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  * All rights reserved.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. The name of the author may not be used to endorse or promote products
  20  *    derived from this software without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include "defs.h"
  35 #include "native_defs.h"
  36 #include <sys/param.h>
  37 #include <signal.h>
  38
  39 /* for struct iovec */
  40 #include <sys/uio.h>
  41
  42 #include "regs.h"
  43 #include "ptrace.h"
  44
  45 #if defined(SPARC64)
  46 # undef PTRACE_GETREGS
  47 # define PTRACE_GETREGS PTRACE_GETREGS64
  48 # undef PTRACE_SETREGS
  49 # define PTRACE_SETREGS PTRACE_SETREGS64
  50 #endif
  51
  52 #if defined SPARC64
  53 # include <asm/psrcompat.h>
  54 #elif defined SPARC
  55 # include <asm/psr.h>
  56 #endif
  57
  58 #ifdef IA64
  59 # include <asm/rse.h>
  60 #endif
  61
  62 #ifndef NT_PRSTATUS
  63 # define NT_PRSTATUS 1
  64 #endif
  65
  66 #ifndef NSIG
  67 # warning: NSIG is not defined, using 32
  68 # define NSIG 32
  69 #endif
  70
  71 #include "syscall.h"
  72
  73 /* Define these shorthand notations to simplify the syscallent files. */
  74 #define TD TRACE_DESC
  75 #define TF TRACE_FILE
  76 #define TI TRACE_IPC
  77 #define TN TRACE_NETWORK
  78 #define TP TRACE_PROCESS
  79 #define TS TRACE_SIGNAL
  80 #define TM TRACE_MEMORY
  81 #define NF SYSCALL_NEVER_FAILS
  82 #define MA MAX_ARGS
  83 #define SI STACKTRACE_INVALIDATE_CACHE
  84 #define SE STACKTRACE_CAPTURE_ON_ENTER
  85
  86 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  87
  88 const struct_sysent sysent0[] = {
  89 #include "syscallent.h"
  90 };
  91
  92 #if SUPPORTED_PERSONALITIES > 1
  93 # include PERSONALITY1_INCLUDE_FUNCS
  94 static const struct_sysent sysent1[] = {
  95 # include "syscallent1.h"
  96 };
  97 #endif
  98
  99 #if SUPPORTED_PERSONALITIES > 2
 100 # include PERSONALITY2_INCLUDE_FUNCS
 101 static const struct_sysent sysent2[] = {
 102 # include "syscallent2.h"
 103 };
 104 #endif
 105
 106 /* Now undef them since short defines cause wicked namespace pollution. */
 107 #undef SEN
 108 #undef TD
 109 #undef TF
 110 #undef TI
 111 #undef TN
 112 #undef TP
 113 #undef TS
 114 #undef TM
 115 #undef NF
 116 #undef MA
 117 #undef SI
 118 #undef SE
 119
 120 /*
 121  * `ioctlent[012].h' files are automatically generated by the auxiliary
 122  * program `ioctlsort', such that the list is sorted by the `code' field.
 123  * This has the side-effect of resolving the _IO.. macros into
 124  * plain integers, eliminating the need to include here everything
 125  * in "/usr/include".
 126  */
 127
 128 const char *const errnoent0[] = {
 129 #include "errnoent.h"
 130 };
 131 const char *const signalent0[] = {
 132 #include "signalent.h"
 133 };
 134 const struct_ioctlent ioctlent0[] = {
 135 #include "ioctlent0.h"
 136 };
 137
 138 #if SUPPORTED_PERSONALITIES > 1
 139 static const char *const errnoent1[] = {
 140 # include "errnoent1.h"
 141 };
 142 static const char *const signalent1[] = {
 143 # include "signalent1.h"
 144 };
 145 static const struct_ioctlent ioctlent1[] = {
 146 # include "ioctlent1.h"
 147 };
 148 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 149 static const struct_printers printers0 = {
 150 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 151 };
 152 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 153 static const struct_printers printers1 = {
 154 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 155 };
 156 #endif
 157
 158 #if SUPPORTED_PERSONALITIES > 2
 159 static const char *const errnoent2[] = {
 160 # include "errnoent2.h"
 161 };
 162 static const char *const signalent2[] = {
 163 # include "signalent2.h"
 164 };
 165 static const struct_ioctlent ioctlent2[] = {
 166 # include "ioctlent2.h"
 167 };
 168 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 169 static const struct_printers printers2 = {
 170 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 171 };
 172 #endif
 173
 174 enum {
 175         nsyscalls0 = ARRAY_SIZE(sysent0)
 176 #if SUPPORTED_PERSONALITIES > 1
 177         , nsyscalls1 = ARRAY_SIZE(sysent1)
 178 # if SUPPORTED_PERSONALITIES > 2
 179         , nsyscalls2 = ARRAY_SIZE(sysent2)
 180 # endif
 181 #endif
 182 };
 183
 184 enum {
 185         nerrnos0 = ARRAY_SIZE(errnoent0)
 186 #if SUPPORTED_PERSONALITIES > 1
 187         , nerrnos1 = ARRAY_SIZE(errnoent1)
 188 # if SUPPORTED_PERSONALITIES > 2
 189         , nerrnos2 = ARRAY_SIZE(errnoent2)
 190 # endif
 191 #endif
 192 };
 193
 194 enum {
 195         nsignals0 = ARRAY_SIZE(signalent0)
 196 #if SUPPORTED_PERSONALITIES > 1
 197         , nsignals1 = ARRAY_SIZE(signalent1)
 198 # if SUPPORTED_PERSONALITIES > 2
 199         , nsignals2 = ARRAY_SIZE(signalent2)
 200 # endif
 201 #endif
 202 };
 203
 204 enum {
 205         nioctlents0 = ARRAY_SIZE(ioctlent0)
 206 #if SUPPORTED_PERSONALITIES > 1
 207         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 208 # if SUPPORTED_PERSONALITIES > 2
 209         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 210 # endif
 211 #endif
 212 };
 213
 214 #if SUPPORTED_PERSONALITIES > 1
 215 const struct_sysent *sysent = sysent0;
 216 const char *const *errnoent = errnoent0;
 217 const char *const *signalent = signalent0;
 218 const struct_ioctlent *ioctlent = ioctlent0;
 219 const struct_printers *printers = &printers0;
 220 #endif
 221
 222 unsigned nsyscalls = nsyscalls0;
 223 unsigned nerrnos = nerrnos0;
 224 unsigned nsignals = nsignals0;
 225 unsigned nioctlents = nioctlents0;
 226
 227 unsigned num_quals;
 228 qualbits_t *qual_vec[SUPPORTED_PERSONALITIES];
 229
 230 static const unsigned nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 231         nsyscalls0,
 232 #if SUPPORTED_PERSONALITIES > 1
 233         nsyscalls1,
 234 #endif
 235 #if SUPPORTED_PERSONALITIES > 2
 236         nsyscalls2,
 237 #endif
 238 };
 239 static const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 240         sysent0,
 241 #if SUPPORTED_PERSONALITIES > 1
 242         sysent1,
 243 #endif
 244 #if SUPPORTED_PERSONALITIES > 2
 245         sysent2,
 246 #endif
 247 };
 248
 249 enum {
 250         MAX_NSYSCALLS1 = (nsyscalls0
 251 #if SUPPORTED_PERSONALITIES > 1
 252                         > nsyscalls1 ? nsyscalls0 : nsyscalls1
 253 #endif
 254                         ),
 255         MAX_NSYSCALLS2 = (MAX_NSYSCALLS1
 256 #if SUPPORTED_PERSONALITIES > 2
 257                         > nsyscalls2 ? MAX_NSYSCALLS1 : nsyscalls2
 258 #endif
 259                         ),
 260         MAX_NSYSCALLS = MAX_NSYSCALLS2,
 261         /* We are ready for arches with up to 255 signals,
 262          * even though the largest known signo is on MIPS and it is 128.
