]> granicus.if.org Git - strace/blob - syscall.c
Update copyright headers
[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  * Copyright (c) 1999-2018 The strace developers.
10  * All rights reserved.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. The name of the author may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34
35 #include "defs.h"
36 #include "native_defs.h"
37 #include "ptrace.h"
38 #include "nsig.h"
39 #include "number_set.h"
40 #include <limits.h>
41
42 /* for struct iovec */
43 #include <sys/uio.h>
44
45 /* for __X32_SYSCALL_BIT */
46 #include <asm/unistd.h>
47
48 #include "regs.h"
49
50 #if defined(SPARC64)
51 # undef PTRACE_GETREGS
52 # define PTRACE_GETREGS PTRACE_GETREGS64
53 # undef PTRACE_SETREGS
54 # define PTRACE_SETREGS PTRACE_SETREGS64
55 #endif
56
57 #ifndef NT_PRSTATUS
58 # define NT_PRSTATUS 1
59 #endif
60
61 #include "syscall.h"
62 #include "xstring.h"
63
64 /* Define these shorthand notations to simplify the syscallent files. */
65 #include "sysent_shorthand_defs.h"
66
67 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
68
69 const struct_sysent sysent0[] = {
70 #include "syscallent.h"
71 };
72
73 #if SUPPORTED_PERSONALITIES > 1
74 # include PERSONALITY1_INCLUDE_FUNCS
75 static const struct_sysent sysent1[] = {
76 # include "syscallent1.h"
77 };
78 #endif
79
80 #if SUPPORTED_PERSONALITIES > 2
81 # include PERSONALITY2_INCLUDE_FUNCS
82 static const struct_sysent sysent2[] = {
83 # include "syscallent2.h"
84 };
85 #endif
86
87 /* Now undef them since short defines cause wicked namespace pollution. */
88 #include "sysent_shorthand_undefs.h"
89
90 /*
91  * `ioctlent[012].h' files are automatically generated by the auxiliary
92  * program `ioctlsort', such that the list is sorted by the `code' field.
93  * This has the side-effect of resolving the _IO.. macros into
94  * plain integers, eliminating the need to include here everything
95  * in "/usr/include".
96  */
97
98 const char *const errnoent0[] = {
99 #include "errnoent.h"
100 };
101 const char *const signalent0[] = {
102 #include "signalent.h"
103 };
104 const struct_ioctlent ioctlent0[] = {
105 #include "ioctlent0.h"
106 };
107
108 #if SUPPORTED_PERSONALITIES > 1
109 static const char *const errnoent1[] = {
110 # include "errnoent1.h"
111 };
112 static const char *const signalent1[] = {
113 # include "signalent1.h"
114 };
115 static const struct_ioctlent ioctlent1[] = {
116 # include "ioctlent1.h"
117 };
118 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
119 static const struct_printers printers0 = {
120 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
121 };
122 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
123 static const struct_printers printers1 = {
124 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
125 };
126 #endif
127
128 #if SUPPORTED_PERSONALITIES > 2
129 static const char *const errnoent2[] = {
130 # include "errnoent2.h"
131 };
132 static const char *const signalent2[] = {
133 # include "signalent2.h"
134 };
135 static const struct_ioctlent ioctlent2[] = {
136 # include "ioctlent2.h"
137 };
138 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
139 static const struct_printers printers2 = {
140 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
141 };
142 #endif
143
144 enum {
145         nsyscalls0 = ARRAY_SIZE(sysent0)
146 #if SUPPORTED_PERSONALITIES > 1
147         , nsyscalls1 = ARRAY_SIZE(sysent1)
148 # if SUPPORTED_PERSONALITIES > 2
149         , nsyscalls2 = ARRAY_SIZE(sysent2)
150 # endif
151 #endif
152 };
153
154 enum {
155         nerrnos0 = ARRAY_SIZE(errnoent0)
156 #if SUPPORTED_PERSONALITIES > 1
157         , nerrnos1 = ARRAY_SIZE(errnoent1)
158 # if SUPPORTED_PERSONALITIES > 2
159         , nerrnos2 = ARRAY_SIZE(errnoent2)
160 # endif
161 #endif
162 };
163
164 enum {
165         nsignals0 = ARRAY_SIZE(signalent0)
166 #if SUPPORTED_PERSONALITIES > 1
167         , nsignals1 = ARRAY_SIZE(signalent1)
168 # if SUPPORTED_PERSONALITIES > 2
169         , nsignals2 = ARRAY_SIZE(signalent2)
170 # endif
171 #endif
172 };
173
174 enum {
175         nioctlents0 = ARRAY_SIZE(ioctlent0)
176 #if SUPPORTED_PERSONALITIES > 1
177         , nioctlents1 = ARRAY_SIZE(ioctlent1)
178 # if SUPPORTED_PERSONALITIES > 2
179         , nioctlents2 = ARRAY_SIZE(ioctlent2)
180 # endif
181 #endif
182 };
183
184 #if SUPPORTED_PERSONALITIES > 1
185 const struct_sysent *sysent = sysent0;
186 const char *const *errnoent = errnoent0;
187 const char *const *signalent = signalent0;
188 const struct_ioctlent *ioctlent = ioctlent0;
189 const struct_printers *printers = &printers0;
190 #endif
191
192 unsigned nsyscalls = nsyscalls0;
193 unsigned nerrnos = nerrnos0;
194 unsigned nsignals = nsignals0;
195 unsigned nioctlents = nioctlents0;
196
197 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
198         nsyscalls0,
199 #if SUPPORTED_PERSONALITIES > 1
200         nsyscalls1,
201 #endif
202 #if SUPPORTED_PERSONALITIES > 2
203         nsyscalls2,
204 #endif
205 };
206 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
207         sysent0,
208 #if SUPPORTED_PERSONALITIES > 1
209         sysent1,
210 #endif
211 #if SUPPORTED_PERSONALITIES > 2
212         sysent2,
213 #endif
214 };
215
216 const char *const personality_names[] =
217 # if defined X86_64
218         {"64 bit", "32 bit", "x32"}
219 # elif defined X32
220         {"x32", "32 bit"}
221 # elif SUPPORTED_PERSONALITIES == 2
222         {"64 bit", "32 bit"}
