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) 1999-2019 The strace developers.
  10  * All rights reserved.
  11  *
  12  * SPDX-License-Identifier: LGPL-2.1-or-later
  13  */
  14
  15 #include "defs.h"
  16 #include "get_personality.h"
  17 #include "mmap_notify.h"
  18 #include "native_defs.h"
  19 #include "ptrace.h"
  20 #include "ptrace_syscall_info.h"
  21 #include "nsig.h"
  22 #include "number_set.h"
  23 #include "delay.h"
  24 #include "retval.h"
  25 #include <limits.h>
  26
  27 /* for struct iovec */
  28 #include <sys/uio.h>
  29
  30 /* for __X32_SYSCALL_BIT */
  31 #include "scno.h"
  32
  33 #include "regs.h"
  34
  35 #if defined(SPARC64)
  36 # undef PTRACE_GETREGS
  37 # define PTRACE_GETREGS PTRACE_GETREGS64
  38 # undef PTRACE_SETREGS
  39 # define PTRACE_SETREGS PTRACE_SETREGS64
  40 #endif
  41
  42 #include "syscall.h"
  43 #include "xstring.h"
  44 #include "syscallent_base_nr.h"
  45
  46 /* Define these shorthand notations to simplify the syscallent files. */
  47 #include "sysent_shorthand_defs.h"
  48
  49 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  50
  51 const struct_sysent sysent0[] = {
  52 #include "syscallent.h"
  53 };
  54
  55 #if SUPPORTED_PERSONALITIES > 1
  56 # include PERSONALITY1_INCLUDE_FUNCS
  57 static const struct_sysent sysent1[] = {
  58 # include "syscallent1.h"
  59 };
  60 #endif
  61
  62 #if SUPPORTED_PERSONALITIES > 2
  63 # include PERSONALITY2_INCLUDE_FUNCS
  64 static const struct_sysent sysent2[] = {
  65 # include "syscallent2.h"
  66 };
  67 #endif
  68
  69 /* Now undef them since short defines cause wicked namespace pollution. */
  70 #include "sysent_shorthand_undefs.h"
  71
  72 const char *const errnoent[] = {
  73 #include "errnoent.h"
  74 };
  75 const char *const signalent[] = {
  76 #include "signalent.h"
  77
  78 };
  79 /*
  80  * `ioctlent[012].h' files are automatically generated by the auxiliary
  81  * program `ioctlsort', such that the list is sorted by the `code' field.
  82  * This has the side-effect of resolving the _IO.. macros into
  83  * plain integers, eliminating the need to include here everything
  84  * in "/usr/include".
  85  */
  86
  87 const struct_ioctlent ioctlent0[] = {
  88 #include "ioctlent0.h"
  89 };
  90
  91 #if SUPPORTED_PERSONALITIES > 1
  92 static const struct_ioctlent ioctlent1[] = {
  93 # include "ioctlent1.h"
  94 };
  95 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
  96 static const struct_printers printers0 = {
  97 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
  98 };
  99 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 100 static const struct_printers printers1 = {
 101 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 102 };
 103 #endif
 104
 105 #if SUPPORTED_PERSONALITIES > 2
 106 static const struct_ioctlent ioctlent2[] = {
 107 # include "ioctlent2.h"
 108 };
 109 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 110 static const struct_printers printers2 = {
 111 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 112 };
 113 #endif
 114
 115 enum {
 116         nsyscalls0 = ARRAY_SIZE(sysent0)
 117 #if SUPPORTED_PERSONALITIES > 1
 118         , nsyscalls1 = ARRAY_SIZE(sysent1)
 119 # if SUPPORTED_PERSONALITIES > 2
 120         , nsyscalls2 = ARRAY_SIZE(sysent2)
 121 # endif
 122 #endif
 123 };
 124
 125 enum {
 126         nioctlents0 = ARRAY_SIZE(ioctlent0)
 127 #if SUPPORTED_PERSONALITIES > 1
 128         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 129 # if SUPPORTED_PERSONALITIES > 2
 130         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 131 # endif
 132 #endif
 133 };
 134
 135 #if SUPPORTED_PERSONALITIES > 1
 136 const struct_sysent *sysent = sysent0;
 137 const struct_ioctlent *ioctlent = ioctlent0;
 138 const struct_printers *printers = &printers0;
 139 #endif
 140
 141 const unsigned int nerrnos = ARRAY_SIZE(errnoent);
 142 const unsigned int nsignals = ARRAY_SIZE(signalent);
 143 unsigned nsyscalls = nsyscalls0;
 144 unsigned nioctlents = nioctlents0;
 145
 146 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 147         nsyscalls0,
 148 #if SUPPORTED_PERSONALITIES > 1
 149         nsyscalls1,
 150 #endif
 151 #if SUPPORTED_PERSONALITIES > 2
 152         nsyscalls2,
 153 #endif
 154 };
 155 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 156         sysent0,
 157 #if SUPPORTED_PERSONALITIES > 1
 158         sysent1,
 159 #endif
 160 #if SUPPORTED_PERSONALITIES > 2
 161         sysent2,
 162 #endif
 163 };
 164
 165 const char *const personality_names[] =
 166 #if defined X86_64
 167         {"64 bit", "32 bit", "x32"}
 168 #elif defined X32
 169         {"x32", "32 bit"}
 170 #elif SUPPORTED_PERSONALITIES == 2
 171         {"64 bit", "32 bit"}
 172 #else
 173         {STRINGIFY_VAL(__WORDSIZE) " bit"}
 174 #endif
 175         ;
 176
 177 const char *const personality_designators[] =
 178 #if defined X86_64
 179         { "64", "32", "x32" }
 180 #elif defined X32
 181         { "x32", "32" }
 182 #elif SUPPORTED_PERSONALITIES == 2
 183         { "64", "32" }
 184 #else
 185         { STRINGIFY_VAL(__WORDSIZE) }
 186 #endif
 187         ;
 188
 189 #if SUPPORTED_PERSONALITIES > 1
 190
 191 unsigned current_personality;
 192
 193 # ifndef current_wordsize
 194 unsigned current_wordsize = PERSONALITY0_WORDSIZE;
 195 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 196         PERSONALITY0_WORDSIZE,
 197         PERSONALITY1_WORDSIZE,
 198 #  if SUPPORTED_PERSONALITIES > 2
 199         PERSONALITY2_WORDSIZE,
 200 #  endif
 201 };
 202 # endif
 203
 204 # ifndef current_klongsize
 205 unsigned current_klongsize = PERSONALITY0_KLONGSIZE;
 206 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 207         PERSONALITY0_KLONGSIZE,
 208         PERSONALITY1_KLONGSIZE,
 209 #  if SUPPORTED_PERSONALITIES > 2
 210         PERSONALITY2_KLONGSIZE,
 211 #  endif
 212 };
 213 # endif
 214
 215 void
 216 set_personality(unsigned int personality)
 217 {
 218         if (personality == current_personality)
 219                 return;
 220
 221         if (personality >= SUPPORTED_PERSONALITIES)
 222                 error_msg_and_die("Requested switch to unsupported personality "
 223                                   "%u", personality);
 224
 225         nsyscalls = nsyscall_vec[personality];
 226         sysent = sysent_vec[personality];
 227
 228         switch (personality) {
 229         case 0:
 230                 ioctlent = ioctlent0;