 263          * The number of existing syscalls on all arches is
 264          * larger that 255 anyway, so it is just a pedantic matter.
 265          */
 266         MIN_QUALS = MAX_NSYSCALLS > 255 ? MAX_NSYSCALLS : 255
 267 };
 268
 269 #if SUPPORTED_PERSONALITIES > 1
 270 unsigned current_personality;
 271
 272 # ifndef current_wordsize
 273 unsigned current_wordsize;
 274 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 275         PERSONALITY0_WORDSIZE,
 276         PERSONALITY1_WORDSIZE,
 277 # if SUPPORTED_PERSONALITIES > 2
 278         PERSONALITY2_WORDSIZE,
 279 # endif
 280 };
 281 # endif
 282
 283 void
 284 set_personality(int personality)
 285 {
 286         nsyscalls = nsyscall_vec[personality];
 287         sysent = sysent_vec[personality];
 288
 289         switch (personality) {
 290         case 0:
 291                 errnoent = errnoent0;
 292                 nerrnos = nerrnos0;
 293                 ioctlent = ioctlent0;
 294                 nioctlents = nioctlents0;
 295                 signalent = signalent0;
 296                 nsignals = nsignals0;
 297                 printers = &printers0;
 298                 break;
 299
 300         case 1:
 301                 errnoent = errnoent1;
 302                 nerrnos = nerrnos1;
 303                 ioctlent = ioctlent1;
 304                 nioctlents = nioctlents1;
 305                 signalent = signalent1;
 306                 nsignals = nsignals1;
 307                 printers = &printers1;
 308                 break;
 309
 310 # if SUPPORTED_PERSONALITIES > 2
 311         case 2:
 312                 errnoent = errnoent2;
 313                 nerrnos = nerrnos2;
 314                 ioctlent = ioctlent2;
 315                 nioctlents = nioctlents2;
 316                 signalent = signalent2;
 317                 nsignals = nsignals2;
 318                 printers = &printers2;
 319                 break;
 320 # endif
 321         }
 322
 323         current_personality = personality;
 324 # ifndef current_wordsize
 325         current_wordsize = personality_wordsize[personality];
 326 # endif
 327 }
 328
 329 static void
 330 update_personality(struct tcb *tcp, unsigned int personality)
 331 {
 332         if (personality == current_personality)
 333                 return;
 334         set_personality(personality);
 335
 336         if (personality == tcp->currpers)
 337                 return;
 338         tcp->currpers = personality;
 339
 340 # undef PERSONALITY_NAMES
 341 # if defined POWERPC64
 342 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 343 # elif defined X86_64
 344 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
 345 # elif defined X32
 346 #  define PERSONALITY_NAMES {"x32", "32 bit"}
 347 # elif defined AARCH64
 348 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 349 # elif defined TILE
 350 #  define PERSONALITY_NAMES {"64-bit", "32-bit"}
 351 # endif
 352 # ifdef PERSONALITY_NAMES
 353         if (!qflag) {
 354                 static const char *const names[] = PERSONALITY_NAMES;
 355                 error_msg("[ Process PID=%d runs in %s mode. ]",
 356                           tcp->pid, names[personality]);
 357         }
 358 # endif
 359 }
 360 #endif
 361
 362 static int qual_desc(const char *, unsigned int, int);
 363 static int qual_fault(const char *, unsigned int, int);
 364 static int qual_signal(const char *, unsigned int, int);
 365 static int qual_syscall(const char *, unsigned int, int);
 366
 367 static const struct qual_options {
 368         unsigned int bitflag;
 369         const char *option_name;
 370         int (*qualify)(const char *, unsigned int, int);
 371         const char *argument_name;
 372 } qual_options[] = {
 373         { QUAL_TRACE,   "trace",        qual_syscall,   "system call"   },
 374         { QUAL_TRACE,   "t",            qual_syscall,   "system call"   },
 375         { QUAL_ABBREV,  "abbrev",       qual_syscall,   "system call"   },
 376         { QUAL_ABBREV,  "a",            qual_syscall,   "system call"   },
 377         { QUAL_VERBOSE, "verbose",      qual_syscall,   "system call"   },
 378         { QUAL_VERBOSE, "v",            qual_syscall,   "system call"   },
 379         { QUAL_RAW,     "raw",          qual_syscall,   "system call"   },
 380         { QUAL_RAW,     "x",            qual_syscall,   "system call"   },
 381         { QUAL_SIGNAL,  "signal",       qual_signal,    "signal"        },
 382         { QUAL_SIGNAL,  "signals",      qual_signal,    "signal"        },
 383         { QUAL_SIGNAL,  "s",            qual_signal,    "signal"        },
 384         { QUAL_READ,    "read",         qual_desc,      "descriptor"    },
 385         { QUAL_READ,    "reads",        qual_desc,      "descriptor"    },
 386         { QUAL_READ,    "r",            qual_desc,      "descriptor"    },
 387         { QUAL_WRITE,   "write",        qual_desc,      "descriptor"    },
 388         { QUAL_WRITE,   "writes",       qual_desc,      "descriptor"    },
 389         { QUAL_WRITE,   "w",            qual_desc,      "descriptor"    },
 390         { QUAL_FAULT,   "fault",        qual_fault,     "fault argument"},
 391         { 0,            NULL,           NULL,           NULL            },
 392 };
 393
 394 static void
 395 reallocate_vec(void **vec, unsigned int old_nmemb,
 396                size_t size, unsigned int new_nmemb)
 397 {
 398         unsigned int p;
 399
 400         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 401                 vec[p] = xreallocarray(vec[p], new_nmemb, size);
 402                 memset(vec[p] + size * old_nmemb, 0,
 403                        (new_nmemb - old_nmemb) * size);
 404         }
 405 }
 406
 407 static void
 408 reallocate_qual(const unsigned int n)
 409 {
 410         reallocate_vec((void **) qual_vec, num_quals, sizeof(qualbits_t), n);
 411         num_quals = n;
 412 }
 413
 414 struct fault_opts {
 415         uint16_t first;
 416         uint16_t step;
 417         uint16_t err;
 418 };
 419
 420 static unsigned int num_faults;
 421 static struct fault_opts *fault_vec[SUPPORTED_PERSONALITIES];
 422
 423 static inline void
 424 reallocate_fault(const unsigned int n)
 425 {
 426         reallocate_vec((void **) fault_vec, num_faults,
 427                        sizeof(struct fault_opts), n);
 428         num_faults = n;
 429 }
 430
 431 static void
 432 qualify_one(const unsigned int n, unsigned int bitflag, const int not,
 433             const int pers, const struct fault_opts *fopts)
 434 {
 435         int p;
 436
 437         if (num_quals <= n) {
 438                 reallocate_qual(n + 1);
 439                 reallocate_fault(n + 1);
 440         }
 441
 442         for (p = 0; p < SUPPORTED_PERSONALITIES; p++) {
 443                 if (pers == p || pers < 0) {
 444                         if (not)
