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_fault(const char *, unsigned int, int);
 363 static int qual_signal(const char *, unsigned int, int);
 364 static int qual_syscall(const char *, unsigned int, int);
 365
 366 static const struct qual_options {
 367         unsigned int bitflag;
 368         const char *option_name;
 369         int (*qualify)(const char *, unsigned int, int);
 370         const char *argument_name;
 371 } qual_options[] = {
 372         { QUAL_TRACE,   "trace",        qual_syscall,   "system call"   },
 373         { QUAL_TRACE,   "t",            qual_syscall,   "system call"   },
 374         { QUAL_ABBREV,  "abbrev",       qual_syscall,   "system call"   },
 375         { QUAL_ABBREV,  "a",            qual_syscall,   "system call"   },
 376         { QUAL_VERBOSE, "verbose",      qual_syscall,   "system call"   },
 377         { QUAL_VERBOSE, "v",            qual_syscall,   "system call"   },
 378         { QUAL_RAW,     "raw",          qual_syscall,   "system call"   },
 379         { QUAL_RAW,     "x",            qual_syscall,   "system call"   },
 380         { QUAL_SIGNAL,  "signal",       qual_signal,    "signal"        },
 381         { QUAL_SIGNAL,  "signals",      qual_signal,    "signal"        },
 382         { QUAL_SIGNAL,  "s",            qual_signal,    "signal"        },
 383         { QUAL_READ,    "read",         NULL,           "descriptor"    },
 384         { QUAL_READ,    "reads",        NULL,           "descriptor"    },
 385         { QUAL_READ,    "r",            NULL,           "descriptor"    },
 386         { QUAL_WRITE,   "write",        NULL,           "descriptor"    },
 387         { QUAL_WRITE,   "writes",       NULL,           "descriptor"    },
 388         { QUAL_WRITE,   "w",            NULL,           "descriptor"    },
 389         { QUAL_FAULT,   "fault",        qual_fault,     "fault argument"},
 390         { 0,            NULL,           NULL,           NULL            },
 391 };
 392
 393 static void
 394 reallocate_vec(void **vec, unsigned int old_nmemb,
 395                size_t size, unsigned int new_nmemb)
 396 {
 397         unsigned int p;
 398
 399         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 400                 vec[p] = xreallocarray(vec[p], new_nmemb, size);
 401                 memset(vec[p] + size * old_nmemb, 0,
 402                        (new_nmemb - old_nmemb) * size);
 403         }
 404 }
 405
 406 static void
 407 reallocate_qual(const unsigned int n)
 408 {
 409         reallocate_vec((void **) qual_vec, num_quals, sizeof(qualbits_t), n);
 410         num_quals = n;
 411 }
 412
 413 struct fault_opts {
 414         uint16_t first;
 415         uint16_t step;
 416         uint16_t err;
 417 };
 418
 419 static unsigned int num_faults;
 420 static struct fault_opts *fault_vec[SUPPORTED_PERSONALITIES];
 421
 422 static inline void
 423 reallocate_fault(const unsigned int n)
 424 {
 425         reallocate_vec((void **) fault_vec, num_faults,
 426                        sizeof(struct fault_opts), n);
 427         num_faults = n;
 428 }
 429
 430 static void
 431 qualify_one(const unsigned int n, unsigned int bitflag, const int not,
 432             const int pers, const struct fault_opts *fopts)
 433 {
 434         int p;
 435
 436         if (num_quals <= n) {
 437                 reallocate_qual(n + 1);
 438                 reallocate_fault(n + 1);
 439         }
 440
 441         for (p = 0; p < SUPPORTED_PERSONALITIES; p++) {
 442                 if (pers == p || pers < 0) {
 443                         if (not)
 444                                 qual_vec[p][n] &= ~bitflag;
 445                         else {
 446                                 qual_vec[p][n] |= bitflag;
 447                                 if (fopts)
 448                                         memcpy(&fault_vec[p][n], fopts,
 449                                                sizeof(*fopts));
 450                         }
 451                 }
 452         }
 453 }
 454
 455 static bool
 456 qualify_scno(const char *const s, const unsigned int bitflag,
 457              const int not, const struct fault_opts *const fopts)
 458 {
 459         int n = string_to_uint_upto(s, MAX_NSYSCALLS - 1);
 460         if (n < 0)
 461                 return false;
 462
 463         if (not && fopts) {
 464                 /* set bitflag for all syscall numbers except n */
 465                 unsigned int p;
 466                 for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 467                         unsigned int i;
 468
 469                         for (i = 0; i < nsyscall_vec[p]; ++i) {
 470                                 if (i != (unsigned int) n
 471                                     && sysent_vec[p][i].sys_name) {
 472                                         qualify_one(i, bitflag, 0, p, fopts);
 473                                 }
 474                         }
 475                 }
 476         } else {
 477                 qualify_one(n, bitflag, not, -1, fopts);
 478         }
 479
 480         return true;
 481 }
 482
 483 static int
 484 lookup_class(const char *s)
 485 {
 486         if (strcmp(s, "all") == 0)
 487                 return 0;
 488         if (strcmp(s, "file") == 0)
 489                 return TRACE_FILE;
 490         if (strcmp(s, "ipc") == 0)
 491                 return TRACE_IPC;
 492         if (strcmp(s, "network") == 0)
 493                 return TRACE_NETWORK;
 494         if (strcmp(s, "process") == 0)
 495                 return TRACE_PROCESS;
 496         if (strcmp(s, "signal") == 0)
 497                 return TRACE_SIGNAL;
 498         if (strcmp(s, "desc") == 0)
 499                 return TRACE_DESC;
 500         if (strcmp(s, "memory") == 0)
 501                 return TRACE_MEMORY;
 502         return -1;
 503 }
 504
 505 static bool
 506 qualify_syscall_class(const char *const s, const unsigned int bitflag,
 507                       const int not, const struct fault_opts *const fopts)
 508 {
 509         unsigned int p;
 510         const int n = lookup_class(s);
 511
 512         if (n < 0)
 513                 return false;
 514
 515         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 516                 unsigned int i;
 517
 518                 for (i = 0; i < nsyscall_vec[p]; ++i) {
