]> granicus.if.org Git - strace/blob - syscall.c
x32: fix decoding of flags argument of preadv2 and pwritev2 syscalls
[strace] / 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 "nsig.h"
37 #include <sys/param.h>
38
39 /* for struct iovec */
40 #include <sys/uio.h>
41
42 /* for __X32_SYSCALL_BIT */
43 #include <asm/unistd.h>
44
45 #include "regs.h"
46 #include "ptrace.h"
47
48 #if defined(SPARC64)
49 # undef PTRACE_GETREGS
50 # define PTRACE_GETREGS PTRACE_GETREGS64
51 # undef PTRACE_SETREGS
52 # define PTRACE_SETREGS PTRACE_SETREGS64
53 #endif
54
55 #if defined SPARC64
56 # include <asm/psrcompat.h>
57 #elif defined SPARC
58 # include <asm/psr.h>
59 #endif
60
61 #ifdef IA64
62 # include <asm/rse.h>
63 #endif
64
65 #ifndef NT_PRSTATUS
66 # define NT_PRSTATUS 1
67 #endif
68
69 #include "syscall.h"
70
71 /* Define these shorthand notations to simplify the syscallent files. */
72 #define TD TRACE_DESC
73 #define TF TRACE_FILE
74 #define TI TRACE_IPC
75 #define TN TRACE_NETWORK
76 #define TP TRACE_PROCESS
77 #define TS TRACE_SIGNAL
78 #define TM TRACE_MEMORY
79 #define NF SYSCALL_NEVER_FAILS
80 #define MA MAX_ARGS
81 #define SI STACKTRACE_INVALIDATE_CACHE
82 #define SE STACKTRACE_CAPTURE_ON_ENTER
83 #define CST COMPAT_SYSCALL_TYPES
84
85 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
86
87 const struct_sysent sysent0[] = {
88 #include "syscallent.h"
89 };
90
91 #if SUPPORTED_PERSONALITIES > 1
92 # include PERSONALITY1_INCLUDE_FUNCS
93 static const struct_sysent sysent1[] = {
94 # include "syscallent1.h"
95 };
96 #endif
97
98 #if SUPPORTED_PERSONALITIES > 2
99 # include PERSONALITY2_INCLUDE_FUNCS
100 static const struct_sysent sysent2[] = {
101 # include "syscallent2.h"
102 };
103 #endif
104
105 /* Now undef them since short defines cause wicked namespace pollution. */
106 #undef SEN
107 #undef TD
108 #undef TF
109 #undef TI
110 #undef TN
111 #undef TP
112 #undef TS
113 #undef TM
114 #undef NF
115 #undef MA
116 #undef SI
117 #undef SE
118 #undef CST
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 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
228         nsyscalls0,
229 #if SUPPORTED_PERSONALITIES > 1
230         nsyscalls1,
231 #endif
232 #if SUPPORTED_PERSONALITIES > 2
233         nsyscalls2,
234 #endif
235 };
236 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
237         sysent0,
238 #if SUPPORTED_PERSONALITIES > 1
239         sysent1,
240 #endif
241 #if SUPPORTED_PERSONALITIES > 2
242         sysent2,
243 #endif
244 };
245
246 #if SUPPORTED_PERSONALITIES > 1
247 unsigned current_personality;
248
249 # ifndef current_wordsize
250 unsigned current_wordsize;
251 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
252         PERSONALITY0_WORDSIZE,
253         PERSONALITY1_WORDSIZE,
254 # if SUPPORTED_PERSONALITIES > 2
255         PERSONALITY2_WORDSIZE,
256 # endif
257 };
258 # endif
259
260 # ifndef current_klongsize
261 unsigned current_klongsize;
262 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
263         PERSONALITY0_KLONGSIZE,
264         PERSONALITY1_KLONGSIZE,
265 #  if SUPPORTED_PERSONALITIES > 2
266         PERSONALITY2_KLONGSIZE,
267 #  endif
268 };
269 # endif
270
271 void
272 set_personality(int personality)
273 {
274         nsyscalls = nsyscall_vec[personality];
275         sysent = sysent_vec[personality];
276
277         switch (personality) {
278         case 0:
279                 errnoent = errnoent0;
280                 nerrnos = nerrnos0;
281                 ioctlent = ioctlent0;
282                 nioctlents = nioctlents0;
283                 signalent = signalent0;
284                 nsignals = nsignals0;
285                 printers = &printers0;
286                 break;
287
288         case 1:
289                 errnoent = errnoent1;
290                 nerrnos = nerrnos1;
291                 ioctlent = ioctlent1;
292                 nioctlents = nioctlents1;
293                 signalent = signalent1;
294                 nsignals = nsignals1;
295                 printers = &printers1;
296                 break;
297
298 # if SUPPORTED_PERSONALITIES > 2
299         case 2:
300                 errnoent = errnoent2;
301                 nerrnos = nerrnos2;
302                 ioctlent = ioctlent2;
303                 nioctlents = nioctlents2;
304                 signalent = signalent2;
305                 nsignals = nsignals2;
306                 printers = &printers2;
307                 break;
308 # endif
309         }
310
311         current_personality = personality;
312 # ifndef current_wordsize
313         current_wordsize = personality_wordsize[personality];
314 # endif
315 # ifndef current_klongsize
316         current_klongsize = personality_klongsize[personality];
317 # endif
318 }
319
320 static void
321 update_personality(struct tcb *tcp, unsigned int personality)
322 {
323         if (personality == current_personality)
324                 return;
325         set_personality(personality);
326
327         if (personality == tcp->currpers)
328                 return;
329         tcp->currpers = personality;
330
331 # undef PERSONALITY_NAMES
332 # if defined POWERPC64
333 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
334 # elif defined X86_64
335 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
336 # elif defined X32
337 #  define PERSONALITY_NAMES {"x32", "32 bit"}