 445                                 qual_vec[p][n] &= ~bitflag;
 446                         else {
 447                                 qual_vec[p][n] |= bitflag;
 448                                 if (fopts)
 449                                         memcpy(&fault_vec[p][n], fopts,
 450                                                sizeof(*fopts));
 451                         }
 452                 }
 453         }
 454 }
 455
 456 static bool
 457 qualify_scno(const char *const s, const unsigned int bitflag,
 458              const int not, const struct fault_opts *const fopts)
 459 {
 460         int i = string_to_uint_upto(s, MAX_NSYSCALLS - 1);
 461         if (i < 0)
 462                 return false;
 463
 464         qualify_one(i, bitflag, not, -1, fopts);
 465         return true;
 466 }
 467
 468 static int
 469 lookup_class(const char *s)
 470 {
 471         if (strcmp(s, "all") == 0)
 472                 return 0;
 473         if (strcmp(s, "file") == 0)
 474                 return TRACE_FILE;
 475         if (strcmp(s, "ipc") == 0)
 476                 return TRACE_IPC;
 477         if (strcmp(s, "network") == 0)
 478                 return TRACE_NETWORK;
 479         if (strcmp(s, "process") == 0)
 480                 return TRACE_PROCESS;
 481         if (strcmp(s, "signal") == 0)
 482                 return TRACE_SIGNAL;
 483         if (strcmp(s, "desc") == 0)
 484                 return TRACE_DESC;
 485         if (strcmp(s, "memory") == 0)
 486                 return TRACE_MEMORY;
 487         return -1;
 488 }
 489
 490 static bool
 491 qualify_syscall_class(const char *const s, const unsigned int bitflag,
 492                       const int not, const struct fault_opts *const fopts)
 493 {
 494         unsigned int p;
 495         const int n = lookup_class(s);
 496
 497         if (n < 0)
 498                 return false;
 499
 500         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 501                 unsigned int i;
 502
 503                 for (i = 0; i < nsyscall_vec[p]; ++i) {
 504                         if (sysent_vec[p][i].sys_name
 505                             && (sysent_vec[p][i].sys_flags & n) == n) {
 506                                 qualify_one(i, bitflag, not, p, fopts);
 507                         }
 508                 }
 509         }
 510
 511         return true;
 512 }
 513
 514 static bool
 515 qualify_syscall_name(const char *const s, const unsigned int bitflag,
 516                      const int not, const struct fault_opts *const fopts)
 517 {
 518         bool found = false;
 519         unsigned int p;
 520
 521         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 522                 unsigned int i;
 523
 524                 for (i = 0; i < nsyscall_vec[p]; ++i) {
 525                         if (sysent_vec[p][i].sys_name
 526                             && strcmp(s, sysent_vec[p][i].sys_name) == 0) {
 527                                 qualify_one(i, bitflag, not, p, fopts);
 528                                 found = true;
 529                         }
 530                 }
 531         }
 532
 533         return found;
 534 }
 535
 536 static int
 537 qual_syscall_ex(const char *const s, const unsigned int bitflag,
 538                 const int not, const struct fault_opts *const fopts)
 539 {
 540         if (qualify_scno(s, bitflag, not, fopts)
 541             || qualify_syscall_class(s, bitflag, not, fopts)
 542             || qualify_syscall_name(s, bitflag, not, fopts)) {
 543                 return 0;
 544         }
 545
 546         return -1;
 547 }
 548
 549 static int
 550 qual_syscall(const char *const s, const unsigned int bitflag, const int not)
 551 {
 552         return qual_syscall_ex(s, bitflag, not, NULL);
 553 }
 554
 555 /*
 556  * Returns NULL if STR does not start with PREFIX,
 557  * or a pointer to the first char in STR after PREFIX.
 558  */
 559 static const char *
 560 strip_prefix(const char *prefix, const char *str)
 561 {
 562         size_t len = strlen(prefix);
 563
 564         return (len > strlen(str) || memcmp(prefix, str, len))
 565                ? NULL : str + len;
 566 }
 567
 568 static int
 569 find_errno_by_name(const char *name)
 570 {
 571         unsigned int i;
 572
 573         for (i = 1; i < nerrnos; ++i) {
 574                 if (errnoent[i] && (strcmp(name, errnoent[i]) == 0))
 575                         return i;
 576         }
 577
 578         return -1;
 579 }
 580
 581 static bool
 582 parse_fault_token(const char *const token, struct fault_opts *const fopts)
 583 {
 584         const char *val;
 585         int intval;
 586
 587         if ((val = strip_prefix("when=", token))) {
 588                 /*
 589                  *      == 1+1
 590                  * F    == F+0
 591                  * F+   == F+1
 592                  * F+S
 593                  */
 594                 char *end;
 595                 intval = string_to_uint_ex(val, &end, 0xffff, "+");
 596                 if (intval < 1)
 597                         return false;
 598
 599                 fopts->first = intval;
 600
 601                 if (*end) {
 602                         val = end + 1;
 603                         if (*val) {
 604                                 /* F+S */
 605                                 intval = string_to_uint_upto(val, 0xffff);
 606                                 if (intval < 1)
 607                                         return false;
 608                                 fopts->step = intval;
 609                         } else {
 610                                 /* F+ == F+1 */
 611                                 fopts->step = 1;
 612                         }
 613                 } else {
 614                         /* F == F+0 */
 615                         fopts->step = 0;
 616                 }
 617         } else if ((val = strip_prefix("error=", token))) {
 618                 intval = string_to_uint_upto(val, 4095);
 619                 if (intval < 0)
 620                         intval = find_errno_by_name(val);
 621                 if (intval < 1)
 622                         return false;
 623                 fopts->err = intval;
 624         } else {
 625                 return false;
 626         }
 627
 628         return true;
 629 }
 630
 631 static const char *
 632 parse_fault_expression(const char *const s, char **buf,
 633                        struct fault_opts *const fopts)
 634 {
 635         const char *name;
 636         const char *token;
 637         char *saveptr = NULL;
 638
 639         *buf = xstrdup(s);
 640         name = strtok_r(*buf, ":", &saveptr);
 641         if (!name || !*name)
 642                 goto parse_error;
 643
 644         while ((token = strtok_r(NULL, ":", &saveptr))) {
 645                 if (!parse_fault_token(token, fopts))
 646                         goto parse_error;
