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 #if defined SPARC64
  58 # include <asm/psrcompat.h>
  59 #elif defined SPARC
  60 # include <asm/psr.h>
  61 #endif
  62
  63 #ifdef IA64
  64 # include <asm/rse.h>
  65 #endif
  66
  67 #ifndef NT_PRSTATUS
  68 # define NT_PRSTATUS 1
  69 #endif
  70
  71 #include "syscall.h"
  72
  73 /* Define these shorthand notations to simplify the syscallent files. */
  74 #include "sysent_shorthand_defs.h"
  75
  76 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  77
  78 const struct_sysent sysent0[] = {
  79 #include "syscallent.h"
  80 };
  81
  82 #if SUPPORTED_PERSONALITIES > 1
  83 # include PERSONALITY1_INCLUDE_FUNCS
  84 static const struct_sysent sysent1[] = {
  85 # include "syscallent1.h"
  86 };
  87 #endif
  88
  89 #if SUPPORTED_PERSONALITIES > 2
  90 # include PERSONALITY2_INCLUDE_FUNCS
  91 static const struct_sysent sysent2[] = {
  92 # include "syscallent2.h"
  93 };
  94 #endif
  95
  96 /* Now undef them since short defines cause wicked namespace pollution. */
  97 #include "sysent_shorthand_undefs.h"
  98
  99 /*
 100  * `ioctlent[012].h' files are automatically generated by the auxiliary
 101  * program `ioctlsort', such that the list is sorted by the `code' field.
 102  * This has the side-effect of resolving the _IO.. macros into
 103  * plain integers, eliminating the need to include here everything
 104  * in "/usr/include".
 105  */
 106
 107 const char *const errnoent0[] = {
 108 #include "errnoent.h"
 109 };
 110 const char *const signalent0[] = {
 111 #include "signalent.h"
 112 };
 113 const struct_ioctlent ioctlent0[] = {
 114 #include "ioctlent0.h"
 115 };
 116
 117 #if SUPPORTED_PERSONALITIES > 1
 118 static const char *const errnoent1[] = {
 119 # include "errnoent1.h"
 120 };
 121 static const char *const signalent1[] = {
 122 # include "signalent1.h"
 123 };
 124 static const struct_ioctlent ioctlent1[] = {
 125 # include "ioctlent1.h"
 126 };
 127 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 128 static const struct_printers printers0 = {
 129 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 130 };
 131 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 132 static const struct_printers printers1 = {
 133 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 134 };
 135 #endif
 136
 137 #if SUPPORTED_PERSONALITIES > 2
 138 static const char *const errnoent2[] = {
 139 # include "errnoent2.h"
 140 };
 141 static const char *const signalent2[] = {
 142 # include "signalent2.h"
 143 };
 144 static const struct_ioctlent ioctlent2[] = {
 145 # include "ioctlent2.h"
 146 };
 147 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 148 static const struct_printers printers2 = {
 149 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 150 };
 151 #endif
 152
 153 enum {
 154         nsyscalls0 = ARRAY_SIZE(sysent0)
 155 #if SUPPORTED_PERSONALITIES > 1
 156         , nsyscalls1 = ARRAY_SIZE(sysent1)
 157 # if SUPPORTED_PERSONALITIES > 2
 158         , nsyscalls2 = ARRAY_SIZE(sysent2)
 159 # endif
 160 #endif
 161 };
 162
 163 enum {
 164         nerrnos0 = ARRAY_SIZE(errnoent0)
 165 #if SUPPORTED_PERSONALITIES > 1
 166         , nerrnos1 = ARRAY_SIZE(errnoent1)
 167 # if SUPPORTED_PERSONALITIES > 2
 168         , nerrnos2 = ARRAY_SIZE(errnoent2)
 169 # endif
 170 #endif
 171 };
 172
 173 enum {
 174         nsignals0 = ARRAY_SIZE(signalent0)
 175 #if SUPPORTED_PERSONALITIES > 1
 176         , nsignals1 = ARRAY_SIZE(signalent1)
 177 # if SUPPORTED_PERSONALITIES > 2
 178         , nsignals2 = ARRAY_SIZE(signalent2)
 179 # endif
 180 #endif
 181 };
 182
 183 enum {
 184         nioctlents0 = ARRAY_SIZE(ioctlent0)
 185 #if SUPPORTED_PERSONALITIES > 1
 186         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 187 # if SUPPORTED_PERSONALITIES > 2
 188         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 189 # endif
 190 #endif
 191 };
 192
 193 #if SUPPORTED_PERSONALITIES > 1
 194 const struct_sysent *sysent = sysent0;
 195 const char *const *errnoent = errnoent0;
 196 const char *const *signalent = signalent0;
 197 const struct_ioctlent *ioctlent = ioctlent0;
 198 const struct_printers *printers = &printers0;
 199 #endif
 200
 201 unsigned nsyscalls = nsyscalls0;
 202 unsigned nerrnos = nerrnos0;
 203 unsigned nsignals = nsignals0;
 204 unsigned nioctlents = nioctlents0;
 205
 206 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 207         nsyscalls0,
 208 #if SUPPORTED_PERSONALITIES > 1
 209         nsyscalls1,
 210 #endif
 211 #if SUPPORTED_PERSONALITIES > 2
 212         nsyscalls2,
 213 #endif
 214 };
 215 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 216         sysent0,
 217 #if SUPPORTED_PERSONALITIES > 1
 218         sysent1,
 219 #endif
 220 #if SUPPORTED_PERSONALITIES > 2
 221         sysent2,
 222 #endif
