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
  63 /* Define these shorthand notations to simplify the syscallent files. */
  64 #include "sysent_shorthand_defs.h"
  65
  66 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  67
  68 const struct_sysent sysent0[] = {
  69 #include "syscallent.h"
  70 };
  71
  72 #if SUPPORTED_PERSONALITIES > 1
  73 # include PERSONALITY1_INCLUDE_FUNCS
  74 static const struct_sysent sysent1[] = {
  75 # include "syscallent1.h"
  76 };
  77 #endif
  78
  79 #if SUPPORTED_PERSONALITIES > 2
  80 # include PERSONALITY2_INCLUDE_FUNCS
  81 static const struct_sysent sysent2[] = {
  82 # include "syscallent2.h"
  83 };
  84 #endif
  85
  86 /* Now undef them since short defines cause wicked namespace pollution. */
  87 #include "sysent_shorthand_undefs.h"
  88
  89 /*
  90  * `ioctlent[012].h' files are automatically generated by the auxiliary
  91  * program `ioctlsort', such that the list is sorted by the `code' field.
  92  * This has the side-effect of resolving the _IO.. macros into
  93  * plain integers, eliminating the need to include here everything
  94  * in "/usr/include".
  95  */
  96
  97 const char *const errnoent0[] = {
  98 #include "errnoent.h"
  99 };
 100 const char *const signalent0[] = {
 101 #include "signalent.h"
 102 };
 103 const struct_ioctlent ioctlent0[] = {
 104 #include "ioctlent0.h"
 105 };
 106
 107 #if SUPPORTED_PERSONALITIES > 1
 108 static const char *const errnoent1[] = {
 109 # include "errnoent1.h"
 110 };
 111 static const char *const signalent1[] = {
 112 # include "signalent1.h"
 113 };
 114 static const struct_ioctlent ioctlent1[] = {
 115 # include "ioctlent1.h"
 116 };
 117 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 118 static const struct_printers printers0 = {
 119 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 120 };
 121 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 122 static const struct_printers printers1 = {
 123 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 124 };
 125 #endif
 126
 127 #if SUPPORTED_PERSONALITIES > 2
 128 static const char *const errnoent2[] = {
 129 # include "errnoent2.h"
 130 };
 131 static const char *const signalent2[] = {
 132 # include "signalent2.h"
 133 };
 134 static const struct_ioctlent ioctlent2[] = {
 135 # include "ioctlent2.h"
 136 };
 137 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 138 static const struct_printers printers2 = {
 139 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 140 };
 141 #endif
 142
 143 enum {
 144         nsyscalls0 = ARRAY_SIZE(sysent0)
 145 #if SUPPORTED_PERSONALITIES > 1
 146         , nsyscalls1 = ARRAY_SIZE(sysent1)
 147 # if SUPPORTED_PERSONALITIES > 2
 148         , nsyscalls2 = ARRAY_SIZE(sysent2)
 149 # endif
 150 #endif
 151 };
 152
 153 enum {
 154         nerrnos0 = ARRAY_SIZE(errnoent0)
 155 #if SUPPORTED_PERSONALITIES > 1
 156         , nerrnos1 = ARRAY_SIZE(errnoent1)
 157 # if SUPPORTED_PERSONALITIES > 2
 158         , nerrnos2 = ARRAY_SIZE(errnoent2)
 159 # endif
 160 #endif
 161 };
 162
 163 enum {
 164         nsignals0 = ARRAY_SIZE(signalent0)
 165 #if SUPPORTED_PERSONALITIES > 1
 166         , nsignals1 = ARRAY_SIZE(signalent1)
 167 # if SUPPORTED_PERSONALITIES > 2
 168         , nsignals2 = ARRAY_SIZE(signalent2)
 169 # endif
 170 #endif
 171 };
 172
 173 enum {
 174         nioctlents0 = ARRAY_SIZE(ioctlent0)
 175 #if SUPPORTED_PERSONALITIES > 1
 176         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 177 # if SUPPORTED_PERSONALITIES > 2
 178         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 179 # endif
 180 #endif
 181 };
 182
 183 #if SUPPORTED_PERSONALITIES > 1
 184 const struct_sysent *sysent = sysent0;
 185 const char *const *errnoent = errnoent0;
 186 const char *const *signalent = signalent0;
 187 const struct_ioctlent *ioctlent = ioctlent0;
 188 const struct_printers *printers = &printers0;
 189 #endif
 190
 191 unsigned nsyscalls = nsyscalls0;
 192 unsigned nerrnos = nerrnos0;
 193 unsigned nsignals = nsignals0;
 194 unsigned nioctlents = nioctlents0;
 195
 196 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 197         nsyscalls0,
 198 #if SUPPORTED_PERSONALITIES > 1
 199         nsyscalls1,
 200 #endif
 201 #if SUPPORTED_PERSONALITIES > 2
 202         nsyscalls2,
 203 #endif
 204 };
 205 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 206         sysent0,
 207 #if SUPPORTED_PERSONALITIES > 1
 208         sysent1,
 209 #endif
 210 #if SUPPORTED_PERSONALITIES > 2
 211         sysent2,
 212 #endif
 213 };
 214
 215 #if SUPPORTED_PERSONALITIES > 1
 216 unsigned current_personality;
 217
 218 # ifndef current_wordsize
 219 unsigned current_wordsize;