 647         }
 648
 649         return name;
 650
 651 parse_error:
 652         free(*buf);
 653         return *buf = NULL;
 654 }
 655
 656 static int
 657 qual_fault(const char *const s, const unsigned int bitflag, const int not)
 658 {
 659         struct fault_opts opts = {
 660                 .first = 1,
 661                 .step = 1,
 662                 .err = 0
 663         };
 664
 665         char *buf = NULL;
 666         const char *name = parse_fault_expression(s, &buf, &opts);
 667
 668         if (!name)
 669                 return -1;
 670
 671         int rc = qual_syscall_ex(name, bitflag, not, &opts);
 672
 673         free(buf);
 674         return rc;
 675 }
 676
 677 static int
 678 qual_signal(const char *s, const unsigned int bitflag, const int not)
 679 {
 680         unsigned int i;
 681
 682         if (*s >= '0' && *s <= '9') {
 683                 int signo = string_to_uint_upto(s, 255);
 684                 if (signo < 0)
 685                         return -1;
 686                 qualify_one(signo, bitflag, not, -1, NULL);
 687                 return 0;
 688         }
 689         if (strncasecmp(s, "SIG", 3) == 0)
 690                 s += 3;
 691         for (i = 0; i <= NSIG; i++) {
 692                 if (strcasecmp(s, signame(i) + 3) == 0) {
 693                         qualify_one(i, bitflag, not, -1, NULL);
 694                         return 0;
 695                 }
 696         }
 697         return -1;
 698 }
 699
 700 static int
 701 qual_desc(const char *s, const unsigned int bitflag, const int not)
 702 {
 703         int desc = string_to_uint_upto(s, 0x7fff);
 704         if (desc < 0)
 705                 return -1;
 706         qualify_one(desc, bitflag, not, -1, NULL);
 707         return 0;
 708 }
 709
 710 void
 711 qualify(const char *s)
 712 {
 713         const struct qual_options *opt;
 714         char *copy;
 715         const char *p;
 716         int not;
 717         unsigned int i;
 718
 719         if (num_quals == 0) {
 720                 reallocate_qual(MIN_QUALS);
 721                 reallocate_fault(MIN_QUALS);
 722         }
 723
 724         opt = &qual_options[0];
 725         for (i = 0; (p = qual_options[i].option_name); i++) {
 726                 unsigned int len = strlen(p);
 727                 if (strncmp(s, p, len) == 0 && s[len] == '=') {
 728                         opt = &qual_options[i];
 729                         s += len + 1;
 730                         break;
 731                 }
 732         }
 733         not = 0;
 734         if (*s == '!') {
 735                 not = 1;
 736                 s++;
 737         }
 738         if (strcmp(s, "none") == 0) {
 739                 not = 1 - not;
 740                 s = "all";
 741         }
 742         if (opt->bitflag != QUAL_FAULT && strcmp(s, "all") == 0) {
 743                 for (i = 0; i < num_quals; i++) {
 744                         qualify_one(i, opt->bitflag, not, -1, NULL);
 745                 }
 746                 return;
 747         }
 748         for (i = 0; i < num_quals; i++) {
 749                 qualify_one(i, opt->bitflag, !not, -1, NULL);
 750         }
 751         copy = xstrdup(s);
 752         for (p = strtok(copy, ","); p; p = strtok(NULL, ",")) {
 753                 if (opt->qualify(p, opt->bitflag, not)) {
 754                         error_msg_and_die("invalid %s '%s'",
 755                                 opt->argument_name, p);
 756                 }
 757         }
 758         free(copy);
 759         return;
 760 }
 761
 762 #ifdef SYS_socket_subcall
 763 static void
 764 decode_socket_subcall(struct tcb *tcp)
 765 {
 766         const int call = tcp->u_arg[0];
 767
 768         if (call < 1 || call >= SYS_socket_nsubcalls)
 769                 return;
 770
 771         const unsigned long scno = SYS_socket_subcall + call;
 772         const unsigned int nargs = sysent[scno].nargs;
 773         uint64_t buf[nargs];
 774
 775         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 776                 return;
 777
 778         tcp->scno = scno;
 779         tcp->qual_flg = qual_flags[scno];
 780         tcp->s_ent = &sysent[scno];
 781
 782         unsigned int i;
 783         for (i = 0; i < nargs; ++i)
 784                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 785                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 786 }
 787 #endif
 788
 789 #ifdef SYS_ipc_subcall
 790 static void
 791 decode_ipc_subcall(struct tcb *tcp)
 792 {
 793         unsigned int call = tcp->u_arg[0];
 794         const unsigned int version = call >> 16;
 795
 796         if (version) {
 797 # if defined S390 || defined S390X
 798                 return;
 799 # else
 800 #  ifdef SPARC64
 801                 if (current_wordsize == 8)
 802                         return;
 803 #  endif
 804                 set_tcb_priv_ulong(tcp, version);
 805                 call &= 0xffff;
 806 # endif
 807         }
 808
 809         switch (call) {
 810                 case  1: case  2: case  3: case  4:
 811                 case 11: case 12: case 13: case 14:
 812                 case 21: case 22: case 23: case 24:
 813                         break;
 814                 default:
 815                         return;
 816         }
 817
 818         tcp->scno = SYS_ipc_subcall + call;
 819         tcp->qual_flg = qual_flags[tcp->scno];
 820         tcp->s_ent = &sysent[tcp->scno];
 821
 822         const unsigned int n = tcp->s_ent->nargs;
 823         unsigned int i;
 824         for (i = 0; i < n; i++)
 825                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 826 }
 827 #endif
 828
 829 #ifdef LINUX_MIPSO32
 830 static void
 831 decode_mips_subcall(struct tcb *tcp)
 832 {
 833         if (!SCNO_IS_VALID(tcp->u_arg[0]))
 834                 return;
 835         tcp->scno = tcp->u_arg[0];
 836         tcp->qual_flg = qual_flags[tcp->scno];
 837         tcp->s_ent = &sysent[tcp->scno];
 838         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 839                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 840         /*
 841          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 842          * and sync_file_range) requires additional code,
 843          * see linux/mips/get_syscall_args.c
 844          */
 845         if (tcp->s_ent->nargs == MAX_ARGS) {
 846                 if (umoven(tcp,
 847                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 848                            sizeof(tcp->u_arg[0]),
 849                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 850                 tcp->u_arg[MAX_ARGS - 1] = 0;
 851         }
 852 }
 853 #endif /* LINUX_MIPSO32 */
 854
 855 static void
 856 dumpio(struct tcb *tcp)
 857 {
 858         int sen;
 859
 860         if (syserror(tcp))
 861                 return;