 231                 nioctlents = nioctlents0;
 232                 printers = &printers0;
 233                 break;
 234
 235         case 1:
 236                 ioctlent = ioctlent1;
 237                 nioctlents = nioctlents1;
 238                 printers = &printers1;
 239                 break;
 240
 241 # if SUPPORTED_PERSONALITIES > 2
 242         case 2:
 243                 ioctlent = ioctlent2;
 244                 nioctlents = nioctlents2;
 245                 printers = &printers2;
 246                 break;
 247 # endif
 248         }
 249
 250         current_personality = personality;
 251 # ifndef current_wordsize
 252         current_wordsize = personality_wordsize[personality];
 253 # endif
 254 # ifndef current_klongsize
 255         current_klongsize = personality_klongsize[personality];
 256 # endif
 257 }
 258
 259 static void
 260 update_personality(struct tcb *tcp, unsigned int personality)
 261 {
 262         static bool need_mpers_warning[] =
 263                 { false, !HAVE_PERSONALITY_1_MPERS, !HAVE_PERSONALITY_2_MPERS };
 264
 265         set_personality(personality);
 266
 267         if (personality == tcp->currpers)
 268                 return;
 269         tcp->currpers = personality;
 270
 271         if (!qflag) {
 272                 error_msg("[ Process PID=%d runs in %s mode. ]",
 273                           tcp->pid, personality_names[personality]);
 274         }
 275
 276         if (need_mpers_warning[personality]) {
 277                 error_msg("WARNING: Proper structure decoding for this "
 278                           "personality is not supported, please consider "
 279                           "building strace with mpers support enabled.");
 280                 need_mpers_warning[personality] = false;
 281         }
 282 }
 283 #endif
 284
 285 #ifdef SYS_socket_subcall
 286 static void
 287 decode_socket_subcall(struct tcb *tcp)
 288 {
 289         const int call = tcp->u_arg[0];
 290
 291         if (call < 1 || call >= SYS_socket_nsubcalls)
 292                 return;
 293
 294         const kernel_ulong_t scno = SYS_socket_subcall + call;
 295         const unsigned int nargs = sysent[scno].nargs;
 296         uint64_t buf[nargs];
 297
 298         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 299                 return;
 300
 301         tcp->scno = scno;
 302         tcp->qual_flg = qual_flags(scno);
 303         tcp->s_ent = &sysent[scno];
 304
 305         unsigned int i;
 306         for (i = 0; i < nargs; ++i)
 307                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 308                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 309 }
 310 #endif /* SYS_socket_subcall */
 311
 312 #ifdef SYS_ipc_subcall
 313 static void
 314 decode_ipc_subcall(struct tcb *tcp)
 315 {
 316         unsigned int call = tcp->u_arg[0];
 317         const unsigned int version = call >> 16;
 318
 319         if (version) {
 320 # if defined S390 || defined S390X
 321                 return;
 322 # else
 323 #  ifdef SPARC64
 324                 if (current_wordsize == 8)
 325                         return;
 326 #  endif
 327                 set_tcb_priv_ulong(tcp, version);
 328                 call &= 0xffff;
 329 # endif
 330         }
 331
 332         switch (call) {
 333                 case  1: case  2: case  3: case  4:
 334                 case 11: case 12: case 13: case 14:
 335                 case 21: case 22: case 23: case 24:
 336                         break;
 337                 default:
 338                         return;
 339         }
 340
 341         tcp->scno = SYS_ipc_subcall + call;
 342         tcp->qual_flg = qual_flags(tcp->scno);
 343         tcp->s_ent = &sysent[tcp->scno];
 344
 345         const unsigned int n = n_args(tcp);
 346         unsigned int i;
 347         for (i = 0; i < n; i++)
 348                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 349 }
 350 #endif /* SYS_ipc_subcall */
 351
 352 #ifdef SYS_syscall_subcall
 353 /* The implementation is architecture specific.  */
 354 static void decode_syscall_subcall(struct tcb *);
 355 #endif /* SYS_syscall_subcall */
 356
 357 static void
 358 dumpio(struct tcb *tcp)
 359 {
 360         int fd = tcp->u_arg[0];
 361         if (fd < 0)
 362                 return;
 363
 364         if (is_number_in_set(fd, write_set)) {
 365                 switch (tcp_sysent(tcp)->sen) {
 366                 case SEN_write:
 367                 case SEN_pwrite:
 368                 case SEN_send:
 369                 case SEN_sendto:
 370                 case SEN_mq_timedsend_time32:
 371                 case SEN_mq_timedsend_time64:
 372                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 373                         break;
 374                 case SEN_writev:
 375                 case SEN_pwritev:
 376                 case SEN_pwritev2:
 377                 case SEN_vmsplice:
 378                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 379                         break;
 380                 case SEN_sendmsg:
 381                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 382                         break;
 383                 case SEN_sendmmsg:
 384                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 385                         break;
 386                 }
 387         }
 388
 389         if (syserror(tcp))
 390                 return;
 391
 392         if (is_number_in_set(fd, read_set)) {
 393                 switch (tcp_sysent(tcp)->sen) {
 394                 case SEN_read:
 395                 case SEN_pread:
 396                 case SEN_recv:
 397                 case SEN_recvfrom:
 398                 case SEN_mq_timedreceive_time32:
 399                 case SEN_mq_timedreceive_time64:
 400                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 401                         return;
 402                 case SEN_readv:
 403                 case SEN_preadv:
 404                 case SEN_preadv2:
 405                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 406                                      tcp->u_rval);
 407                         return;
 408                 case SEN_recvmsg:
 409                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 410                         return;
 411                 case SEN_recvmmsg:
 412                 case SEN_recvmmsg_time32:
 413                 case SEN_recvmmsg_time64:
 414                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 415                         return;
 416                 }
 417         }
 418 }
 419
 420 static const char *
 421 err_name(uint64_t err)
 422 {
 423         return err < nerrnos ? errnoent[err] : NULL;
 424 }
 425
 426 void
 427 print_err(int64_t err, bool negated)
 428 {
 429         const char *str = err_name(negated ? -err : err);
 430
 431         if (!str || xlat_verbose(xlat_verbosity) != XLAT_STYLE_ABBREV)
 432                 tprintf(negated ? "%" PRId64 : "%" PRIu64, err);
 433         if (!str || xlat_verbose(xlat_verbosity) == XLAT_STYLE_RAW)
 434                 return;
 435         (xlat_verbose(xlat_verbosity) == XLAT_STYLE_ABBREV
 436                 ? tprintf : tprintf_comment)("%s%s",
 437                                              negated ? "-" : "", str);
 438 }
 439
 440 static void
 441 print_err_ret(kernel_ulong_t ret, unsigned long u_error)
 442 {
 443         const char *u_error_str = err_name(u_error);
 444
 445         if (u_error_str)
 446                 tprintf("= %" PRI_kld " %s (%s)",
 447                         ret, u_error_str, strerror(u_error));
 448         else
 449                 tprintf("= %" PRI_kld " (errno %lu)", ret, u_error);
 450 }
 451
 452 static long get_regs(struct tcb *);
 453 static int get_syscall_args(struct tcb *);
 454 static int get_syscall_result(struct tcb *);
 455 static void get_error(struct tcb *, bool);
 456 static void set_error(struct tcb *, unsigned long);
 457 static void set_success(struct tcb *, kernel_long_t);
 458 static int arch_get_scno(struct tcb *);
 459 static int arch_check_scno(struct tcb *);
 460 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 461 static int arch_get_syscall_args(struct tcb *);
 462 static void arch_get_error(struct tcb *, bool);
 463 static int arch_set_error(struct tcb *);
 464 static int arch_set_success(struct tcb *);
 465 #if MAX_ARGS > 6
 466 static void arch_get_syscall_args_extra(struct tcb *, unsigned int);
 467 #endif
 468
 469 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
 470
 471 static struct inject_opts *
 472 tcb_inject_opts(struct tcb *tcp)
 473 {
 474         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
 475                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
 476 }
 477
 478
 479 static long
 480 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
 481 {
 482         if (!tcp->inject_vec[current_personality]) {
 483                 tcp->inject_vec[current_personality] =
 484                         xcalloc(nsyscalls, sizeof(**inject_vec));
 485                 memcpy(tcp->inject_vec[current_personality],
 486                        inject_vec[current_personality],
 487                        nsyscalls * sizeof(**inject_vec));
 488         }
 489
 490         struct inject_opts *opts = tcb_inject_opts(tcp);
 491
 492         if (!opts || opts->first == 0)
 493                 return 0;
 494
 495         --opts->first;
 496
 497         if (opts->first != 0)
 498                 return 0;
 499
 500         opts->first = opts->step;
 501
 502         if (!recovering(tcp)) {
 503                 if (opts->data.flags & INJECT_F_SIGNAL)
 504                         *signo = opts->data.signo;
 505                 if (opts->data.flags & (INJECT_F_ERROR | INJECT_F_RETVAL)) {
 506                         kernel_long_t scno =
 507                                 (opts->data.flags & INJECT_F_SYSCALL)
 508                                 ? (kernel_long_t) shuffle_scno(opts->data.scno)
 509                                 : -1;
 510
 511                         if (!arch_set_scno(tcp, scno)) {
 512                                 tcp->flags |= TCB_TAMPERED;
 513                                 if (scno != -1)
 514                                         tcp->flags |= TCB_TAMPERED_NO_FAIL;
 515 #if ARCH_NEEDS_SET_ERROR_FOR_SCNO_TAMPERING
 516                                 /*
 517                                  * So far it's just a workaround for hppa,
 518                                  * but let's pretend it could be used elsewhere.
 519                                  */
 520                                 else {
 521                                         kernel_long_t rval =
 522                                                 (opts->data.flags & INJECT_F_RETVAL) ?
 523                                                 ENOSYS : retval_get(opts->data.rval_idx);
 524
 525                                         tcp->u_error = 0; /* force reset */
 526                                         set_error(tcp, rval);
 527                                 }
 528 #endif
 529                         }
 530                 }
 531                 if (opts->data.flags & INJECT_F_DELAY_ENTER)
 532                         delay_tcb(tcp, opts->data.delay_idx, true);
 533                 if (opts->data.flags & INJECT_F_DELAY_EXIT)
 534                         tcp->flags |= TCB_INJECT_DELAY_EXIT;
 535         }
 536
 537         return 0;
 538 }
 539
 540 static long
 541 tamper_with_syscall_exiting(struct tcb *tcp)
 542 {
 543         struct inject_opts *opts = tcb_inject_opts(tcp);
 544         if (!opts)
 545                 return 0;
 546
 547         if (inject_delay_exit(tcp))
 548                 delay_tcb(tcp, opts->data.delay_idx, false);
 549
 550         if (!syscall_tampered(tcp))
 551                 return 0;
 552
 553         if (!syserror(tcp) ^ !!syscall_tampered_nofail(tcp)) {
 554                 error_msg("Failed to tamper with process %d: unexpectedly got"
 555                           " %serror (return value %#" PRI_klx ", error %lu)",
 556                           tcp->pid, syscall_tampered_nofail(tcp) ? "" : "no ",
 557                           tcp->u_rval, tcp->u_error);
 558
 559                 return 1;
 560         }
 561
 562         if (opts->data.flags & INJECT_F_RETVAL)
 563                 set_success(tcp, retval_get(opts->data.rval_idx));
 564         else
 565                 set_error(tcp, retval_get(opts->data.rval_idx));
 566
 567         return 0;
 568 }
 569
 570 /*
 571  * Returns:
 572  * 0: "ignore this ptrace stop", bail out silently.
 573  * 1: ok, decoded; call
 574  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
 575  * other: error; call syscall_entering_finish(tcp, res), where res is the value
 576  *    returned.