 519                         if (!sysent_vec[p][i].sys_name)
 520                                 continue;
 521                         const bool match = (sysent_vec[p][i].sys_flags & n) == n;
 522                         if (match ^ (not && fopts)) {
 523                                 qualify_one(i, bitflag, not && !fopts, p, fopts);
 524                         }
 525                 }
 526         }
 527
 528         return true;
 529 }
 530
 531 static bool
 532 qualify_syscall_name(const char *const s, const unsigned int bitflag,
 533                      const int not, const struct fault_opts *const fopts)
 534 {
 535         bool found = false;
 536         unsigned int p;
 537
 538         for (p = 0; p < SUPPORTED_PERSONALITIES; ++p) {
 539                 unsigned int i;
 540
 541                 for (i = 0; i < nsyscall_vec[p]; ++i) {
 542                         if (!sysent_vec[p][i].sys_name)
 543                                 continue;
 544                         const bool match = !strcmp(s, sysent_vec[p][i].sys_name);
 545                         found = found || match;
 546                         if (match ^ (not && fopts)) {
 547                                 qualify_one(i, bitflag, not && !fopts, p, fopts);
 548                         }
 549                 }
 550         }
 551
 552         return found;
 553 }
 554
 555 static int
 556 qual_syscall_ex(const char *const s, const unsigned int bitflag,
 557                 const int not, const struct fault_opts *const fopts)
 558 {
 559         if (qualify_scno(s, bitflag, not, fopts)
 560             || qualify_syscall_class(s, bitflag, not, fopts)
 561             || qualify_syscall_name(s, bitflag, not, fopts)) {
 562                 return 0;
 563         }
 564
 565         return -1;
 566 }
 567
 568 static int
 569 qual_syscall(const char *const s, const unsigned int bitflag, const int not)
 570 {
 571         return qual_syscall_ex(s, bitflag, not, NULL);
 572 }
 573
 574 /*
 575  * Returns NULL if STR does not start with PREFIX,
 576  * or a pointer to the first char in STR after PREFIX.
 577  */
 578 static const char *
 579 strip_prefix(const char *prefix, const char *str)
 580 {
 581         size_t len = strlen(prefix);
 582
 583         return strncmp(prefix, str, len) ? NULL : str + len;
 584 }
 585
 586 static int
 587 find_errno_by_name(const char *name)
 588 {
 589         unsigned int i;
 590
 591         for (i = 1; i < nerrnos; ++i) {
 592                 if (errnoent[i] && (strcmp(name, errnoent[i]) == 0))
 593                         return i;
 594         }
 595
 596         return -1;
 597 }
 598
 599 static bool
 600 parse_fault_token(const char *const token, struct fault_opts *const fopts)
 601 {
 602         const char *val;
 603         int intval;
 604
 605         if ((val = strip_prefix("when=", token))) {
 606                 /*
 607                  *      == 1+1
 608                  * F    == F+0
 609                  * F+   == F+1
 610                  * F+S
 611                  */
 612                 char *end;
 613                 intval = string_to_uint_ex(val, &end, 0xffff, "+");
 614                 if (intval < 1)
 615                         return false;
 616
 617                 fopts->first = intval;
 618
 619                 if (*end) {
 620                         val = end + 1;
 621                         if (*val) {
 622                                 /* F+S */
 623                                 intval = string_to_uint_upto(val, 0xffff);
 624                                 if (intval < 1)
 625                                         return false;
 626                                 fopts->step = intval;
 627                         } else {
 628                                 /* F+ == F+1 */
 629                                 fopts->step = 1;
 630                         }
 631                 } else {
 632                         /* F == F+0 */
 633                         fopts->step = 0;
 634                 }
 635         } else if ((val = strip_prefix("error=", token))) {
 636                 intval = string_to_uint_upto(val, 4095);
 637                 if (intval < 0)
 638                         intval = find_errno_by_name(val);
 639                 if (intval < 1)
 640                         return false;
 641                 fopts->err = intval;
 642         } else {
 643                 return false;
 644         }
 645
 646         return true;
 647 }
 648
 649 static char *
 650 parse_fault_expression(const char *const s, char **buf,
 651                        struct fault_opts *const fopts)
 652 {
 653         char *saveptr = NULL;
 654         char *name = NULL;
 655         char *token;
 656
 657         *buf = xstrdup(s);
 658         for (token = strtok_r(*buf, ":", &saveptr); token;
 659              token = strtok_r(NULL, ":", &saveptr)) {
 660                 if (!name)
 661                         name = token;
 662                 else if (!parse_fault_token(token, fopts))
 663                         goto parse_error;
 664         }
 665
 666         if (name)
 667                 return name;
 668
 669 parse_error:
 670         free(*buf);
 671         return *buf = NULL;
 672 }
 673
 674 static int
 675 qual_fault(const char *const s, const unsigned int bitflag, const int not)
 676 {
 677         struct fault_opts opts = {
 678                 .first = 1,
 679                 .step = 1,
 680                 .err = 0
 681         };
 682
 683         char *buf = NULL;
 684         char *name = parse_fault_expression(s, &buf, &opts);
 685         char *saveptr = NULL;
 686         const char *token;
 687         int rc = -1;
 688
 689         if (!name)
 690                 return -1;
 691
 692         for (token = strtok_r(name, ",", &saveptr); token;
 693              token = strtok_r(NULL, ",", &saveptr)) {
 694                 rc = qual_syscall_ex(token, bitflag, not, &opts);
 695                 if (rc)
 696                         break;
 697         }
 698
 699         free(buf);
 700         return rc;
 701 }
 702
 703 static int
 704 qual_signal(const char *s, const unsigned int bitflag, const int not)
 705 {
 706         int i;
 707
 708         if (*s >= '0' && *s <= '9') {
 709                 i = string_to_uint_upto(s, 255);
 710                 if (i < 0)
 711                         return -1;