338 # elif defined AARCH64
339 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
340 # elif defined TILE
341 #  define PERSONALITY_NAMES {"64-bit", "32-bit"}
342 # endif
343 # ifdef PERSONALITY_NAMES
344         if (!qflag) {
345                 static const char *const names[] = PERSONALITY_NAMES;
346                 error_msg("[ Process PID=%d runs in %s mode. ]",
347                           tcp->pid, names[personality]);
348         }
349 # endif
350 }
351 #endif
352
353 #ifdef SYS_socket_subcall
354 static void
355 decode_socket_subcall(struct tcb *tcp)
356 {
357         const int call = tcp->u_arg[0];
358
359         if (call < 1 || call >= SYS_socket_nsubcalls)
360                 return;
361
362         const kernel_ulong_t scno = SYS_socket_subcall + call;
363         const unsigned int nargs = sysent[scno].nargs;
364         uint64_t buf[nargs];
365
366         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
367                 return;
368
369         tcp->scno = scno;
370         tcp->qual_flg = qual_flags(scno);
371         tcp->s_ent = &sysent[scno];
372
373         unsigned int i;
374         for (i = 0; i < nargs; ++i)
375                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
376                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
377 }
378 #endif
379
380 #ifdef SYS_ipc_subcall
381 static void
382 decode_ipc_subcall(struct tcb *tcp)
383 {
384         unsigned int call = tcp->u_arg[0];
385         const unsigned int version = call >> 16;
386
387         if (version) {
388 # if defined S390 || defined S390X
389                 return;
390 # else
391 #  ifdef SPARC64
392                 if (current_wordsize == 8)
393                         return;
394 #  endif
395                 set_tcb_priv_ulong(tcp, version);
396                 call &= 0xffff;
397 # endif
398         }
399
400         switch (call) {
401                 case  1: case  2: case  3: case  4:
402                 case 11: case 12: case 13: case 14:
403                 case 21: case 22: case 23: case 24:
404                         break;
405                 default:
406                         return;
407         }
408
409         tcp->scno = SYS_ipc_subcall + call;
410         tcp->qual_flg = qual_flags(tcp->scno);
411         tcp->s_ent = &sysent[tcp->scno];
412
413         const unsigned int n = tcp->s_ent->nargs;
414         unsigned int i;
415         for (i = 0; i < n; i++)
416                 tcp->u_arg[i] = tcp->u_arg[i + 1];
417 }
418 #endif
419
420 #ifdef LINUX_MIPSO32
421 static void
422 decode_mips_subcall(struct tcb *tcp)
423 {
424         if (!scno_is_valid(tcp->u_arg[0]))
425                 return;
426         tcp->scno = tcp->u_arg[0];
427         tcp->qual_flg = qual_flags(tcp->scno);
428         tcp->s_ent = &sysent[tcp->scno];
429         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
430                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
431         /*
432          * Fetching the last arg of 7-arg syscalls (fadvise64_64
433          * and sync_file_range) requires additional code,
434          * see linux/mips/get_syscall_args.c
435          */
436         if (tcp->s_ent->nargs == MAX_ARGS) {
437                 if (umoven(tcp,
438                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
439                            sizeof(tcp->u_arg[0]),
440                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
441                 tcp->u_arg[MAX_ARGS - 1] = 0;
442         }
443 }
444 #endif /* LINUX_MIPSO32 */
445
446 static void
447 dumpio(struct tcb *tcp)
448 {
449         if (syserror(tcp))
450                 return;
451
452         int fd = tcp->u_arg[0];
453         if (fd < 0)
454                 return;
455
456         if (is_number_in_set(fd, &read_set)) {
457                 switch (tcp->s_ent->sen) {
458                 case SEN_read:
459                 case SEN_pread:
460                 case SEN_recv:
461                 case SEN_recvfrom:
462                 case SEN_mq_timedreceive:
463                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
464                         return;
465                 case SEN_readv:
466                 case SEN_preadv:
467                 case SEN_preadv2:
468                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
469                                      tcp->u_rval);
470                         return;
471                 case SEN_recvmsg:
472                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
473                         return;
474                 case SEN_recvmmsg:
475                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
476                         return;
477                 }
478         }
479         if (is_number_in_set(fd, &write_set)) {
480                 switch (tcp->s_ent->sen) {
481                 case SEN_write:
482                 case SEN_pwrite:
483                 case SEN_send:
484                 case SEN_sendto:
485                 case SEN_mq_timedsend:
486                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
487                         break;
488                 case SEN_writev:
489                 case SEN_pwritev:
490                 case SEN_pwritev2:
491                 case SEN_vmsplice:
492                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
493                         break;
494                 case SEN_sendmsg:
495                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
496                         break;
497                 case SEN_sendmmsg:
498                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
499                         break;
500                 }
501         }
502 }
503
504 /*
505  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
506  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
507  */
508 static kernel_ulong_t
509 shuffle_scno(kernel_ulong_t scno)
510 {
511 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
512         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
513                 return scno;
514
515         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
516         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
517                 return 0x000ffff0;
518         if (scno == 0x000ffff0)
519                 return ARM_FIRST_SHUFFLED_SYSCALL;
520
521 # define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
522         /*
523          * Is it ARM specific syscall?
524          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
525          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
526          */
527         if (scno >= 0x000f0000 &&
528             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
529                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
530         }
531         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
532                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
533         }
534 #endif /* ARM || AARCH64 */
535
536         return scno;
537 }
538
539 const char *
540 err_name(unsigned long err)
541 {
542         if ((err < nerrnos) && errnoent[err])
543                 return errnoent[err];
544
545         return NULL;
546 }
547
548 static long get_regs_error;
549
550 void
551 clear_regs(void)
552 {
553         get_regs_error = -1;
554 }
555
556 static int get_syscall_args(struct tcb *);
557 static int get_syscall_result(struct tcb *);
558 static int arch_get_scno(struct tcb *tcp);
559 static int arch_set_scno(struct tcb *, kernel_ulong_t);
560 static void get_error(struct tcb *, const bool);
561 static int arch_set_error(struct tcb *);
562 static int arch_set_success(struct tcb *);
563
564 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
565
566 static struct inject_opts *
567 tcb_inject_opts(struct tcb *tcp)
568 {
569         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
570                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
571 }
572
573
574 static long
575 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
576 {
577         if (!tcp->inject_vec[current_personality]) {
578                 tcp->inject_vec[current_personality] =
579                         xcalloc(nsyscalls, sizeof(**inject_vec));
580                 memcpy(tcp->inject_vec[current_personality],
581                        inject_vec[current_personality],
582                        nsyscalls * sizeof(**inject_vec));
583         }
584
585         struct inject_opts *opts = tcb_inject_opts(tcp);
586
587         if (!opts || opts->first == 0)
588                 return 0;
589
590         --opts->first;
591
592         if (opts->first != 0)
593                 return 0;
594
595         opts->first = opts->step;
596
597         if (opts->signo > 0)
598                 *signo = opts->signo;
599         if (opts->rval != INJECT_OPTS_RVAL_DEFAULT && !arch_set_scno(tcp, -1))
600                 tcp->flags |= TCB_TAMPERED;
601
602         return 0;
603 }
604
605 static long
606 tamper_with_syscall_exiting(struct tcb *tcp)
607 {
608         struct inject_opts *opts = tcb_inject_opts(tcp);
609
610         if (!opts)
611                 return 0;
612
613         if (opts->rval >= 0) {
614                 kernel_long_t u_rval = tcp->u_rval;
615
616                 tcp->u_rval = opts->rval;
617                 if (arch_set_success(tcp)) {
618                         tcp->u_rval = u_rval;
619                 } else {
620                         tcp->u_error = 0;
621                 }
622         } else {
623                 unsigned long new_error = -opts->rval;
624
625                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
626                         unsigned long u_error = tcp->u_error;
627
628                         tcp->u_error = new_error;
629                         if (arch_set_error(tcp)) {
630                                 tcp->u_error = u_error;
631                         }
632                 }
633         }
634
635         return 0;
636 }
637
638 static int
639 trace_syscall_entering(struct tcb *tcp, unsigned int *sig)
640 {
641         int res, scno_good;
642
643         scno_good = res = get_scno(tcp);
644         if (res == 0)
645                 return res;
646         if (res == 1)
647                 res = get_syscall_args(tcp);
648
649         if (res != 1) {
650                 printleader(tcp);
651                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
652                 /*
653                  * " <unavailable>" will be added later by the code which
654                  * detects ptrace errors.