 862         if ((unsigned long) tcp->u_arg[0] >= num_quals)
 863                 return;
 864         sen = tcp->s_ent->sen;
 865         if (SEN_printargs == sen)
 866                 return;
 867         if (qual_flags[tcp->u_arg[0]] & QUAL_READ) {
 868                 switch (sen) {
 869                 case SEN_read:
 870                 case SEN_pread:
 871                 case SEN_recv:
 872                 case SEN_recvfrom:
 873                 case SEN_mq_timedreceive:
 874                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 875                         return;
 876                 case SEN_readv:
 877                 case SEN_preadv:
 878                 case SEN_preadv2:
 879                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 880                                      tcp->u_rval);
 881                         return;
 882                 case SEN_recvmsg:
 883                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 884                         return;
 885                 case SEN_recvmmsg:
 886                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 887                         return;
 888                 }
 889         }
 890         if (qual_flags[tcp->u_arg[0]] & QUAL_WRITE) {
 891                 switch (sen) {
 892                 case SEN_write:
 893                 case SEN_pwrite:
 894                 case SEN_send:
 895                 case SEN_sendto:
 896                 case SEN_mq_timedsend:
 897                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 898                         break;
 899                 case SEN_writev:
 900                 case SEN_pwritev:
 901                 case SEN_pwritev2:
 902                 case SEN_vmsplice:
 903                         dumpiov(tcp, tcp->u_arg[2], tcp->u_arg[1]);
 904                         break;
 905                 case SEN_sendmsg:
 906                         dumpiov_in_msghdr(tcp, tcp->u_arg[1],
 907                                           (unsigned long) -1L);
 908                         break;
 909                 case SEN_sendmmsg:
 910                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 911                         break;
 912                 }
 913         }
 914 }
 915
 916 /*
 917  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 918  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 919  */
 920 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
 921 static long
 922 shuffle_scno(unsigned long scno)
 923 {
 924         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 925                 return scno;
 926
 927         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 928         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 929                 return 0x000ffff0;
 930         if (scno == 0x000ffff0)
 931                 return ARM_FIRST_SHUFFLED_SYSCALL;
 932
 933 #define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 934         /*
 935          * Is it ARM specific syscall?
 936          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 937          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 938          */
 939         if (scno >= 0x000f0000 &&
 940             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 941                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 942         }
 943         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 944                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 945         }
 946
 947         return scno;
 948 }
 949 #else
 950 # define shuffle_scno(scno) ((long)(scno))
 951 #endif
 952
 953 const char *
 954 syscall_name(long scno)
 955 {
 956         static char buf[sizeof("syscall_%lu") + sizeof(long)*3];
 957
 958         if (SCNO_IS_VALID(scno))
 959                 return sysent[scno].sys_name;
 960         else {
 961                 sprintf(buf, "syscall_%lu", scno);
 962                 return buf;
 963         }
 964 }
 965
 966 const char *
 967 err_name(unsigned long err)
 968 {
 969         if ((err < nerrnos) && errnoent[err])
 970                 return errnoent[err];
 971
 972         return NULL;
 973 }
 974
 975 static long get_regs_error;
 976
 977 void
 978 clear_regs(void)
 979 {
 980         get_regs_error = -1;
 981 }
 982
 983 static int get_syscall_args(struct tcb *);
 984 static int get_syscall_result(struct tcb *);
 985 static int arch_get_scno(struct tcb *tcp);
 986 static int arch_set_scno(struct tcb *, long);
 987 static void get_error(struct tcb *, const bool);
 988 static int arch_set_error(struct tcb *);
 989
 990 static struct fault_opts *
 991 tcb_fault_opts(struct tcb *tcp)
 992 {
 993         return (SCNO_IN_RANGE(tcp->scno) && tcp->fault_vec[current_personality])
 994                ? &tcp->fault_vec[current_personality][tcp->scno] : NULL;
 995 }
 996
 997
 998 static long
 999 inject_syscall_fault_entering(struct tcb *tcp)
1000 {
1001         if (!tcp->fault_vec[current_personality]) {
1002                 tcp->fault_vec[current_personality] =
1003                         xreallocarray(NULL, num_faults,
1004                                       sizeof(struct fault_opts));
1005                 memcpy(tcp->fault_vec[current_personality],
1006                        fault_vec[current_personality],
1007                        num_faults * sizeof(struct fault_opts));
1008         }
1009
1010         struct fault_opts *opts = tcb_fault_opts(tcp);
1011
1012         if (opts->first == 0)
1013                 return 0;
1014
1015         --opts->first;
1016
1017         if (opts->first != 0)
1018                 return 0;
1019
1020         opts->first = opts->step;
1021
1022         if (!arch_set_scno(tcp, -1))
1023                 tcp->flags |= TCB_FAULT_INJ;
1024
1025         return 0;
1026 }
1027
1028 static long
1029 update_syscall_fault_exiting(struct tcb *tcp)
1030 {
1031         struct fault_opts *opts = tcb_fault_opts(tcp);
1032
1033         if (opts && opts->err && tcp->u_error != opts->err) {
1034                 unsigned long u_error = tcp->u_error;
1035                 tcp->u_error = opts->err;
1036                 if (arch_set_error(tcp))
1037                         tcp->u_error = u_error;
1038         }
1039
1040         return 0;
1041 }
1042
1043 static int
1044 trace_syscall_entering(struct tcb *tcp)
1045 {
1046         int res, scno_good;
1047
1048         scno_good = res = get_scno(tcp);
1049         if (res == 0)
1050                 return res;
1051         if (res == 1)
1052                 res = get_syscall_args(tcp);
1053
1054         if (res != 1) {
1055                 printleader(tcp);
1056                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
1057                 /*
1058                  * " <unavailable>" will be added later by the code which
1059                  * detects ptrace errors.