 220 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 221         PERSONALITY0_WORDSIZE,
 222         PERSONALITY1_WORDSIZE,
 223 # if SUPPORTED_PERSONALITIES > 2
 224         PERSONALITY2_WORDSIZE,
 225 # endif
 226 };
 227 # endif
 228
 229 # ifndef current_klongsize
 230 unsigned current_klongsize;
 231 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 232         PERSONALITY0_KLONGSIZE,
 233         PERSONALITY1_KLONGSIZE,
 234 #  if SUPPORTED_PERSONALITIES > 2
 235         PERSONALITY2_KLONGSIZE,
 236 #  endif
 237 };
 238 # endif
 239
 240 void
 241 set_personality(int personality)
 242 {
 243         nsyscalls = nsyscall_vec[personality];
 244         sysent = sysent_vec[personality];
 245
 246         switch (personality) {
 247         case 0:
 248                 errnoent = errnoent0;
 249                 nerrnos = nerrnos0;
 250                 ioctlent = ioctlent0;
 251                 nioctlents = nioctlents0;
 252                 signalent = signalent0;
 253                 nsignals = nsignals0;
 254                 printers = &printers0;
 255                 break;
 256
 257         case 1:
 258                 errnoent = errnoent1;
 259                 nerrnos = nerrnos1;
 260                 ioctlent = ioctlent1;
 261                 nioctlents = nioctlents1;
 262                 signalent = signalent1;
 263                 nsignals = nsignals1;
 264                 printers = &printers1;
 265                 break;
 266
 267 # if SUPPORTED_PERSONALITIES > 2
 268         case 2:
 269                 errnoent = errnoent2;
 270                 nerrnos = nerrnos2;
 271                 ioctlent = ioctlent2;
 272                 nioctlents = nioctlents2;
 273                 signalent = signalent2;
 274                 nsignals = nsignals2;
 275                 printers = &printers2;
 276                 break;
 277 # endif
 278         }
 279
 280         current_personality = personality;
 281 # ifndef current_wordsize
 282         current_wordsize = personality_wordsize[personality];
 283 # endif
 284 # ifndef current_klongsize
 285         current_klongsize = personality_klongsize[personality];
 286 # endif
 287 }
 288
 289 static void
 290 update_personality(struct tcb *tcp, unsigned int personality)
 291 {
 292         if (personality == current_personality)
 293                 return;
 294         set_personality(personality);
 295
 296         if (personality == tcp->currpers)
 297                 return;
 298         tcp->currpers = personality;
 299
 300 # undef PERSONALITY_NAMES
 301 # if defined X86_64
 302 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
 303 # elif defined X32
 304 #  define PERSONALITY_NAMES {"x32", "32 bit"}
 305 # elif SUPPORTED_PERSONALITIES == 2
 306 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 307 # endif
 308 # ifdef PERSONALITY_NAMES
 309         if (!qflag) {
 310                 static const char *const names[] = PERSONALITY_NAMES;
 311                 error_msg("[ Process PID=%d runs in %s mode. ]",
 312                           tcp->pid, names[personality]);
 313         }
 314 # endif
 315 }
 316 #endif
 317
 318 #ifdef SYS_socket_subcall
 319 static void
 320 decode_socket_subcall(struct tcb *tcp)
 321 {
 322         const int call = tcp->u_arg[0];
 323
 324         if (call < 1 || call >= SYS_socket_nsubcalls)
 325                 return;
 326
 327         const kernel_ulong_t scno = SYS_socket_subcall + call;
 328         const unsigned int nargs = sysent[scno].nargs;
 329         uint64_t buf[nargs];
 330
 331         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 332                 return;
 333
 334         tcp->scno = scno;
 335         tcp->qual_flg = qual_flags(scno);
 336         tcp->s_ent = &sysent[scno];
 337
 338         unsigned int i;
 339         for (i = 0; i < nargs; ++i)
 340                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 341                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 342 }
 343 #endif /* SYS_socket_subcall */
 344
 345 #ifdef SYS_ipc_subcall
 346 static void
 347 decode_ipc_subcall(struct tcb *tcp)
 348 {
 349         unsigned int call = tcp->u_arg[0];
 350         const unsigned int version = call >> 16;
 351
 352         if (version) {
 353 # if defined S390 || defined S390X
 354                 return;
 355 # else
 356 #  ifdef SPARC64
 357                 if (current_wordsize == 8)
 358                         return;
 359 #  endif
 360                 set_tcb_priv_ulong(tcp, version);
 361                 call &= 0xffff;
 362 # endif
 363         }
 364
 365         switch (call) {
 366                 case  1: case  2: case  3: case  4:
 367                 case 11: case 12: case 13: case 14:
 368                 case 21: case 22: case 23: case 24:
 369                         break;
 370                 default:
 371                         return;
 372         }
 373
 374         tcp->scno = SYS_ipc_subcall + call;
 375         tcp->qual_flg = qual_flags(tcp->scno);
 376         tcp->s_ent = &sysent[tcp->scno];
 377
 378         const unsigned int n = tcp->s_ent->nargs;
 379         unsigned int i;
 380         for (i = 0; i < n; i++)
 381                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 382 }
 383 #endif /* SYS_ipc_subcall */
 384
 385 #ifdef SYS_syscall_subcall
 386 static void
 387 decode_syscall_subcall(struct tcb *tcp)
 388 {
 389         if (!scno_is_valid(tcp->u_arg[0]))
 390                 return;
 391         tcp->scno = tcp->u_arg[0];
 392         tcp->qual_flg = qual_flags(tcp->scno);
 393         tcp->s_ent = &sysent[tcp->scno];
 394         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 395                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 396 # ifdef LINUX_MIPSO32
 397         /*
 398          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 399          * and sync_file_range) requires additional code,
 400          * see linux/mips/get_syscall_args.c
 401          */
 402         if (tcp->s_ent->nargs == MAX_ARGS) {
 403                 if (umoven(tcp,
 404                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 405                            sizeof(tcp->u_arg[0]),
 406                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 407                 tcp->u_arg[MAX_ARGS - 1] = 0;
 408         }
 409 # endif /* LINUX_MIPSO32 */
 410 }
 411 #endif /* SYS_syscall_subcall */
 412
 413 static void
 414 dumpio(struct tcb *tcp)
 415 {
 416         if (syserror(tcp))
 417                 return;
 418
 419         int fd = tcp->u_arg[0];
 420         if (fd < 0)
 421                 return;
 422
 423         if (is_number_in_set(fd, read_set)) {
 424                 switch (tcp->s_ent->sen) {
 425                 case SEN_read:
 426                 case SEN_pread:
 427                 case SEN_recv:
 428                 case SEN_recvfrom:
 429                 case SEN_mq_timedreceive:
 430                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 431                         return;
 432                 case SEN_readv:
 433                 case SEN_preadv:
 434                 case SEN_preadv2:
 435                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 436                                      tcp->u_rval);
 437                         return;
 438                 case SEN_recvmsg:
 439                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 440                         return;
 441                 case SEN_recvmmsg:
 442                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 443                         return;
 444                 }
 445         }
 446         if (is_number_in_set(fd, write_set)) {
 447                 switch (tcp->s_ent->sen) {
 448                 case SEN_write:
 449                 case SEN_pwrite:
 450                 case SEN_send:
 451                 case SEN_sendto:
 452                 case SEN_mq_timedsend:
 453                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 454                         break;
 455                 case SEN_writev:
 456                 case SEN_pwritev:
 457                 case SEN_pwritev2:
 458                 case SEN_vmsplice:
 459                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 460                         break;
 461                 case SEN_sendmsg:
 462                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 463                         break;
 464                 case SEN_sendmmsg:
 465                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 466                         break;
 467                 }
 468         }
 469 }
 470
 471 /*
 472  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 473  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 474  */
 475 static kernel_ulong_t
 476 shuffle_scno(kernel_ulong_t scno)
 477 {
 478 #ifdef ARM_FIRST_SHUFFLED_SYSCALL       /* So far only 32-bit ARM needs this */
 479         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 480                 return scno;
 481
 482         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 483         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 484                 return 0x000ffff0;
 485         if (scno == 0x000ffff0)
 486                 return ARM_FIRST_SHUFFLED_SYSCALL;
 487
 488 # define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 489         /*
 490          * Is it ARM specific syscall?