655                  */
656                 goto ret;
657         }
658
659 #ifdef LINUX_MIPSO32
660         if (SEN_syscall == tcp->s_ent->sen)
661                 decode_mips_subcall(tcp);
662 #endif
663
664 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
665         switch (tcp->s_ent->sen) {
666 # ifdef SYS_socket_subcall
667                 case SEN_socketcall:
668                         decode_socket_subcall(tcp);
669                         break;
670 # endif
671 # ifdef SYS_ipc_subcall
672                 case SEN_ipc:
673                         decode_ipc_subcall(tcp);
674                         break;
675 # endif
676         }
677 #endif
678
679         /* Restrain from fault injection while the trace executes strace code. */
680         if (hide_log(tcp)) {
681                 tcp->qual_flg &= ~QUAL_INJECT;
682         }
683
684         switch (tcp->s_ent->sen) {
685                 case SEN_execve:
686                 case SEN_execveat:
687 #if defined SPARC || defined SPARC64
688                 case SEN_execv:
689 #endif
690                         tcp->flags &= ~TCB_HIDE_LOG;
691                         break;
692         }
693
694         if (!(tcp->qual_flg & QUAL_TRACE)
695          || (tracing_paths && !pathtrace_match(tcp))
696         ) {
697                 tcp->flags |= TCB_INSYSCALL | TCB_FILTERED;
698                 tcp->sys_func_rval = 0;
699                 return 0;
700         }
701
702         tcp->flags &= ~TCB_FILTERED;
703
704         if (hide_log(tcp)) {
705                 res = 0;
706                 goto ret;
707         }
708
709         if (tcp->qual_flg & QUAL_INJECT)
710                 tamper_with_syscall_entering(tcp, sig);
711
712         if (cflag == CFLAG_ONLY_STATS) {
713                 res = 0;
714                 goto ret;
715         }
716
717 #ifdef USE_LIBUNWIND
718         if (stack_trace_enabled) {
719                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
720                         unwind_capture_stacktrace(tcp);
721         }
722 #endif
723
724         printleader(tcp);
725         tprintf("%s(", tcp->s_ent->sys_name);
726         if (tcp->qual_flg & QUAL_RAW)
727                 res = printargs(tcp);
728         else
729                 res = tcp->s_ent->sys_func(tcp);
730
731         fflush(tcp->outf);
732  ret:
733         tcp->flags |= TCB_INSYSCALL;
734         tcp->sys_func_rval = res;
735         /* Measure the entrance time as late as possible to avoid errors. */
736         if (Tflag || cflag)
737                 gettimeofday(&tcp->etime, NULL);
738         return res;
739 }
740
741 static bool
742 syscall_tampered(struct tcb *tcp)
743 {
744         return tcp->flags & TCB_TAMPERED;
745 }
746
747 static int
748 trace_syscall_exiting(struct tcb *tcp)
749 {
750         int sys_res;
751         struct timeval tv;
752         int res;
753         unsigned long u_error;
754         const char *u_error_str;
755
756         /* Measure the exit time as early as possible to avoid errors. */
757         if (Tflag || cflag)
758                 gettimeofday(&tv, NULL);
759
760 #ifdef USE_LIBUNWIND
761         if (stack_trace_enabled) {
762                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
763                         unwind_cache_invalidate(tcp);
764         }
765 #endif
766
767 #if SUPPORTED_PERSONALITIES > 1
768         update_personality(tcp, tcp->currpers);
769 #endif
770         res = (get_regs_error ? -1 : get_syscall_result(tcp));
771         if (filtered(tcp) || hide_log(tcp))
772                 goto ret;
773
774         if (syserror(tcp) && syscall_tampered(tcp))
775                 tamper_with_syscall_exiting(tcp);
776
777         if (cflag) {
778                 count_syscall(tcp, &tv);
779                 if (cflag == CFLAG_ONLY_STATS) {
780                         goto ret;
781                 }
782         }
783
784         /* If not in -ff mode, and printing_tcp != tcp,
785          * then the log currently does not end with output
786          * of _our syscall entry_, but with something else.