1060                  */
1061                 goto ret;
1062         }
1063
1064 #ifdef LINUX_MIPSO32
1065         if (SEN_syscall == tcp->s_ent->sen)
1066                 decode_mips_subcall(tcp);
1067 #endif
1068
1069         if (   SEN_execve == tcp->s_ent->sen
1070 # if defined(SPARC) || defined(SPARC64)
1071             || SEN_execv == tcp->s_ent->sen
1072 # endif
1073            ) {
1074                 hide_log_until_execve = 0;
1075         }
1076
1077 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
1078         switch (tcp->s_ent->sen) {
1079 # ifdef SYS_socket_subcall
1080                 case SEN_socketcall:
1081                         decode_socket_subcall(tcp);
1082                         break;
1083 # endif
1084 # ifdef SYS_ipc_subcall
1085                 case SEN_ipc:
1086                         decode_ipc_subcall(tcp);
1087                         break;
1088 # endif
1089         }
1090 #endif
1091
1092         if (!(tcp->qual_flg & QUAL_TRACE)
1093          || (tracing_paths && !pathtrace_match(tcp))
1094         ) {
1095                 tcp->flags |= TCB_INSYSCALL | TCB_FILTERED;
1096                 tcp->sys_func_rval = 0;
1097                 return 0;
1098         }
1099
1100         tcp->flags &= ~TCB_FILTERED;
1101
1102         if (hide_log_until_execve) {
1103                 res = 0;
1104                 goto ret;
1105         }
1106
1107         if (tcp->qual_flg & QUAL_FAULT)
1108                 inject_syscall_fault_entering(tcp);
1109
1110         if (cflag == CFLAG_ONLY_STATS) {
1111                 res = 0;
1112                 goto ret;
1113         }
1114
1115 #ifdef USE_LIBUNWIND
1116         if (stack_trace_enabled) {
1117                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
1118                         unwind_capture_stacktrace(tcp);
1119         }
1120 #endif
1121
1122         printleader(tcp);
1123         tprintf("%s(", tcp->s_ent->sys_name);
1124         if ((tcp->qual_flg & QUAL_RAW) && SEN_exit != tcp->s_ent->sen)
1125                 res = printargs(tcp);
1126         else
1127                 res = tcp->s_ent->sys_func(tcp);
1128
1129         fflush(tcp->outf);
1130  ret:
1131         tcp->flags |= TCB_INSYSCALL;
1132         tcp->sys_func_rval = res;
1133         /* Measure the entrance time as late as possible to avoid errors. */
1134         if (Tflag || cflag)
1135                 gettimeofday(&tcp->etime, NULL);
1136         return res;
1137 }
1138
1139 static bool
1140 syscall_fault_injected(struct tcb *tcp)
1141 {
1142         return tcp->flags & TCB_FAULT_INJ;
1143 }
1144
1145 static int
1146 trace_syscall_exiting(struct tcb *tcp)
1147 {
1148         int sys_res;
1149         struct timeval tv;
1150         int res;
1151         unsigned long u_error;
1152         const char *u_error_str;
1153
1154         /* Measure the exit time as early as possible to avoid errors. */
1155         if (Tflag || cflag)
1156                 gettimeofday(&tv, NULL);
1157
1158 #ifdef USE_LIBUNWIND
1159         if (stack_trace_enabled) {
1160                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
1161                         unwind_cache_invalidate(tcp);
1162         }
1163 #endif
1164
1165 #if SUPPORTED_PERSONALITIES > 1
1166         update_personality(tcp, tcp->currpers);
1167 #endif
1168         res = (get_regs_error ? -1 : get_syscall_result(tcp));
1169         if (filtered(tcp) || hide_log_until_execve)
1170                 goto ret;
1171
1172         if (syserror(tcp) && syscall_fault_injected(tcp))
1173                 update_syscall_fault_exiting(tcp);
1174
1175         if (cflag) {
1176                 count_syscall(tcp, &tv);
1177                 if (cflag == CFLAG_ONLY_STATS) {
1178                         goto ret;
1179                 }
1180         }
1181
1182         /* If not in -ff mode, and printing_tcp != tcp,
1183          * then the log currently does not end with output
1184          * of _our syscall entry_, but with something else.
1185          * We need to say which syscall's return is this.
1186          *
1187          * Forced reprinting via TCB_REPRINT is used only by
1188          * "strace -ff -oLOG test/threaded_execve" corner case.
1189          * It's the only case when -ff mode needs reprinting.
1190          */
1191         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
1192                 tcp->flags &= ~TCB_REPRINT;
1193                 printleader(tcp);
1194                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
1195         }
1196         printing_tcp = tcp;
1197
1198         tcp->s_prev_ent = NULL;
1199         if (res != 1) {
1200                 /* There was error in one of prior ptrace ops */
1201                 tprints(") ");
1202                 tabto();
1203                 tprints("= ? <unavailable>\n");
1204                 line_ended();
1205                 tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
1206                 tcp->sys_func_rval = 0;
1207                 free_tcb_priv_data(tcp);
1208                 return res;
1209         }
1210         tcp->s_prev_ent = tcp->s_ent;
1211
1212         sys_res = 0;
1213         if (tcp->qual_flg & QUAL_RAW) {
1214                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
1215         } else {
1216         /* FIXME: not_failing_only (IOW, option -z) is broken:
1217          * failure of syscall is known only after syscall return.
1218          * Thus we end up with something like this on, say, ENOENT:
1219          *     open("doesnt_exist", O_RDONLY <unfinished ...>
1220          *     {next syscall decode}
1221          * whereas the intended result is that open(...) line
1222          * is not shown at all.
1223          */
1224                 if (not_failing_only && tcp->u_error)
1225                         goto ret;       /* ignore failed syscalls */
1226                 if (tcp->sys_func_rval & RVAL_DECODED)
1227                         sys_res = tcp->sys_func_rval;
1228                 else
1229                         sys_res = tcp->s_ent->sys_func(tcp);
1230         }
1231
1232         tprints(") ");
1233         tabto();
1234         u_error = tcp->u_error;
1235
1236         if (tcp->qual_flg & QUAL_RAW) {
1237                 if (u_error) {
1238                         tprintf("= -1 (errno %lu)", u_error);
1239                         if (syscall_fault_injected(tcp))
1240                                 tprints(" (INJECTED)");
1241                 } else {
1242                         tprintf("= %#lx", tcp->u_rval);
1243                 }
1244         }
1245         else if (!(sys_res & RVAL_NONE) && u_error) {
1246                 switch (u_error) {
1247                 /* Blocked signals do not interrupt any syscalls.
1248                  * In this case syscalls don't return ERESTARTfoo codes.
1249                  *
1250                  * Deadly signals set to SIG_DFL interrupt syscalls
1251                  * and kill the process regardless of which of the codes below
1252                  * is returned by the interrupted syscall.
1253                  * In some cases, kernel forces a kernel-generated deadly
1254                  * signal to be unblocked and set to SIG_DFL (and thus cause
1255                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
1256                  * or SIGILL. (The alternative is to leave process spinning
1257                  * forever on the faulty instruction - not useful).
1258                  *
1259                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
1260                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
1261                  * but kernel will always restart them.
1262                  */
1263                 case ERESTARTSYS:
1264                         /* Most common type of signal-interrupted syscall exit code.
1265                          * The system call will be restarted with the same arguments
1266                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
1267                          */
1268                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
1269                         break;
1270                 case ERESTARTNOINTR:
1271                         /* Rare. For example, fork() returns this if interrupted.
1272                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
1273                          */
1274                         tprints("= ? ERESTARTNOINTR (To be restarted)");
1275                         break;
1276                 case ERESTARTNOHAND:
1277                         /* pause(), rt_sigsuspend() etc use this code.
1278                          * SA_RESTART is ignored (assumed not set):
1279                          * syscall won't restart (will return EINTR instead)
1280                          * even after signal with SA_RESTART set. However,
1281                          * after SIG_IGN or SIG_DFL signal it will restart
1282                          * (thus the name "restart only if has no handler").