 223 };
 224
 225 #if SUPPORTED_PERSONALITIES > 1
 226 unsigned current_personality;
 227
 228 # ifndef current_wordsize
 229 unsigned current_wordsize;
 230 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 231         PERSONALITY0_WORDSIZE,
 232         PERSONALITY1_WORDSIZE,
 233 # if SUPPORTED_PERSONALITIES > 2
 234         PERSONALITY2_WORDSIZE,
 235 # endif
 236 };
 237 # endif
 238
 239 # ifndef current_klongsize
 240 unsigned current_klongsize;
 241 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 242         PERSONALITY0_KLONGSIZE,
 243         PERSONALITY1_KLONGSIZE,
 244 #  if SUPPORTED_PERSONALITIES > 2
 245         PERSONALITY2_KLONGSIZE,
 246 #  endif
 247 };
 248 # endif
 249
 250 void
 251 set_personality(int personality)
 252 {
 253         nsyscalls = nsyscall_vec[personality];
 254         sysent = sysent_vec[personality];
 255
 256         switch (personality) {
 257         case 0:
 258                 errnoent = errnoent0;
 259                 nerrnos = nerrnos0;
 260                 ioctlent = ioctlent0;
 261                 nioctlents = nioctlents0;
 262                 signalent = signalent0;
 263                 nsignals = nsignals0;
 264                 printers = &printers0;
 265                 break;
 266
 267         case 1:
 268                 errnoent = errnoent1;
 269                 nerrnos = nerrnos1;
 270                 ioctlent = ioctlent1;
 271                 nioctlents = nioctlents1;
 272                 signalent = signalent1;
 273                 nsignals = nsignals1;
 274                 printers = &printers1;
 275                 break;
 276
 277 # if SUPPORTED_PERSONALITIES > 2
 278         case 2:
 279                 errnoent = errnoent2;
 280                 nerrnos = nerrnos2;
 281                 ioctlent = ioctlent2;
 282                 nioctlents = nioctlents2;
 283                 signalent = signalent2;
 284                 nsignals = nsignals2;
 285                 printers = &printers2;
 286                 break;
 287 # endif
 288         }
 289
 290         current_personality = personality;
 291 # ifndef current_wordsize
 292         current_wordsize = personality_wordsize[personality];
 293 # endif
 294 # ifndef current_klongsize
 295         current_klongsize = personality_klongsize[personality];
 296 # endif
 297 }
 298
 299 static void
 300 update_personality(struct tcb *tcp, unsigned int personality)
 301 {
 302         if (personality == current_personality)
 303                 return;
 304         set_personality(personality);
 305
 306         if (personality == tcp->currpers)
 307                 return;
 308         tcp->currpers = personality;
 309
 310 # undef PERSONALITY_NAMES
 311 # if defined POWERPC64
 312 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 313 # elif defined X86_64
 314 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
 315 # elif defined X32
 316 #  define PERSONALITY_NAMES {"x32", "32 bit"}
 317 # elif defined AARCH64
 318 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 319 # elif defined TILE
 320 #  define PERSONALITY_NAMES {"64-bit", "32-bit"}
 321 # endif
 322 # ifdef PERSONALITY_NAMES
 323         if (!qflag) {
 324                 static const char *const names[] = PERSONALITY_NAMES;
 325                 error_msg("[ Process PID=%d runs in %s mode. ]",
 326                           tcp->pid, names[personality]);
 327         }
 328 # endif
 329 }
 330 #endif
 331
 332 #ifdef SYS_socket_subcall
 333 static void
 334 decode_socket_subcall(struct tcb *tcp)
 335 {
 336         const int call = tcp->u_arg[0];
 337
 338         if (call < 1 || call >= SYS_socket_nsubcalls)
 339                 return;
 340
 341         const kernel_ulong_t scno = SYS_socket_subcall + call;
 342         const unsigned int nargs = sysent[scno].nargs;
 343         uint64_t buf[nargs];
 344
 345         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 346                 return;
 347
 348         tcp->scno = scno;
 349         tcp->qual_flg = qual_flags(scno);
 350         tcp->s_ent = &sysent[scno];
 351
 352         unsigned int i;
 353         for (i = 0; i < nargs; ++i)
 354                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 355                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 356 }
 357 #endif
 358
 359 #ifdef SYS_ipc_subcall
 360 static void
 361 decode_ipc_subcall(struct tcb *tcp)
 362 {
 363         unsigned int call = tcp->u_arg[0];
 364         const unsigned int version = call >> 16;
 365
 366         if (version) {
 367 # if defined S390 || defined S390X
 368                 return;
 369 # else
 370 #  ifdef SPARC64
 371                 if (current_wordsize == 8)
 372                         return;
 373 #  endif
 374                 set_tcb_priv_ulong(tcp, version);