787          * We need to say which syscall's return is this.
788          *
789          * Forced reprinting via TCB_REPRINT is used only by
790          * "strace -ff -oLOG test/threaded_execve" corner case.
791          * It's the only case when -ff mode needs reprinting.
792          */
793         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
794                 tcp->flags &= ~TCB_REPRINT;
795                 printleader(tcp);
796                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
797         }
798         printing_tcp = tcp;
799
800         tcp->s_prev_ent = NULL;
801         if (res != 1) {
802                 /* There was error in one of prior ptrace ops */
803                 tprints(") ");
804                 tabto();
805                 tprints("= ? <unavailable>\n");
806                 line_ended();
807                 tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
808                 tcp->sys_func_rval = 0;
809                 free_tcb_priv_data(tcp);
810                 return res;
811         }
812         tcp->s_prev_ent = tcp->s_ent;
813
814         sys_res = 0;
815         if (tcp->qual_flg & QUAL_RAW) {
816                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
817         } else {
818         /* FIXME: not_failing_only (IOW, option -z) is broken:
819          * failure of syscall is known only after syscall return.
820          * Thus we end up with something like this on, say, ENOENT:
821          *     open("doesnt_exist", O_RDONLY <unfinished ...>
822          *     {next syscall decode}
823          * whereas the intended result is that open(...) line
824          * is not shown at all.
825          */
826                 if (not_failing_only && tcp->u_error)
827                         goto ret;       /* ignore failed syscalls */
828                 if (tcp->sys_func_rval & RVAL_DECODED)
829                         sys_res = tcp->sys_func_rval;
830                 else
831                         sys_res = tcp->s_ent->sys_func(tcp);
832         }
833
834         tprints(") ");
835         tabto();
836         u_error = tcp->u_error;
837
838         if (tcp->qual_flg & QUAL_RAW) {
839                 if (u_error) {
840                         tprintf("= -1 (errno %lu)", u_error);
841                 } else {
842                         tprintf("= %#" PRI_klx, tcp->u_rval);
843                 }
844                 if (syscall_tampered(tcp))
845                         tprints(" (INJECTED)");
846         }
847         else if (!(sys_res & RVAL_NONE) && u_error) {
848                 switch (u_error) {
849                 /* Blocked signals do not interrupt any syscalls.
850                  * In this case syscalls don't return ERESTARTfoo codes.
851                  *
852                  * Deadly signals set to SIG_DFL interrupt syscalls
853                  * and kill the process regardless of which of the codes below
854                  * is returned by the interrupted syscall.
855                  * In some cases, kernel forces a kernel-generated deadly
856                  * signal to be unblocked and set to SIG_DFL (and thus cause
857                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
858                  * or SIGILL. (The alternative is to leave process spinning
859                  * forever on the faulty instruction - not useful).
860                  *
861                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
862                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
863                  * but kernel will always restart them.
864                  */
865                 case ERESTARTSYS:
866                         /* Most common type of signal-interrupted syscall exit code.
867                          * The system call will be restarted with the same arguments
868                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
869                          */
870                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
871                         break;
872                 case ERESTARTNOINTR:
873                         /* Rare. For example, fork() returns this if interrupted.
874                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
875                          */
876                         tprints("= ? ERESTARTNOINTR (To be restarted)");
877                         break;
878                 case ERESTARTNOHAND:
879                         /* pause(), rt_sigsuspend() etc use this code.
880                          * SA_RESTART is ignored (assumed not set):
881                          * syscall won't restart (will return EINTR instead)
882                          * even after signal with SA_RESTART set. However,
883                          * after SIG_IGN or SIG_DFL signal it will restart
884                          * (thus the name "restart only if has no handler").