223 # else
224         {STRINGIFY_VAL(__WORDSIZE) " bit"}
225 # endif
226         ;
227
228 #if SUPPORTED_PERSONALITIES > 1
229
230 unsigned current_personality;
231
232 # ifndef current_wordsize
233 unsigned current_wordsize = PERSONALITY0_WORDSIZE;
234 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
235         PERSONALITY0_WORDSIZE,
236         PERSONALITY1_WORDSIZE,
237 # if SUPPORTED_PERSONALITIES > 2
238         PERSONALITY2_WORDSIZE,
239 # endif
240 };
241 # endif
242
243 # ifndef current_klongsize
244 unsigned current_klongsize = PERSONALITY0_KLONGSIZE;
245 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
246         PERSONALITY0_KLONGSIZE,
247         PERSONALITY1_KLONGSIZE,
248 #  if SUPPORTED_PERSONALITIES > 2
249         PERSONALITY2_KLONGSIZE,
250 #  endif
251 };
252 # endif
253
254 void
255 set_personality(unsigned int personality)
256 {
257         if (personality == current_personality)
258                 return;
259
260         if (personality >= SUPPORTED_PERSONALITIES)
261                 error_msg_and_die("Requested switch to unsupported personality "
262                                   "%u", personality);
263
264         nsyscalls = nsyscall_vec[personality];
265         sysent = sysent_vec[personality];
266
267         switch (personality) {
268         case 0:
269                 errnoent = errnoent0;
270                 nerrnos = nerrnos0;
271                 ioctlent = ioctlent0;
272                 nioctlents = nioctlents0;
273                 signalent = signalent0;
274                 nsignals = nsignals0;
275                 printers = &printers0;
276                 break;
277
278         case 1:
279                 errnoent = errnoent1;
280                 nerrnos = nerrnos1;
281                 ioctlent = ioctlent1;
282                 nioctlents = nioctlents1;
283                 signalent = signalent1;
284                 nsignals = nsignals1;
285                 printers = &printers1;
286                 break;
287
288 # if SUPPORTED_PERSONALITIES > 2
289         case 2:
290                 errnoent = errnoent2;
291                 nerrnos = nerrnos2;
292                 ioctlent = ioctlent2;
293                 nioctlents = nioctlents2;
294                 signalent = signalent2;
295                 nsignals = nsignals2;
296                 printers = &printers2;
297                 break;
298 # endif
299         }
300
301         current_personality = personality;
302 # ifndef current_wordsize
303         current_wordsize = personality_wordsize[personality];
304 # endif
305 # ifndef current_klongsize
306         current_klongsize = personality_klongsize[personality];
307 # endif
308 }
309
310 static void
311 update_personality(struct tcb *tcp, unsigned int personality)
312 {
313         static bool need_mpers_warning[] =
314                 { false, !HAVE_PERSONALITY_1_MPERS, !HAVE_PERSONALITY_2_MPERS };
315
316         set_personality(personality);
317
318         if (personality == tcp->currpers)
319                 return;
320         tcp->currpers = personality;
321
322         if (!qflag) {
323                 error_msg("[ Process PID=%d runs in %s mode. ]",
324                           tcp->pid, personality_names[personality]);
325         }
326
327         if (need_mpers_warning[personality]) {
328                 error_msg("WARNING: Proper structure decoding for this "
329                           "personality is not supported, please consider "
330                           "building strace with mpers support enabled.");
331                 need_mpers_warning[personality] = false;
332         }
333 }
334 #endif
335
336 #ifdef SYS_socket_subcall
337 static void
338 decode_socket_subcall(struct tcb *tcp)
339 {
340         const int call = tcp->u_arg[0];
341
342         if (call < 1 || call >= SYS_socket_nsubcalls)
343                 return;
344
345         const kernel_ulong_t scno = SYS_socket_subcall + call;
346         const unsigned int nargs = sysent[scno].nargs;
347         uint64_t buf[nargs];
348
349         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
350                 return;
351
352         tcp->scno = scno;
353         tcp->qual_flg = qual_flags(scno);
354         tcp->s_ent = &sysent[scno];
355
356         unsigned int i;
357         for (i = 0; i < nargs; ++i)
358                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
359                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
360 }
361 #endif /* SYS_socket_subcall */
362
363 #ifdef SYS_ipc_subcall
364 static void
365 decode_ipc_subcall(struct tcb *tcp)
366 {
367         unsigned int call = tcp->u_arg[0];
368         const unsigned int version = call >> 16;
369
370         if (version) {
371 # if defined S390 || defined S390X
372                 return;
373 # else
374 #  ifdef SPARC64
375                 if (current_wordsize == 8)
376                         return;
377 #  endif
378                 set_tcb_priv_ulong(tcp, version);
379                 call &= 0xffff;
380 # endif
381         }
382
383         switch (call) {
384                 case  1: case  2: case  3: case  4:
385                 case 11: case 12: case 13: case 14:
386                 case 21: case 22: case 23: case 24:
387                         break;
388                 default:
389                         return;
390         }
391
392         tcp->scno = SYS_ipc_subcall + call;
393         tcp->qual_flg = qual_flags(tcp->scno);