 375                 call &= 0xffff;
 376 # endif
 377         }
 378
 379         switch (call) {
 380                 case  1: case  2: case  3: case  4:
 381                 case 11: case 12: case 13: case 14:
 382                 case 21: case 22: case 23: case 24:
 383                         break;
 384                 default:
 385                         return;
 386         }
 387
 388         tcp->scno = SYS_ipc_subcall + call;
 389         tcp->qual_flg = qual_flags(tcp->scno);
 390         tcp->s_ent = &sysent[tcp->scno];
 391
 392         const unsigned int n = tcp->s_ent->nargs;
 393         unsigned int i;
 394         for (i = 0; i < n; i++)
 395                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 396 }
 397 #endif
 398
 399 #ifdef LINUX_MIPSO32
 400 static void
 401 decode_mips_subcall(struct tcb *tcp)
 402 {
 403         if (!scno_is_valid(tcp->u_arg[0]))
 404                 return;
 405         tcp->scno = tcp->u_arg[0];
 406         tcp->qual_flg = qual_flags(tcp->scno);
 407         tcp->s_ent = &sysent[tcp->scno];
 408         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 409                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 410         /*
 411          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 412          * and sync_file_range) requires additional code,
 413          * see linux/mips/get_syscall_args.c
 414          */
 415         if (tcp->s_ent->nargs == MAX_ARGS) {
 416                 if (umoven(tcp,
 417                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 418                            sizeof(tcp->u_arg[0]),
 419                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 420                 tcp->u_arg[MAX_ARGS - 1] = 0;
 421         }
 422 }
 423 #endif /* LINUX_MIPSO32 */
 424
 425 static void
 426 dumpio(struct tcb *tcp)
 427 {
 428         if (syserror(tcp))
 429                 return;
 430
 431         int fd = tcp->u_arg[0];
 432         if (fd < 0)
 433                 return;
 434
 435         if (is_number_in_set(fd, read_set)) {
 436                 switch (tcp->s_ent->sen) {
 437                 case SEN_read:
 438                 case SEN_pread:
 439                 case SEN_recv:
 440                 case SEN_recvfrom:
 441                 case SEN_mq_timedreceive:
 442                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 443                         return;
 444                 case SEN_readv:
 445                 case SEN_preadv:
 446                 case SEN_preadv2:
 447                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 448                                      tcp->u_rval);
 449                         return;
 450                 case SEN_recvmsg:
 451                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 452                         return;
 453                 case SEN_recvmmsg:
 454                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 455                         return;
 456                 }
 457         }
 458         if (is_number_in_set(fd, write_set)) {
 459                 switch (tcp->s_ent->sen) {
 460                 case SEN_write:
 461                 case SEN_pwrite:
 462                 case SEN_send:
 463                 case SEN_sendto:
 464                 case SEN_mq_timedsend:
 465                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 466                         break;
 467                 case SEN_writev:
 468                 case SEN_pwritev:
 469                 case SEN_pwritev2:
 470                 case SEN_vmsplice:
 471                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 472                         break;
 473                 case SEN_sendmsg:
 474                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 475                         break;
 476                 case SEN_sendmmsg:
 477                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 478                         break;
 479                 }
 480         }
 481 }
 482
 483 /*
 484  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 485  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 486  */
 487 static kernel_ulong_t
 488 shuffle_scno(kernel_ulong_t scno)
 489 {
 490 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
 491         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 492                 return scno;
 493
 494         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 495         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 496                 return 0x000ffff0;
 497         if (scno == 0x000ffff0)
 498                 return ARM_FIRST_SHUFFLED_SYSCALL;
 499
 500 # define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 501         /*
 502          * Is it ARM specific syscall?