 712                 qualify_one(i, bitflag, not, -1, NULL);
 713                 return 0;
 714         }
 715         if (strncasecmp(s, "SIG", 3) == 0)
 716                 s += 3;
 717         for (i = 0; i <= NSIG; ++i) {
 718                 const char *name = signame(i);
 719                 if (strncasecmp(name, "SIG", 3) != 0)
 720                         continue;
 721                 name += 3;
 722
 723                 if (strcasecmp(name, s) != 0)
 724                         continue;
 725                 qualify_one(i, bitflag, not, -1, NULL);
 726                 return 0;
 727         }
 728         return -1;
 729 }
 730
 731 void
 732 qualify(const char *s)
 733 {
 734         const struct qual_options *opt;
 735         char *copy;
 736         const char *p;
 737         int not;
 738         unsigned int i;
 739
 740         if (num_quals == 0) {
 741                 reallocate_qual(MIN_QUALS);
 742                 reallocate_fault(MIN_QUALS);
 743         }
 744
 745         opt = &qual_options[0];
 746         for (i = 0; (p = qual_options[i].option_name); i++) {
 747                 unsigned int len = strlen(p);
 748                 if (strncmp(s, p, len) == 0 && s[len] == '=') {
 749                         opt = &qual_options[i];
 750                         s += len + 1;
 751                         break;
 752                 }
 753         }
 754
 755         switch (opt->bitflag) {
 756                 case QUAL_READ:
 757                         qualify_read(s);
 758                         return;
 759                 case QUAL_WRITE:
 760                         qualify_write(s);
 761                         return;
 762         }
 763
 764         not = 0;
 765         if (*s == '!') {
 766                 not = 1;
 767                 s++;
 768         }
 769         if (strcmp(s, "none") == 0) {
 770                 not = 1 - not;
 771                 s = "all";
 772         }
 773         if (opt->bitflag == QUAL_FAULT) {
 774                 if (opt->qualify(s, opt->bitflag, not)) {
 775                         error_msg_and_die("invalid %s '%s'",
 776                                 opt->argument_name, s);
 777                 }
 778                 return;
 779         }
 780         if (strcmp(s, "all") == 0) {
 781                 for (i = 0; i < num_quals; ++i) {
 782                         qualify_one(i, opt->bitflag, not, -1, NULL);
 783                 }
 784                 return;
 785         }
 786         for (i = 0; i < num_quals; ++i) {
 787                 qualify_one(i, opt->bitflag, !not, -1, NULL);
 788         }
 789         copy = xstrdup(s);
 790         for (p = strtok(copy, ","); p; p = strtok(NULL, ",")) {
 791                 if (opt->qualify(p, opt->bitflag, not)) {
 792                         error_msg_and_die("invalid %s '%s'",
 793                                 opt->argument_name, p);
 794                 }
 795         }
 796         free(copy);
 797         return;
 798 }
 799
 800 #ifdef SYS_socket_subcall
 801 static void
 802 decode_socket_subcall(struct tcb *tcp)
 803 {
 804         const int call = tcp->u_arg[0];
 805
 806         if (call < 1 || call >= SYS_socket_nsubcalls)
 807                 return;
 808
 809         const unsigned long scno = SYS_socket_subcall + call;
 810         const unsigned int nargs = sysent[scno].nargs;
 811         uint64_t buf[nargs];
 812
 813         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 814                 return;
 815
 816         tcp->scno = scno;
 817         tcp->qual_flg = qual_flags[scno];
 818         tcp->s_ent = &sysent[scno];
 819
 820         unsigned int i;
 821         for (i = 0; i < nargs; ++i)
 822                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 823                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 824 }
 825 #endif
 826
 827 #ifdef SYS_ipc_subcall
 828 static void
 829 decode_ipc_subcall(struct tcb *tcp)
 830 {
 831         unsigned int call = tcp->u_arg[0];
 832         const unsigned int version = call >> 16;
 833
 834         if (version) {
 835 # if defined S390 || defined S390X
 836                 return;
 837 # else
 838 #  ifdef SPARC64
 839                 if (current_wordsize == 8)
 840                         return;
 841 #  endif
 842                 set_tcb_priv_ulong(tcp, version);
 843                 call &= 0xffff;
 844 # endif
 845         }
 846
 847         switch (call) {
 848                 case  1: case  2: case  3: case  4:
 849                 case 11: case 12: case 13: case 14:
 850                 case 21: case 22: case 23: case 24:
 851                         break;
 852                 default:
 853                         return;
 854         }
 855
 856         tcp->scno = SYS_ipc_subcall + call;
 857         tcp->qual_flg = qual_flags[tcp->scno];
 858         tcp->s_ent = &sysent[tcp->scno];
 859
 860         const unsigned int n = tcp->s_ent->nargs;
 861         unsigned int i;
 862         for (i = 0; i < n; i++)
 863                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 864 }
 865 #endif
 866
 867 #ifdef LINUX_MIPSO32
 868 static void
 869 decode_mips_subcall(struct tcb *tcp)
 870 {
 871         if (!SCNO_IS_VALID(tcp->u_arg[0]))
 872                 return;
 873         tcp->scno = tcp->u_arg[0];
 874         tcp->qual_flg = qual_flags[tcp->scno];
 875         tcp->s_ent = &sysent[tcp->scno];
 876         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 877                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 878         /*
 879          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 880          * and sync_file_range) requires additional code,
 881          * see linux/mips/get_syscall_args.c
 882          */
 883         if (tcp->s_ent->nargs == MAX_ARGS) {
 884                 if (umoven(tcp,
 885                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 886                            sizeof(tcp->u_arg[0]),
 887                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 888                 tcp->u_arg[MAX_ARGS - 1] = 0;
 889         }
 890 }
 891 #endif /* LINUX_MIPSO32 */
 892
 893 static void
 894 dumpio(struct tcb *tcp)
 895 {
 896         if (syserror(tcp))
 897                 return;
 898
 899         int fd = tcp->u_arg[0];
 900         if (fd < 0)
 901                 return;
