualify("verbose=all"); 1639#if DEFAULT_QUAL_FLAGS != (QUAL_TRACE | QUAL_ABBREV | QUAL_VERBOSE) 1640# error Bug in DEFAULT_QUAL_FLAGS 1641#endif 1642 qualify("signal=all"); 1643 while ((c = getopt(argc, argv, 1644 "+b:cCdfFhiqrtTvVxyz" 1645 "D" 1646 "a:e:o:O:p:s:S:u:E:P:I:")) != EOF) { 1647 switch (c) { 1648 case 'b': 1649 if (strcmp(optarg, "execve") != 0) 1650 error_msg_and_die("Syscall '%s' for -b isn't supported", 1651 optarg); 1652 detach_on_execve = 1; 1653 break; 1654 case 'c': 1655 if (cflag == CFLAG_BOTH) { 1656 error_msg_and_die("-c and -C are mutually exclusive"); 1657 } 1658 cflag = CFLAG_ONLY_STATS; 1659 break; 1660 case 'C': 1661 if (cflag == CFLAG_ONLY_STATS) { 1662 error_msg_and_die("-c and -C are mutually exclusive"); 1663 } 1664 cflag = CFLAG_BOTH; 1665 break; 1666 case 'd': 1667 debug_flag = 1; 1668 break; 1669 case 'D': 1670 daemonized_tracer = 1; 1671 break; 1672 case 'F': 1673 optF = 1; 1674 break; 1675 case 'f': 1676 followfork++; 1677 break; 1678 case 'h': 1679 usage(stdout, 0); 1680 break; 1681 case 'i': 1682 iflag = 1; 1683 break; 1684 case 'q': 1685 qflag++; 1686 break; 1687 case 'r': 1688 rflag = 1; 1689 /* fall through to tflag++ */ 1690 case 't': 1691 tflag++; 1692 break; 1693 case 'T': 1694 Tflag = 1; 1695 break; 1696 case 'x': 1697 xflag++; 1698 break; 1699 case 'y': 1700 show_fd_path = 1; 1701 break; 1702 case 'v': 1703 qualify("abbrev=none"); 1704 break; 1705 case 'V': 1706 printf("%s -- version %s\n", PACKAGE_NAME, VERSION); 1707 exit(0); 1708 break; 1709 case 'z': 1710 not_failing_only = 1; 1711 break; 1712 case 'a': 1713 acolumn = string_to_uint(optarg); 1714 if (acolumn < 0) 1715 error_opt_arg(c, optarg); 1716 break; 1717 case 'e': 1718 qualify(optarg); 1719 break; 1720 case 'o': 1721 outfname = strdup(optarg); 1722 break; 1723 case 'O': 1724 i = string_to_uint(optarg); 1725 if (i < 0) 1726 error_opt_arg(c, optarg); 1727 set_overhead(i); 1728 break; 1729 case 'p': 1730 process_opt_p_list(optarg); 1731 break; 1732 case 'P': 1733 pathtrace_select(optarg); 1734 break; 1735 case 's': 1736 i = string_to_uint(optarg); 1737 if (i < 0) 1738 error_opt_arg(c, optarg); 1739 max_strlen = i; 1740 break; 1741 case 'S': 1742 set_sortby(optarg); 1743 break; 1744 case 'u': 1745 username = strdup(optarg); 1746 break; 1747 case 'E': 1748 if (putenv(optarg) < 0) 1749 die_out_of_memory(); 1750 break; 1751 case 'I': 1752 opt_intr = string_to_uint(optarg); 1753 if (opt_intr <= 0 || opt_intr >= NUM_INTR_OPTS) 1754 error_opt_arg(c, optarg); 1755 break; 1756 default: 1757 usage(stderr, 1); 1758 break; 1759 } 1760 } 1761 argv += optind; 1762 /* argc -= optind; - no need, argc is not used below */ 1763 1764 acolumn_spaces = malloc(acolumn + 1); 1765 if (!acolumn_spaces) 1766 die_out_of_memory(); 1767 memset(acolumn_spaces, ' ', acolumn); 1768 acolumn_spaces[acolumn] = '\0'; 1769 1770 /* Must have