granicus.if.org Git - strace/blob - syscall.c

   1 /*
   2  * Copyright (c) 1991, 1992 Paul Kranenburg <pk@cs.few.eur.nl>
   3  * Copyright (c) 1993 Branko Lankester <branko@hacktic.nl>
   4  * Copyright (c) 1993, 1994, 1995, 1996 Rick Sladkey <jrs@world.std.com>
   5  * Copyright (c) 1996-1999 Wichert Akkerman <wichert@cistron.nl>
   6  * Copyright (c) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
   7  *                     Linux for s390 port by D.J. Barrow
   8  *                    <barrow_dj@mail.yahoo.com,djbarrow@de.ibm.com>
   9  * Copyright (c) 1999-2017 The strace developers.
  10  * All rights reserved.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions and the following disclaimer.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  * 3. The name of the author may not be used to endorse or promote products
  21  *    derived from this software without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include "defs.h"
  36 #include "native_defs.h"
  37 #include "nsig.h"
  38 #include "number_set.h"
  39 #include <sys/param.h>
  40
  41 /* for struct iovec */
  42 #include <sys/uio.h>
  43
  44 /* for __X32_SYSCALL_BIT */
  45 #include <asm/unistd.h>
  46
  47 #include "regs.h"
  48 #include "ptrace.h"
  49
  50 #if defined(SPARC64)
  51 # undef PTRACE_GETREGS
  52 # define PTRACE_GETREGS PTRACE_GETREGS64
  53 # undef PTRACE_SETREGS
  54 # define PTRACE_SETREGS PTRACE_SETREGS64
  55 #endif
  56
  57 #ifndef NT_PRSTATUS
  58 # define NT_PRSTATUS 1
  59 #endif
  60
  61 #include "syscall.h"
  62 #include "xstring.h"
  63
  64 /* Define these shorthand notations to simplify the syscallent files. */
  65 #include "sysent_shorthand_defs.h"
  66
  67 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  68
  69 const struct_sysent sysent0[] = {
  70 #include "syscallent.h"
  71 };
  72
  73 #if SUPPORTED_PERSONALITIES > 1
  74 # include PERSONALITY1_INCLUDE_FUNCS
  75 static const struct_sysent sysent1[] = {
  76 # include "syscallent1.h"
  77 };
  78 #endif
  79
  80 #if SUPPORTED_PERSONALITIES > 2
  81 # include PERSONALITY2_INCLUDE_FUNCS
  82 static const struct_sysent sysent2[] = {
  83 # include "syscallent2.h"
  84 };
  85 #endif
  86
  87 /* Now undef them since short defines cause wicked namespace pollution. */
  88 #include "sysent_shorthand_undefs.h"
  89
  90 /*
  91  * `ioctlent[012].h' files are automatically generated by the auxiliary
  92  * program `ioctlsort', such that the list is sorted by the `code' field.
  93  * This has the side-effect of resolving the _IO.. macros into
  94  * plain integers, eliminating the need to include here everything
  95  * in "/usr/include".
  96  */
  97
  98 const char *const errnoent0[] = {
  99 #include "errnoent.h"
 100 };
 101 const char *const signalent0[] = {
 102 #include "signalent.h"
 103 };
 104 const struct_ioctlent ioctlent0[] = {
 105 #include "ioctlent0.h"
 106 };
 107
 108 #if SUPPORTED_PERSONALITIES > 1
 109 static const char *const errnoent1[] = {
 110 # include "errnoent1.h"
 111 };
 112 static const char *const signalent1[] = {
 113 # include "signalent1.h"
 114 };
 115 static const struct_ioctlent ioctlent1[] = {
 116 # include "ioctlent1.h"
 117 };
 118 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 119 static const struct_printers printers0 = {
 120 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 121 };
 122 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 123 static const struct_printers printers1 = {
 124 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 125 };
 126 #endif
 127
 128 #if SUPPORTED_PERSONALITIES > 2
 129 static const char *const errnoent2[] = {
 130 # include "errnoent2.h"
 131 };
 132 static const char *const signalent2[] = {
 133 # include "signalent2.h"
 134 };
 135 static const struct_ioctlent ioctlent2[] = {
 136 # include "ioctlent2.h"
 137 };
 138 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 139 static const struct_printers printers2 = {
 140 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 141 };
 142 #endif
 143
 144 enum {
 145         nsyscalls0 = ARRAY_SIZE(sysent0)
 146 #if SUPPORTED_PERSONALITIES > 1
 147         , nsyscalls1 = ARRAY_SIZE(sysent1)
 148 # if SUPPORTED_PERSONALITIES > 2
 149         , nsyscalls2 = ARRAY_SIZE(sysent2)
 150 # endif
 151 #endif
 152 };
 153
 154 enum {
 155         nerrnos0 = ARRAY_SIZE(errnoent0)
 156 #if SUPPORTED_PERSONALITIES > 1
 157         , nerrnos1 = ARRAY_SIZE(errnoent1)
 158 # if SUPPORTED_PERSONALITIES > 2
 159         , nerrnos2 = ARRAY_SIZE(errnoent2)
 160 # endif
 161 #endif
 162 };
 163
 164 enum {
 165         nsignals0 = ARRAY_SIZE(signalent0)
 166 #if SUPPORTED_PERSONALITIES > 1
 167         , nsignals1 = ARRAY_SIZE(signalent1)
 168 # if SUPPORTED_PERSONALITIES > 2
 169         , nsignals2 = ARRAY_SIZE(signalent2)
 170 # endif
 171 #endif
 172 };
 173
 174 enum {
 175         nioctlents0 = ARRAY_SIZE(ioctlent0)
 176 #if SUPPORTED_PERSONALITIES > 1
 177         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 178 # if SUPPORTED_PERSONALITIES > 2
 179         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 180 # endif
 181 #endif
 182 };
 183
 184 #if SUPPORTED_PERSONALITIES > 1
 185 const struct_sysent *sysent = sysent0;
 186 const char *const *errnoent = errnoent0;
 187 const char *const *signalent = signalent0;
 188 const struct_ioctlent *ioctlent = ioctlent0;
 189 const struct_printers *printers = &printers0;
 190 #endif
 191
 192 unsigned nsyscalls = nsyscalls0;
 193 unsigned nerrnos = nerrnos0;
 194 unsigned nsignals = nsignals0;
 195 unsigned nioctlents = nioctlents0;
 196