885                          */
886                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
887                         break;
888                 case ERESTART_RESTARTBLOCK:
889                         /* Syscalls like nanosleep(), poll() which can't be
890                          * restarted with their original arguments use this
891                          * code. Kernel will execute restart_syscall() instead,
892                          * which changes arguments before restarting syscall.
893                          * SA_RESTART is ignored (assumed not set) similarly
894                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
895                          * since restart data is saved in "restart block"
896                          * in task struct, and if signal handler uses a syscall
897                          * which in turn saves another such restart block,
898                          * old data is lost and restart becomes impossible)
899                          */
900                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
901                         break;
902                 default:
903                         u_error_str = err_name(u_error);
904                         if (u_error_str)
905                                 tprintf("= -1 %s (%s)",
906                                         u_error_str, strerror(u_error));
907                         else
908                                 tprintf("= -1 %lu (%s)",
909                                         u_error, strerror(u_error));
910                         break;
911                 }
912                 if (syscall_tampered(tcp))
913                         tprints(" (INJECTED)");
914                 if ((sys_res & RVAL_STR) && tcp->auxstr)
915                         tprintf(" (%s)", tcp->auxstr);
916         }
917         else {
918                 if (sys_res & RVAL_NONE)
919                         tprints("= ?");
920                 else {
921                         switch (sys_res & RVAL_MASK) {
922                         case RVAL_HEX:
923 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
924                                 if (current_wordsize < sizeof(tcp->u_rval)) {
925                                         tprintf("= %#x",
926                                                 (unsigned int) tcp->u_rval);
927                                 } else
928 #endif
929                                 {
930                                         tprintf("= %#" PRI_klx, tcp->u_rval);
931                                 }
932                                 break;
933                         case RVAL_OCTAL:
934                                 tprints("= ");
935                                 print_numeric_long_umask(tcp->u_rval);
936                                 break;
937                         case RVAL_UDECIMAL:
938 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
939                                 if (current_wordsize < sizeof(tcp->u_rval)) {
940                                         tprintf("= %u",
941                                                 (unsigned int) tcp->u_rval);
942                                 } else
943 #endif
944                                 {
945                                         tprintf("= %" PRI_klu, tcp->u_rval);
946                                 }
947                                 break;
948                         case RVAL_DECIMAL:
949                                 tprintf("= %" PRI_kld, tcp->u_rval);
950                                 break;
951                         case RVAL_FD:
952                                 if (show_fd_path) {
953                                         tprints("= ");
954                                         printfd(tcp, tcp->u_rval);
955                                 }
956                                 else
957                                         tprintf("= %" PRI_kld, tcp->u_rval);
958                                 break;
959                         default:
960                                 error_msg("invalid rval format");
961                                 break;
962                         }
963                 }
964                 if ((sys_res & RVAL_STR) && tcp->auxstr)
965                         tprintf(" (%s)", tcp->auxstr);
966                 if (syscall_tampered(tcp))
967                         tprints(" (INJECTED)");
968         }
969         if (Tflag) {
970                 tv_sub(&tv, &tv, &tcp->etime);
971                 tprintf(" <%ld.%06ld>",
972                         (long) tv.tv_sec, (long) tv.tv_usec);
973         }
974         tprints("\n");
975         dumpio(tcp);
976         line_ended();
977
978 #ifdef USE_LIBUNWIND
979         if (stack_trace_enabled)
980                 unwind_print_stacktrace(tcp);
981 #endif
982
983  ret:
984         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
985         tcp->sys_func_rval = 0;
986         free_tcb_priv_data(tcp);
987         return 0;
988 }
989
990 int
991 trace_syscall(struct tcb *tcp, unsigned int *signo)
992 {
993         return exiting(tcp) ?
994                 trace_syscall_exiting(tcp) : trace_syscall_entering(tcp, signo);
995 }
996
997 bool
998 is_erestart(struct tcb *tcp)
999 {
1000         switch (tcp->u_error) {
1001                 case ERESTARTSYS:
1002                 case ERESTARTNOINTR:
1003                 case ERESTARTNOHAND:
1004                 case ERESTART_RESTARTBLOCK:
1005                         return true;
1006                 default:
1007                         return false;
1008         }
1009 }
1010
1011 static unsigned long saved_u_error;
1012
1013 void
1014 temporarily_clear_syserror(struct tcb *tcp)
1015 {
1016         saved_u_error = tcp->u_error;
1017         tcp->u_error = 0;
1018 }
1019
1020 void
1021 restore_cleared_syserror(struct tcb *tcp)
1022 {
1023         tcp->u_error = saved_u_error;
1024 }
1025
1026 /*
1027  * Check the syscall return value register value for whether it is
1028  * a negated errno code indicating an error, or a success return value.