1283                          */
1284                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
1285                         break;
1286                 case ERESTART_RESTARTBLOCK:
1287                         /* Syscalls like nanosleep(), poll() which can't be
1288                          * restarted with their original arguments use this
1289                          * code. Kernel will execute restart_syscall() instead,
1290                          * which changes arguments before restarting syscall.
1291                          * SA_RESTART is ignored (assumed not set) similarly
1292                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
1293                          * since restart data is saved in "restart block"
1294                          * in task struct, and if signal handler uses a syscall
1295                          * which in turn saves another such restart block,
1296                          * old data is lost and restart becomes impossible)
1297                          */
1298                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
1299                         break;
1300                 default:
1301                         u_error_str = err_name(u_error);
1302                         if (u_error_str)
1303                                 tprintf("= -1 %s (%s)",
1304                                         u_error_str, strerror(u_error));
1305                         else
1306                                 tprintf("= -1 %lu (%s)",
1307                                         u_error, strerror(u_error));
1308                         break;
1309                 }
1310                 if (syscall_fault_injected(tcp))
1311                         tprintf(" (INJECTED)");
1312                 if ((sys_res & RVAL_STR) && tcp->auxstr)
1313                         tprintf(" (%s)", tcp->auxstr);
1314         }
1315         else {
1316                 if (sys_res & RVAL_NONE)
1317                         tprints("= ?");
1318                 else {
1319                         switch (sys_res & RVAL_MASK) {
1320                         case RVAL_HEX:
1321 #if SUPPORTED_PERSONALITIES > 1
1322                                 if (current_wordsize < sizeof(long))
1323                                         tprintf("= %#x",
1324                                                 (unsigned int) tcp->u_rval);
1325                                 else
1326 #endif
1327                                         tprintf("= %#lx", tcp->u_rval);
1328                                 break;
1329                         case RVAL_OCTAL:
1330                                 tprints("= ");
1331                                 print_numeric_long_umask(tcp->u_rval);
1332                                 break;
1333                         case RVAL_UDECIMAL:
1334 #if SUPPORTED_PERSONALITIES > 1
1335                                 if (current_wordsize < sizeof(long))
1336                                         tprintf("= %u",
1337                                                 (unsigned int) tcp->u_rval);
1338                                 else
1339 #endif
1340                                         tprintf("= %lu", tcp->u_rval);
1341                                 break;
1342                         case RVAL_DECIMAL:
1343                                 tprintf("= %ld", tcp->u_rval);
1344                                 break;
1345                         case RVAL_FD:
1346                                 if (show_fd_path) {
1347                                         tprints("= ");
1348                                         printfd(tcp, tcp->u_rval);
1349                                 }
1350                                 else
1351                                         tprintf("= %ld", tcp->u_rval);
1352                                 break;
1353 #if HAVE_STRUCT_TCB_EXT_ARG
1354                         /*
1355                         case RVAL_LHEX:
1356                                 tprintf("= %#llx", tcp->u_lrval);
1357                                 break;
1358                         case RVAL_LOCTAL:
1359                                 tprintf("= %#llo", tcp->u_lrval);
1360                                 break;
1361                         */
1362                         case RVAL_LUDECIMAL:
1363                                 tprintf("= %llu", tcp->u_lrval);
1364                                 break;
1365                         /*
1366                         case RVAL_LDECIMAL:
1367                                 tprintf("= %lld", tcp->u_lrval);
1368                                 break;
1369                         */
1370 #endif /* HAVE_STRUCT_TCB_EXT_ARG */
1371                         default:
1372                                 error_msg("invalid rval format");
1373                                 break;
1374                         }
1375                 }
1376                 if ((sys_res & RVAL_STR) && tcp->auxstr)
1377                         tprintf(" (%s)", tcp->auxstr);
1378         }
1379         if (Tflag) {
1380                 tv_sub(&tv, &tv, &tcp->etime);
1381                 tprintf(" <%ld.%06ld>",
1382                         (long) tv.tv_sec, (long) tv.tv_usec);
1383         }
1384         tprints("\n");
1385         dumpio(tcp);
1386         line_ended();
1387
1388 #ifdef USE_LIBUNWIND
1389         if (stack_trace_enabled)
1390                 unwind_print_stacktrace(tcp);
1391 #endif
1392
1393  ret:
1394         tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
1395         tcp->sys_func_rval = 0;
1396         free_tcb_priv_data(tcp);
1397         return 0;
1398 }
1399
1400 int
1401 trace_syscall(struct tcb *tcp)
1402 {
1403         return exiting(tcp) ?
1404                 trace_syscall_exiting(tcp) : trace_syscall_entering(tcp);
1405 }
1406
1407 bool
1408 is_erestart(struct tcb *tcp)
1409 {
1410         switch (tcp->u_error) {
1411                 case ERESTARTSYS:
1412                 case ERESTARTNOINTR:
1413                 case ERESTARTNOHAND:
1414                 case ERESTART_RESTARTBLOCK:
1415                         return true;
1416                 default:
1417                         return false;
1418         }
1419 }
1420
1421 static unsigned long saved_u_error;
1422
1423 void
1424 temporarily_clear_syserror(struct tcb *tcp)
1425 {
1426         saved_u_error = tcp->u_error;
1427         tcp->u_error = 0;
1428 }
1429
1430 void
1431 restore_cleared_syserror(struct tcb *tcp)
1432 {
1433         tcp->u_error = saved_u_error;
1434 }
1435
1436 #include "kernel_types.h"
1437
1438 /*
1439  * Check the syscall return value register value for whether it is
1440  * a negated errno code indicating an error, or a success return value.
1441  */
1442 static inline bool
1443 is_negated_errno(kernel_ulong_t val)
1444 {
1445         /* Linux kernel defines MAX_ERRNO to 4095. */
1446         kernel_ulong_t max = -(kernel_long_t) 4095;
1447
1448 #if defined X86_64 || defined X32
1449         /*
1450          * current_wordsize is 4 for x32 personality
1451          * but truncation _must not_ be done in it, so
1452          * check current_personality instead.