 197 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 198         nsyscalls0,
 199 #if SUPPORTED_PERSONALITIES > 1
 200         nsyscalls1,
 201 #endif
 202 #if SUPPORTED_PERSONALITIES > 2
 203         nsyscalls2,
 204 #endif
 205 };
 206 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 207         sysent0,
 208 #if SUPPORTED_PERSONALITIES > 1
 209         sysent1,
 210 #endif
 211 #if SUPPORTED_PERSONALITIES > 2
 212         sysent2,
 213 #endif
 214 };
 215
 216 const char *const personality_names[] =
 217 # if defined X86_64
 218         {"64 bit", "32 bit", "x32"}
 219 # elif defined X32
 220         {"x32", "32 bit"}
 221 # elif SUPPORTED_PERSONALITIES == 2
 222         {"64 bit", "32 bit"}
 223 # else
 224         {STRINGIFY_VAL(__WORDSIZE) " bit"}
 225 # endif
 226         ;
 227
 228 #if SUPPORTED_PERSONALITIES > 1
 229
 230 unsigned current_personality;
 231
 232 # ifndef current_wordsize
 233 unsigned current_wordsize = PERSONALITY0_WORDSIZE;
 234 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 235         PERSONALITY0_WORDSIZE,
 236         PERSONALITY1_WORDSIZE,
 237 # if SUPPORTED_PERSONALITIES > 2
 238         PERSONALITY2_WORDSIZE,
 239 # endif
 240 };
 241 # endif
 242
 243 # ifndef current_klongsize
 244 unsigned current_klongsize = PERSONALITY0_KLONGSIZE;
 245 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 246         PERSONALITY0_KLONGSIZE,
 247         PERSONALITY1_KLONGSIZE,
 248 #  if SUPPORTED_PERSONALITIES > 2
 249         PERSONALITY2_KLONGSIZE,
 250 #  endif
 251 };
 252 # endif
 253
 254 void
 255 set_personality(unsigned int personality)
 256 {
 257         if (personality == current_personality)
 258                 return;
 259
 260         if (personality >= SUPPORTED_PERSONALITIES)
 261                 error_msg_and_die("Requested switch to unsupported personality "
 262                                   "%u", personality);
 263
 264         nsyscalls = nsyscall_vec[personality];
 265         sysent = sysent_vec[personality];
 266
 267         switch (personality) {
 268         case 0:
 269                 errnoent = errnoent0;
 270                 nerrnos = nerrnos0;
 271                 ioctlent = ioctlent0;
 272                 nioctlents = nioctlents0;
 273                 signalent = signalent0;
 274                 nsignals = nsignals0;
 275                 printers = &printers0;
 276                 break;
 277
 278         case 1:
 279                 errnoent = errnoent1;
 280                 nerrnos = nerrnos1;
 281                 ioctlent = ioctlent1;
 282                 nioctlents = nioctlents1;
 283                 signalent = signalent1;
 284                 nsignals = nsignals1;
 285                 printers = &printers1;
 286                 break;
 287
 288 # if SUPPORTED_PERSONALITIES > 2
 289         case 2:
 290                 errnoent = errnoent2;
 291                 nerrnos = nerrnos2;
 292                 ioctlent = ioctlent2;
 293                 nioctlents = nioctlents2;
 294                 signalent = signalent2;
 295                 nsignals = nsignals2;
 296                 printers = &printers2;
 297                 break;
 298 # endif
 299         }
 300
 301         current_personality = personality;
 302 # ifndef current_wordsize
 303         current_wordsize = personality_wordsize[personality];
 304 # endif
 305 # ifndef current_klongsize
 306         current_klongsize = personality_klongsize[personality];
 307 # endif
 308 }
 309
 310 static void
 311 update_personality(struct tcb *tcp, unsigned int personality)
 312 {
 313         static bool need_mpers_warning[] =
 314                 { false, !HAVE_PERSONALITY_1_MPERS, !HAVE_PERSONALITY_2_MPERS };
 315
 316         set_personality(personality);
 317
 318         if (personality == tcp->currpers)
 319                 return;
 320         tcp->currpers = personality;
 321
 322         if (!qflag) {
 323                 error_msg("[ Process PID=%d runs in %s mode. ]",
 324                           tcp->pid, personality_names[personality]);
 325         }
 326
 327         if (need_mpers_warning[personality]) {
 328                 error_msg("WARNING: Proper structure decoding for this "
 329                           "personality is not supported, please consider "
 330                           "building strace with mpers support enabled.");
 331                 need_mpers_warning[personality] = false;
 332         }
 333 }
 334 #endif
 335
 336 #ifdef SYS_socket_subcall
 337 static void
 338 decode_socket_subcall(struct tcb *tcp)
 339 {
 340         const int call = tcp->u_arg[0];
 341
 342         if (call < 1 || call >= SYS_socket_nsubcalls)
 343                 return;
 344
 345         const kernel_ulong_t scno = SYS_socket_subcall + call;
 346         const unsigned int nargs = sysent[scno].nargs;
 347         uint64_t buf[nargs];
 348
 349         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 350                 return;
 351
 352         tcp->scno = scno;
 353         tcp->qual_flg = qual_flags(scno);
 354         tcp->s_ent = &sysent[scno];
 355
 356         unsigned int i;
 357         for (i = 0; i < nargs; ++i)
 358                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 359                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 360 }
 361 #endif /* SYS_socket_subcall */
 362
 363 #ifdef SYS_ipc_subcall
 364 static void
 365 decode_ipc_subcall(struct tcb *tcp)
 366 {
 367         unsigned int call = tcp->u_arg[0];
 368         const unsigned int version = call >> 16;
 369
 370         if (version) {
 371 # if defined S390 || defined S390X
 372                 return;