394         tcp->s_ent = &sysent[tcp->scno];
395
396         const unsigned int n = tcp->s_ent->nargs;
397         unsigned int i;
398         for (i = 0; i < n; i++)
399                 tcp->u_arg[i] = tcp->u_arg[i + 1];
400 }
401 #endif /* SYS_ipc_subcall */
402
403 #ifdef SYS_syscall_subcall
404 static void
405 decode_syscall_subcall(struct tcb *tcp)
406 {
407         if (!scno_is_valid(tcp->u_arg[0]))
408                 return;
409         tcp->scno = tcp->u_arg[0];
410         tcp->qual_flg = qual_flags(tcp->scno);
411         tcp->s_ent = &sysent[tcp->scno];
412         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
413                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
414 # ifdef LINUX_MIPSO32
415         /*
416          * Fetching the last arg of 7-arg syscalls (fadvise64_64
417          * and sync_file_range) requires additional code,
418          * see linux/mips/get_syscall_args.c
419          */
420         if (tcp->s_ent->nargs == MAX_ARGS) {
421                 if (umoven(tcp,
422                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
423                            sizeof(tcp->u_arg[0]),
424                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
425                 tcp->u_arg[MAX_ARGS - 1] = 0;
426         }
427 # endif /* LINUX_MIPSO32 */
428 }
429 #endif /* SYS_syscall_subcall */
430
431 static void
432 dumpio(struct tcb *tcp)
433 {
434         if (syserror(tcp))
435                 return;
436
437         int fd = tcp->u_arg[0];
438         if (fd < 0)
439                 return;
440
441         if (is_number_in_set(fd, read_set)) {
442                 switch (tcp->s_ent->sen) {
443                 case SEN_read:
444                 case SEN_pread:
445                 case SEN_recv:
446                 case SEN_recvfrom:
447                 case SEN_mq_timedreceive:
448                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
449                         return;
450                 case SEN_readv:
451                 case SEN_preadv:
452                 case SEN_preadv2:
453                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
454                                      tcp->u_rval);
455                         return;
456                 case SEN_recvmsg:
457                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
458                         return;
459                 case SEN_recvmmsg:
460                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
461                         return;
462                 }
463         }
464         if (is_number_in_set(fd, write_set)) {
465                 switch (tcp->s_ent->sen) {
466                 case SEN_write:
467                 case SEN_pwrite:
468                 case SEN_send:
469                 case SEN_sendto:
470                 case SEN_mq_timedsend:
471                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
472                         break;
473                 case SEN_writev:
474                 case SEN_pwritev:
475                 case SEN_pwritev2:
476                 case SEN_vmsplice:
477                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
478                         break;
479                 case SEN_sendmsg:
480                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
481                         break;
482                 case SEN_sendmmsg:
483                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
484                         break;
485                 }
486         }
487 }
488
489 const char *
490 err_name(unsigned long err)
491 {
492         if ((err < nerrnos) && errnoent[err])
493                 return errnoent[err];
494
495         return NULL;
496 }
497
498 static long get_regs(struct tcb *);
499 static int get_syscall_args(struct tcb *);
500 static int get_syscall_result(struct tcb *);
501 static int arch_get_scno(struct tcb *tcp);
502 static int arch_set_scno(struct tcb *, kernel_ulong_t);
503 static void get_error(struct tcb *, const bool);
504 static int arch_set_error(struct tcb *);
505 static int arch_set_success(struct tcb *);
506
507 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
508
509 static struct inject_opts *
510 tcb_inject_opts(struct tcb *tcp)
511 {
512         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
513                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
514 }
515
516
517 static long
518 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
519 {
520         if (!tcp->inject_vec[current_personality]) {
521                 tcp->inject_vec[current_personality] =
522                         xcalloc(nsyscalls, sizeof(**inject_vec));
523                 memcpy(tcp->inject_vec[current_personality],
524                        inject_vec[current_personality],
525                        nsyscalls * sizeof(**inject_vec));
526         }
527
528         struct inject_opts *opts = tcb_inject_opts(tcp);
529
530         if (!opts || opts->first == 0)
531                 return 0;
532
533         --opts->first;
534
535         if (opts->first != 0)
536                 return 0;
537
538         opts->first = opts->step;
539
540         if (!recovering(tcp)) {
541                 if (opts->data.flags & INJECT_F_SIGNAL)
542                         *signo = opts->data.signo;
543                 if (opts->data.flags & INJECT_F_RETVAL &&