PROG [ARGS], or -p PID. Not both. */ 1771 if (!argv[0] == !nprocs) 1772 usage(stderr, 1); 1773 1774 if (nprocs != 0 && daemonized_tracer) { 1775 error_msg_and_die("-D and -p are mutually exclusive"); 1776 } 1777 1778 if (!followfork) 1779 followfork = optF; 1780 1781 if (followfork >= 2 && cflag) { 1782 error_msg_and_die("(-c or -C) and -ff are mutually exclusive"); 1783 } 1784 1785 /* See if they want to run as another user. */ 1786 if (username != NULL) { 1787 struct passwd *pent; 1788 1789 if (getuid() != 0 || geteuid() != 0) { 1790 error_msg_and_die("You must be root to use the -u option"); 1791 } 1792 pent = getpwnam(username); 1793 if (pent == NULL) { 1794 error_msg_and_die("Cannot find user '%s'", username); 1795 } 1796 run_uid = pent->pw_uid; 1797 run_gid = pent->pw_gid; 1798 } 1799 else { 1800 run_uid = getuid(); 1801 run_gid = getgid(); 1802 } 1803 1804 /* 1805 * On any reasonably recent Linux kernel (circa about 2.5.46) 1806 * need_fork_exec_workarounds should stay 0 after these tests: 1807 */ 1808 /*need_fork_exec_workarounds = 0; - already is */ 1809 if (followfork) 1810 need_fork_exec_workarounds = test_ptrace_setoptions_followfork(); 1811 need_fork_exec_workarounds |= test_ptrace_setoptions_for_all(); 1812 test_ptrace_seize(); 1813 1814 /* Check if they want to redirect the output. */ 1815 if (outfname) { 1816 /* See if they want to pipe the output. */ 1817 if (outfname[0] == '|' || outfname[0] == '!') { 1818 /* 1819 * We can't do the <outfname>.PID funny business 1820 * when using popen, so prohibit it. 1821 */ 1822 if (followfork >= 2) 1823 error_msg_and_die("Piping the output and -ff are mutually exclusive"); 1824 shared_log = strace_popen(outfname + 1); 1825 } 1826 else if (followfork < 2) 1827 shared_log = strace_fopen(outfname); 1828 } else { 1829 /* -ff without -o FILE is the same as single -f */ 1830 if (followfork >= 2) 1831 followfork = 1; 1832 } 1833 1834 if (!outfname || outfname[0] == '|' || outfname[0] == '!') { 1835 char *buf = malloc(BUFSIZ); 1836 if (!buf) 1837 die_out_of_memory(); 1838 setvbuf(shared_log, buf, _IOLBF, BUFSIZ); 1839 } 1840 if (outfname && argv[0]) { 1841 if (!opt_intr) 1842 opt_intr = INTR_NEVER; 1843 qflag = 1; 1844 } 1845 if (!opt_intr) 1846 opt_intr = INTR_WHILE_WAIT; 1847 1848 /* argv[0] -pPID -oFILE Default interactive setting 1849 * yes 0 0 INTR_WHILE_WAIT 1850 * no 1 0 INTR_WHILE_WAIT 1851 * yes 0 1 INTR_NEVER 1852 * no 1 1 INTR_WHILE_WAIT 1853 */ 1854 1855 sigemptyset(&empty_set); 1856 sigemptyset(&blocked_set); 1857 1858 /* startup_child() must be called before the signal handlers get 1859 * installed below as they are inherited into the spawned process. 1860 * Also we do not need to be protected by them as during interruption 1861 * in the startup_child() mode we kill the spawned process anyway. 