 577  */
 578 int
 579 syscall_entering_decode(struct tcb *tcp)
 580 {
 581         int res = get_scno(tcp);
 582         if (res == 0)
 583                 return res;
 584         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
 585                 printleader(tcp);
 586                 tprintf("%s(", tcp_sysent(tcp)->sys_name);
 587                 /*
 588                  * " <unavailable>" will be added later by the code which
 589                  * detects ptrace errors.
 590                  */
 591                 return res;
 592         }
 593
 594 #ifdef SYS_syscall_subcall
 595         if (tcp_sysent(tcp)->sen == SEN_syscall)
 596                 decode_syscall_subcall(tcp);
 597 #endif
 598 #if defined SYS_ipc_subcall     \
 599  || defined SYS_socket_subcall
 600         switch (tcp_sysent(tcp)->sen) {
 601 # ifdef SYS_ipc_subcall
 602                 case SEN_ipc:
 603                         decode_ipc_subcall(tcp);
 604                         break;
 605 # endif
 606 # ifdef SYS_socket_subcall
 607                 case SEN_socketcall:
 608                         decode_socket_subcall(tcp);
 609                         break;
 610 # endif
 611         }
 612 #endif
 613
 614         return 1;
 615 }
 616
 617 int
 618 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
 619 {
 620         if (hide_log(tcp)) {
 621                 /*
 622                  * Restrain from fault injection
 623                  * while the trace executes strace code.
 624                  */
 625                 tcp->qual_flg &= ~QUAL_INJECT;
 626
 627                 switch (tcp_sysent(tcp)->sen) {
 628                         case SEN_execve:
 629                         case SEN_execveat:
 630                         case SEN_execv:
 631                                 /*
 632                                  * First exec* syscall makes the log visible.
 633                                  */
 634                                 tcp->flags &= ~TCB_HIDE_LOG;
 635                                 /*
 636                                  * Check whether this exec* syscall succeeds.
 637                                  */
 638                                 tcp->flags |= TCB_CHECK_EXEC_SYSCALL;
 639                                 break;
 640                 }
 641         }
 642
 643         if (hide_log(tcp) || !traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
 644                 tcp->flags |= TCB_FILTERED;
 645                 return 0;
 646         }
 647
 648         tcp->flags &= ~TCB_FILTERED;
 649
 650         if (inject(tcp))
 651                 tamper_with_syscall_entering(tcp, sig);
 652
 653         if (cflag == CFLAG_ONLY_STATS) {
 654                 return 0;
 655         }
 656
 657 #ifdef ENABLE_STACKTRACE
 658         if (stack_trace_enabled) {
 659                 if (tcp_sysent(tcp)->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 660                         unwind_tcb_capture(tcp);
 661         }
 662 #endif
 663
 664         if (!is_complete_set(status_set, NUMBER_OF_STATUSES))
 665                 strace_open_memstream(tcp);
 666
 667         printleader(tcp);
 668         tprintf("%s(", tcp_sysent(tcp)->sys_name);
 669         int res = raw(tcp) ? printargs(tcp) : tcp_sysent(tcp)->sys_func(tcp);
 670         fflush(tcp->outf);
 671         return res;
 672 }
 673
 674 void
 675 syscall_entering_finish(struct tcb *tcp, int res)
 676 {
 677         tcp->flags |= TCB_INSYSCALL;
 678         tcp->sys_func_rval = res;
 679
 680         /* Measure the entrance time as late as possible to avoid errors. */
 681         if ((Tflag || cflag) && !filtered(tcp))
 682                 clock_gettime(CLOCK_MONOTONIC, &tcp->etime);
 683
 684         /* Start tracking system time */
 685         if (cflag) {
 686                 if (debug_flag) {
 687                         struct timespec dt;
 688
 689                         ts_sub(&dt, &tcp->stime, &tcp->ltime);
 690
 691                         if (ts_nz(&dt))
 692                                 debug_func_msg("pid %d: %.9f seconds of system "
 693                                                "time spent since the last "
 694                                                "syscall exit",
 695                                                tcp->pid, ts_float(&dt));
 696                 }
 697
 698                 tcp->ltime = tcp->stime;
 699         }
 700 }
 701
 702 /* Returns:
 703  * 0: "bail out".
 704  * 1: ok.
 705  * -1: error in one of ptrace ops.
 706  *
 707  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
 708  *    value. Anyway, call syscall_exiting_finish(tcp) then.
 709  */
 710 int
 711 syscall_exiting_decode(struct tcb *tcp, struct timespec *pts)
 712 {
 713         /* Measure the exit time as early as possible to avoid errors. */
 714         if ((Tflag || cflag) && !filtered(tcp))
 715                 clock_gettime(CLOCK_MONOTONIC, pts);
 716
 717         if (tcp_sysent(tcp)->sys_flags & MEMORY_MAPPING_CHANGE)
 718                 mmap_notify_report(tcp);
 719
 720         if (filtered(tcp))
 721                 return 0;
 722
 723         if (check_exec_syscall(tcp)) {
 724                 /* The check failed, hide the log.  */
 725                 tcp->flags |= TCB_HIDE_LOG;
 726         }
 727
 728 #if SUPPORTED_PERSONALITIES > 1
 729         update_personality(tcp, tcp->currpers);
 730 #endif
 731
 732         return get_syscall_result(tcp);
 733 }
 734
 735 void
 736 print_syscall_resume(struct tcb *tcp)
 737 {
 738         /* If not in -ff mode, and printing_tcp != tcp,
 739          * then the log currently does not end with output
 740          * of _our syscall entry_, but with something else.
 741          * We need to say which syscall's return is this.
 742          *
 743          * Forced reprinting via TCB_REPRINT is used only by
 744          * "strace -ff -oLOG test/threaded_execve" corner case.
 745          * It's the only case when -ff mode needs reprinting.