544                     !arch_set_scno(tcp, -1))
545                         tcp->flags |= TCB_TAMPERED;
546         }
547
548         return 0;
549 }
550
551 static long
552 tamper_with_syscall_exiting(struct tcb *tcp)
553 {
554         if (!syserror(tcp)) {
555                 error_msg("Failed to tamper with process %d: got no error "
556                           "(return value %#" PRI_klx ")",
557                           tcp->pid, tcp->u_rval);
558
559                 return 1;
560         }
561
562         struct inject_opts *opts = tcb_inject_opts(tcp);
563         bool update_tcb = false;
564
565         if (!opts)
566                 return 0;
567
568         if (opts->data.rval >= 0) {
569                 kernel_long_t u_rval = tcp->u_rval;
570
571                 tcp->u_rval = opts->data.rval;
572                 if (arch_set_success(tcp)) {
573                         tcp->u_rval = u_rval;
574                 } else {
575                         update_tcb = true;
576                         tcp->u_error = 0;
577                 }
578         } else {
579                 unsigned long new_error = -opts->data.rval;
580
581                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
582                         unsigned long u_error = tcp->u_error;
583
584                         tcp->u_error = new_error;
585                         if (arch_set_error(tcp)) {
586                                 tcp->u_error = u_error;
587                         } else {
588                                 update_tcb = true;
589                         }
590                 }
591         }
592
593         if (update_tcb) {
594                 tcp->u_error = 0;
595                 get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
596         }
597
598         return 0;
599 }
600
601 /*
602  * Returns:
603  * 0: "ignore this ptrace stop", bail out silently.
604  * 1: ok, decoded; call
605  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
606  * other: error; call syscall_entering_finish(tcp, res), where res is the value
607  *    returned.
608  */
609 int
610 syscall_entering_decode(struct tcb *tcp)
611 {
612         int res = get_scno(tcp);
613         if (res == 0)
614                 return res;
615         int scno_good = res;
616         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
617                 printleader(tcp);
618                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
619                 /*
620                  * " <unavailable>" will be added later by the code which
621                  * detects ptrace errors.
622                  */
623                 return res;
624         }
625
626 #if defined SYS_ipc_subcall     \
627  || defined SYS_socket_subcall  \
628  || defined SYS_syscall_subcall
629         for (;;) {
630                 switch (tcp->s_ent->sen) {
631 # ifdef SYS_ipc_subcall
632                 case SEN_ipc:
633                         decode_ipc_subcall(tcp);
634                         break;
635 # endif
636 # ifdef SYS_socket_subcall
637                 case SEN_socketcall:
638                         decode_socket_subcall(tcp);
639                         break;
640 # endif
641 # ifdef SYS_syscall_subcall
642                 case SEN_syscall:
643                         decode_syscall_subcall(tcp);
644                         if (tcp->s_ent->sen != SEN_syscall)
645                                 continue;
646                         break;
647 # endif
648                 }
649                 break;
650         }
651 #endif
652
653         return 1;
654 }
655
656 int
657 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
658 {
659         /* Restrain from fault injection while the trace executes strace code. */
660         if (hide_log(tcp)) {
661                 tcp->qual_flg &= ~QUAL_INJECT;
662         }
663
664         switch (tcp->s_ent->sen) {
665                 case SEN_execve:
666                 case SEN_execveat:
667 #if defined SPARC || defined SPARC64
668                 case SEN_execv:
669 #endif
670                         tcp->flags &= ~TCB_HIDE_LOG;
671                         break;
672         }
673
674         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
675                 tcp->flags |= TCB_FILTERED;
676                 return 0;
677         }
678
679         tcp->flags &= ~TCB_FILTERED;
680
681         if (hide_log(tcp)) {
682                 return 0;
683         }
684
685         if (inject(tcp))
686                 tamper_with_syscall_entering(tcp, sig);
687
688         if (cflag == CFLAG_ONLY_STATS) {
689                 return 0;
690         }
691
692 #ifdef USE_LIBUNWIND
693         if (stack_trace_enabled) {
694                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
695                         unwind_capture_stacktrace(tcp);
696         }
697 #endif
698
699         printleader(tcp);
700         tprintf("%s(", tcp->s_ent->sys_name);
701         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
702         fflush(tcp->outf);
703         return res;
704 }
705
706 void
707 syscall_entering_finish(struct tcb *tcp, int res)
708 {
709         tcp->flags |= TCB_INSYSCALL;
710         tcp->sys_func_rval = res;
711         /* Measure the entrance time as late as possible to avoid errors. */
712         if ((Tflag || cflag) && !filtered(tcp))
713                 gettimeofday(&tcp->etime, NULL);
714 }
715
716 /* Returns:
717  * 0: "bail out".