1862 */ 1863 if (argv[0]) { 1864 if (!NOMMU_SYSTEM || daemonized_tracer) 1865 hide_log_until_execve = 1; 1866 skip_one_b_execve = 1; 1867 startup_child(argv); 1868 } 1869 1870 sa.sa_handler = SIG_IGN; 1871 sigemptyset(&sa.sa_mask); 1872 sa.sa_flags = 0; 1873 sigaction(SIGTTOU, &sa, NULL); /* SIG_IGN */ 1874 sigaction(SIGTTIN, &sa, NULL); /* SIG_IGN */ 1875 if (opt_intr != INTR_ANYWHERE) { 1876 if (opt_intr == INTR_BLOCK_TSTP_TOO) 1877 sigaction(SIGTSTP, &sa, NULL); /* SIG_IGN */ 1878 /* 1879 * In interactive mode (if no -o OUTFILE, or -p PID is used), 1880 * fatal signals are blocked while syscall stop is processed, 1881 * and acted on in between, when waiting for new syscall stops. 1882 * In non-interactive mode, signals are ignored. 1883 */ 1884 if (opt_intr == INTR_WHILE_WAIT) { 1885 sigaddset(&blocked_set, SIGHUP); 1886 sigaddset(&blocked_set, SIGINT); 1887 sigaddset(&blocked_set, SIGQUIT); 1888 sigaddset(&blocked_set, SIGPIPE); 1889 sigaddset(&blocked_set, SIGTERM); 1890 sa.sa_handler = interrupt; 1891 } 1892 /* SIG_IGN, or set handler for these */ 1893 sigaction(SIGHUP, &sa, NULL); 1894 sigaction(SIGINT, &sa, NULL); 1895 sigaction(SIGQUIT, &sa, NULL); 1896 sigaction(SIGPIPE, &sa, NULL); 1897 sigaction(SIGTERM, &sa, NULL); 1898 } 1899 if (nprocs != 0 || daemonized_tracer) 1900 startup_attach(); 1901 1902 /* Do we want pids printed in our -o OUTFILE? 1903 * -ff: no (every pid has its own file); or 1904 * -f: yes (there can be more pids in the future); or 1905 * -p PID1,PID2: yes (there are already more than one pid) 1906 */ 1907 print_pid_pfx = (outfname && followfork < 2 && (followfork == 1 || nprocs > 1)); 1908} 1909 1910static struct tcb * 1911pid2tcb(int pid) 1912{ 1913 int i; 1914 1915 if (pid <= 0) 1916 return NULL; 1917 1918 for (i = 0; i < tcbtabsize; i++) { 1919 struct tcb *tcp = tcbtab[i]; 1920 if (tcp->pid == pid) 1921 return tcp; 1922 } 1923 1924 return NULL; 1925} 1926 1927static void 1928cleanup(void) 1929{ 1930 int i; 1931 struct tcb *tcp; 1932 int fatal_sig; 1933 1934 /* 'interrupted' is a volatile object, fetch it only once */ 1935 fatal_sig = interrupted; 1936 if (!fatal_sig) 1937 fatal_sig = SIGTERM; 1938 1939 for (i = 0; i < tcbtabsize; i++) { 1940 tcp = tcbtab[i]; 1941 if (!tcp->pid) 1942 continue; 1943 if (debug_flag) 1944 fprintf(stderr, 1945 "cleanup: looking at pid %u\n", tcp->pid); 1946 if (tcp->pid == strace_child) { 1947 kill(tcp->pid, SIGCONT); 1948 kill(tcp->pid, fatal_sig); 1949 } 1950 detach(tcp); 1951 } 1952 if (cflag) 1953 call_summary(shared_log); 1954} 1955 1956static void 1957interrupt(int sig) 1958{ 1959 interrupted = sig; 1960} 1961 1962static void 1963trace(void) 1964{ 1965 struct rusage ru; 1966 1967 while (nprocs != 0) { 1968 int pid; 1969 int wait_errno; 1970 int status, sig; 1971 int stopped; 1972 struct tcb *tcp; 1973 unsigned event; 1974 1975 if (interrupted) 1976 return; 1977 1978 if (interactive) 1979 