 503          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 504          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 505          */
 506         if (scno >= 0x000f0000 &&
 507             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 508                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 509         }
 510         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 511                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 512         }
 513 #endif /* ARM || AARCH64 */
 514
 515         return scno;
 516 }
 517
 518 const char *
 519 err_name(unsigned long err)
 520 {
 521         if ((err < nerrnos) && errnoent[err])
 522                 return errnoent[err];
 523
 524         return NULL;
 525 }
 526
 527 static long get_regs_error;
 528
 529 void
 530 clear_regs(void)
 531 {
 532         get_regs_error = -1;
 533 }
 534
 535 static void get_regs(pid_t pid);
 536 static int get_syscall_args(struct tcb *);
 537 static int get_syscall_result(struct tcb *);
 538 static int arch_get_scno(struct tcb *tcp);
 539 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 540 static void get_error(struct tcb *, const bool);
 541 static int arch_set_error(struct tcb *);
 542 static int arch_set_success(struct tcb *);
 543
 544 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
 545
 546 static struct inject_opts *
 547 tcb_inject_opts(struct tcb *tcp)
 548 {
 549         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
 550                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
 551 }
 552
 553
 554 static long
 555 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
 556 {
 557         if (!tcp->inject_vec[current_personality]) {
 558                 tcp->inject_vec[current_personality] =
 559                         xcalloc(nsyscalls, sizeof(**inject_vec));
 560                 memcpy(tcp->inject_vec[current_personality],
 561                        inject_vec[current_personality],
 562                        nsyscalls * sizeof(**inject_vec));
 563         }
 564
 565         struct inject_opts *opts = tcb_inject_opts(tcp);
 566
 567         if (!opts || opts->first == 0)
 568                 return 0;
 569
 570         --opts->first;
 571
 572         if (opts->first != 0)
 573                 return 0;
 574
 575         opts->first = opts->step;
 576
 577         if (opts->signo > 0)
 578                 *signo = opts->signo;
 579         if (opts->rval != INJECT_OPTS_RVAL_DEFAULT && !arch_set_scno(tcp, -1))
 580                 tcp->flags |= TCB_TAMPERED;
 581
 582         return 0;
 583 }
 584
 585 static long
 586 tamper_with_syscall_exiting(struct tcb *tcp)
 587 {
 588         struct inject_opts *opts = tcb_inject_opts(tcp);
 589
 590         if (!opts)
 591                 return 0;
 592
 593         if (opts->rval >= 0) {
 594                 kernel_long_t u_rval = tcp->u_rval;
 595
 596                 tcp->u_rval = opts->rval;
 597                 if (arch_set_success(tcp)) {
 598                         tcp->u_rval = u_rval;
 599                 } else {
 600                         tcp->u_error = 0;
 601                 }
 602         } else {
 603                 unsigned long new_error = -opts->rval;
 604
 605                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
 606                         unsigned long u_error = tcp->u_error;
 607
 608                         tcp->u_error = new_error;
 609                         if (arch_set_error(tcp)) {
 610                                 tcp->u_error = u_error;
 611                         }
 612                 }
 613         }
 614
 615         return 0;
 616 }
 617
 618 /*
 619  * Returns:
 620  * 0: "ignore this ptrace stop", bail out silently.
 621  * 1: ok, decoded; call
 622  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
 623  * other: error; call syscall_entering_finish(tcp, res), where res is the value
 624  *    returned.
 625  */
 626 int
 627 syscall_entering_decode(struct tcb *tcp)
 628 {
 629         int res = get_scno(tcp);
 630         if (res == 0)
 631                 return res;
 632         int scno_good = res;
 633         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
 634                 printleader(tcp);
 635                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
 636                 /*
 637                  * " <unavailable>" will be added later by the code which
 638                  * detects ptrace errors.
 639                  */
 640                 return res;
 641         }
 642
 643 #ifdef LINUX_MIPSO32
 644         if (SEN_syscall == tcp->s_ent->sen)
 645                 decode_mips_subcall(tcp);
 646 #endif
 647
 648 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
 649         switch (tcp->s_ent->sen) {
 650 # ifdef SYS_socket_subcall
 651                 case SEN_socketcall:
 652                         decode_socket_subcall(tcp);
 653                         break;
 654 # endif
 655 # ifdef SYS_ipc_subcall
 656                 case SEN_ipc:
 657                         decode_ipc_subcall(tcp);
 658                         break;
 659 # endif
 660         }
 661 #endif
 662
 663         return 1;
 664 }
 665
 666 int
 667 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
 668 {
 669         /* Restrain from fault injection while the trace executes strace code. */
 670         if (hide_log(tcp)) {
 671                 tcp->qual_flg &= ~QUAL_INJECT;
 672         }
 673
 674         switch (tcp->s_ent->sen) {
 675                 case SEN_execve:
 676                 case SEN_execveat:
 677 #if defined SPARC || defined SPARC64
 678                 case SEN_execv:
 679 #endif
 680                         tcp->flags &= ~TCB_HIDE_LOG;
 681                         break;
 682         }
 683
 684         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
 685                 tcp->flags |= TCB_FILTERED;
 686                 return 0;
 687         }
 688
 689         tcp->flags &= ~TCB_FILTERED;
 690
 691         if (hide_log(tcp)) {
 692                 return 0;
 693         }
 694
 695         if (inject(tcp))
 696                 tamper_with_syscall_entering(tcp, sig);
 697
 698         if (cflag == CFLAG_ONLY_STATS) {
 699                 return 0;
 700         }
 701
 702 #ifdef USE_LIBUNWIND
 703         if (stack_trace_enabled) {
 704                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 705                         unwind_capture_stacktrace(tcp);
 706         }
 707 #endif
 708
 709         printleader(tcp);
 710         tprintf("%s(", tcp->s_ent->sys_name);
 711         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
 712         fflush(tcp->outf);
 713         return res;
 714 }
 715
 716 void
 717 syscall_entering_finish(struct tcb *tcp, int res)
 718 {
 719         tcp->flags |= TCB_INSYSCALL;
 720         tcp->sys_func_rval = res;
 721         /* Measure the entrance time as late as possible to avoid errors. */
 722         if ((Tflag || cflag) && !filtered(tcp))
 723                 gettimeofday(&tcp->etime, NULL);
 724 }
 725
 726 static bool
 727 syscall_tampered(struct tcb *tcp)
 728 {
 729         return tcp->flags & TCB_TAMPERED;
 730 }
 731
 732 /* Returns:
 733  * 0: "bail out".