 491          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 492          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 493          */
 494         if (scno >= 0x000f0000 &&
 495             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 496                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 497         }
 498         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 499                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 500         }
 501 #endif /* ARM_FIRST_SHUFFLED_SYSCALL */
 502
 503         return scno;
 504 }
 505
 506 const char *
 507 err_name(unsigned long err)
 508 {
 509         if ((err < nerrnos) && errnoent[err])
 510                 return errnoent[err];
 511
 512         return NULL;
 513 }
 514
 515 static long get_regs(struct tcb *);
 516 static int get_syscall_args(struct tcb *);
 517 static int get_syscall_result(struct tcb *);
 518 static int arch_get_scno(struct tcb *tcp);
 519 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 520 static void get_error(struct tcb *, const bool);
 521 static int arch_set_error(struct tcb *);
 522 static int arch_set_success(struct tcb *);
 523
 524 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
 525
 526 static struct inject_opts *
 527 tcb_inject_opts(struct tcb *tcp)
 528 {
 529         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
 530                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
 531 }
 532
 533
 534 static long
 535 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
 536 {
 537         if (!tcp->inject_vec[current_personality]) {
 538                 tcp->inject_vec[current_personality] =
 539                         xcalloc(nsyscalls, sizeof(**inject_vec));
 540                 memcpy(tcp->inject_vec[current_personality],
 541                        inject_vec[current_personality],
 542                        nsyscalls * sizeof(**inject_vec));
 543         }
 544
 545         struct inject_opts *opts = tcb_inject_opts(tcp);
 546
 547         if (!opts || opts->first == 0)
 548                 return 0;
 549
 550         --opts->first;
 551
 552         if (opts->first != 0)
 553                 return 0;
 554
 555         opts->first = opts->step;
 556
 557         if (opts->data.flags & INJECT_F_SIGNAL)
 558                 *signo = opts->data.signo;
 559         if (opts->data.flags & INJECT_F_RETVAL && !arch_set_scno(tcp, -1))
 560                 tcp->flags |= TCB_TAMPERED;
 561
 562         return 0;
 563 }
 564
 565 static long
 566 tamper_with_syscall_exiting(struct tcb *tcp)
 567 {
 568         struct inject_opts *opts = tcb_inject_opts(tcp);
 569
 570         if (!opts)
 571                 return 0;
 572
 573         if (opts->data.rval >= 0) {
 574                 kernel_long_t u_rval = tcp->u_rval;
 575
 576                 tcp->u_rval = opts->data.rval;
 577                 if (arch_set_success(tcp)) {
 578                         tcp->u_rval = u_rval;
 579                 } else {
 580                         tcp->u_error = 0;
 581                 }
 582         } else {
 583                 unsigned long new_error = -opts->data.rval;
 584
 585                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
 586                         unsigned long u_error = tcp->u_error;
 587
 588                         tcp->u_error = new_error;
 589                         if (arch_set_error(tcp)) {
 590                                 tcp->u_error = u_error;
 591                         }
 592                 }
 593         }
 594
 595         return 0;
 596 }
 597
 598 /*
 599  * Returns:
 600  * 0: "ignore this ptrace stop", bail out silently.
 601  * 1: ok, decoded; call
 602  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
 603  * other: error; call syscall_entering_finish(tcp, res), where res is the value
 604  *    returned.
 605  */
 606 int
 607 syscall_entering_decode(struct tcb *tcp)
 608 {
 609         int res = get_scno(tcp);
 610         if (res == 0)
 611                 return res;
 612         int scno_good = res;
 613         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
 614                 printleader(tcp);
 615                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
 616                 /*
 617                  * " <unavailable>" will be added later by the code which
 618                  * detects ptrace errors.
 619                  */
 620                 return res;
 621         }
 622
 623 #if defined SYS_ipc_subcall     \
 624  || defined SYS_socket_subcall  \
 625  || defined SYS_syscall_subcall
 626         for (;;) {
 627                 switch (tcp->s_ent->sen) {
 628 # ifdef SYS_ipc_subcall
 629                 case SEN_ipc:
 630                         decode_ipc_subcall(tcp);
 631                         break;
 632 # endif
 633 # ifdef SYS_socket_subcall
 634                 case SEN_socketcall:
 635                         decode_socket_subcall(tcp);
 636                         break;
 637 # endif
 638 # ifdef SYS_syscall_subcall
 639                 case SEN_syscall:
 640                         decode_syscall_subcall(tcp);
 641                         if (tcp->s_ent->sen != SEN_syscall)
 642                                 continue;
 643                         break;