 373 # else
 374 #  ifdef SPARC64
 375                 if (current_wordsize == 8)
 376                         return;
 377 #  endif
 378                 set_tcb_priv_ulong(tcp, version);
 379                 call &= 0xffff;
 380 # endif
 381         }
 382
 383         switch (call) {
 384                 case  1: case  2: case  3: case  4:
 385                 case 11: case 12: case 13: case 14:
 386                 case 21: case 22: case 23: case 24:
 387                         break;
 388                 default:
 389                         return;
 390         }
 391
 392         tcp->scno = SYS_ipc_subcall + call;
 393         tcp->qual_flg = qual_flags(tcp->scno);
 394         tcp->s_ent = &sysent[tcp->scno];
 395
 396         const unsigned int n = tcp->s_ent->nargs;
 397         unsigned int i;
 398         for (i = 0; i < n; i++)
 399                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 400 }
 401 #endif /* SYS_ipc_subcall */
 402
 403 #ifdef SYS_syscall_subcall
 404 static void
 405 decode_syscall_subcall(struct tcb *tcp)
 406 {
 407         if (!scno_is_valid(tcp->u_arg[0]))
 408                 return;
 409         tcp->scno = tcp->u_arg[0];
 410         tcp->qual_flg = qual_flags(tcp->scno);
 411         tcp->s_ent = &sysent[tcp->scno];
 412         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 413                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 414 # ifdef LINUX_MIPSO32
 415         /*
 416          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 417          * and sync_file_range) requires additional code,
 418          * see linux/mips/get_syscall_args.c
 419          */
 420         if (tcp->s_ent->nargs == MAX_ARGS) {
 421                 if (umoven(tcp,
 422                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 423                            sizeof(tcp->u_arg[0]),
 424                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 425                 tcp->u_arg[MAX_ARGS - 1] = 0;
 426         }
 427 # endif /* LINUX_MIPSO32 */
 428 }
 429 #endif /* SYS_syscall_subcall */
 430
 431 static void
 432 dumpio(struct tcb *tcp)
 433 {
 434         if (syserror(tcp))
 435                 return;
 436
 437         int fd = tcp->u_arg[0];
 438         if (fd < 0)
 439                 return;
 440
 441         if (is_number_in_set(fd, read_set)) {
 442                 switch (tcp->s_ent->sen) {
 443                 case SEN_read:
 444                 case SEN_pread:
 445                 case SEN_recv:
 446                 case SEN_recvfrom:
 447                 case SEN_mq_timedreceive:
 448                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 449                         return;
 450                 case SEN_readv:
 451                 case SEN_preadv:
 452                 case SEN_preadv2:
 453                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 454                                      tcp->u_rval);
 455                         return;
 456                 case SEN_recvmsg:
 457                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 458                         return;
 459                 case SEN_recvmmsg:
 460                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 461                         return;
 462                 }
 463         }
 464         if (is_number_in_set(fd, write_set)) {
 465                 switch (tcp->s_ent->sen) {
 466                 case SEN_write:
 467                 case SEN_pwrite:
 468                 case SEN_send:
 469                 case SEN_sendto:
 470                 case SEN_mq_timedsend:
 471                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 472                         break;
 473                 case SEN_writev:
 474                 case SEN_pwritev:
 475                 case SEN_pwritev2:
 476                 case SEN_vmsplice:
 477                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 478                         break;
 479                 case SEN_sendmsg:
 480                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 481                         break;
 482                 case SEN_sendmmsg:
 483                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 484                         break;
 485                 }
 486         }
 487 }
 488
 489 const char *
 490 err_name(unsigned long err)
 491 {
 492         if ((err < nerrnos) && errnoent[err])
 493                 return errnoent[err];
 494
 495         return NULL;
 496 }
 497
 498 static long get_regs(struct tcb *);
 499 static int get_syscall_args(struct tcb *);
 500 static int get_syscall_result(struct tcb *);
 501 static int arch_get_scno(struct tcb *tcp);
 502 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 503 static void get_error(struct tcb *, const bool);
 504 static int arch_set_error(struct tcb *);
 505 static int arch_set_success(struct tcb *);
 506
 507 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
 508
 509 static struct inject_opts *
 510 tcb_inject_opts(struct tcb *tcp)
 511 {
 512         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
 513                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
 514 }
 515
 516
 517 static long
 518 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
 519 {
 520         if (!tcp->inject_vec[current_personality]) {
 521                 tcp->inject_vec[current_personality] =
 522                         xcalloc(nsyscalls, sizeof(**inject_vec));
 523                 memcpy(tcp->inject_vec[current_personality],