 902
 903         if (is_number_in_set(fd, &read_set)) {
 904                 switch (tcp->s_ent->sen) {
 905                 case SEN_read:
 906                 case SEN_pread:
 907                 case SEN_recv:
 908                 case SEN_recvfrom:
 909                 case SEN_mq_timedreceive:
 910                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 911                         return;
 912                 case SEN_readv:
 913                 case SEN_preadv:
 914                 case SEN_preadv2:
 915                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 916                                      tcp->u_rval);
 917                         return;
 918                 case SEN_recvmsg:
 919                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 920                         return;
 921                 case SEN_recvmmsg:
 922                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 923                         return;
 924                 }
 925         }
 926         if (is_number_in_set(fd, &write_set)) {
 927                 switch (tcp->s_ent->sen) {
 928                 case SEN_write:
 929                 case SEN_pwrite:
 930                 case SEN_send:
 931                 case SEN_sendto:
 932                 case SEN_mq_timedsend:
 933                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 934                         break;
 935                 case SEN_writev:
 936                 case SEN_pwritev:
 937                 case SEN_pwritev2:
 938                 case SEN_vmsplice:
 939                         dumpiov(tcp, tcp->u_arg[2], tcp->u_arg[1]);
 940                         break;
 941                 case SEN_sendmsg:
 942                         dumpiov_in_msghdr(tcp, tcp->u_arg[1],
 943                                           (unsigned long) -1L);
 944                         break;
 945                 case SEN_sendmmsg:
 946                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 947                         break;
 948                 }
 949         }
 950 }
 951
 952 /*
 953  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 954  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 955  */
 956 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
 957 static long
 958 shuffle_scno(unsigned long scno)
 959 {
 960         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 961                 return scno;
 962
 963         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 964         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 965                 return 0x000ffff0;
 966         if (scno == 0x000ffff0)
 967                 return ARM_FIRST_SHUFFLED_SYSCALL;
 968
 969 #define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 970         /*
 971          * Is it ARM specific syscall?
 972          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 973          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 974          */
 975         if (scno >= 0x000f0000 &&
 976             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 977                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 978         }
 979         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 980                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 981         }
 982
 983         return scno;
 984 }
 985 #else
 986 # define shuffle_scno(scno) ((long)(scno))
 987 #endif
 988
 989 const char *
 990 err_name(unsigned long err)
 991 {
 992         if ((err < nerrnos) && errnoent[err])
 993                 return errnoent[err];
 994
 995         return NULL;
 996 }
 997
 998 static long get_regs_error;
 999
1000 void
1001 clear_regs(void)
1002 {
1003         get_regs_error = -1;
1004 }
1005
1006 static int get_syscall_args(struct tcb *);
1007 static int get_syscall_result(struct tcb *);
1008 static int arch_get_scno(struct tcb *tcp);
1009 static int arch_set_scno(struct tcb *, long);
1010 static void get_error(struct tcb *, const bool);
1011 static int arch_set_error(struct tcb *);
1012
1013 static struct fault_opts *
1014 tcb_fault_opts(struct tcb *tcp)
1015 {
1016         return (SCNO_IN_RANGE(tcp->scno) && tcp->fault_vec[current_personality])
1017                ? &tcp->fault_vec[current_personality][tcp->scno] : NULL;
1018 }
1019
1020
1021 static long
1022 inject_syscall_fault_entering(struct tcb *tcp)
1023 {
1024         if (!tcp->fault_vec[current_personality]) {
1025                 tcp->fault_vec[current_personality] =
1026                         xreallocarray(NULL, num_faults,
1027                                       sizeof(struct fault_opts));
1028                 memcpy(tcp->fault_vec[current_personality],
1029                        fault_vec[current_personality],
1030                        num_faults * sizeof(struct fault_opts));
1031         }
1032
1033         struct fault_opts *opts = tcb_fault_opts(tcp);
1034
1035         if (opts->first == 0)
1036                 return 0;
1037
1038         --opts->first;
1039
1040         if (opts->first != 0)
1041                 return 0;
1042
1043         opts->first = opts->step;
1044
1045         if (!arch_set_scno(tcp, -1))
1046                 tcp->flags |= TCB_FAULT_INJ;
1047
1048         return 0;
1049 }
1050
1051 static long
1052 update_syscall_fault_exiting(struct tcb *tcp)
1053 {
1054         struct fault_opts *opts = tcb_fault_opts(tcp);
1055
1056         if (opts && opts->err && tcp->u_error != opts->err) {
1057                 unsigned long u_error = tcp->u_error;
1058                 tcp->u_error = opts->err;
1059                 if (arch_set_error(tcp))
1060                         tcp->u_error = u_error;
1061         }
1062
1063         return 0;
1064 }
1065
1066 static int
1067 trace_syscall_entering(struct tcb *tcp)
1068 {
1069         int res, scno_good;
1070
1071         scno_good = res = get_scno(tcp);
1072         if (res == 0)
1073                 return res;
1074         if (res == 1)
1075                 res = get_syscall_args(tcp);
1076
1077         if (res != 1) {
1078                 printleader(tcp);
1079                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
1080                 /*
1081                  * " <unavailable>" will be added later by the code which
1082                  * detects ptrace errors.