 644 # endif
 645                 }
 646                 break;
 647         }
 648 #endif
 649
 650         return 1;
 651 }
 652
 653 int
 654 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
 655 {
 656         /* Restrain from fault injection while the trace executes strace code. */
 657         if (hide_log(tcp)) {
 658                 tcp->qual_flg &= ~QUAL_INJECT;
 659         }
 660
 661         switch (tcp->s_ent->sen) {
 662                 case SEN_execve:
 663                 case SEN_execveat:
 664 #if defined SPARC || defined SPARC64
 665                 case SEN_execv:
 666 #endif
 667                         tcp->flags &= ~TCB_HIDE_LOG;
 668                         break;
 669         }
 670
 671         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
 672                 tcp->flags |= TCB_FILTERED;
 673                 return 0;
 674         }
 675
 676         tcp->flags &= ~TCB_FILTERED;
 677
 678         if (hide_log(tcp)) {
 679                 return 0;
 680         }
 681
 682         if (inject(tcp))
 683                 tamper_with_syscall_entering(tcp, sig);
 684
 685         if (cflag == CFLAG_ONLY_STATS) {
 686                 return 0;
 687         }
 688
 689 #ifdef USE_LIBUNWIND
 690         if (stack_trace_enabled) {
 691                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 692                         unwind_capture_stacktrace(tcp);
 693         }
 694 #endif
 695
 696         printleader(tcp);
 697         tprintf("%s(", tcp->s_ent->sys_name);
 698         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
 699         fflush(tcp->outf);
 700         return res;
 701 }
 702
 703 void
 704 syscall_entering_finish(struct tcb *tcp, int res)
 705 {
 706         tcp->flags |= TCB_INSYSCALL;
 707         tcp->sys_func_rval = res;
 708         /* Measure the entrance time as late as possible to avoid errors. */
 709         if ((Tflag || cflag) && !filtered(tcp))
 710                 gettimeofday(&tcp->etime, NULL);
 711 }
 712
 713 static bool
 714 syscall_tampered(struct tcb *tcp)
 715 {
 716         return tcp->flags & TCB_TAMPERED;
 717 }
 718
 719 /* Returns:
 720  * 0: "bail out".
 721  * 1: ok.
 722  * -1: error in one of ptrace ops.
 723  *
 724  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
 725  *    value. Anyway, call syscall_exiting_finish(tcp) then.
 726  */
 727 int
 728 syscall_exiting_decode(struct tcb *tcp, struct timeval *ptv)
 729 {
 730         /* Measure the exit time as early as possible to avoid errors. */
 731         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
 732                 gettimeofday(ptv, NULL);
 733
 734 #ifdef USE_LIBUNWIND
 735         if (stack_trace_enabled) {
 736                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 737                         unwind_cache_invalidate(tcp);
 738         }
 739 #endif
 740
 741         if (filtered(tcp) || hide_log(tcp))
 742                 return 0;
 743
 744 #if SUPPORTED_PERSONALITIES > 1
 745         update_personality(tcp, tcp->currpers);
 746 #endif
 747
 748         return get_syscall_result(tcp);
 749 }
 750
 751 int
 752 syscall_exiting_trace(struct tcb *tcp, struct timeval tv, int res)
 753 {
 754         if (syserror(tcp) && syscall_tampered(tcp))
 755                 tamper_with_syscall_exiting(tcp);
 756
 757         if (cflag) {
 758                 count_syscall(tcp, &tv);
 759                 if (cflag == CFLAG_ONLY_STATS) {
 760                         return 0;
 761                 }
 762         }
 763
 764         /* If not in -ff mode, and printing_tcp != tcp,
 765          * then the log currently does not end with output
 766          * of _our syscall entry_, but with something else.
 767          * We need to say which syscall's return is this.
 768          *
 769          * Forced reprinting via TCB_REPRINT is used only by
 770          * "strace -ff -oLOG test/threaded_execve" corner case.
 771          * It's the only case when -ff mode needs reprinting.
 772          */
 773         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 774                 tcp->flags &= ~TCB_REPRINT;
 775                 printleader(tcp);
 776                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 777         }
 778         printing_tcp = tcp;
 779
 780         tcp->s_prev_ent = NULL;
 781         if (res != 1) {
 782                 /* There was error in one of prior ptrace ops */
 783                 tprints(") ");
 784                 tabto();
 785                 tprints("= ? <unavailable>\n");
 786                 line_ended();
 787                 return res;
 788         }
 789         tcp->s_prev_ent = tcp->s_ent;
 790
 791         int sys_res = 0;
 792         if (raw(tcp)) {
 793                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 794         } else {
 795         /* FIXME: not_failing_only (IOW, option -z) is broken:
 796          * failure of syscall is known only after syscall return.
 797          * Thus we end up with something like this on, say, ENOENT:
 798          *     open("does_not_exist", O_RDONLY <unfinished ...>
 799          *     {next syscall decode}
 800          * whereas the intended result is that open(...) line
 801          * is not shown at all.