 734  * 1: ok.
 735  * -1: error in one of ptrace ops.
 736  *
 737  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
 738  *    value. Anyway, call syscall_exiting_finish(tcp) then.
 739  */
 740 int
 741 syscall_exiting_decode(struct tcb *tcp, struct timeval *ptv)
 742 {
 743         /* Measure the exit time as early as possible to avoid errors. */
 744         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
 745                 gettimeofday(ptv, NULL);
 746
 747 #ifdef USE_LIBUNWIND
 748         if (stack_trace_enabled) {
 749                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 750                         unwind_cache_invalidate(tcp);
 751         }
 752 #endif
 753
 754         if (filtered(tcp) || hide_log(tcp))
 755                 return 0;
 756
 757         get_regs(tcp->pid);
 758 #if SUPPORTED_PERSONALITIES > 1
 759         update_personality(tcp, tcp->currpers);
 760 #endif
 761         return get_regs_error ? -1 : get_syscall_result(tcp);
 762 }
 763
 764 int
 765 syscall_exiting_trace(struct tcb *tcp, struct timeval tv, int res)
 766 {
 767         if (syserror(tcp) && syscall_tampered(tcp))
 768                 tamper_with_syscall_exiting(tcp);
 769
 770         if (cflag) {
 771                 count_syscall(tcp, &tv);
 772                 if (cflag == CFLAG_ONLY_STATS) {
 773                         return 0;
 774                 }
 775         }
 776
 777         /* If not in -ff mode, and printing_tcp != tcp,
 778          * then the log currently does not end with output
 779          * of _our syscall entry_, but with something else.
 780          * We need to say which syscall's return is this.
 781          *
 782          * Forced reprinting via TCB_REPRINT is used only by
 783          * "strace -ff -oLOG test/threaded_execve" corner case.
 784          * It's the only case when -ff mode needs reprinting.
 785          */
 786         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 787                 tcp->flags &= ~TCB_REPRINT;
 788                 printleader(tcp);
 789                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 790         }
 791         printing_tcp = tcp;
 792
 793         tcp->s_prev_ent = NULL;
 794         if (res != 1) {
 795                 /* There was error in one of prior ptrace ops */
 796                 tprints(") ");
 797                 tabto();
 798                 tprints("= ? <unavailable>\n");
 799                 line_ended();
 800                 return res;
 801         }
 802         tcp->s_prev_ent = tcp->s_ent;
 803
 804         int sys_res = 0;
 805         if (raw(tcp)) {
 806                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 807         } else {
 808         /* FIXME: not_failing_only (IOW, option -z) is broken:
 809          * failure of syscall is known only after syscall return.
 810          * Thus we end up with something like this on, say, ENOENT:
 811          *     open("does_not_exist", O_RDONLY <unfinished ...>
 812          *     {next syscall decode}
 813          * whereas the intended result is that open(...) line
 814          * is not shown at all.
 815          */
 816                 if (not_failing_only && tcp->u_error)
 817                         return 0;       /* ignore failed syscalls */
 818                 if (tcp->sys_func_rval & RVAL_DECODED)
 819                         sys_res = tcp->sys_func_rval;
 820                 else
 821                         sys_res = tcp->s_ent->sys_func(tcp);
 822         }
 823
 824         tprints(") ");
 825         tabto();
 826         unsigned long u_error = tcp->u_error;
 827
 828         if (raw(tcp)) {
 829                 if (u_error) {
 830                         tprintf("= -1 (errno %lu)", u_error);
 831                 } else {
 832                         tprintf("= %#" PRI_klx, tcp->u_rval);
 833                 }
 834                 if (syscall_tampered(tcp))
 835                         tprints(" (INJECTED)");
 836         } else if (!(sys_res & RVAL_NONE) && u_error) {
 837                 const char *u_error_str;
 838
 839                 switch (u_error) {
 840                 /* Blocked signals do not interrupt any syscalls.