718  * 1: ok.
719  * -1: error in one of ptrace ops.
720  *
721  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
722  *    value. Anyway, call syscall_exiting_finish(tcp) then.
723  */
724 int
725 syscall_exiting_decode(struct tcb *tcp, struct timeval *ptv)
726 {
727         /* Measure the exit time as early as possible to avoid errors. */
728         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
729                 gettimeofday(ptv, NULL);
730
731 #ifdef USE_LIBUNWIND
732         if (stack_trace_enabled) {
733                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
734                         unwind_cache_invalidate(tcp);
735         }
736 #endif
737
738         if (filtered(tcp) || hide_log(tcp))
739                 return 0;
740
741 #if SUPPORTED_PERSONALITIES > 1
742         update_personality(tcp, tcp->currpers);
743 #endif
744
745         return get_syscall_result(tcp);
746 }
747
748 int
749 syscall_exiting_trace(struct tcb *tcp, struct timeval tv, int res)
750 {
751         if (syscall_tampered(tcp))
752                 tamper_with_syscall_exiting(tcp);
753
754         if (cflag) {
755                 count_syscall(tcp, &tv);
756                 if (cflag == CFLAG_ONLY_STATS) {
757                         return 0;
758                 }
759         }
760
761         /* If not in -ff mode, and printing_tcp != tcp,
762          * then the log currently does not end with output
763          * of _our syscall entry_, but with something else.
764          * We need to say which syscall's return is this.
765          *
766          * Forced reprinting via TCB_REPRINT is used only by
767          * "strace -ff -oLOG test/threaded_execve" corner case.
768          * It's the only case when -ff mode needs reprinting.
769          */
770         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
771                 tcp->flags &= ~TCB_REPRINT;
772                 printleader(tcp);
773                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
774         }
775         printing_tcp = tcp;
776
777         tcp->s_prev_ent = NULL;
778         if (res != 1) {
779                 /* There was error in one of prior ptrace ops */
780                 tprints(") ");
781                 tabto();
782                 tprints("= ? <unavailable>\n");
783                 line_ended();
784                 return res;
785         }
786         tcp->s_prev_ent = tcp->s_ent;
787
788         int sys_res = 0;
789         if (raw(tcp)) {
790                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
791         } else {
792         /* FIXME: not_failing_only (IOW, option -z) is broken:
793          * failure of syscall is known only after syscall return.
794          * Thus we end up with something like this on, say, ENOENT:
795          *     open("does_not_exist", O_RDONLY <unfinished ...>
796          *     {next syscall decode}
797          * whereas the intended result is that open(...) line
798          * is not shown at all.
799          */
800                 if (not_failing_only && tcp->u_error)
801                         return 0;       /* ignore failed syscalls */
802                 if (tcp->sys_func_rval & RVAL_DECODED)
803                         sys_res = tcp->sys_func_rval;
804                 else
805                         sys_res = tcp->s_ent->sys_func(tcp);
806         }
807
808         tprints(") ");
809         tabto();
810         unsigned long u_error = tcp->u_error;
811         kernel_long_t u_rval;
812
813         if (raw(tcp)) {
814                 if (u_error) {
815                         tprintf("= -1 (errno %lu)", u_error);
816                 } else {
817                         tprintf("= %#" PRI_klx, tcp->u_rval);
818                 }
819                 if (syscall_tampered(tcp))
820                         tprints(" (INJECTED)");
821         } else if (!(sys_res & RVAL_NONE) && u_error) {
822                 const char *u_error_str;
823
824                 switch (u_error) {
825                 /* Blocked signals do not interrupt any syscalls.
826                  * In this case syscalls don't return ERESTARTfoo codes.
827                  *
828                  * Deadly signals set to SIG_DFL interrupt syscalls
829                  * and kill the process regardless of which of the codes below
830                  * is returned by the interrupted syscall.
831                  * In some cases, kernel forces a kernel-generated deadly
832                  * signal to be unblocked and set to SIG_DFL (and thus cause
833                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
834                  * or SIGILL. (The alternative is to leave process spinning
835                  * forever on the faulty instruction - not useful).
836                  *
837                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
838                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
839                  * but kernel will always restart them.