1083                  */
1084                 goto ret;
1085         }
1086
1087 #ifdef LINUX_MIPSO32
1088         if (SEN_syscall == tcp->s_ent->sen)
1089                 decode_mips_subcall(tcp);
1090 #endif
1091
1092 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
1093         switch (tcp->s_ent->sen) {
1094 # ifdef SYS_socket_subcall
1095                 case SEN_socketcall:
1096                         decode_socket_subcall(tcp);
1097                         break;
1098 # endif
1099 # ifdef SYS_ipc_subcall
1100                 case SEN_ipc:
1101                         decode_ipc_subcall(tcp);
1102                         break;
1103 # endif
1104         }
1105 #endif
1106
1107         /* Restrain from fault injection while the trace executes strace code. */
1108         if (hide_log(tcp)) {
1109                 tcp->qual_flg &= ~QUAL_FAULT;
1110         }
1111
1112         switch (tcp->s_ent->sen) {
1113                 case SEN_execve:
1114                 case SEN_execveat:
1115 #if defined SPARC || defined SPARC64
1116                 case SEN_execv:
1117 #endif
1118                         tcp->flags &= ~TCB_HIDE_LOG;
1119                         break;
1120         }
1121
1122         if (!(tcp->qual_flg & QUAL_TRACE)
1123          || (tracing_paths && !pathtrace_match(tcp))
1124         ) {
1125                 tcp->flags |= TCB_INSYSCALL | TCB_FILTERED;
1126                 tcp->sys_func_rval = 0;
1127                 return 0;
1128         }
1129
1130         tcp->flags &= ~TCB_FILTERED;
1131
1132         if (hide_log(tcp)) {
1133                 res = 0;
1134                 goto ret;
1135         }
1136
1137         if (tcp->qual_flg & QUAL_FAULT)
1138                 inject_syscall_fault_entering(tcp);
1139
1140         if (cflag == CFLAG_ONLY_STATS) {
1141                 res = 0;
1142                 goto ret;
1143         }
1144
1145 #ifdef USE_LIBUNWIND
1146         if (stack_trace_enabled) {
1147                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
1148                         unwind_capture_stacktrace(tcp);
1149         }
1150 #endif
1151
1152         printleader(tcp);
1153         tprintf("%s(", tcp->s_ent->sys_name);
1154         if (tcp->qual_flg & QUAL_RAW)
1155                 res = printargs(tcp);
1156         else
1157                 res = tcp->s_ent->sys_func(tcp);
1158
1159         fflush(tcp->outf);
1160  ret:
1161         tcp->flags |= TCB_INSYSCALL;
1162         tcp->sys_func_rval = res;
1163         /* Measure the entrance time as late as possible to avoid errors. */
1164         if (Tflag || cflag)
1165                 gettimeofday(&tcp->etime, NULL);
1166         return res;
1167 }
1168
1169 static bool
1170 syscall_fault_injected(struct tcb *tcp)
1171 {
1172         return tcp->flags & TCB_FAULT_INJ;
1173 }
1174
1175 static int
1176 trace_syscall_exiting(struct tcb *tcp)
1177 {
1178         int sys_res;
1179         struct timeval tv;
1180         int res;
1181         unsigned long u_error;
1182         const char *u_error_str;
1183
1184         /* Measure the exit time as early as possible to avoid errors. */
1185         if (Tflag || cflag)
1186                 gettimeofday(&tv, NULL);
1187
1188 #ifdef USE_LIBUNWIND
1189         if (stack_trace_enabled) {
1190                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
1191                         unwind_cache_invalidate(tcp);
1192         }
1193 #endif
1194
1195 #if SUPPORTED_PERSONALITIES > 1
1196         update_personality(tcp, tcp->currpers);
1197 #endif
1198         res = (get_regs_error ? -1 : get_syscall_result(tcp));
1199         if (filtered(tcp) || hide_log(tcp))
1200                 goto ret;
1201
1202         if (syserror(tcp) && syscall_fault_injected(tcp))
1203                 update_syscall_fault_exiting(tcp);
1204
1205         if (cflag) {
1206                 count_syscall(tcp, &tv);
1207                 if (cflag == CFLAG_ONLY_STATS) {
1208                         goto ret;
1209                 }
1210         }
1211
1212         /* If not in -ff mode, and printing_tcp != tcp,
1213          * then the log currently does not end with output
1214          * of _our syscall entry_, but with something else.
1215          * We need to say which syscall's return is this.
1216          *
1217          * Forced reprinting via TCB_REPRINT is used only by
1218          * "strace -ff -oLOG test/threaded_execve" corner case.
1219          * It's the only case when -ff mode needs reprinting.
1220          */
1221         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
1222                 tcp->flags &= ~TCB_REPRINT;
1223                 printleader(tcp);
1224                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
1225         }
1226         printing_tcp = tcp;
1227
1228         tcp->s_prev_ent = NULL;
1229         if (res != 1) {
1230                 /* There was error in one of prior ptrace ops */
1231                 tprints(") ");
1232                 tabto();
1233                 tprints("= ? <unavailable>\n");
1234                 line_ended();
1235                 tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
1236                 tcp->sys_func_rval = 0;
1237                 free_tcb_priv_data(tcp);
1238                 return res;
1239         }
1240         tcp->s_prev_ent = tcp->s_ent;
1241
1242         sys_res = 0;
1243         if (tcp->qual_flg & QUAL_RAW) {
1244                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
1245         } else {
1246         /* FIXME: not_failing_only (IOW, option -z) is broken:
1247          * failure of syscall is known only after syscall return.