 841                  * In this case syscalls don't return ERESTARTfoo codes.
 842                  *
 843                  * Deadly signals set to SIG_DFL interrupt syscalls
 844                  * and kill the process regardless of which of the codes below
 845                  * is returned by the interrupted syscall.
 846                  * In some cases, kernel forces a kernel-generated deadly
 847                  * signal to be unblocked and set to SIG_DFL (and thus cause
 848                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 849                  * or SIGILL. (The alternative is to leave process spinning
 850                  * forever on the faulty instruction - not useful).
 851                  *
 852                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 853                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 854                  * but kernel will always restart them.
 855                  */
 856                 case ERESTARTSYS:
 857                         /* Most common type of signal-interrupted syscall exit code.
 858                          * The system call will be restarted with the same arguments
 859                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 860                          */
 861                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 862                         break;
 863                 case ERESTARTNOINTR:
 864                         /* Rare. For example, fork() returns this if interrupted.
 865                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 866                          */
 867                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 868                         break;
 869                 case ERESTARTNOHAND:
 870                         /* pause(), rt_sigsuspend() etc use this code.
 871                          * SA_RESTART is ignored (assumed not set):
 872                          * syscall won't restart (will return EINTR instead)
 873                          * even after signal with SA_RESTART set. However,
 874                          * after SIG_IGN or SIG_DFL signal it will restart
 875                          * (thus the name "restart only if has no handler").
 876                          */
 877                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 878                         break;
 879                 case ERESTART_RESTARTBLOCK:
 880                         /* Syscalls like nanosleep(), poll() which can't be
 881                          * restarted with their original arguments use this
 882                          * code. Kernel will execute restart_syscall() instead,
 883                          * which changes arguments before restarting syscall.
 884                          * SA_RESTART is ignored (assumed not set) similarly
 885                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 886                          * since restart data is saved in "restart block"
 887                          * in task struct, and if signal handler uses a syscall
 888                          * which in turn saves another such restart block,
 889                          * old data is lost and restart becomes impossible)
 890                          */
 891                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 892                         break;
 893                 default:
 894                         u_error_str = err_name(u_error);
 895                         if (u_error_str)
 896                                 tprintf("= -1 %s (%s)",
 897                                         u_error_str, strerror(u_error));
 898                         else
 899                                 tprintf("= -1 %lu (%s)",
 900                                         u_error, strerror(u_error));
 901                         break;
 902                 }
 903                 if (syscall_tampered(tcp))
 904                         tprints(" (INJECTED)");
 905                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 906                         tprintf(" (%s)", tcp->auxstr);
 907         } else {
 908                 if (sys_res & RVAL_NONE)
 909                         tprints("= ?");
 910                 else {
 911                         switch (sys_res & RVAL_MASK) {
 912                         case RVAL_HEX:
 913 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 914                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 915                                         tprintf("= %#x",
 916                                                 (unsigned int) tcp->u_rval);
 917                                 } else
 918 #endif
 919                                 {
 920                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 921                                 }
 922                                 break;
 923                         case RVAL_OCTAL:
 924                                 tprints("= ");
 925                                 print_numeric_long_umask(tcp->u_rval);
 926                                 break;
 927                         case RVAL_UDECIMAL:
 928 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 929                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 930                                         tprintf("= %u",
 931                                                 (unsigned int) tcp->u_rval);
 932                                 } else
 933 #endif
 934                                 {
 935                                         tprintf("= %" PRI_klu, tcp->u_rval);
 936                                 }
 937                                 break;
 938                         case RVAL_DECIMAL:
 939                                 tprintf("= %" PRI_kld, tcp->u_rval);
 940                                 break;
 941                         case RVAL_FD:
 942                                 if (show_fd_path) {