 524                        inject_vec[current_personality],
 525                        nsyscalls * sizeof(**inject_vec));
 526         }
 527
 528         struct inject_opts *opts = tcb_inject_opts(tcp);
 529
 530         if (!opts || opts->first == 0)
 531                 return 0;
 532
 533         --opts->first;
 534
 535         if (opts->first != 0)
 536                 return 0;
 537
 538         opts->first = opts->step;
 539
 540         if (!recovering(tcp)) {
 541                 if (opts->data.flags & INJECT_F_SIGNAL)
 542                         *signo = opts->data.signo;
 543                 if (opts->data.flags & INJECT_F_RETVAL &&
 544                     !arch_set_scno(tcp, -1))
 545                         tcp->flags |= TCB_TAMPERED;
 546         }
 547
 548         return 0;
 549 }
 550
 551 static long
 552 tamper_with_syscall_exiting(struct tcb *tcp)
 553 {
 554         struct inject_opts *opts = tcb_inject_opts(tcp);
 555
 556         if (!opts)
 557                 return 0;
 558
 559         if (opts->data.rval >= 0) {
 560                 kernel_long_t u_rval = tcp->u_rval;
 561
 562                 tcp->u_rval = opts->data.rval;
 563                 if (arch_set_success(tcp)) {
 564                         tcp->u_rval = u_rval;
 565                 } else {
 566                         tcp->u_error = 0;
 567                 }
 568         } else {
 569                 unsigned long new_error = -opts->data.rval;
 570
 571                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
 572                         unsigned long u_error = tcp->u_error;
 573
 574                         tcp->u_error = new_error;
 575                         if (arch_set_error(tcp)) {
 576                                 tcp->u_error = u_error;
 577                         }
 578                 }
 579         }
 580
 581         return 0;
 582 }
 583
 584 /*
 585  * Returns:
 586  * 0: "ignore this ptrace stop", bail out silently.
 587  * 1: ok, decoded; call
 588  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
 589  * other: error; call syscall_entering_finish(tcp, res), where res is the value
 590  *    returned.
 591  */
 592 int
 593 syscall_entering_decode(struct tcb *tcp)
 594 {
 595         int res = get_scno(tcp);
 596         if (res == 0)
 597                 return res;
 598         int scno_good = res;
 599         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
 600                 printleader(tcp);
 601                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
 602                 /*
 603                  * " <unavailable>" will be added later by the code which
 604                  * detects ptrace errors.
 605                  */
 606                 return res;
 607         }
 608
 609 #if defined SYS_ipc_subcall     \
 610  || defined SYS_socket_subcall  \
 611  || defined SYS_syscall_subcall
 612         for (;;) {
 613                 switch (tcp->s_ent->sen) {
 614 # ifdef SYS_ipc_subcall
 615                 case SEN_ipc:
 616                         decode_ipc_subcall(tcp);
 617                         break;
 618 # endif
 619 # ifdef SYS_socket_subcall
 620                 case SEN_socketcall:
 621                         decode_socket_subcall(tcp);
 622                         break;
 623 # endif
 624 # ifdef SYS_syscall_subcall
 625                 case SEN_syscall:
 626                         decode_syscall_subcall(tcp);
 627                         if (tcp->s_ent->sen != SEN_syscall)
 628                                 continue;
 629                         break;
 630 # endif
 631                 }
 632                 break;
 633         }
 634 #endif
 635
 636         return 1;
 637 }
 638
 639 int
 640 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
 641 {
 642         /* Restrain from fault injection while the trace executes strace code. */
 643         if (hide_log(tcp)) {
 644                 tcp->qual_flg &= ~QUAL_INJECT;
 645         }
 646
 647         switch (tcp->s_ent->sen) {
 648                 case SEN_execve:
 649                 case SEN_execveat:
 650 #if defined SPARC || defined SPARC64
 651                 case SEN_execv:
 652 #endif
 653                         tcp->flags &= ~TCB_HIDE_LOG;
 654                         break;
 655         }
 656
 657         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
 658                 tcp->flags |= TCB_FILTERED;
 659                 return 0;
 660         }
 661
 662         tcp->flags &= ~TCB_FILTERED;
 663
 664         if (hide_log(tcp)) {
 665                 return 0;
 666         }
 667
 668         if (inject(tcp))
 669                 tamper_with_syscall_entering(tcp, sig);
 670
 671         if (cflag == CFLAG_ONLY_STATS) {
 672                 return 0;
 673         }
 674
 675 #ifdef USE_LIBUNWIND
 676         if (stack_trace_enabled) {
 677                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 678                         unwind_capture_stacktrace(tcp);
 679         }
 680 #endif
 681
 682         printleader(tcp);
 683         tprintf("%s(", tcp->s_ent->sys_name);
 684         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
 685         fflush(tcp->outf);
 686         return res;
 687 }
 688
 689 void
 690 syscall_entering_finish(struct tcb *tcp, int res)
 691 {
 692         tcp->flags |= TCB_INSYSCALL;
 693         tcp->sys_func_rval = res;
 694         /* Measure the entrance time as late as possible to avoid errors. */
 695         if ((Tflag || cflag) && !filtered(tcp))
 696                 gettimeofday(&tcp->etime, NULL);
 697 }
 698
 699 /* Returns:
 700  * 0: "bail out".