sigprocmask(SIG_SETMASK, &empty_set, NULL); 1980 pid = wait4(-1, &status, __WALL, (cflag ? &ru : NULL)); 1981 wait_errno = errno; 1982 if (interactive) 1983 sigprocmask(SIG_BLOCK, &blocked_set, NULL); 1984 1985 if (pid < 0) { 1986 if (wait_errno == EINTR) 1987 continue; 1988 if (wait_errno == ECHILD) 1989 /* Should not happen since nprocs > 0 */ 1990 return; 1991 errno = wait_errno; 1992 perror_msg_and_die("wait4(__WALL)"); 1993 } 1994 1995 if (pid == popen_pid) { 1996 if (!WIFSTOPPED(status)) 1997 popen_pid = 0; 1998 continue; 1999 } 2000 2001 event = ((unsigned)status >> 16); 2002 if (debug_flag) { 2003 char buf[sizeof("WIFEXITED,exitcode=%u") + sizeof(int)*3 /*paranoia:*/ + 16]; 2004 char evbuf[sizeof(",EVENT_VFORK_DONE (%u)") + sizeof(int)*3 /*paranoia:*/ + 16]; 2005 strcpy(buf, "???"); 2006 if (WIFSIGNALED(status)) 2007#ifdef WCOREDUMP 2008 sprintf(buf, "WIFSIGNALED,%ssig=%s", 2009 WCOREDUMP(status) ? "core," : "", 2010 signame(WTERMSIG(status))); 2011#else 2012 sprintf(buf, "WIFSIGNALED,sig=%s", 2013 signame(WTERMSIG(status))); 2014#endif 2015 if (WIFEXITED(status)) 2016 sprintf(buf, "WIFEXITED,exitcode=%u", WEXITSTATUS(status)); 2017 if (WIFSTOPPED(status)) 2018 sprintf(buf, "WIFSTOPPED,sig=%s", signame(WSTOPSIG(status))); 2019#ifdef WIFCONTINUED 2020 /* Should never be seen */ 2021 if (WIFCONTINUED(status)) 2022 strcpy(buf, "WIFCONTINUED"); 2023#endif 2024 evbuf[0] = '\0'; 2025 if (event != 0) { 2026 static const char *const event_names[] = { 2027 [PTRACE_EVENT_CLONE] = "CLONE", 2028 [PTRACE_EVENT_FORK] = "FORK", 2029 [PTRACE_EVENT_VFORK] = "VFORK", 2030 [PTRACE_EVENT_VFORK_DONE] = "VFORK_DONE", 2031 [PTRACE_EVENT_EXEC] = "EXEC", 2032 [PTRACE_EVENT_EXIT] = "EXIT", 2033 /* [PTRACE_EVENT_STOP (=128)] would make biggish array */ 2034 }; 2035 const char *e = "??"; 2036 if (event < ARRAY_SIZE(event_names)) 2037 e = event_names[event]; 2038 else if (event == PTRACE_EVENT_STOP) 2039 e = "STOP"; 2040 sprintf(evbuf, ",EVENT_%s (%u)", e, event); 2041 } 2042 fprintf(stderr, " [wait(0x%06x) = %u] %s%s\n", status, pid, buf, evbuf); 2043 } 2044 2045 /* Look up 'pid' in our table. */ 2046 tcp = pid2tcb(pid); 2047 2048 if (!tcp) { 2049 if (!WIFSTOPPED(status)) { 2050 /* This can happen if we inherited 2051 * an unknown child. Example: 2052 * (sleep 1 & exec strace sleep 2) 2053 */ 2054 error_msg("Exit of unknown pid %u seen", pid); 2055 continue; 2056 } 2057 if (followfork) { 2058 /* We assume it's a fork/vfork/clone child */ 2059 tcp = alloctcb(pid); 2060 tcp->flags |= TCB_ATTACHED | TCB_STARTUP | post_attach_sigstop; 2061 newoutf(tcp); 2062 if (!qflag) 2063 fprintf(stderr, "Process %d attached\n", 2064 pid); 2065 } else { 2066 /* This can happen if a clone call used 2067 * CLONE_PTRACE itself. 