 943                                         tprints("= ");
 944                                         printfd(tcp, tcp->u_rval);
 945                                 } else
 946                                         tprintf("= %" PRI_kld, tcp->u_rval);
 947                                 break;
 948                         default:
 949                                 error_msg("invalid rval format");
 950                                 break;
 951                         }
 952                 }
 953                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 954                         tprintf(" (%s)", tcp->auxstr);
 955                 if (syscall_tampered(tcp))
 956                         tprints(" (INJECTED)");
 957         }
 958         if (Tflag) {
 959                 tv_sub(&tv, &tv, &tcp->etime);
 960                 tprintf(" <%ld.%06ld>",
 961                         (long) tv.tv_sec, (long) tv.tv_usec);
 962         }
 963         tprints("\n");
 964         dumpio(tcp);
 965         line_ended();
 966
 967 #ifdef USE_LIBUNWIND
 968         if (stack_trace_enabled)
 969                 unwind_print_stacktrace(tcp);
 970 #endif
 971         return 0;
 972 }
 973
 974 void
 975 syscall_exiting_finish(struct tcb *tcp)
 976 {
 977         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
 978         tcp->sys_func_rval = 0;
 979         free_tcb_priv_data(tcp);
 980 }
 981
 982 bool
 983 is_erestart(struct tcb *tcp)
 984 {
 985         switch (tcp->u_error) {
 986                 case ERESTARTSYS:
 987                 case ERESTARTNOINTR:
 988                 case ERESTARTNOHAND:
 989                 case ERESTART_RESTARTBLOCK:
 990                         return true;
 991                 default:
 992                         return false;
 993         }
 994 }
 995
 996 static unsigned long saved_u_error;
 997
 998 void
 999 temporarily_clear_syserror(struct tcb *tcp)
1000 {
1001         saved_u_error = tcp->u_error;
1002         tcp->u_error = 0;
1003 }
1004
1005 void
1006 restore_cleared_syserror(struct tcb *tcp)
1007 {
1008         tcp->u_error = saved_u_error;
1009 }
1010
1011 #include "arch_regs.c"
1012
1013 #ifdef HAVE_GETRVAL2
1014 # include "arch_getrval2.c"
1015 #endif
1016
1017 void
1018 print_pc(struct tcb *tcp)
1019 {
1020 #if defined ARCH_PC_REG
1021 # define ARCH_GET_PC 0
1022 #elif defined ARCH_PC_PEEK_ADDR
1023         kernel_ulong_t pc;
1024 # define ARCH_PC_REG pc
1025 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1026 #else
1027 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1028 #endif
1029         get_regs(tcp->pid);
1030         if (get_regs_error || ARCH_GET_PC)
1031                 tprints(current_wordsize == 4 ? "[????????] "
1032                                               : "[????????????????] ");
1033         else
1034                 tprintf(current_wordsize == 4
1035                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1036                         (kernel_ulong_t) ARCH_PC_REG);
1037 }
1038
1039 #include "getregs_old.h"
1040
1041 #undef ptrace_getregset_or_getregs
1042 #undef ptrace_setregset_or_setregs
1043 #ifdef ARCH_REGS_FOR_GETREGSET
1044
1045 # define ptrace_getregset_or_getregs ptrace_getregset
1046 static long
1047 ptrace_getregset(pid_t pid)
1048 {
1049 # ifdef ARCH_IOVEC_FOR_GETREGSET
1050         /* variable iovec */
1051         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1052         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1053                       &ARCH_IOVEC_FOR_GETREGSET);
1054 # else
1055         /* constant iovec */
1056         static struct iovec io = {
1057                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1058                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1059         };
1060         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1061
1062 # endif
1063 }
1064
1065 # ifndef HAVE_GETREGS_OLD
1066 #  define ptrace_setregset_or_setregs ptrace_setregset
1067 static int
1068 ptrace_setregset(pid_t pid)
1069 {
1070 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1071         /* variable iovec */
1072         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1073                       &ARCH_IOVEC_FOR_GETREGSET);
1074 #  else
1075         /* constant iovec */
1076         static struct iovec io = {
1077                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1078                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1079         };
1080         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1081 #  endif
1082 }
1083 # endif /* !HAVE_GETREGS_OLD */
1084
1085 #elif defined ARCH_REGS_FOR_GETREGS
1086
1087 # define ptrace_getregset_or_getregs ptrace_getregs
1088 static long
1089 ptrace_getregs(pid_t pid)
1090 {
1091 # if defined SPARC || defined SPARC64
1092         /* SPARC systems have the meaning of data and addr reversed */
1093         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1094 # else
1095         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1096 # endif
1097 }
1098
1099 # ifndef HAVE_GETREGS_OLD
1100 #  define ptrace_setregset_or_setregs ptrace_setregs
1101 static int
1102 ptrace_setregs(pid_t pid)
1103 {
1104 #  if defined SPARC || defined SPARC64
1105         /* SPARC systems have the meaning of data and addr reversed */
1106         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1107 #  else
1108         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1109 #  endif
1110 }
1111 # endif /* !HAVE_GETREGS_OLD */
1112
1113 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1114
1115 static void
1116 get_regs(pid_t pid)
1117 {
1118 #undef USE_GET_SYSCALL_RESULT_REGS
1119 #ifdef ptrace_getregset_or_getregs
1120
1121         if (get_regs_error != -1)
1122                 return;
1123
1124 # ifdef HAVE_GETREGS_OLD
1125         /*
1126          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1127          * fallback to getregs_old.