 701  * 1: ok.
 702  * -1: error in one of ptrace ops.
 703  *
 704  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
 705  *    value. Anyway, call syscall_exiting_finish(tcp) then.
 706  */
 707 int
 708 syscall_exiting_decode(struct tcb *tcp, struct timeval *ptv)
 709 {
 710         /* Measure the exit time as early as possible to avoid errors. */
 711         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
 712                 gettimeofday(ptv, NULL);
 713
 714 #ifdef USE_LIBUNWIND
 715         if (stack_trace_enabled) {
 716                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 717                         unwind_cache_invalidate(tcp);
 718         }
 719 #endif
 720
 721         if (filtered(tcp) || hide_log(tcp))
 722                 return 0;
 723
 724 #if SUPPORTED_PERSONALITIES > 1
 725         update_personality(tcp, tcp->currpers);
 726 #endif
 727
 728         return get_syscall_result(tcp);
 729 }
 730
 731 int
 732 syscall_exiting_trace(struct tcb *tcp, struct timeval tv, int res)
 733 {
 734         if (syserror(tcp) && syscall_tampered(tcp))
 735                 tamper_with_syscall_exiting(tcp);
 736
 737         if (cflag) {
 738                 count_syscall(tcp, &tv);
 739                 if (cflag == CFLAG_ONLY_STATS) {
 740                         return 0;
 741                 }
 742         }
 743
 744         /* If not in -ff mode, and printing_tcp != tcp,
 745          * then the log currently does not end with output
 746          * of _our syscall entry_, but with something else.
 747          * We need to say which syscall's return is this.
 748          *
 749          * Forced reprinting via TCB_REPRINT is used only by
 750          * "strace -ff -oLOG test/threaded_execve" corner case.
 751          * It's the only case when -ff mode needs reprinting.
 752          */
 753         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 754                 tcp->flags &= ~TCB_REPRINT;
 755                 printleader(tcp);
 756                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 757         }
 758         printing_tcp = tcp;
 759
 760         tcp->s_prev_ent = NULL;
 761         if (res != 1) {
 762                 /* There was error in one of prior ptrace ops */
 763                 tprints(") ");
 764                 tabto();
 765                 tprints("= ? <unavailable>\n");
 766                 line_ended();
 767                 return res;
 768         }
 769         tcp->s_prev_ent = tcp->s_ent;
 770
 771         int sys_res = 0;
 772         if (raw(tcp)) {
 773                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 774         } else {
 775         /* FIXME: not_failing_only (IOW, option -z) is broken:
 776          * failure of syscall is known only after syscall return.
 777          * Thus we end up with something like this on, say, ENOENT:
 778          *     open("does_not_exist", O_RDONLY <unfinished ...>
 779          *     {next syscall decode}
 780          * whereas the intended result is that open(...) line
 781          * is not shown at all.
 782          */
 783                 if (not_failing_only && tcp->u_error)
 784                         return 0;       /* ignore failed syscalls */
 785                 if (tcp->sys_func_rval & RVAL_DECODED)
 786                         sys_res = tcp->sys_func_rval;
 787                 else
 788                         sys_res = tcp->s_ent->sys_func(tcp);
 789         }
 790
 791         tprints(") ");
 792         tabto();
 793         unsigned long u_error = tcp->u_error;
 794         kernel_long_t u_rval;
 795
 796         if (raw(tcp)) {
 797                 if (u_error) {
 798                         tprintf("= -1 (errno %lu)", u_error);
 799                 } else {
 800                         tprintf("= %#" PRI_klx, tcp->u_rval);
 801                 }
 802                 if (syscall_tampered(tcp))
 803                         tprints(" (INJECTED)");
 804         } else if (!(sys_res & RVAL_NONE) && u_error) {
 805                 const char *u_error_str;
 806
 807                 switch (u_error) {
 808                 /* Blocked signals do not interrupt any syscalls.