1248          * Thus we end up with something like this on, say, ENOENT:
1249          *     open("doesnt_exist", O_RDONLY <unfinished ...>
1250          *     {next syscall decode}
1251          * whereas the intended result is that open(...) line
1252          * is not shown at all.
1253          */
1254                 if (not_failing_only && tcp->u_error)
1255                         goto ret;       /* ignore failed syscalls */
1256                 if (tcp->sys_func_rval & RVAL_DECODED)
1257                         sys_res = tcp->sys_func_rval;
1258                 else
1259                         sys_res = tcp->s_ent->sys_func(tcp);
1260         }
1261
1262         tprints(") ");
1263         tabto();
1264         u_error = tcp->u_error;
1265
1266         if (tcp->qual_flg & QUAL_RAW) {
1267                 if (u_error) {
1268                         tprintf("= -1 (errno %lu)", u_error);
1269                         if (syscall_fault_injected(tcp))
1270                                 tprints(" (INJECTED)");
1271                 } else {
1272                         tprintf("= %#lx", tcp->u_rval);
1273                 }
1274         }
1275         else if (!(sys_res & RVAL_NONE) && u_error) {
1276                 switch (u_error) {
1277                 /* Blocked signals do not interrupt any syscalls.
1278                  * In this case syscalls don't return ERESTARTfoo codes.
1279                  *
1280                  * Deadly signals set to SIG_DFL interrupt syscalls
1281                  * and kill the process regardless of which of the codes below
1282                  * is returned by the interrupted syscall.
1283                  * In some cases, kernel forces a kernel-generated deadly
1284                  * signal to be unblocked and set to SIG_DFL (and thus cause
1285                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
1286                  * or SIGILL. (The alternative is to leave process spinning
1287                  * forever on the faulty instruction - not useful).
1288                  *
1289                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
1290                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
1291                  * but kernel will always restart them.
1292                  */
1293                 case ERESTARTSYS:
1294                         /* Most common type of signal-interrupted syscall exit code.
1295                          * The system call will be restarted with the same arguments
1296                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
1297                          */
1298                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
1299                         break;
1300                 case ERESTARTNOINTR:
1301                         /* Rare. For example, fork() returns this if interrupted.
1302                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
1303                          */
1304                         tprints("= ? ERESTARTNOINTR (To be restarted)");
1305                         break;
1306                 case ERESTARTNOHAND:
1307                         /* pause(), rt_sigsuspend() etc use this code.
1308                          * SA_RESTART is ignored (assumed not set):
1309                          * syscall won't restart (will return EINTR instead)
1310                          * even after signal with SA_RESTART set. However,
1311                          * after SIG_IGN or SIG_DFL signal it will restart
1312                          * (thus the name "restart only if has no handler").
1313                          */
1314                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
1315                         break;
1316                 case ERESTART_RESTARTBLOCK:
1317                         /* Syscalls like nanosleep(), poll() which can't be
1318                          * restarted with their original arguments use this
1319                          * code. Kernel will execute restart_syscall() instead,
1320                          * which changes arguments before restarting syscall.
1321                          * SA_RESTART is ignored (assumed not set) similarly
1322                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
1323                          * since restart data is saved in "restart block"
1324                          * in task struct, and if signal handler uses a syscall
1325                          * which in turn saves another such restart block,
1326                          * old data is lost and restart becomes impossible)
1327                          */
1328                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
1329                         break;
1330                 default:
1331                         u_error_str = err_name(u_error);
1332                         if (u_error_str)
1333                                 tprintf("= -1 %s (%s)",
1334                                         u_error_str, strerror(u_error));
1335                         else
1336                                 tprintf("= -1 %lu (%s)",
1337                                         u_error, strerror(u_error));
1338                         break;
1339                 }
1340                 if (syscall_fault_injected(tcp))
1341                         tprintf(" (INJECTED)");
1342                 if ((sys_res & RVAL_STR) && tcp->auxstr)
1343                         tprintf(" (%s)", tcp->auxstr);
1344         }
1345         else {
1346                 if (sys_res & RVAL_NONE)
1347                         tprints("= ?");
1348                 else {
1349                         switch (sys_res & RVAL_MASK) {
1350                         case RVAL_HEX:
1351 #if SUPPORTED_PERSONALITIES > 1
1352                                 if (current_wordsize < sizeof(long))
1353                                         tprintf("= %#x",
1354                                                 (unsigned int) tcp->u_rval);
1355                                 else
1356 #endif
1357                                         tprintf("= %#lx", tcp->u_rval);
1358                                 break;
1359                         case RVAL_OCTAL:
1360                                 tprints("= ");
1361                                 print_numeric_long_umask(tcp->u_rval);
1362                                 break;
1363                         case RVAL_UDECIMAL:
1364 #if SUPPORTED_PERSONALITIES > 1
1365                                 if (current_wordsize < sizeof(long))
1366                                         tprintf("= %u",
1367                                                 (unsigned int) tcp->u_rval);
1368                                 else
1369 #endif
1370                                         tprintf("= %lu", tcp->u_rval);
1371                                 break;
1372                         case RVAL_DECIMAL:
1373                                 tprintf("= %ld", tcp->u_rval);
1374                                 break;
1375                         case RVAL_FD:
1376                                 if (show_fd_path) {
1377                                         tprints("= ");
1378                                         printfd(tcp, tcp->u_rval);
1379                                 }
1380                                 else
1381                                         tprintf("= %ld", tcp->u_rval);
1382                                 break;
1383 #if HAVE_STRUCT_TCB_EXT_ARG
1384                         /*
1385                         case RVAL_LHEX:
1386                                 tprintf("= %#llx", tcp->u_lrval);
1387                                 break;
1388                         case RVAL_LOCTAL:
1389                                 tprintf("= %#llo", tcp->u_lrval);
1390                                 break;
1391                         */
1392                         case RVAL_LUDECIMAL:
1393                                 tprintf("= %llu", tcp->u_lrval);
1394                                 break;
1395                         /*
1396                         case RVAL_LDECIMAL:
1397                                 tprintf("= %lld", tcp->u_lrval);
1398                                 break;
1399                         */
1400 #endif /* HAVE_STRUCT_TCB_EXT_ARG */
1401                         default:
1402                                 error_msg("invalid rval format");
1403                                 break;
1404                         }
1405                 }
1406                 if ((sys_res & RVAL_STR) && tcp->auxstr)
1407                         tprintf(" (%s)", tcp->auxstr);
1408         }
1409         if (Tflag) {
1410                 tv_sub(&tv, &tv, &tcp->etime);
1411                 tprintf(" <%ld.%06ld>",
1412                         (long) tv.tv_sec, (long) tv.tv_usec);
1413         }
1414         tprints("\n");
1415         dumpio(tcp);
1416         line_ended();
1417
1418 #ifdef USE_LIBUNWIND
1419         if (stack_trace_enabled)
1420                 unwind_print_stacktrace(tcp);
1421 #endif
1422
1423  ret:
1424         tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
1425         tcp->sys_func_rval = 0;
1426         free_tcb_priv_data(tcp);
1427         return 0;
1428 }
1429
1430 int
1431 trace_syscall(struct tcb *tcp)
1432 {
1433         return exiting(tcp) ?