840                  */
841                 case ERESTARTSYS:
842                         /* Most common type of signal-interrupted syscall exit code.
843                          * The system call will be restarted with the same arguments
844                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
845                          */
846                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
847                         break;
848                 case ERESTARTNOINTR:
849                         /* Rare. For example, fork() returns this if interrupted.
850                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
851                          */
852                         tprints("= ? ERESTARTNOINTR (To be restarted)");
853                         break;
854                 case ERESTARTNOHAND:
855                         /* pause(), rt_sigsuspend() etc use this code.
856                          * SA_RESTART is ignored (assumed not set):
857                          * syscall won't restart (will return EINTR instead)
858                          * even after signal with SA_RESTART set. However,
859                          * after SIG_IGN or SIG_DFL signal it will restart
860                          * (thus the name "restart only if has no handler").
861                          */
862                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
863                         break;
864                 case ERESTART_RESTARTBLOCK:
865                         /* Syscalls like nanosleep(), poll() which can't be
866                          * restarted with their original arguments use this
867                          * code. Kernel will execute restart_syscall() instead,
868                          * which changes arguments before restarting syscall.
869                          * SA_RESTART is ignored (assumed not set) similarly
870                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
871                          * since restart data is saved in "restart block"
872                          * in task struct, and if signal handler uses a syscall
873                          * which in turn saves another such restart block,
874                          * old data is lost and restart becomes impossible)
875                          */
876                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
877                         break;
878                 default:
879                         u_rval = sys_res & RVAL_PRINT_ERR_VAL ?
880                                  tcp->u_rval : -1;
881                         u_error_str = err_name(u_error);
882                         if (u_error_str)
883                                 tprintf("= %" PRI_kld " %s (%s)",
884                                         u_rval, u_error_str, strerror(u_error));
885                         else
886                                 tprintf("= %" PRI_kld " %lu (%s)",
887                                         u_rval, u_error, strerror(u_error));
888                         break;
889                 }
890                 if (syscall_tampered(tcp))
891                         tprints(" (INJECTED)");
892                 if ((sys_res & RVAL_STR) && tcp->auxstr)
893                         tprintf(" (%s)", tcp->auxstr);
894         } else {
895                 if (sys_res & RVAL_NONE)
896                         tprints("= ?");
897                 else {
898                         switch (sys_res & RVAL_MASK) {
899                         case RVAL_HEX:
900 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
901                                 if (current_klongsize < sizeof(tcp->u_rval)) {
902                                         tprintf("= %#x",
903                                                 (unsigned int) tcp->u_rval);
904                                 } else
905 #endif
906                                 {
907                                         tprintf("= %#" PRI_klx, tcp->u_rval);
908                                 }
909                                 break;
910                         case RVAL_OCTAL:
911                                 tprints("= ");
912                                 print_numeric_long_umask(tcp->u_rval);
913                                 break;
914                         case RVAL_UDECIMAL:
915 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
916                                 if (current_klongsize < sizeof(tcp->u_rval)) {
917                                         tprintf("= %u",
918                                                 (unsigned int) tcp->u_rval);
919                                 } else
920 #endif
921                                 {
922                                         tprintf("= %" PRI_klu, tcp->u_rval);
923                                 }
924                                 break;
925                         case RVAL_DECIMAL:
926                                 tprintf("= %" PRI_kld, tcp->u_rval);
927                                 break;
928                         case RVAL_FD:
929                                 if (show_fd_path) {
930                                         tprints("= ");
931                                         printfd(tcp, tcp->u_rval);
932                                 } else
933                                         tprintf("= %" PRI_kld, tcp->u_rval);
934                                 break;
935                         default:
936                                 error_msg("invalid rval format");
937                                 break;
938                         }
939                 }