 809                  * In this case syscalls don't return ERESTARTfoo codes.
 810                  *
 811                  * Deadly signals set to SIG_DFL interrupt syscalls
 812                  * and kill the process regardless of which of the codes below
 813                  * is returned by the interrupted syscall.
 814                  * In some cases, kernel forces a kernel-generated deadly
 815                  * signal to be unblocked and set to SIG_DFL (and thus cause
 816                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 817                  * or SIGILL. (The alternative is to leave process spinning
 818                  * forever on the faulty instruction - not useful).
 819                  *
 820                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 821                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 822                  * but kernel will always restart them.
 823                  */
 824                 case ERESTARTSYS:
 825                         /* Most common type of signal-interrupted syscall exit code.
 826                          * The system call will be restarted with the same arguments
 827                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 828                          */
 829                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 830                         break;
 831                 case ERESTARTNOINTR:
 832                         /* Rare. For example, fork() returns this if interrupted.
 833                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 834                          */
 835                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 836                         break;
 837                 case ERESTARTNOHAND:
 838                         /* pause(), rt_sigsuspend() etc use this code.
 839                          * SA_RESTART is ignored (assumed not set):
 840                          * syscall won't restart (will return EINTR instead)
 841                          * even after signal with SA_RESTART set. However,
 842                          * after SIG_IGN or SIG_DFL signal it will restart
 843                          * (thus the name "restart only if has no handler").
 844                          */
 845                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 846                         break;
 847                 case ERESTART_RESTARTBLOCK:
 848                         /* Syscalls like nanosleep(), poll() which can't be
 849                          * restarted with their original arguments use this
 850                          * code. Kernel will execute restart_syscall() instead,
 851                          * which changes arguments before restarting syscall.
 852                          * SA_RESTART is ignored (assumed not set) similarly
 853                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 854                          * since restart data is saved in "restart block"
 855                          * in task struct, and if signal handler uses a syscall
 856                          * which in turn saves another such restart block,
 857                          * old data is lost and restart becomes impossible)
 858                          */
 859                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 860                         break;
 861                 default:
 862                         u_rval = sys_res & RVAL_PRINT_ERR_VAL ?
 863                                  tcp->u_rval : -1;
 864                         u_error_str = err_name(u_error);
 865                         if (u_error_str)
 866                                 tprintf("= %" PRI_kld " %s (%s)",
 867                                         u_rval, u_error_str, strerror(u_error));
 868                         else
 869                                 tprintf("= %" PRI_kld " %lu (%s)",
 870                                         u_rval, u_error, strerror(u_error));
 871                         break;
 872                 }
 873                 if (syscall_tampered(tcp))
 874                         tprints(" (INJECTED)");
 875                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 876                         tprintf(" (%s)", tcp->auxstr);
 877         } else {
 878                 if (sys_res & RVAL_NONE)
 879                         tprints("= ?");
 880                 else {
 881                         switch (sys_res & RVAL_MASK) {
 882                         case RVAL_HEX:
 883 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 884                                 if (current_klongsize < sizeof(tcp->u_rval)) {
 885                                         tprintf("= %#x",
 886                                                 (unsigned int) tcp->u_rval);
 887                                 } else
 888 #endif
 889                                 {
 890                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 891                                 }
 892                                 break;
 893                         case RVAL_OCTAL:
 894                                 tprints("= ");
 895                                 print_numeric_long_umask(tcp->u_rval);
 896                                 break;
 897                         case RVAL_UDECIMAL:
 898 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 899                                 if (current_klongsize < sizeof(tcp->u_rval)) {
 900                                         tprintf("= %u",
 901                                                 (unsigned int) tcp->u_rval);
 902                                 } else
 903 #endif
 904                                 {
 905                                         tprintf("= %" PRI_klu, tcp->u_rval);
 906                                 }
 907                                 break;
 908                         case RVAL_DECIMAL:
 909                                 tprintf("= %" PRI_kld, tcp->u_rval);
 910                                 break;
 911                         case RVAL_FD:
 912                                 if (show_fd_path) {