2068 */ 2069 ptrace(PTRACE_CONT, pid, (char *) 0, 0); 2070 error_msg("Stop of unknown pid %u seen, PTRACE_CONTed it", pid); 2071 continue; 2072 } 2073 } 2074 2075 clear_regs(); 2076 if (WIFSTOPPED(status)) 2077 get_regs(pid); 2078 2079 /* Under Linux, execve changes pid to thread leader's pid, 2080 * and we see this changed pid on EVENT_EXEC and later, 2081 * execve sysexit. Leader "disappears" without exit 2082 * notification. Let user know that, drop leader's tcb, 2083 * and fix up pid in execve thread's tcb. 2084 * Effectively, execve thread's tcb replaces leader's tcb. 2085 * 2086 * BTW, leader is 'stuck undead' (doesn't report WIFEXITED 2087 * on exit syscall) in multithreaded programs exactly 2088 * in order to handle this case. 2089 * 2090 * PTRACE_GETEVENTMSG returns old pid starting from Linux 3.0. 2091 * On 2.6 and earlier, it can return garbage. 2092 */ 2093 if (event == PTRACE_EVENT_EXEC && os_release >= KERNEL_VERSION(3,0,0)) { 2094 FILE *fp; 2095 struct tcb *execve_thread; 2096 long old_pid = 0; 2097 2098 if (ptrace(PTRACE_GETEVENTMSG, pid, NULL, (long) &old_pid) < 0) 2099 goto dont_switch_tcbs; 2100 /* Avoid truncation in pid2tcb() param passing */ 2101 if (old_pid > UINT_MAX) 2102 goto dont_switch_tcbs; 2103 if (old_pid <= 0 || old_pid == pid) 2104 goto dont_switch_tcbs; 2105 execve_thread = pid2tcb(old_pid); 2106 /* It should be !NULL, but I feel paranoid */ 2107 if (!execve_thread) 2108 goto dont_switch_tcbs; 2109 2110 if (execve_thread->curcol != 0) { 2111 /* 2112 * One case we are here is -ff: 2113 * try "strace -oLOG -ff test/threaded_execve" 2114 */ 2115 fprintf(execve_thread->outf, " <pid changed to %d ...>\n", pid); 2116 /*execve_thread->curcol = 0; - no need, see code below */ 2117 } 2118 /* Swap output FILEs (needed for -ff) */ 2119 fp = execve_thread->outf; 2120 execve_thread->outf = tcp->outf; 2121 tcp->outf = fp; 2122 /* And their column positions */ 2123 execve_thread->curcol = tcp->curcol; 2124 tcp->curcol = 0; 2125 /* Drop leader, but close execve'd thread outfile (if -ff) */ 2126 droptcb(tcp); 2127 /* Switch to the thread, reusing leader's outfile and pid */ 2128 tcp = execve_thread; 2129 tcp->pid = pid; 2130 if (cflag != CFLAG_ONLY_STATS) { 2131 printleader(tcp); 2132 tprintf("+++ superseded by execve in pid %lu +++\n", old_pid); 2133 line_ended(); 2134 tcp->flags |= TCB_REPRINT; 2135 } 2136 } 2137 dont_switch_tcbs: 2138 2139 if (event == PTRACE_EVENT_EXEC) { 2140 if (detach_on_execve && !skip_one_b_execve) 2141 detach(tcp); /* do "-b execve" thingy */ 2142 skip_one_b_execve = 0; 2143 } 2144 2145 /* Set current output file */ 2146 current_tcp = tcp; 2147 2148 if (cflag) { 2149 tv_sub(&tcp->dtime, &ru.ru_stime, &tcp->stime); 2150 tcp->stime = ru.ru_stime; 2151 } 2152 2153 if (WIFSIGNALED(status)) { 2154 if (pid == strace_child) 2155 exit_code = 0x100 | WTERMSIG(status); 2156 if (cflag != CFLAG_ONLY_STATS 2157 && (qual_flags[WTERMSIG(status)] & QUAL_SIGNAL) 2158 ) { 2159 printleader(tcp); 2160#ifdef