940                 if ((sys_res & RVAL_STR) && tcp->auxstr)
941                         tprintf(" (%s)", tcp->auxstr);
942                 if (syscall_tampered(tcp))
943                         tprints(" (INJECTED)");
944         }
945         if (Tflag) {
946                 tv_sub(&tv, &tv, &tcp->etime);
947                 tprintf(" <%ld.%06ld>",
948                         (long) tv.tv_sec, (long) tv.tv_usec);
949         }
950         tprints("\n");
951         dumpio(tcp);
952         line_ended();
953
954 #ifdef USE_LIBUNWIND
955         if (stack_trace_enabled)
956                 unwind_print_stacktrace(tcp);
957 #endif
958         return 0;
959 }
960
961 void
962 syscall_exiting_finish(struct tcb *tcp)
963 {
964         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
965         tcp->sys_func_rval = 0;
966         free_tcb_priv_data(tcp);
967 }
968
969 bool
970 is_erestart(struct tcb *tcp)
971 {
972         switch (tcp->u_error) {
973                 case ERESTARTSYS:
974                 case ERESTARTNOINTR:
975                 case ERESTARTNOHAND:
976                 case ERESTART_RESTARTBLOCK:
977                         return true;
978                 default:
979                         return false;
980         }
981 }
982
983 static unsigned long saved_u_error;
984
985 void
986 temporarily_clear_syserror(struct tcb *tcp)
987 {
988         saved_u_error = tcp->u_error;
989         tcp->u_error = 0;
990 }
991
992 void
993 restore_cleared_syserror(struct tcb *tcp)
994 {
995         tcp->u_error = saved_u_error;
996 }
997
998 #include "arch_regs.c"
999
1000 #if HAVE_ARCH_GETRVAL2
1001 # include "arch_getrval2.c"
1002 #endif
1003
1004 void
1005 print_pc(struct tcb *tcp)
1006 {
1007 #if defined ARCH_PC_REG
1008 # define ARCH_GET_PC 0
1009 #elif defined ARCH_PC_PEEK_ADDR
1010         kernel_ulong_t pc;
1011 # define ARCH_PC_REG pc
1012 # define ARCH_GET_PC upeek(tcp, ARCH_PC_PEEK_ADDR, &pc)
1013 #else
1014 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1015 #endif
1016         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1017                 tprints(current_wordsize == 4 ? "[????????] "
1018                                               : "[????????????????] ");
1019         else
1020                 tprintf(current_wordsize == 4
1021                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1022                         (kernel_ulong_t) ARCH_PC_REG);
1023 }
1024
1025 #include "getregs_old.h"
1026
1027 #undef ptrace_getregset_or_getregs
1028 #undef ptrace_setregset_or_setregs
1029 #ifdef ARCH_REGS_FOR_GETREGSET
1030
1031 # define ptrace_getregset_or_getregs ptrace_getregset
1032 static long
1033 ptrace_getregset(pid_t pid)
1034 {
1035 # ifdef ARCH_IOVEC_FOR_GETREGSET
1036         /* variable iovec */
1037         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1038         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1039                       &ARCH_IOVEC_FOR_GETREGSET);
1040 # else
1041         /* constant iovec */
1042         static struct iovec io = {
1043                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1044                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1045         };
1046         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1047
1048 # endif
1049 }
1050
1051 # ifndef HAVE_GETREGS_OLD
1052 #  define ptrace_setregset_or_setregs ptrace_setregset
1053 static int
1054 ptrace_setregset(pid_t pid)
1055 {
1056 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1057         /* variable iovec */
1058         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1059                       &ARCH_IOVEC_FOR_GETREGSET);
1060 #  else
1061         /* constant iovec */
1062         static struct iovec io = {
1063                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1064                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1065         };
1066         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1067 #  endif
1068 }
1069 # endif /* !HAVE_GETREGS_OLD */
1070
1071 #elif defined ARCH_REGS_FOR_GETREGS
1072
1073 # define ptrace_getregset_or_getregs ptrace_getregs
1074 static long
1075 ptrace_getregs(pid_t pid)
1076 {
1077 # if defined SPARC || defined SPARC64
1078         /* SPARC systems have the meaning of data and addr reversed */
1079         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1080 # else
1081         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1082 # endif
1083 }
1084
1085 # ifndef HAVE_GETREGS_OLD
1086 #  define ptrace_setregset_or_setregs ptrace_setregs
1087 static int
1088 ptrace_setregs(pid_t pid)
1089 {
1090 #  if defined SPARC || defined SPARC64
1091         /* SPARC systems have the meaning of data and addr reversed */
1092         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1093 #  else
1094         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1095 #  endif
1096 }
1097 # endif /* !HAVE_GETREGS_OLD */
1098
1099 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1100
1101 #ifdef ptrace_getregset_or_getregs
1102 static long get_regs_error;
1103 #endif
1104
1105 void
1106 clear_regs(struct tcb *tcp)
1107 {
1108 #ifdef ptrace_getregset_or_getregs
1109         get_regs_error = -1;
1110 #endif
1111 }
1112
1113 static long
1114 get_regs(struct tcb *const tcp)
1115 {
1116 #ifdef ptrace_getregset_or_getregs
1117
1118         if (get_regs_error != -1)
1119                 return get_regs_error;
1120
1121 # ifdef HAVE_GETREGS_OLD
1122         /*
1123          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1124          * fallback to getregs_old.