1029  */
1030 static inline bool
1031 is_negated_errno(kernel_ulong_t val)
1032 {
1033         /* Linux kernel defines MAX_ERRNO to 4095. */
1034         kernel_ulong_t max = -(kernel_long_t) 4095;
1035
1036 #ifndef current_klongsize
1037         if (current_klongsize < sizeof(val)) {
1038                 val = (uint32_t) val;
1039                 max = (uint32_t) max;
1040         }
1041 #endif /* !current_klongsize */
1042
1043         return val >= max;
1044 }
1045
1046 #include "arch_regs.c"
1047
1048 #ifdef HAVE_GETRVAL2
1049 # include "arch_getrval2.c"
1050 #endif
1051
1052 void
1053 print_pc(struct tcb *tcp)
1054 {
1055 #if defined ARCH_PC_REG
1056 # define ARCH_GET_PC 0
1057 #elif defined ARCH_PC_PEEK_ADDR
1058         kernel_ulong_t pc;
1059 # define ARCH_PC_REG pc
1060 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1061 #else
1062 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1063 #endif
1064         if (get_regs_error || ARCH_GET_PC)
1065                 tprints(current_wordsize == 4 ? "[????????] "
1066                                               : "[????????????????] ");
1067         else
1068                 tprintf(current_wordsize == 4
1069                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1070                         (kernel_ulong_t) ARCH_PC_REG);
1071 }
1072
1073 #include "getregs_old.h"
1074
1075 #undef ptrace_getregset_or_getregs
1076 #undef ptrace_setregset_or_setregs
1077 #ifdef ARCH_REGS_FOR_GETREGSET
1078
1079 # define ptrace_getregset_or_getregs ptrace_getregset
1080 static long
1081 ptrace_getregset(pid_t pid)
1082 {
1083 # ifdef ARCH_IOVEC_FOR_GETREGSET
1084         /* variable iovec */
1085         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1086         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1087                       &ARCH_IOVEC_FOR_GETREGSET);
1088 # else
1089         /* constant iovec */
1090         static struct iovec io = {
1091                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1092                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1093         };
1094         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1095
1096 # endif
1097 }
1098
1099 # ifndef HAVE_GETREGS_OLD
1100 #  define ptrace_setregset_or_setregs ptrace_setregset
1101 static int
1102 ptrace_setregset(pid_t pid)
1103 {
1104 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1105         /* variable iovec */
1106         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1107                       &ARCH_IOVEC_FOR_GETREGSET);
1108 #  else
1109         /* constant iovec */
1110         static struct iovec io = {
1111                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1112                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1113         };
1114         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1115 #  endif
1116 }
1117 # endif /* !HAVE_GETREGS_OLD */
1118
1119 #elif defined ARCH_REGS_FOR_GETREGS
1120
1121 # define ptrace_getregset_or_getregs ptrace_getregs
1122 static long
1123 ptrace_getregs(pid_t pid)
1124 {
1125 # if defined SPARC || defined SPARC64
1126         /* SPARC systems have the meaning of data and addr reversed */
1127         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1128 # else
1129         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1130 # endif
1131 }
1132
1133 # ifndef HAVE_GETREGS_OLD
1134 #  define ptrace_setregset_or_setregs ptrace_setregs
1135 static int
1136 ptrace_setregs(pid_t pid)
1137 {
1138 #  if defined SPARC || defined SPARC64
1139         /* SPARC systems have the meaning of data and addr reversed */
1140         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1141 #  else
1142         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1143 #  endif
1144 }
1145 # endif /* !HAVE_GETREGS_OLD */
1146
1147 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1148
1149 void
1150 get_regs(pid_t pid)
1151 {
1152 #undef USE_GET_SYSCALL_RESULT_REGS
1153 #ifdef ptrace_getregset_or_getregs
1154
1155 # ifdef HAVE_GETREGS_OLD
1156         /*
1157          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1158          * fallback to getregs_old.