 746          */
 747         if ((followfork < 2 && printing_tcp != tcp && !tcp->staged_output_data)
 748             || (tcp->flags & TCB_REPRINT)) {
 749                 tcp->flags &= ~TCB_REPRINT;
 750                 printleader(tcp);
 751                 tprintf("<... %s resumed>", tcp_sysent(tcp)->sys_name);
 752         }
 753 }
 754
 755 int
 756 syscall_exiting_trace(struct tcb *tcp, struct timespec *ts, int res)
 757 {
 758         if (syscall_tampered(tcp) || inject_delay_exit(tcp))
 759                 tamper_with_syscall_exiting(tcp);
 760
 761         if (cflag) {
 762                 count_syscall(tcp, ts);
 763                 if (cflag == CFLAG_ONLY_STATS) {
 764                         return 0;
 765                 }
 766         }
 767
 768         print_syscall_resume(tcp);
 769         printing_tcp = tcp;
 770
 771         tcp->s_prev_ent = NULL;
 772         if (res != 1) {
 773                 /* There was error in one of prior ptrace ops */
 774                 tprints(") ");
 775                 tabto();
 776                 tprints("= ? <unavailable>\n");
 777                 if (!is_complete_set(status_set, NUMBER_OF_STATUSES)) {
 778                         bool publish = is_number_in_set(STATUS_UNAVAILABLE,
 779                                                         status_set);
 780                         strace_close_memstream(tcp, publish);
 781                 }
 782                 line_ended();
 783                 return res;
 784         }
 785         tcp->s_prev_ent = tcp->s_ent;
 786
 787         int sys_res = 0;
 788         if (raw(tcp)) {
 789                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 790         } else {
 791                 if (tcp->sys_func_rval & RVAL_DECODED)
 792                         sys_res = tcp->sys_func_rval;
 793                 else
 794                         sys_res = tcp_sysent(tcp)->sys_func(tcp);
 795         }
 796
 797         if (!is_complete_set(status_set, NUMBER_OF_STATUSES)) {
 798                 bool publish = syserror(tcp)
 799                                && is_number_in_set(STATUS_FAILED, status_set);
 800                 publish |= !syserror(tcp)
 801                            && is_number_in_set(STATUS_SUCCESSFUL, status_set);
 802                 strace_close_memstream(tcp, publish);
 803                 if (!publish) {
 804                         line_ended();
 805                         return 0;
 806                 }
 807         }
 808
 809         tprints(") ");
 810         tabto();
 811
 812         if (raw(tcp)) {
 813                 if (tcp->u_error)
 814                         print_err_ret(tcp->u_rval, tcp->u_error);
 815                 else
 816                         tprintf("= %#" PRI_klx, tcp->u_rval);
 817
 818                 if (syscall_tampered(tcp))
 819                         tprints(" (INJECTED)");
 820         } else if (!(sys_res & RVAL_NONE) && tcp->u_error) {
 821                 switch (tcp->u_error) {
 822                 /* Blocked signals do not interrupt any syscalls.
 823                  * In this case syscalls don't return ERESTARTfoo codes.
 824                  *
 825                  * Deadly signals set to SIG_DFL interrupt syscalls
 826                  * and kill the process regardless of which of the codes below
 827                  * is returned by the interrupted syscall.
 828                  * In some cases, kernel forces a kernel-generated deadly
 829                  * signal to be unblocked and set to SIG_DFL (and thus cause
 830                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 831                  * or SIGILL. (The alternative is to leave process spinning
 832                  * forever on the faulty instruction - not useful).
 833                  *
 834                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 835                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 836                  * but kernel will always restart them.
 837                  */
 838                 case ERESTARTSYS:
 839                         /* Most common type of signal-interrupted syscall exit code.
 840                          * The system call will be restarted with the same arguments
 841                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 842                          */
 843                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 844                         break;
 845                 case ERESTARTNOINTR:
 846                         /* Rare. For example, fork() returns this if interrupted.
 847                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 848                          */
 849                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 850                         break;
 851                 case ERESTARTNOHAND:
 852                         /* pause(), rt_sigsuspend() etc use this code.
 853                          * SA_RESTART is ignored (assumed not set):
 854                          * syscall won't restart (will return EINTR instead)
 855                          * even after signal with SA_RESTART set. However,
 856                          * after SIG_IGN or SIG_DFL signal it will restart
 857                          * (thus the name "restart only if has no handler").
 858                          */
 859                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 860                         break;
 861                 case ERESTART_RESTARTBLOCK:
 862                         /* Syscalls like nanosleep(), poll() which can't be
 863                          * restarted with their original arguments use this
 864                          * code. Kernel will execute restart_syscall() instead,
 865                          * which changes arguments before restarting syscall.
 866                          * SA_RESTART is ignored (assumed not set) similarly
 867                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 868                          * since restart data is saved in "restart block"
 869                          * in task struct, and if signal handler uses a syscall
 870                          * which in turn saves another such restart block,
 871                          * old data is lost and restart becomes impossible)
 872                          */
 873                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 874                         break;
 875                 default:
 876                         print_err_ret(tcp->u_rval, tcp->u_error);
 877                         break;
 878                 }
 879                 if (syscall_tampered(tcp))
 880                         tprints(" (INJECTED)");
 881                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 882                         tprintf(" (%s)", tcp->auxstr);
 883         } else {
 884                 if (sys_res & RVAL_NONE)
 885                         tprints("= ?");
 886                 else {
 887                         switch (sys_res & RVAL_MASK) {
 888                         case RVAL_HEX:
 889 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 890                                 if (current_klongsize < sizeof(tcp->u_rval)) {
 891                                         tprintf("= %#x",