WCOREDUMP 2161 tprintf("+++ killed by %s %s+++\n", 2162 signame(WTERMSIG(status)), 2163 WCOREDUMP(status) ? "(core dumped) " : ""); 2164#else 2165 tprintf("+++ killed by %s +++\n", 2166 signame(WTERMSIG(status))); 2167#endif 2168 line_ended(); 2169 } 2170 droptcb(tcp); 2171 continue; 2172 } 2173 if (WIFEXITED(status)) { 2174 if (pid == strace_child) 2175 exit_code = WEXITSTATUS(status); 2176 if (cflag != CFLAG_ONLY_STATS && 2177 qflag < 2) { 2178 printleader(tcp); 2179 tprintf("+++ exited with %d +++\n", WEXITSTATUS(status)); 2180 line_ended(); 2181 } 2182 droptcb(tcp); 2183 continue; 2184 } 2185 if (!WIFSTOPPED(status)) { 2186 fprintf(stderr, "PANIC: pid %u not stopped\n", pid); 2187 droptcb(tcp); 2188 continue; 2189 } 2190 2191 /* Is this the very first time we see this tracee stopped? */ 2192 if (tcp->flags & TCB_STARTUP) { 2193 if (debug_flag) 2194 fprintf(stderr, "pid %d has TCB_STARTUP, initializing it\n", tcp->pid); 2195 tcp->flags &= ~TCB_STARTUP; 2196 if (tcp->flags & TCB_BPTSET) { 2197 /* 2198 * One example is a breakpoint inherited from 2199 * parent through fork(). 2200 */ 2201 if (clearbpt(tcp) < 0) { 2202 /* Pretty fatal */ 2203 droptcb(tcp); 2204 exit_code = 1; 2205 return; 2206 } 2207 } 2208 if (ptrace_setoptions) { 2209 if (debug_flag) 2210 fprintf(stderr, "setting opts %x on pid %d\n", ptrace_setoptions, tcp->pid); 2211 if (ptrace(PTRACE_SETOPTIONS, tcp->pid, NULL, ptrace_setoptions) < 0) { 2212 if (errno != ESRCH) { 2213 /* Should never happen, really */ 2214 perror_msg_and_die("PTRACE_SETOPTIONS"); 2215 } 2216 } 2217 } 2218 } 2219 2220 sig = WSTOPSIG(status); 2221 2222 if (event != 0) { 2223 /* Ptrace event */ 2224#if USE_SEIZE 2225 if (event == PTRACE_EVENT_STOP) { 2226 /* 2227 * PTRACE_INTERRUPT-stop or group-stop. 2228 * PTRACE_INTERRUPT-stop has sig == SIGTRAP here. 2229 */ 2230 if (sig == SIGSTOP 2231 || sig == SIGTSTP 2232 || sig == SIGTTIN 2233 || sig == SIGTTOU 2234 ) { 2235 stopped = 1; 2236 goto show_stopsig; 2237 } 2238 } 2239#endif 2240 goto restart_tracee_with_sig_0; 2241 } 2242 2243 /* Is this post-attach SIGSTOP? 2244 * Interestingly, the process may stop 2245 * with STOPSIG equal to some other signal 2246 * than SIGSTOP if we happend to attach 2247 * just before the process takes a signal. 2248 */ 2249 if (sig == SIGSTOP && (tcp->flags & TCB_IGNORE_ONE_SIGSTOP)) { 2250 if (debug_flag) 2251 fprintf(stderr, "ignored SIGSTOP on pid %d\n", tcp->pid); 2252 tcp->flags &= ~TCB_IGNORE_ONE_SIGSTOP; 2253 goto restart_tracee_with_sig_0; 2254 } 2255 2256 if (sig != syscall_trap_sig) { 2257 siginfo_t si; 2258 2259 /* Nonzero (true) if tracee is stopped by signal 2260 * (as opposed to "tracee received signal"). 