 913                                         tprints("= ");
 914                                         printfd(tcp, tcp->u_rval);
 915                                 } else
 916                                         tprintf("= %" PRI_kld, tcp->u_rval);
 917                                 break;
 918                         default:
 919                                 error_msg("invalid rval format");
 920                                 break;
 921                         }
 922                 }
 923                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 924                         tprintf(" (%s)", tcp->auxstr);
 925                 if (syscall_tampered(tcp))
 926                         tprints(" (INJECTED)");
 927         }
 928         if (Tflag) {
 929                 tv_sub(&tv, &tv, &tcp->etime);
 930                 tprintf(" <%ld.%06ld>",
 931                         (long) tv.tv_sec, (long) tv.tv_usec);
 932         }
 933         tprints("\n");
 934         dumpio(tcp);
 935         line_ended();
 936
 937 #ifdef USE_LIBUNWIND
 938         if (stack_trace_enabled)
 939                 unwind_print_stacktrace(tcp);
 940 #endif
 941         return 0;
 942 }
 943
 944 void
 945 syscall_exiting_finish(struct tcb *tcp)
 946 {
 947         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
 948         tcp->sys_func_rval = 0;
 949         free_tcb_priv_data(tcp);
 950 }
 951
 952 bool
 953 is_erestart(struct tcb *tcp)
 954 {
 955         switch (tcp->u_error) {
 956                 case ERESTARTSYS:
 957                 case ERESTARTNOINTR:
 958                 case ERESTARTNOHAND:
 959                 case ERESTART_RESTARTBLOCK:
 960                         return true;
 961                 default:
 962                         return false;
 963         }
 964 }
 965
 966 static unsigned long saved_u_error;
 967
 968 void
 969 temporarily_clear_syserror(struct tcb *tcp)
 970 {
 971         saved_u_error = tcp->u_error;
 972         tcp->u_error = 0;
 973 }
 974
 975 void
 976 restore_cleared_syserror(struct tcb *tcp)
 977 {
 978         tcp->u_error = saved_u_error;
 979 }
 980
 981 #include "arch_regs.c"
 982
 983 #if HAVE_ARCH_GETRVAL2
 984 # include "arch_getrval2.c"
 985 #endif
 986
 987 void
 988 print_pc(struct tcb *tcp)
 989 {
 990 #if defined ARCH_PC_REG
 991 # define ARCH_GET_PC 0
 992 #elif defined ARCH_PC_PEEK_ADDR
 993         kernel_ulong_t pc;
 994 # define ARCH_PC_REG pc
 995 # define ARCH_GET_PC upeek(tcp, ARCH_PC_PEEK_ADDR, &pc)
 996 #else
 997 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
 998 #endif
 999         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1000                 tprints(current_wordsize == 4 ? "[????????] "
1001                                               : "[????????????????] ");
1002         else
1003                 tprintf(current_wordsize == 4
1004                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1005                         (kernel_ulong_t) ARCH_PC_REG);
1006 }
1007
1008 #include "getregs_old.h"
1009
1010 #undef ptrace_getregset_or_getregs
1011 #undef ptrace_setregset_or_setregs
1012 #ifdef ARCH_REGS_FOR_GETREGSET
1013
1014 # define ptrace_getregset_or_getregs ptrace_getregset
1015 static long
1016 ptrace_getregset(pid_t pid)
1017 {
1018 # ifdef ARCH_IOVEC_FOR_GETREGSET
1019         /* variable iovec */
1020         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1021         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1022                       &ARCH_IOVEC_FOR_GETREGSET);
1023 # else
1024         /* constant iovec */
1025         static struct iovec io = {
1026                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1027                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1028         };
1029         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1030
1031 # endif
1032 }
1033
1034 # ifndef HAVE_GETREGS_OLD
1035 #  define ptrace_setregset_or_setregs ptrace_setregset
1036 static int
1037 ptrace_setregset(pid_t pid)
1038 {
1039 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1040         /* variable iovec */
1041         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1042                       &ARCH_IOVEC_FOR_GETREGSET);
1043 #  else
1044         /* constant iovec */
1045         static struct iovec io = {
1046                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1047                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1048         };
1049         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1050 #  endif
1051 }
1052 # endif /* !HAVE_GETREGS_OLD */
1053
1054 #elif defined ARCH_REGS_FOR_GETREGS
1055
1056 # define ptrace_getregset_or_getregs ptrace_getregs
1057 static long
1058 ptrace_getregs(pid_t pid)
1059 {
1060 # if defined SPARC || defined SPARC64
1061         /* SPARC systems have the meaning of data and addr reversed */
1062         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1063 # else
1064         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1065 # endif
1066 }
1067
1068 # ifndef HAVE_GETREGS_OLD
1069 #  define ptrace_setregset_or_setregs ptrace_setregs
1070 static int
1071 ptrace_setregs(pid_t pid)
1072 {
1073 #  if defined SPARC || defined SPARC64
1074         /* SPARC systems have the meaning of data and addr reversed */
1075         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1076 #  else
1077         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1078 #  endif
1079 }
1080 # endif /* !HAVE_GETREGS_OLD */
1081
1082 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1083
1084 #ifdef ptrace_getregset_or_getregs
1085 static long get_regs_error;
1086 #endif
1087
1088 void
1089 clear_regs(struct tcb *tcp)
1090 {
1091 #ifdef ptrace_getregset_or_getregs
1092         get_regs_error = -1;
1093 #endif
1094 }
1095
1096 static long
1097 get_regs(struct tcb *const tcp)
1098 {
1099 #ifdef ptrace_getregset_or_getregs
1100
1101         if (get_regs_error != -1)
1102                 return get_regs_error;
1103
1104 # ifdef HAVE_GETREGS_OLD
1105         /*
1106          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1107          * fallback to getregs_old.