 802          */
 803                 if (not_failing_only && tcp->u_error)
 804                         return 0;       /* ignore failed syscalls */
 805                 if (tcp->sys_func_rval & RVAL_DECODED)
 806                         sys_res = tcp->sys_func_rval;
 807                 else
 808                         sys_res = tcp->s_ent->sys_func(tcp);
 809         }
 810
 811         tprints(") ");
 812         tabto();
 813         unsigned long u_error = tcp->u_error;
 814
 815         if (raw(tcp)) {
 816                 if (u_error) {
 817                         tprintf("= -1 (errno %lu)", u_error);
 818                 } else {
 819                         tprintf("= %#" PRI_klx, tcp->u_rval);
 820                 }
 821                 if (syscall_tampered(tcp))
 822                         tprints(" (INJECTED)");
 823         } else if (!(sys_res & RVAL_NONE) && u_error) {
 824                 const char *u_error_str;
 825
 826                 switch (u_error) {
 827                 /* Blocked signals do not interrupt any syscalls.
 828                  * In this case syscalls don't return ERESTARTfoo codes.
 829                  *
 830                  * Deadly signals set to SIG_DFL interrupt syscalls
 831                  * and kill the process regardless of which of the codes below
 832                  * is returned by the interrupted syscall.
 833                  * In some cases, kernel forces a kernel-generated deadly
 834                  * signal to be unblocked and set to SIG_DFL (and thus cause
 835                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 836                  * or SIGILL. (The alternative is to leave process spinning
 837                  * forever on the faulty instruction - not useful).
 838                  *
 839                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 840                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 841                  * but kernel will always restart them.
 842                  */
 843                 case ERESTARTSYS:
 844                         /* Most common type of signal-interrupted syscall exit code.
 845                          * The system call will be restarted with the same arguments
 846                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 847                          */
 848                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 849                         break;
 850                 case ERESTARTNOINTR:
 851                         /* Rare. For example, fork() returns this if interrupted.
 852                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 853                          */
 854                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 855                         break;
 856                 case ERESTARTNOHAND:
 857                         /* pause(), rt_sigsuspend() etc use this code.
 858                          * SA_RESTART is ignored (assumed not set):
 859                          * syscall won't restart (will return EINTR instead)
 860                          * even after signal with SA_RESTART set. However,
 861                          * after SIG_IGN or SIG_DFL signal it will restart
 862                          * (thus the name "restart only if has no handler").
 863                          */
 864                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 865                         break;
 866                 case ERESTART_RESTARTBLOCK:
 867                         /* Syscalls like nanosleep(), poll() which can't be
 868                          * restarted with their original arguments use this
 869                          * code. Kernel will execute restart_syscall() instead,
 870                          * which changes arguments before restarting syscall.
 871                          * SA_RESTART is ignored (assumed not set) similarly
 872                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 873                          * since restart data is saved in "restart block"
 874                          * in task struct, and if signal handler uses a syscall
 875                          * which in turn saves another such restart block,
 876                          * old data is lost and restart becomes impossible)
 877                          */
 878                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 879                         break;
 880                 default:
 881                         u_error_str = err_name(u_error);
 882                         if (u_error_str)
 883                                 tprintf("= -1 %s (%s)",
 884                                         u_error_str, strerror(u_error));
 885                         else
 886                                 tprintf("= -1 %lu (%s)",
 887                                         u_error, strerror(u_error));
 888                         break;
 889                 }
 890                 if (syscall_tampered(tcp))
 891                         tprints(" (INJECTED)");
 892                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 893                         tprintf(" (%s)", tcp->auxstr);
 894         } else {
 895                 if (sys_res & RVAL_NONE)
 896                         tprints("= ?");
 897                 else {
 898                         switch (sys_res & RVAL_MASK) {
 899                         case RVAL_HEX:
 900 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 901                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 902                                         tprintf("= %#x",
 903                                                 (unsigned int) tcp->u_rval);
 904                                 } else
 905 #endif
 906                                 {
 907                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 908                                 }
 909                                 break;
 910                         case RVAL_OCTAL:
 911                                 tprints("= ");
 912                                 print_numeric_long_umask(tcp->u_rval);
 913                                 break;
 914                         case RVAL_UDECIMAL:
 915 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 916                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 917                                         tprintf("= %u",
 918                                                 (unsigned int) tcp->u_rval);
 919                                 } else
 920 #endif
 921                                 {