1434                 trace_syscall_exiting(tcp) : trace_syscall_entering(tcp);
1435 }
1436
1437 bool
1438 is_erestart(struct tcb *tcp)
1439 {
1440         switch (tcp->u_error) {
1441                 case ERESTARTSYS:
1442                 case ERESTARTNOINTR:
1443                 case ERESTARTNOHAND:
1444                 case ERESTART_RESTARTBLOCK:
1445                         return true;
1446                 default:
1447                         return false;
1448         }
1449 }
1450
1451 static unsigned long saved_u_error;
1452
1453 void
1454 temporarily_clear_syserror(struct tcb *tcp)
1455 {
1456         saved_u_error = tcp->u_error;
1457         tcp->u_error = 0;
1458 }
1459
1460 void
1461 restore_cleared_syserror(struct tcb *tcp)
1462 {
1463         tcp->u_error = saved_u_error;
1464 }
1465
1466 #include "kernel_types.h"
1467
1468 /*
1469  * Check the syscall return value register value for whether it is
1470  * a negated errno code indicating an error, or a success return value.
1471  */
1472 static inline bool
1473 is_negated_errno(kernel_ulong_t val)
1474 {
1475         /* Linux kernel defines MAX_ERRNO to 4095. */
1476         kernel_ulong_t max = -(kernel_long_t) 4095;
1477
1478 #if defined X86_64 || defined X32
1479         /*
1480          * current_wordsize is 4 for x32 personality
1481          * but truncation _must not_ be done in it, so
1482          * check current_personality instead.
1483          */
1484         if (current_personality == 1) {
1485                 val = (uint32_t) val;
1486                 max = (uint32_t) max;
1487         }
1488 #elif SUPPORTED_PERSONALITIES > 1 && SIZEOF_LONG > 4
1489         if (current_wordsize < sizeof(val)) {
1490                 val = (uint32_t) val;
1491                 max = (uint32_t) max;
1492         }
1493 #endif
1494
1495         return val >= max;
1496 }
1497
1498 #include "arch_regs.c"
1499
1500 #ifdef HAVE_GETRVAL2
1501 # include "arch_getrval2.c"
1502 #endif
1503
1504 void
1505 print_pc(struct tcb *tcp)
1506 {
1507 #if defined ARCH_PC_REG
1508 # define ARCH_GET_PC 0
1509 #elif defined ARCH_PC_PEEK_ADDR
1510         long pc;
1511 # define ARCH_PC_REG pc
1512 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1513 #else
1514 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1515 #endif
1516         if (get_regs_error || ARCH_GET_PC)
1517                 tprints(current_wordsize == 4 ? "[????????] "
1518                                               : "[????????????????] ");
1519         else
1520                 tprintf(current_wordsize == 4 ? "[%08lx] " : "[%016lx] ",
1521                         (unsigned long) ARCH_PC_REG);
1522 }
1523
1524 #include "getregs_old.h"
1525
1526 #undef ptrace_getregset_or_getregs
1527 #undef ptrace_setregset_or_setregs
1528 #ifdef ARCH_REGS_FOR_GETREGSET
1529
1530 # define ptrace_getregset_or_getregs ptrace_getregset
1531 static long
1532 ptrace_getregset(pid_t pid)
1533 {
1534 # ifdef ARCH_IOVEC_FOR_GETREGSET
1535         /* variable iovec */
1536         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1537         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1538                       &ARCH_IOVEC_FOR_GETREGSET);
1539 # else
1540         /* constant iovec */
1541         static struct iovec io = {
1542                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1543                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1544         };
1545         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1546
1547 # endif
1548 }
1549
1550 # ifndef HAVE_GETREGS_OLD
1551 #  define ptrace_setregset_or_setregs ptrace_setregset
1552 static int
1553 ptrace_setregset(pid_t pid)
1554 {
1555 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1556         /* variable iovec */
1557         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1558                       &ARCH_IOVEC_FOR_GETREGSET);
1559 #  else
1560         /* constant iovec */
1561         static struct iovec io = {
1562                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1563                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1564         };
1565         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1566 #  endif
1567 }
1568 # endif /* !HAVE_GETREGS_OLD */
1569
1570 #elif defined ARCH_REGS_FOR_GETREGS
1571
1572 # define ptrace_getregset_or_getregs ptrace_getregs
1573 static long
1574 ptrace_getregs(pid_t pid)
1575 {
1576 # if defined SPARC || defined SPARC64
1577         /* SPARC systems have the meaning of data and addr reversed */
1578         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1579 # else
1580         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1581 # endif
1582 }
1583
1584 # ifndef HAVE_GETREGS_OLD
1585 #  define ptrace_setregset_or_setregs ptrace_setregs
1586 static int
1587 ptrace_setregs(pid_t pid)
1588 {
1589 #  if defined SPARC || defined SPARC64
1590         /* SPARC systems have the meaning of data and addr reversed */
1591         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1592 #  else
1593         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1594 #  endif
1595 }
1596 # endif /* !HAVE_GETREGS_OLD */
1597
1598 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1599
1600 void
1601 get_regs(pid_t pid)
1602 {
1603 #undef USE_GET_SYSCALL_RESULT_REGS
1604 #ifdef ptrace_getregset_or_getregs
1605
1606 # ifdef HAVE_GETREGS_OLD
1607         /*
1608          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1609          * fallback to getregs_old.