1108          */
1109         static int use_getregs_old;
1110         if (use_getregs_old < 0) {
1111                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1112         } else if (use_getregs_old == 0) {
1113                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1114                 if (get_regs_error >= 0) {
1115                         use_getregs_old = -1;
1116                         return get_regs_error;
1117                 }
1118                 if (errno == EPERM || errno == ESRCH)
1119                         return get_regs_error;
1120                 use_getregs_old = 1;
1121         }
1122         return get_regs_error = getregs_old(tcp);
1123 # else /* !HAVE_GETREGS_OLD */
1124         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1125         get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1126
1127 #  if defined ARCH_PERSONALITY_0_IOV_SIZE
1128         if (get_regs_error)
1129                 return get_regs_error;
1130
1131         switch (ARCH_IOVEC_FOR_GETREGSET.iov_len) {
1132         case ARCH_PERSONALITY_0_IOV_SIZE:
1133                 update_personality(tcp, 0);
1134                 break;
1135         case ARCH_PERSONALITY_1_IOV_SIZE:
1136                 update_personality(tcp, 1);
1137                 break;
1138         default: {
1139                 static bool printed = false;
1140
1141                 if (!printed) {
1142                         error_msg("Unsupported regset size returned by "
1143                                   "PTRACE_GETREGSET: %zu",
1144                                   ARCH_IOVEC_FOR_GETREGSET.iov_len);
1145
1146                         printed = true;
1147                 }
1148
1149                 update_personality(tcp, 0);
1150         }
1151         }
1152 #  endif /* ARCH_PERSONALITY_0_IOV_SIZE */
1153
1154         return get_regs_error;
1155
1156 # endif /* !HAVE_GETREGS_OLD */
1157
1158 #else /* !ptrace_getregset_or_getregs */
1159
1160 # warning get_regs is not implemented for this architecture yet
1161         return 0;
1162
1163 #endif /* !ptrace_getregset_or_getregs */
1164 }
1165
1166 #ifdef ptrace_setregset_or_setregs
1167 static int
1168 set_regs(pid_t pid)
1169 {
1170         return ptrace_setregset_or_setregs(pid);
1171 }
1172 #endif /* ptrace_setregset_or_setregs */
1173
1174 struct sysent_buf {
1175         struct tcb *tcp;
1176         struct_sysent ent;
1177         char buf[sizeof("syscall_0x") + sizeof(kernel_ulong_t) * 2];
1178 };
1179
1180 static void
1181 free_sysent_buf(void *ptr)
1182 {
1183         struct sysent_buf *s = ptr;
1184         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1185         free(ptr);
1186 }
1187
1188 /*
1189  * Returns:
1190  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1191  *    out silently" code.
1192  * 1: ok, continue in syscall_entering_decode().
1193  * other: error, syscall_entering_decode() should print error indicator
1194  *    ("????" etc) and return an appropriate code.
1195  */
1196 int
1197 get_scno(struct tcb *tcp)
1198 {
1199         if (get_regs(tcp) < 0)
1200                 return -1;
1201
1202         int rc = arch_get_scno(tcp);
1203         if (rc != 1)
1204                 return rc;
1205
1206         tcp->scno = shuffle_scno(tcp->scno);
1207
1208         if (scno_is_valid(tcp->scno)) {
1209                 tcp->s_ent = &sysent[tcp->scno];
1210                 tcp->qual_flg = qual_flags(tcp->scno);
1211         } else {
1212                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1213
1214                 s->tcp = tcp;
1215                 s->ent.nargs = MAX_ARGS;
1216                 s->ent.sen = SEN_printargs;
1217                 s->ent.sys_func = printargs;
1218                 s->ent.sys_name = s->buf;
1219                 xsprintf(s->buf, "syscall_%#" PRI_klx, shuffle_scno(tcp->scno));
1220
1221                 tcp->s_ent = &s->ent;
1222                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1223
1224                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1225
1226                 debug_msg("pid %d invalid syscall %#" PRI_klx,
1227                           tcp->pid, shuffle_scno(tcp->scno));
1228         }
1229
1230         /*
1231          * We refrain from argument decoding during recovering
1232          * as tracee memory mappings has changed and the registers
1233          * are very likely pointing to garbage already.
1234          */
1235         if (recovering(tcp))
1236                 tcp->qual_flg |= QUAL_RAW;
1237
1238         return 1;
1239 }
1240
1241 #ifdef ptrace_getregset_or_getregs
1242 # define get_syscall_result_regs get_regs
1243 #else
1244 static int get_syscall_result_regs(struct tcb *);
1245 #endif
1246
1247 /* Returns:
1248  * 1: ok, continue in syscall_exiting_trace().
1249  * -1: error, syscall_exiting_trace() should print error indicator
1250  *    ("????" etc) and bail out.
1251  */
1252 static int
1253 get_syscall_result(struct tcb *tcp)
1254 {
1255         if (get_syscall_result_regs(tcp) < 0)
1256                 return -1;
1257         tcp->u_error = 0;
1258         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1259
1260         return 1;
1261 }
1262
1263 #include "get_scno.c"
1264 #include "set_scno.c"
1265 #include "get_syscall_args.c"
1266 #ifndef ptrace_getregset_or_getregs
1267 # include "get_syscall_result.c"
1268 #endif
1269 #include "get_error.c"
1270 #include "set_error.c"
1271 #ifdef HAVE_GETREGS_OLD
1272 # include "getregs_old.c"
1273 #endif
1274 #include "shuffle_scno.c"
1275
1276 const char *
1277 syscall_name(kernel_ulong_t scno)
1278 {
1279         scno = shuffle_scno(scno);
1280
1281         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1282 }