 892                                                 (unsigned int) tcp->u_rval);
 893                                 } else
 894 #endif
 895                                 {
 896                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 897                                 }
 898                                 break;
 899                         case RVAL_OCTAL:
 900                                 tprints("= ");
 901                                 print_numeric_long_umask(tcp->u_rval);
 902                                 break;
 903                         case RVAL_UDECIMAL:
 904 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 905                                 if (current_klongsize < sizeof(tcp->u_rval)) {
 906                                         tprintf("= %u",
 907                                                 (unsigned int) tcp->u_rval);
 908                                 } else
 909 #endif
 910                                 {
 911                                         tprintf("= %" PRI_klu, tcp->u_rval);
 912                                 }
 913                                 break;
 914                         case RVAL_FD:
 915                                 if (show_fd_path) {
 916                                         tprints("= ");
 917                                         printfd(tcp, tcp->u_rval);
 918                                 } else
 919                                         tprintf("= %" PRI_kld, tcp->u_rval);
 920                                 break;
 921                         default:
 922                                 error_msg("invalid rval format");
 923                                 break;
 924                         }
 925                 }
 926                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 927                         tprintf(" (%s)", tcp->auxstr);
 928                 if (syscall_tampered(tcp))
 929                         tprints(" (INJECTED)");
 930         }
 931         if (Tflag) {
 932                 ts_sub(ts, ts, &tcp->etime);
 933                 tprintf(" <%ld.%06ld>",
 934                         (long) ts->tv_sec, (long) ts->tv_nsec / 1000);
 935         }
 936         tprints("\n");
 937         dumpio(tcp);
 938         line_ended();
 939
 940 #ifdef ENABLE_STACKTRACE
 941         if (stack_trace_enabled)
 942                 unwind_tcb_print(tcp);
 943 #endif
 944         return 0;
 945 }
 946
 947 void
 948 syscall_exiting_finish(struct tcb *tcp)
 949 {
 950         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED | TCB_INJECT_DELAY_EXIT);
 951         tcp->sys_func_rval = 0;
 952         free_tcb_priv_data(tcp);
 953
 954         if (cflag)
 955                 tcp->ltime = tcp->stime;
 956 }
 957
 958 bool
 959 is_erestart(struct tcb *tcp)
 960 {
 961         switch (tcp->u_error) {
 962                 case ERESTARTSYS:
 963                 case ERESTARTNOINTR:
 964                 case ERESTARTNOHAND:
 965                 case ERESTART_RESTARTBLOCK:
 966                         return true;
 967                 default:
 968                         return false;
 969         }
 970 }
 971
 972 static unsigned long saved_u_error;
 973
 974 void
 975 temporarily_clear_syserror(struct tcb *tcp)
 976 {
 977         saved_u_error = tcp->u_error;
 978         tcp->u_error = 0;
 979 }
 980
 981 void
 982 restore_cleared_syserror(struct tcb *tcp)
 983 {
 984         tcp->u_error = saved_u_error;
 985 }
 986
 987 static struct ptrace_syscall_info ptrace_sci;
 988
 989 static bool
 990 ptrace_syscall_info_is_valid(void)
 991 {
 992         return ptrace_get_syscall_info_supported &&
 993                ptrace_sci.op <= PTRACE_SYSCALL_INFO_SECCOMP;
 994 }
 995
 996 #define XLAT_MACROS_ONLY
 997 #include "xlat/nt_descriptor_types.h"
 998 #undef XLAT_MACROS_ONLY
 999
1000 #define ARCH_MIGHT_USE_SET_REGS 1
1001
1002 #include "arch_regs.c"
1003
1004 #if HAVE_ARCH_GETRVAL2
1005 # include "arch_getrval2.c"
1006 #endif
1007
1008 #include "getregs_old.h"
1009 #ifdef HAVE_GETREGS_OLD
1010 /* Either getregs_old() or set_regs() */
1011 # undef ARCH_MIGHT_USE_SET_REGS
1012 # define ARCH_MIGHT_USE_SET_REGS 0
1013 #endif
1014
1015 #undef ptrace_getregset_or_getregs
1016 #undef ptrace_setregset_or_setregs
1017 #ifdef ARCH_REGS_FOR_GETREGSET
1018
1019 # define ptrace_getregset_or_getregs ptrace_getregset
1020 static long
1021 ptrace_getregset(pid_t pid)
1022 {
1023 # ifdef ARCH_IOVEC_FOR_GETREGSET
1024         /* variable iovec */
1025         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1026         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1027                       &ARCH_IOVEC_FOR_GETREGSET);
1028 # else
1029         /* constant iovec */
1030         static struct iovec io = {
1031                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1032                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1033         };
1034         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1035
1036 # endif
1037 }
1038
1039 # if ARCH_MIGHT_USE_SET_REGS
1040 #  define ptrace_setregset_or_setregs ptrace_setregset
1041 static int
1042 ptrace_setregset(pid_t pid)
1043 {
1044 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1045         /* variable iovec */
1046         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1047                       &ARCH_IOVEC_FOR_GETREGSET);
1048 #  else
1049         /* constant iovec */
1050         static struct iovec io = {
1051                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1052                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1053         };
1054         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1055 #  endif
1056 }
1057 # endif /* ARCH_MIGHT_USE_SET_REGS */
1058
1059 #elif defined ARCH_REGS_FOR_GETREGS
1060
1061 # define ptrace_getregset_or_getregs ptrace_getregs
1062 static long
1063 ptrace_getregs(pid_t pid)
1064 {
1065 # if defined SPARC || defined SPARC64
1066         /* SPARC systems have the meaning of data and addr reversed */
1067         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1068 # else
1069         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1070 # endif
1071 }
1072
1073 # if ARCH_MIGHT_USE_SET_REGS
1074 #  define ptrace_setregset_or_setregs ptrace_setregs
1075 static int
1076 ptrace_setregs(pid_t pid)
1077 {
1078 #  if defined SPARC || defined SPARC64
1079         /* SPARC systems have the meaning of data and addr reversed */
1080         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1081 #  else
1082         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1083 #  endif
1084 }
1085 # endif /* ARCH_MIGHT_USE_SET_REGS */
1086
1087 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1088
1089 static long get_regs_error = -1;
1090
1091 void
1092 clear_regs(struct tcb *tcp)
1093 {
1094         ptrace_sci.op = 0xff;
1095         get_regs_error = -1;
1096 }
1097
1098 static long
1099 get_regs(struct tcb *const tcp)
1100 {
1101 #ifdef ptrace_getregset_or_getregs
1102
1103         if (get_regs_error != -1)
1104                 return get_regs_error;
1105
1106 # ifdef HAVE_GETREGS_OLD
1107         /*
1108          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1109          * fallback to getregs_old.
