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