1610          */
1611         static int use_getregs_old;
1612         if (use_getregs_old < 0) {
1613                 get_regs_error = ptrace_getregset_or_getregs(pid);
1614                 return;
1615         } else if (use_getregs_old == 0) {
1616                 get_regs_error = ptrace_getregset_or_getregs(pid);
1617                 if (get_regs_error >= 0) {
1618                         use_getregs_old = -1;
1619                         return;
1620                 }
1621                 if (errno == EPERM || errno == ESRCH)
1622                         return;
1623                 use_getregs_old = 1;
1624         }
1625         get_regs_error = getregs_old(pid);
1626 # else /* !HAVE_GETREGS_OLD */
1627         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1628         get_regs_error = ptrace_getregset_or_getregs(pid);
1629 # endif /* !HAVE_GETREGS_OLD */
1630
1631 #else /* !ptrace_getregset_or_getregs */
1632
1633 # define USE_GET_SYSCALL_RESULT_REGS 1
1634 # warning get_regs is not implemented for this architecture yet
1635         get_regs_error = 0;
1636
1637 #endif /* !ptrace_getregset_or_getregs */
1638 }
1639
1640 #ifdef ptrace_setregset_or_setregs
1641 static int
1642 set_regs(pid_t pid)
1643 {
1644         return ptrace_setregset_or_setregs(pid);
1645 }
1646 #endif /* ptrace_setregset_or_setregs */
1647
1648 struct sysent_buf {
1649         struct tcb *tcp;
1650         struct_sysent ent;
1651         char buf[sizeof("syscall_%lu") + sizeof(long) * 3];
1652 };
1653
1654 static void
1655 free_sysent_buf(void *ptr)
1656 {
1657         struct sysent_buf *s = ptr;
1658         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1659         free(ptr);
1660 }
1661
1662 /*
1663  * Returns:
1664  * 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
1665  * 1: ok, continue in trace_syscall_entering().
1666  * other: error, trace_syscall_entering() should print error indicator
1667  *    ("????" etc) and bail out.
1668  */
1669 int
1670 get_scno(struct tcb *tcp)
1671 {
1672         if (get_regs_error)
1673                 return -1;
1674
1675         int rc = arch_get_scno(tcp);
1676         if (rc != 1)
1677                 return rc;
1678
1679         if (SCNO_IS_VALID(tcp->scno)) {
1680                 tcp->s_ent = &sysent[tcp->scno];
1681                 tcp->qual_flg = qual_flags[tcp->scno];
1682         } else {
1683                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1684
1685                 s->tcp = tcp;
1686                 s->ent.nargs = MAX_ARGS;
1687                 s->ent.sen = SEN_printargs;
1688                 s->ent.sys_func = printargs;
1689                 s->ent.sys_name = s->buf;
1690                 sprintf(s->buf, "syscall_%lu", shuffle_scno(tcp->scno));
1691
1692                 tcp->s_ent = &s->ent;
1693                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1694
1695                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1696
1697                 if (debug_flag)
1698                         error_msg("pid %d invalid syscall %ld", tcp->pid, tcp->scno);
1699         }
1700         return 1;
1701 }
1702
1703 #ifdef USE_GET_SYSCALL_RESULT_REGS
1704 static int get_syscall_result_regs(struct tcb *);
1705 #endif
1706
1707 /* Returns:
1708  * 1: ok, continue in trace_syscall_exiting().
1709  * -1: error, trace_syscall_exiting() should print error indicator
1710  *    ("????" etc) and bail out.
1711  */
1712 static int
1713 get_syscall_result(struct tcb *tcp)
1714 {
1715 #ifdef USE_GET_SYSCALL_RESULT_REGS
1716         if (get_syscall_result_regs(tcp))
1717                 return -1;
1718 #endif
1719         tcp->u_error = 0;
1720         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1721
1722         return 1;
1723 }
1724
1725 #include "get_scno.c"
1726 #include "set_scno.c"
1727 #include "get_syscall_args.c"
1728 #ifdef USE_GET_SYSCALL_RESULT_REGS
1729 # include "get_syscall_result.c"
1730 #endif
1731 #include "get_error.c"
1732 #include "set_error.c"
1733 #ifdef HAVE_GETREGS_OLD
1734 # include "getregs_old.c"
1735 #endif
1736
1737 const char *
1738 syscall_name(long scno)
1739 {
1740 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1741         if (current_personality == X32_PERSONALITY_NUMBER)
1742                 scno &= ~__X32_SYSCALL_BIT;
1743 #endif
1744         return SCNO_IS_VALID(scno) ? sysent[scno].sys_name: NULL;
1745 }