1110          */
1111         static int use_getregs_old;
1112         if (use_getregs_old < 0) {
1113                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1114         } else if (use_getregs_old == 0) {
1115                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1116                 if (get_regs_error >= 0) {
1117                         use_getregs_old = -1;
1118                         return get_regs_error;
1119                 }
1120                 if (errno == EPERM || errno == ESRCH)
1121                         return get_regs_error;
1122                 use_getregs_old = 1;
1123         }
1124         return get_regs_error = getregs_old(tcp);
1125 # else /* !HAVE_GETREGS_OLD */
1126         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1127         get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1128
1129 #  if defined ARCH_PERSONALITY_0_IOV_SIZE
1130         if (get_regs_error)
1131                 return get_regs_error;
1132
1133         switch (ARCH_IOVEC_FOR_GETREGSET.iov_len) {
1134         case ARCH_PERSONALITY_0_IOV_SIZE:
1135                 update_personality(tcp, 0);
1136                 break;
1137         case ARCH_PERSONALITY_1_IOV_SIZE:
1138                 update_personality(tcp, 1);
1139                 break;
1140         default: {
1141                 static bool printed = false;
1142
1143                 if (!printed) {
1144                         error_msg("Unsupported regset size returned by "
1145                                   "PTRACE_GETREGSET: %zu",
1146                                   ARCH_IOVEC_FOR_GETREGSET.iov_len);
1147
1148                         printed = true;
1149                 }
1150
1151                 update_personality(tcp, 0);
1152         }
1153         }
1154 #  endif /* ARCH_PERSONALITY_0_IOV_SIZE */
1155
1156         return get_regs_error;
1157
1158 # endif /* !HAVE_GETREGS_OLD */
1159
1160 #else /* !ptrace_getregset_or_getregs */
1161
1162 # warning get_regs is not implemented for this architecture yet
1163         return 0;
1164
1165 #endif /* !ptrace_getregset_or_getregs */
1166 }
1167
1168 #ifdef ptrace_setregset_or_setregs
1169 static int
1170 set_regs(pid_t pid)
1171 {
1172         return ptrace_setregset_or_setregs(pid);
1173 }
1174 #endif /* ptrace_setregset_or_setregs */
1175
1176 struct sysent_buf {
1177         struct tcb *tcp;
1178         struct_sysent ent;
1179         char buf[sizeof("syscall_0x") + sizeof(kernel_ulong_t) * 2];
1180 };
1181
1182 static void
1183 free_sysent_buf(void *ptr)
1184 {
1185         struct sysent_buf *s = ptr;
1186         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1187         free(ptr);
1188 }
1189
1190 static bool
1191 strace_get_syscall_info(struct tcb *tcp)
1192 {
1193         /*
1194          * ptrace_get_syscall_info_supported should have been checked
1195          * by the caller.
1196          */
1197         if (ptrace_sci.op == 0xff) {
1198                 const size_t size = sizeof(ptrace_sci);
1199                 if (ptrace(PTRACE_GET_SYSCALL_INFO, tcp->pid,
1200                            (void *) size, &ptrace_sci) < 0) {
1201                         get_regs_error = -2;
1202                         return false;
1203                 }
1204 #if SUPPORTED_PERSONALITIES > 1
1205                 int newpers = get_personality_from_syscall_info(&ptrace_sci);
1206                 if (newpers >= 0)
1207                         update_personality(tcp, newpers);
1208 #endif
1209         }
1210
1211         if (entering(tcp)) {
1212                 if (ptrace_sci.op == PTRACE_SYSCALL_INFO_EXIT) {
1213                         error_msg("pid %d: entering"
1214                                   ", ptrace_syscall_info.op == %u",
1215                                   tcp->pid, ptrace_sci.op);
1216                         /* TODO: handle this.  */
1217                 }
1218         } else {
1219                 if (ptrace_sci.op == PTRACE_SYSCALL_INFO_ENTRY) {
1220                         error_msg("pid %d: exiting"
1221                                   ", ptrace_syscall_info.op == %u",
1222                                   tcp->pid, ptrace_sci.op);
1223                         /* TODO: handle this.  */
1224                 }
1225         }
1226
1227         return true;
1228 }
1229
1230 bool
1231 get_instruction_pointer(struct tcb *tcp, kernel_ulong_t *ip)
1232 {
1233         if (get_regs_error < -1)
1234                 return false;
1235
1236         if (ptrace_get_syscall_info_supported) {
1237                 if (!strace_get_syscall_info(tcp))
1238                         return false;
1239                 *ip = (kernel_ulong_t) ptrace_sci.instruction_pointer;
1240                 return true;
1241         }
1242
1243 #if defined ARCH_PC_REG
1244         if (get_regs(tcp) < 0)
1245                 return false;
1246         *ip = (kernel_ulong_t) ARCH_PC_REG;
1247         return true;
1248 #elif defined ARCH_PC_PEEK_ADDR
1249         if (upeek(tcp, ARCH_PC_PEEK_ADDR, ip) < 0)
1250                 return false;
1251         return true;
1252 #else
1253 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1254 #endif
1255 }
1256
1257 bool
1258 get_stack_pointer(struct tcb *tcp, kernel_ulong_t *sp)
1259 {
1260         if (get_regs_error < -1)
1261                 return false;
1262
1263         if (ptrace_get_syscall_info_supported) {
1264                 if (!strace_get_syscall_info(tcp))
1265                         return false;
1266                 *sp = (kernel_ulong_t) ptrace_sci.stack_pointer;
1267                 return true;
1268         }
1269
1270 #if defined ARCH_SP_REG
1271         if (get_regs(tcp) < 0)
1272                 return false;
1273         *sp = (kernel_ulong_t) ARCH_SP_REG;
1274         return true;
1275 #elif defined ARCH_SP_PEEK_ADDR
1276         if (upeek(tcp, ARCH_SP_PEEK_ADDR, sp) < 0)
1277                 return false;
1278         return true;
1279 #else
1280         return false;
1281 #endif
1282 }
1283
1284 static int
1285 get_syscall_regs(struct tcb *tcp)
1286 {
1287         if (get_regs_error != -1)
1288                 return get_regs_error;
1289
1290         if (ptrace_get_syscall_info_supported)
1291                 return strace_get_syscall_info(tcp) ? 0 : get_regs_error;
1292
1293         return get_regs(tcp);
1294 }
1295
1296 const struct_sysent stub_sysent = {
1297         .nargs = MAX_ARGS,
1298         .sys_flags = MEMORY_MAPPING_CHANGE,
1299         .sen = SEN_printargs,
1300         .sys_func = printargs,
1301         .sys_name = "????",
1302 };
1303
1304 /*
1305  * Returns:
1306  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1307  *    out silently" code.
1308  * 1: ok, continue in syscall_entering_decode().
1309  * other: error, syscall_entering_decode() should print error indicator
1310  *    ("????" etc) and return an appropriate code.
1311  */
1312 int
1313 get_scno(struct tcb *tcp)
1314 {
1315         tcp->scno = -1;
1316         tcp->s_ent = NULL;
1317         tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1318
1319         if (get_syscall_regs(tcp) < 0)
1320                 return -1;
1321
1322         if (ptrace_syscall_info_is_valid()) {
1323                 /*
1324                  * So far it's just a workaround for x32,
1325                  * but let's pretend it could be used elsewhere.