1128          */
1129         static int use_getregs_old;
1130         if (use_getregs_old < 0) {
1131                 get_regs_error = ptrace_getregset_or_getregs(pid);
1132                 return;
1133         } else if (use_getregs_old == 0) {
1134                 get_regs_error = ptrace_getregset_or_getregs(pid);
1135                 if (get_regs_error >= 0) {
1136                         use_getregs_old = -1;
1137                         return;
1138                 }
1139                 if (errno == EPERM || errno == ESRCH)
1140                         return;
1141                 use_getregs_old = 1;
1142         }
1143         get_regs_error = getregs_old(pid);
1144 # else /* !HAVE_GETREGS_OLD */
1145         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1146         get_regs_error = ptrace_getregset_or_getregs(pid);
1147 # endif /* !HAVE_GETREGS_OLD */
1148
1149 #else /* !ptrace_getregset_or_getregs */
1150
1151 # define USE_GET_SYSCALL_RESULT_REGS 1
1152 # warning get_regs is not implemented for this architecture yet
1153         get_regs_error = 0;
1154
1155 #endif /* !ptrace_getregset_or_getregs */
1156 }
1157
1158 #ifdef ptrace_setregset_or_setregs
1159 static int
1160 set_regs(pid_t pid)
1161 {
1162         return ptrace_setregset_or_setregs(pid);
1163 }
1164 #endif /* ptrace_setregset_or_setregs */
1165
1166 struct sysent_buf {
1167         struct tcb *tcp;
1168         struct_sysent ent;
1169         char buf[sizeof("syscall_%lu") + sizeof(kernel_ulong_t) * 3];
1170 };
1171
1172 static void
1173 free_sysent_buf(void *ptr)
1174 {
1175         struct sysent_buf *s = ptr;
1176         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1177         free(ptr);
1178 }
1179
1180 /*
1181  * Returns:
1182  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1183  *    out silently" code.
1184  * 1: ok, continue in syscall_entering_decode().
1185  * other: error, syscall_entering_decode() should print error indicator
1186  *    ("????" etc) and return an appropriate code.
1187  */
1188 int
1189 get_scno(struct tcb *tcp)
1190 {
1191         get_regs(tcp->pid);
1192
1193         if (get_regs_error)
1194                 return -1;
1195
1196         int rc = arch_get_scno(tcp);
1197         if (rc != 1)
1198                 return rc;
1199
1200         if (scno_is_valid(tcp->scno)) {
1201                 tcp->s_ent = &sysent[tcp->scno];
1202                 tcp->qual_flg = qual_flags(tcp->scno);
1203         } else {
1204                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1205
1206                 s->tcp = tcp;
1207                 s->ent.nargs = MAX_ARGS;
1208                 s->ent.sen = SEN_printargs;
1209                 s->ent.sys_func = printargs;
1210                 s->ent.sys_name = s->buf;
1211                 sprintf(s->buf, "syscall_%" PRI_klu, shuffle_scno(tcp->scno));
1212
1213                 tcp->s_ent = &s->ent;
1214                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1215
1216                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1217
1218                 if (debug_flag)
1219                         error_msg("pid %d invalid syscall %" PRI_kld,
1220                                   tcp->pid, tcp->scno);
1221         }
1222         return 1;
1223 }
1224
1225 #ifdef USE_GET_SYSCALL_RESULT_REGS
1226 static int get_syscall_result_regs(struct tcb *);
1227 #endif
1228
1229 /* Returns:
1230  * 1: ok, continue in syscall_exiting_trace().
1231  * -1: error, syscall_exiting_trace() should print error indicator
1232  *    ("????" etc) and bail out.
1233  */
1234 static int
1235 get_syscall_result(struct tcb *tcp)
1236 {
1237 #ifdef USE_GET_SYSCALL_RESULT_REGS
1238         if (get_syscall_result_regs(tcp))
1239                 return -1;
1240 #endif
1241         tcp->u_error = 0;
1242         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1243
1244         return 1;
1245 }
1246
1247 #include "get_scno.c"
1248 #include "set_scno.c"
1249 #include "get_syscall_args.c"
1250 #ifdef USE_GET_SYSCALL_RESULT_REGS
1251 # include "get_syscall_result.c"
1252 #endif
1253 #include "get_error.c"
1254 #include "set_error.c"
1255 #ifdef HAVE_GETREGS_OLD
1256 # include "getregs_old.c"
1257 #endif
1258
1259 const char *
1260 syscall_name(kernel_ulong_t scno)
1261 {
1262 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1263         if (current_personality == X32_PERSONALITY_NUMBER)
1264                 scno &= ~__X32_SYSCALL_BIT;
1265 #endif
1266         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1267 }