1125          */
1126         static int use_getregs_old;
1127         if (use_getregs_old < 0) {
1128                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1129         } else if (use_getregs_old == 0) {
1130                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1131                 if (get_regs_error >= 0) {
1132                         use_getregs_old = -1;
1133                         return get_regs_error;
1134                 }
1135                 if (errno == EPERM || errno == ESRCH)
1136                         return get_regs_error;
1137                 use_getregs_old = 1;
1138         }
1139         return get_regs_error = getregs_old(tcp);
1140 # else /* !HAVE_GETREGS_OLD */
1141         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1142         get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1143
1144 #  if defined ARCH_PERSONALITY_0_IOV_SIZE
1145         if (get_regs_error)
1146                 return get_regs_error;
1147
1148         switch (ARCH_IOVEC_FOR_GETREGSET.iov_len) {
1149         case ARCH_PERSONALITY_0_IOV_SIZE:
1150                 update_personality(tcp, 0);
1151                 break;
1152         case ARCH_PERSONALITY_1_IOV_SIZE:
1153                 update_personality(tcp, 1);
1154                 break;
1155         default: {
1156                 static bool printed = false;
1157
1158                 if (!printed) {
1159                         error_msg("Unsupported regset size returned by "
1160                                   "PTRACE_GETREGSET: %zu",
1161                                   ARCH_IOVEC_FOR_GETREGSET.iov_len);
1162
1163                         printed = true;
1164                 }
1165
1166                 update_personality(tcp, 0);
1167         }
1168         }
1169 #  endif /* ARCH_PERSONALITY_0_IOV_SIZE */
1170
1171         return get_regs_error;
1172
1173 # endif /* !HAVE_GETREGS_OLD */
1174
1175 #else /* !ptrace_getregset_or_getregs */
1176
1177 # warning get_regs is not implemented for this architecture yet
1178         return 0;
1179
1180 #endif /* !ptrace_getregset_or_getregs */
1181 }
1182
1183 #ifdef ptrace_setregset_or_setregs
1184 static int
1185 set_regs(pid_t pid)
1186 {
1187         return ptrace_setregset_or_setregs(pid);
1188 }
1189 #endif /* ptrace_setregset_or_setregs */
1190
1191 struct sysent_buf {
1192         struct tcb *tcp;
1193         struct_sysent ent;
1194         char buf[sizeof("syscall_0x") + sizeof(kernel_ulong_t) * 2];
1195 };
1196
1197 static void
1198 free_sysent_buf(void *ptr)
1199 {
1200         struct sysent_buf *s = ptr;
1201         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1202         free(ptr);
1203 }
1204
1205 /*
1206  * Returns:
1207  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1208  *    out silently" code.
1209  * 1: ok, continue in syscall_entering_decode().
1210  * other: error, syscall_entering_decode() should print error indicator
1211  *    ("????" etc) and return an appropriate code.
1212  */
1213 int
1214 get_scno(struct tcb *tcp)
1215 {
1216         if (get_regs(tcp) < 0)
1217                 return -1;
1218
1219         int rc = arch_get_scno(tcp);
1220         if (rc != 1)
1221                 return rc;
1222
1223         tcp->scno = shuffle_scno(tcp->scno);
1224
1225         if (scno_is_valid(tcp->scno)) {
1226                 tcp->s_ent = &sysent[tcp->scno];
1227                 tcp->qual_flg = qual_flags(tcp->scno);
1228         } else {
1229                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1230
1231                 s->tcp = tcp;
1232                 s->ent.nargs = MAX_ARGS;
1233                 s->ent.sen = SEN_printargs;
1234                 s->ent.sys_func = printargs;
1235                 s->ent.sys_name = s->buf;
1236                 xsprintf(s->buf, "syscall_%#" PRI_klx, shuffle_scno(tcp->scno));
1237
1238                 tcp->s_ent = &s->ent;
1239                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1240
1241                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1242
1243                 debug_msg("pid %d invalid syscall %#" PRI_klx,
1244                           tcp->pid, shuffle_scno(tcp->scno));
1245         }
1246
1247         /*
1248          * We refrain from argument decoding during recovering
1249          * as tracee memory mappings has changed and the registers
1250          * are very likely pointing to garbage already.
1251          */
1252         if (recovering(tcp))
1253                 tcp->qual_flg |= QUAL_RAW;
1254
1255         return 1;
1256 }
1257
1258 #ifdef ptrace_getregset_or_getregs
1259 # define get_syscall_result_regs get_regs
1260 #else
1261 static int get_syscall_result_regs(struct tcb *);
1262 #endif
1263
1264 /* Returns:
1265  * 1: ok, continue in syscall_exiting_trace().
1266  * -1: error, syscall_exiting_trace() should print error indicator
1267  *    ("????" etc) and bail out.
1268  */
1269 static int
1270 get_syscall_result(struct tcb *tcp)
1271 {
1272         if (get_syscall_result_regs(tcp) < 0)
1273                 return -1;
1274         tcp->u_error = 0;
1275         get_error(tcp,
1276                   !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS)
1277                         || syscall_tampered(tcp));
1278
1279         return 1;
1280 }
1281
1282 #include "get_scno.c"
1283 #include "set_scno.c"
1284 #include "get_syscall_args.c"
1285 #ifndef ptrace_getregset_or_getregs
1286 # include "get_syscall_result.c"
1287 #endif
1288 #include "get_error.c"
1289 #include "set_error.c"
1290 #ifdef HAVE_GETREGS_OLD
1291 # include "getregs_old.c"
1292 #endif
1293 #include "shuffle_scno.c"
1294
1295 const char *
1296 syscall_name(kernel_ulong_t scno)
1297 {
1298         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1299 }