1326                  */
1327                 int rc = arch_check_scno(tcp);
1328                 if (rc != 1)
1329                         return rc;
1330                 tcp->scno = ptrace_sci.entry.nr;
1331         } else {
1332                 int rc = arch_get_scno(tcp);
1333                 if (rc != 1)
1334                         return rc;
1335         }
1336
1337         tcp->scno = shuffle_scno(tcp->scno);
1338
1339         if (scno_is_valid(tcp->scno)) {
1340                 tcp->s_ent = &sysent[tcp->scno];
1341                 tcp->qual_flg = qual_flags(tcp->scno);
1342         } else {
1343                 struct sysent_buf *s = xzalloc(sizeof(*s));
1344
1345                 s->tcp = tcp;
1346                 s->ent = stub_sysent;
1347                 s->ent.sys_name = s->buf;
1348                 xsprintf(s->buf, "syscall_%#" PRI_klx, shuffle_scno(tcp->scno));
1349
1350                 tcp->s_ent = &s->ent;
1351
1352                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1353
1354                 debug_msg("pid %d invalid syscall %#" PRI_klx,
1355                           tcp->pid, shuffle_scno(tcp->scno));
1356         }
1357
1358         /*
1359          * We refrain from argument decoding during recovering
1360          * as tracee memory mappings has changed and the registers
1361          * are very likely pointing to garbage already.
1362          */
1363         if (recovering(tcp))
1364                 tcp->qual_flg |= QUAL_RAW;
1365
1366         return 1;
1367 }
1368
1369 static int
1370 get_syscall_args(struct tcb *tcp)
1371 {
1372         if (ptrace_syscall_info_is_valid()) {
1373                 const unsigned int n =
1374                         MIN(ARRAY_SIZE(tcp->u_arg),
1375                             ARRAY_SIZE(ptrace_sci.entry.args));
1376                 for (unsigned int i = 0; i < n; ++i)
1377                         tcp->u_arg[i] = ptrace_sci.entry.args[i];
1378 #if SUPPORTED_PERSONALITIES > 1
1379                 if (tcp_sysent(tcp)->sys_flags & COMPAT_SYSCALL_TYPES) {
1380                         for (unsigned int i = 0; i < n; ++i)
1381                                 tcp->u_arg[i] = (uint32_t) tcp->u_arg[i];
1382                 }
1383 #endif
1384                 /*
1385                  * So far it's just a workaround for mips o32,
1386                  * but let's pretend it could be used elsewhere.
1387                  */
1388 #if MAX_ARGS > 6
1389                 arch_get_syscall_args_extra(tcp, n);
1390 #endif
1391                 return 1;
1392         }
1393         return arch_get_syscall_args(tcp);
1394 }
1395
1396 #ifdef ptrace_getregset_or_getregs
1397 # define get_syscall_result_regs get_syscall_regs
1398 #else
1399 static int get_syscall_result_regs(struct tcb *);
1400 #endif
1401
1402 /* Returns:
1403  * 1: ok, continue in syscall_exiting_trace().
1404  * -1: error, syscall_exiting_trace() should print error indicator
1405  *    ("????" etc) and bail out.
1406  */
1407 static int
1408 get_syscall_result(struct tcb *tcp)
1409 {
1410         if (get_syscall_result_regs(tcp) < 0)
1411                 return -1;
1412         get_error(tcp,
1413                   (!(tcp_sysent(tcp)->sys_flags & SYSCALL_NEVER_FAILS)
1414                         || syscall_tampered(tcp))
1415                   && !syscall_tampered_nofail(tcp));
1416
1417         return 1;
1418 }
1419
1420 static void
1421 get_error(struct tcb *tcp, const bool check_errno)
1422 {
1423         if (ptrace_syscall_info_is_valid()) {
1424                 if (ptrace_sci.exit.is_error) {
1425                         tcp->u_rval = -1;
1426                         tcp->u_error = -ptrace_sci.exit.rval;
1427                 } else {
1428                         tcp->u_error = 0;
1429                         tcp->u_rval = ptrace_sci.exit.rval;
1430                 }
1431         } else {
1432                 tcp->u_error = 0;
1433                 arch_get_error(tcp, check_errno);
1434         }
1435 }
1436
1437 static void
1438 set_error(struct tcb *tcp, unsigned long new_error)
1439 {
1440         const unsigned long old_error = tcp->u_error;
1441
1442         if (new_error == old_error || new_error > MAX_ERRNO_VALUE)
1443                 return;
1444
1445 #ifdef ptrace_setregset_or_setregs
1446         /* if we are going to invoke set_regs, call get_regs first */
1447         if (get_regs(tcp) < 0)
1448                 return;
1449 #endif
1450
1451         tcp->u_error = new_error;
1452         if (arch_set_error(tcp)) {
1453                 tcp->u_error = old_error;
1454                 /* arch_set_error does not update u_rval */
1455         } else {
1456                 if (ptrace_syscall_info_is_valid())
1457                         tcp->u_rval = -1;
1458                 else
1459                         get_error(tcp, !(tcp_sysent(tcp)->sys_flags &
1460                                          SYSCALL_NEVER_FAILS));
1461         }
1462 }
1463
1464 static void
1465 set_success(struct tcb *tcp, kernel_long_t new_rval)
1466 {
1467         const kernel_long_t old_rval = tcp->u_rval;
1468
1469 #ifdef ptrace_setregset_or_setregs
1470         /* if we are going to invoke set_regs, call get_regs first */
1471         if (get_regs(tcp) < 0)
1472                 return;
1473 #endif
1474
1475         tcp->u_rval = new_rval;
1476         if (arch_set_success(tcp)) {
1477                 tcp->u_rval = old_rval;
1478                 /* arch_set_success does not update u_error */
1479         } else {
1480                 if (ptrace_syscall_info_is_valid())
1481                         tcp->u_error = 0;
1482                 else
1483                         get_error(tcp, !(tcp_sysent(tcp)->sys_flags &
1484                                          SYSCALL_NEVER_FAILS));
1485         }
1486 }
1487
1488 #include "get_scno.c"
1489 #include "check_scno.c"
1490 #include "set_scno.c"
1491 #include "get_syscall_args.c"
1492 #ifndef ptrace_getregset_or_getregs
1493 # include "get_syscall_result.c"
1494 #endif
1495 #include "get_error.c"
1496 #include "set_error.c"
1497 #ifdef HAVE_GETREGS_OLD
1498 # include "getregs_old.c"
1499 #endif
1500 #include "shuffle_scno.c"
1501
1502 const char *
1503 syscall_name(kernel_ulong_t scno)
1504 {
1505         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1506 }