2261 * TODO: shouldn't we check for errno == EINVAL too? 2262 * We can get ESRCH instead, you know... 2263 */ 2264 stopped = (ptrace(PTRACE_GETSIGINFO, pid, 0, (long) &si) < 0); 2265#if USE_SEIZE 2266 show_stopsig: 2267#endif 2268 if (cflag != CFLAG_ONLY_STATS 2269 && !hide_log_until_execve 2270 && (qual_flags[sig] & QUAL_SIGNAL) 2271 ) { 2272 printleader(tcp); 2273 if (!stopped) { 2274 tprintf("--- %s ", signame(sig)); 2275 printsiginfo(&si, verbose(tcp)); 2276 tprints(" ---\n"); 2277 } else 2278 tprintf("--- stopped by %s ---\n", 2279 signame(sig)); 2280 line_ended(); 2281 } 2282 2283 if (!stopped) 2284 /* It's signal-delivery-stop. Inject the signal */ 2285 goto restart_tracee; 2286 2287 /* It's group-stop */ 2288 if (use_seize) { 2289 /* 2290 * This ends ptrace-stop, but does *not* end group-stop. 2291 * This makes stopping signals work properly on straced process 2292 * (that is, process really stops. It used to continue to run). 2293 */ 2294 if (ptrace_restart(PTRACE_LISTEN, tcp, 0) < 0) { 2295 /* Note: ptrace_restart emitted error message */ 2296 exit_code = 1; 2297 return; 2298 } 2299 continue; 2300 } 2301 /* We don't have PTRACE_LISTEN support... */ 2302 goto restart_tracee; 2303 } 2304 2305 /* We handled quick cases, we are permitted to interrupt now. */ 2306 if (interrupted) 2307 return; 2308 2309 /* This should be syscall entry or exit. 2310 * (Or it still can be that pesky post-execve SIGTRAP!) 2311 * Handle it. 2312 */ 2313 if (trace_syscall(tcp) < 0) { 2314 /* ptrace() failed in trace_syscall(). 2315 * Likely a result of process disappearing mid-flight. 2316 * Observed case: exit_group() or SIGKILL terminating 2317 * all processes in thread group. 2318 * We assume that ptrace error was caused by process death. 2319 * We used to detach(tcp) here, but since we no longer 2320 * implement "detach before death" policy/hack, 2321 * we can let this process to report its death to us 2322 * normally, via WIFEXITED or WIFSIGNALED wait status. 2323 */ 2324 continue; 2325 } 2326 restart_tracee_with_sig_0: 2327 sig = 0; 2328 restart_tracee: 2329 if (ptrace_restart(PTRACE_SYSCALL, tcp, sig) < 0) { 2330 /* Note: ptrace_restart emitted error message */ 2331 exit_code = 1; 2332 return; 2333 } 2334 } /* while (nprocs != 0) */ 2335} 2336 2337int 2338main(int argc, char *argv[]) 2339{ 2340 init(argc, argv); 2341 2342 /* Run main tracing loop */ 2343 trace(); 2344 2345 cleanup(); 2346 fflush(NULL); 2347 if (shared_log != stderr) 2348 fclose(shared_log); 2349 if (popen_pid) { 2350 while (waitpid(popen_pid, NULL, 0) < 0 && errno == EINTR) 2351 ; 2352 } 2353 if (exit_code > 0xff) { 2354 /* Avoid potential core file clobbering. */ 2355 struct_rlimit rlim = {0, 0}; 2356 set_rlimit(RLIMIT_CORE, &rlim); 2357 2358 /* Child was killed by a signal, mimic that. */ 2359 exit_code &= 0xff; 2360 signal(exit_code, SIG_DFL); 2361 raise(exit_code); 2362 /* Paranoia - what if this signal is not fatal? 2363 Exit with 128 + signo then. */ 2364 exit_code += 128; 2365 } 2366 2367 return exit_code; 2368} 2369