1159          */
1160         static int use_getregs_old;
1161         if (use_getregs_old < 0) {
1162                 get_regs_error = ptrace_getregset_or_getregs(pid);
1163                 return;
1164         } else if (use_getregs_old == 0) {
1165                 get_regs_error = ptrace_getregset_or_getregs(pid);
1166                 if (get_regs_error >= 0) {
1167                         use_getregs_old = -1;
1168                         return;
1169                 }
1170                 if (errno == EPERM || errno == ESRCH)
1171                         return;
1172                 use_getregs_old = 1;
1173         }
1174         get_regs_error = getregs_old(pid);
1175 # else /* !HAVE_GETREGS_OLD */
1176         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1177         get_regs_error = ptrace_getregset_or_getregs(pid);
1178 # endif /* !HAVE_GETREGS_OLD */
1179
1180 #else /* !ptrace_getregset_or_getregs */
1181
1182 # define USE_GET_SYSCALL_RESULT_REGS 1
1183 # warning get_regs is not implemented for this architecture yet
1184         get_regs_error = 0;
1185
1186 #endif /* !ptrace_getregset_or_getregs */
1187 }
1188
1189 #ifdef ptrace_setregset_or_setregs
1190 static int
1191 set_regs(pid_t pid)
1192 {
1193         return ptrace_setregset_or_setregs(pid);
1194 }
1195 #endif /* ptrace_setregset_or_setregs */
1196
1197 struct sysent_buf {
1198         struct tcb *tcp;
1199         struct_sysent ent;
1200         char buf[sizeof("syscall_%lu") + sizeof(kernel_ulong_t) * 3];
1201 };
1202
1203 static void
1204 free_sysent_buf(void *ptr)
1205 {
1206         struct sysent_buf *s = ptr;
1207         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1208         free(ptr);
1209 }
1210
1211 /*
1212  * Returns:
1213  * 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
1214  * 1: ok, continue in trace_syscall_entering().
1215  * other: error, trace_syscall_entering() should print error indicator
1216  *    ("????" etc) and bail out.
1217  */
1218 int
1219 get_scno(struct tcb *tcp)
1220 {
1221         if (get_regs_error)
1222                 return -1;
1223
1224         int rc = arch_get_scno(tcp);
1225         if (rc != 1)
1226                 return rc;
1227
1228         if (scno_is_valid(tcp->scno)) {
1229                 tcp->s_ent = &sysent[tcp->scno];
1230                 tcp->qual_flg = qual_flags(tcp->scno);
1231         } else {
1232                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1233
1234                 s->tcp = tcp;
1235                 s->ent.nargs = MAX_ARGS;
1236                 s->ent.sen = SEN_printargs;
1237                 s->ent.sys_func = printargs;
1238                 s->ent.sys_name = s->buf;
1239                 sprintf(s->buf, "syscall_%" PRI_klu, shuffle_scno(tcp->scno));
1240
1241                 tcp->s_ent = &s->ent;
1242                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1243
1244                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1245
1246                 if (debug_flag)
1247                         error_msg("pid %d invalid syscall %" PRI_kld,
1248                                   tcp->pid, tcp->scno);
1249         }
1250         return 1;
1251 }
1252
1253 #ifdef USE_GET_SYSCALL_RESULT_REGS
1254 static int get_syscall_result_regs(struct tcb *);
1255 #endif
1256
1257 /* Returns:
1258  * 1: ok, continue in trace_syscall_exiting().
1259  * -1: error, trace_syscall_exiting() should print error indicator
1260  *    ("????" etc) and bail out.
1261  */
1262 static int
1263 get_syscall_result(struct tcb *tcp)
1264 {
1265 #ifdef USE_GET_SYSCALL_RESULT_REGS
1266         if (get_syscall_result_regs(tcp))
1267                 return -1;
1268 #endif
1269         tcp->u_error = 0;
1270         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1271
1272         return 1;
1273 }
1274
1275 #include "get_scno.c"
1276 #include "set_scno.c"
1277 #include "get_syscall_args.c"
1278 #ifdef USE_GET_SYSCALL_RESULT_REGS
1279 # include "get_syscall_result.c"
1280 #endif
1281 #include "get_error.c"
1282 #include "set_error.c"
1283 #ifdef HAVE_GETREGS_OLD
1284 # include "getregs_old.c"
1285 #endif
1286
1287 const char *
1288 syscall_name(kernel_ulong_t scno)
1289 {
1290 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1291         if (current_personality == X32_PERSONALITY_NUMBER)
1292                 scno &= ~__X32_SYSCALL_BIT;
1293 #endif
1294         return scno_is_valid(scno) ? sysent[scno].sys_name: NULL;
1295 }