 922                                         tprintf("= %" PRI_klu, tcp->u_rval);
 923                                 }
 924                                 break;
 925                         case RVAL_DECIMAL:
 926                                 tprintf("= %" PRI_kld, tcp->u_rval);
 927                                 break;
 928                         case RVAL_FD:
 929                                 if (show_fd_path) {
 930                                         tprints("= ");
 931                                         printfd(tcp, tcp->u_rval);
 932                                 } else
 933                                         tprintf("= %" PRI_kld, tcp->u_rval);
 934                                 break;
 935                         default:
 936                                 error_msg("invalid rval format");
 937                                 break;
 938                         }
 939                 }
 940                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 941                         tprintf(" (%s)", tcp->auxstr);
 942                 if (syscall_tampered(tcp))
 943                         tprints(" (INJECTED)");
 944         }
 945         if (Tflag) {
 946                 tv_sub(&tv, &tv, &tcp->etime);
 947                 tprintf(" <%ld.%06ld>",
 948                         (long) tv.tv_sec, (long) tv.tv_usec);
 949         }
 950         tprints("\n");
 951         dumpio(tcp);
 952         line_ended();
 953
 954 #ifdef USE_LIBUNWIND
 955         if (stack_trace_enabled)
 956                 unwind_print_stacktrace(tcp);
 957 #endif
 958         return 0;
 959 }
 960
 961 void
 962 syscall_exiting_finish(struct tcb *tcp)
 963 {
 964         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
 965         tcp->sys_func_rval = 0;
 966         free_tcb_priv_data(tcp);
 967 }
 968
 969 bool
 970 is_erestart(struct tcb *tcp)
 971 {
 972         switch (tcp->u_error) {
 973                 case ERESTARTSYS:
 974                 case ERESTARTNOINTR:
 975                 case ERESTARTNOHAND:
 976                 case ERESTART_RESTARTBLOCK:
 977                         return true;
 978                 default:
 979                         return false;
 980         }
 981 }
 982
 983 static unsigned long saved_u_error;
 984
 985 void
 986 temporarily_clear_syserror(struct tcb *tcp)
 987 {
 988         saved_u_error = tcp->u_error;
 989         tcp->u_error = 0;
 990 }
 991
 992 void
 993 restore_cleared_syserror(struct tcb *tcp)
 994 {
 995         tcp->u_error = saved_u_error;
 996 }
 997
 998 #include "arch_regs.c"
 999
1000 #ifdef HAVE_GETRVAL2
1001 # include "arch_getrval2.c"
1002 #endif
1003
1004 void
1005 print_pc(struct tcb *tcp)
1006 {
1007 #if defined ARCH_PC_REG
1008 # define ARCH_GET_PC 0
1009 #elif defined ARCH_PC_PEEK_ADDR
1010         kernel_ulong_t pc;
1011 # define ARCH_PC_REG pc
1012 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1013 #else
1014 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1015 #endif
1016         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1017                 tprints(current_wordsize == 4 ? "[????????] "
1018                                               : "[????????????????] ");
1019         else
1020                 tprintf(current_wordsize == 4
1021                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1022                         (kernel_ulong_t) ARCH_PC_REG);
1023 }
1024
1025 #include "getregs_old.h"
1026
1027 #undef ptrace_getregset_or_getregs
1028 #undef ptrace_setregset_or_setregs
1029 #ifdef ARCH_REGS_FOR_GETREGSET
1030
1031 # define ptrace_getregset_or_getregs ptrace_getregset
1032 static long
1033 ptrace_getregset(pid_t pid)
1034 {
1035 # ifdef ARCH_IOVEC_FOR_GETREGSET
1036         /* variable iovec */
1037         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1038         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1039                       &ARCH_IOVEC_FOR_GETREGSET);
1040 # else
1041         /* constant iovec */
1042         static struct iovec io = {
1043                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1044                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1045         };
1046         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1047
1048 # endif
1049 }
1050
1051 # ifndef HAVE_GETREGS_OLD
1052 #  define ptrace_setregset_or_setregs ptrace_setregset
1053 static int
1054 ptrace_setregset(pid_t pid)
1055 {
1056 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1057         /* variable iovec */
1058         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1059                       &ARCH_IOVEC_FOR_GETREGSET);
1060 #  else
1061         /* constant iovec */
1062         static struct iovec io = {
1063                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1064                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1065         };
1066         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1067 #  endif
1068 }
1069 # endif /* !HAVE_GETREGS_OLD */
1070
1071 #elif defined ARCH_REGS_FOR_GETREGS
1072
1073 # define ptrace_getregset_or_getregs ptrace_getregs
1074 static long
1075 ptrace_getregs(pid_t pid)
1076 {
1077 # if defined SPARC || defined SPARC64
1078         /* SPARC systems have the meaning of data and addr reversed */
1079         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1080 # else
1081         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1082 # endif
1083 }
1084
1085 # ifndef HAVE_GETREGS_OLD
1086 #  define ptrace_setregset_or_setregs ptrace_setregs
1087 static int
1088 ptrace_setregs(pid_t pid)
1089 {
1090 #  if defined SPARC || defined SPARC64
1091         /* SPARC systems have the meaning of data and addr reversed */
1092         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1093 #  else
1094         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1095 #  endif
1096 }
1097 # endif /* !HAVE_GETREGS_OLD */
1098
1099 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1100
1101 #ifdef ptrace_getregset_or_getregs
1102 static long get_regs_error;
1103 #endif
1104
1105 void
1106 clear_regs(void)
1107 {
1108 #ifdef ptrace_getregset_or_getregs
1109         get_regs_error = -1;
1110 #endif
1111 }
1112
1113 static long
1114 get_regs(struct tcb *const tcp)
1115 {
1116 #ifdef ptrace_getregset_or_getregs
1117
1118         if (get_regs_error != -1)
1119                 return get_regs_error;
1120
1121 # ifdef HAVE_GETREGS_OLD
1122         /*
1123          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1124          * fallback to getregs_old.