1453          */
1454         if (current_personality == 1) {
1455                 val = (uint32_t) val;
1456                 max = (uint32_t) max;
1457         }
1458 #elif SUPPORTED_PERSONALITIES > 1 && SIZEOF_LONG > 4
1459         if (current_wordsize < sizeof(val)) {
1460                 val = (uint32_t) val;
1461                 max = (uint32_t) max;
1462         }
1463 #endif
1464
1465         return val >= max;
1466 }
1467
1468 #include "arch_regs.c"
1469
1470 #ifdef HAVE_GETRVAL2
1471 # include "arch_getrval2.c"
1472 #endif
1473
1474 void
1475 print_pc(struct tcb *tcp)
1476 {
1477 #if defined ARCH_PC_REG
1478 # define ARCH_GET_PC 0
1479 #elif defined ARCH_PC_PEEK_ADDR
1480         long pc;
1481 # define ARCH_PC_REG pc
1482 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1483 #else
1484 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1485 #endif
1486         if (get_regs_error || ARCH_GET_PC)
1487                 tprints(current_wordsize == 4 ? "[????????] "
1488                                               : "[????????????????] ");
1489         else
1490                 tprintf(current_wordsize == 4 ? "[%08lx] " : "[%016lx] ",
1491                         (unsigned long) ARCH_PC_REG);
1492 }
1493
1494 #include "getregs_old.h"
1495
1496 #undef ptrace_getregset_or_getregs
1497 #undef ptrace_setregset_or_setregs
1498 #ifdef ARCH_REGS_FOR_GETREGSET
1499
1500 # define ptrace_getregset_or_getregs ptrace_getregset
1501 static long
1502 ptrace_getregset(pid_t pid)
1503 {
1504 # ifdef ARCH_IOVEC_FOR_GETREGSET
1505         /* variable iovec */
1506         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1507         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1508                       &ARCH_IOVEC_FOR_GETREGSET);
1509 # else
1510         /* constant iovec */
1511         static struct iovec io = {
1512                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1513                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1514         };
1515         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1516
1517 # endif
1518 }
1519
1520 # ifndef HAVE_GETREGS_OLD
1521 #  define ptrace_setregset_or_setregs ptrace_setregset
1522 static int
1523 ptrace_setregset(pid_t pid)
1524 {
1525 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1526         /* variable iovec */
1527         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1528                       &ARCH_IOVEC_FOR_GETREGSET);
1529 #  else
1530         /* constant iovec */
1531         static struct iovec io = {
1532                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1533                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1534         };
1535         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1536 #  endif
1537 }
1538 # endif /* !HAVE_GETREGS_OLD */
1539
1540 #elif defined ARCH_REGS_FOR_GETREGS
1541
1542 # define ptrace_getregset_or_getregs ptrace_getregs
1543 static long
1544 ptrace_getregs(pid_t pid)
1545 {
1546 # if defined SPARC || defined SPARC64
1547         /* SPARC systems have the meaning of data and addr reversed */
1548         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1549 # else
1550         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1551 # endif
1552 }
1553
1554 # ifndef HAVE_GETREGS_OLD
1555 #  define ptrace_setregset_or_setregs ptrace_setregs
1556 static int
1557 ptrace_setregs(pid_t pid)
1558 {
1559 #  if defined SPARC || defined SPARC64
1560         /* SPARC systems have the meaning of data and addr reversed */
1561         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1562 #  else
1563         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1564 #  endif
1565 }
1566 # endif /* !HAVE_GETREGS_OLD */
1567
1568 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1569
1570 void
1571 get_regs(pid_t pid)
1572 {
1573 #undef USE_GET_SYSCALL_RESULT_REGS
1574 #ifdef ptrace_getregset_or_getregs
1575
1576 # ifdef HAVE_GETREGS_OLD
1577         /*
1578          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1579          * fallback to getregs_old.
1580          */
1581         static int use_getregs_old;
1582         if (use_getregs_old < 0) {
1583                 get_regs_error = ptrace_getregset_or_getregs(pid);
1584                 return;
1585         } else if (use_getregs_old == 0) {
1586                 get_regs_error = ptrace_getregset_or_getregs(pid);
1587                 if (get_regs_error >= 0) {
1588                         use_getregs_old = -1;
1589                         return;
1590                 }
1591                 if (errno == EPERM || errno == ESRCH)
1592                         return;
1593                 use_getregs_old = 1;
1594         }
1595         get_regs_error = getregs_old(pid);
1596 # else /* !HAVE_GETREGS_OLD */
1597         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1598         get_regs_error = ptrace_getregset_or_getregs(pid);
1599 # endif /* !HAVE_GETREGS_OLD */
1600
1601 #else /* !ptrace_getregset_or_getregs */
1602
1603 # define USE_GET_SYSCALL_RESULT_REGS 1
1604 # warning get_regs is not implemented for this architecture yet
1605         get_regs_error = 0;
1606
1607 #endif /* !ptrace_getregset_or_getregs */
1608 }
1609
1610 #ifdef ptrace_setregset_or_setregs
1611 static int
1612 set_regs(pid_t pid)
1613 {
1614         return ptrace_setregset_or_setregs(pid);
1615 }
1616 #endif /* ptrace_setregset_or_setregs */
1617
1618 struct sysent_buf {
1619         struct tcb *tcp;
1620         struct_sysent ent;
1621         char buf[sizeof("syscall_%lu") + sizeof(long) * 3];
1622 };
1623
1624 static void
1625 free_sysent_buf(void *ptr)
1626 {
1627         struct sysent_buf *s = ptr;
1628         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1629         free(ptr);
1630 }
1631
1632 /*
1633  * Returns:
1634  * 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
1635  * 1: ok, continue in trace_syscall_entering().
1636  * other: error, trace_syscall_entering() should print error indicator
1637  *    ("????" etc) and bail out.
1638  */
1639 int
1640 get_scno(struct tcb *tcp)
1641 {
1642         if (get_regs_error)
1643                 return -1;
1644
1645         int rc = arch_get_scno(tcp);
1646         if (rc != 1)
1647                 return rc;
1648
1649         if (SCNO_IS_VALID(tcp->scno)) {
1650                 tcp->s_ent = &sysent[tcp->scno];
1651                 tcp->qual_flg = qual_flags[tcp->scno];
1652         } else {
1653                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1654
1655                 s->tcp = tcp;
1656                 s->ent.nargs = MAX_ARGS;
1657                 s->ent.sen = SEN_printargs;
1658                 s->ent.sys_func = printargs;
1659                 s->ent.sys_name = s->buf;
1660                 sprintf(s->buf, "syscall_%lu", shuffle_scno(tcp->scno));
1661
1662                 tcp->s_ent = &s->ent;
1663                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1664
1665                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1666
1667                 if (debug_flag)
1668                         error_msg("pid %d invalid syscall %ld", tcp->pid, tcp->scno);
1669         }
1670         return 1;
1671 }
1672
1673 #ifdef USE_GET_SYSCALL_RESULT_REGS
1674 static int get_syscall_result_regs(struct tcb *);
1675 #endif
1676
1677 /* Returns:
1678  * 1: ok, continue in trace_syscall_exiting().
1679  * -1: error, trace_syscall_exiting() should print error indicator
1680  *    ("????" etc) and bail out.
1681  */
1682 static int
1683 get_syscall_result(struct tcb *tcp)
1684 {
1685 #ifdef USE_GET_SYSCALL_RESULT_REGS
1686         if (get_syscall_result_regs(tcp))
1687                 return -1;
1688 #endif
1689         tcp->u_error = 0;
1690         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1691
1692         return 1;
1693 }
1694
1695 #include "get_scno.c"
1696 #include "set_scno.c"
1697 #include "get_syscall_args.c"
1698 #ifdef USE_GET_SYSCALL_RESULT_REGS
1699 # include "get_syscall_result.c"
1700 #endif
1701 #include "get_error.c"
1702 #include "set_error.c"
1703 #ifdef HAVE_GETREGS_OLD
1704 # include "getregs_old.c"
1705 #endif