1125          */
1126         static int use_getregs_old;
1127         if (use_getregs_old < 0) {
1128                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1129         } else if (use_getregs_old == 0) {
1130                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1131                 if (get_regs_error >= 0) {
1132                         use_getregs_old = -1;
1133                         return get_regs_error;
1134                 }
1135                 if (errno == EPERM || errno == ESRCH)
1136                         return get_regs_error;
1137                 use_getregs_old = 1;
1138         }
1139         return get_regs_error = getregs_old(tcp);
1140 # else /* !HAVE_GETREGS_OLD */
1141         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1142         return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1143 # endif /* !HAVE_GETREGS_OLD */
1144
1145 #else /* !ptrace_getregset_or_getregs */
1146
1147 # warning get_regs is not implemented for this architecture yet
1148         return 0;
1149
1150 #endif /* !ptrace_getregset_or_getregs */
1151 }
1152
1153 #ifdef ptrace_setregset_or_setregs
1154 static int
1155 set_regs(pid_t pid)
1156 {
1157         return ptrace_setregset_or_setregs(pid);
1158 }
1159 #endif /* ptrace_setregset_or_setregs */
1160
1161 struct sysent_buf {
1162         struct tcb *tcp;
1163         struct_sysent ent;
1164         char buf[sizeof("syscall_%lu") + sizeof(kernel_ulong_t) * 3];
1165 };
1166
1167 static void
1168 free_sysent_buf(void *ptr)
1169 {
1170         struct sysent_buf *s = ptr;
1171         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1172         free(ptr);
1173 }
1174
1175 /*
1176  * Returns:
1177  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1178  *    out silently" code.
1179  * 1: ok, continue in syscall_entering_decode().
1180  * other: error, syscall_entering_decode() should print error indicator
1181  *    ("????" etc) and return an appropriate code.
1182  */
1183 int
1184 get_scno(struct tcb *tcp)
1185 {
1186         if (get_regs(tcp) < 0)
1187                 return -1;
1188
1189         int rc = arch_get_scno(tcp);
1190         if (rc != 1)
1191                 return rc;
1192
1193         if (scno_is_valid(tcp->scno)) {
1194                 tcp->s_ent = &sysent[tcp->scno];
1195                 tcp->qual_flg = qual_flags(tcp->scno);
1196         } else {
1197                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1198
1199                 s->tcp = tcp;
1200                 s->ent.nargs = MAX_ARGS;
1201                 s->ent.sen = SEN_printargs;
1202                 s->ent.sys_func = printargs;
1203                 s->ent.sys_name = s->buf;
1204                 sprintf(s->buf, "syscall_%" PRI_klu, shuffle_scno(tcp->scno));
1205
1206                 tcp->s_ent = &s->ent;
1207                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1208
1209                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1210
1211                 debug_msg("pid %d invalid syscall %" PRI_kld,
1212                           tcp->pid, tcp->scno);
1213         }
1214         return 1;
1215 }
1216
1217 #ifdef ptrace_getregset_or_getregs
1218 # define get_syscall_result_regs get_regs
1219 #else
1220 static int get_syscall_result_regs(struct tcb *);
1221 #endif
1222
1223 /* Returns:
1224  * 1: ok, continue in syscall_exiting_trace().
1225  * -1: error, syscall_exiting_trace() should print error indicator
1226  *    ("????" etc) and bail out.
1227  */
1228 static int
1229 get_syscall_result(struct tcb *tcp)
1230 {
1231         if (get_syscall_result_regs(tcp) < 0)
1232                 return -1;
1233         tcp->u_error = 0;
1234         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1235
1236         return 1;
1237 }
1238
1239 #include "get_scno.c"
1240 #include "set_scno.c"
1241 #include "get_syscall_args.c"
1242 #ifndef ptrace_getregset_or_getregs
1243 # include "get_syscall_result.c"
1244 #endif
1245 #include "get_error.c"
1246 #include "set_error.c"
1247 #ifdef HAVE_GETREGS_OLD
1248 # include "getregs_old.c"
1249 #endif
1250
1251 const char *
1252 syscall_name(kernel_ulong_t scno)
1253 {
1254 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1255         if (current_personality == X32_PERSONALITY_NUMBER)
1256                 scno &= ~__X32_SYSCALL_BIT;
1257 #endif
1258         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1259 }