xref: /openbmc/qemu/linux-user/signal.c (revision e4fdf9df)
1 /*
2  *  Emulation of Linux signals
3  *
4  *  Copyright (c) 2003 Fabrice Bellard
5  *
6  *  This program is free software; you can redistribute it and/or modify
7  *  it under the terms of the GNU General Public License as published by
8  *  the Free Software Foundation; either version 2 of the License, or
9  *  (at your option) any later version.
10  *
11  *  This program is distributed in the hope that it will be useful,
12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 #include "qemu/osdep.h"
20 #include "qemu/bitops.h"
21 #include "exec/gdbstub.h"
22 #include "hw/core/tcg-cpu-ops.h"
23 
24 #include <sys/ucontext.h>
25 #include <sys/resource.h>
26 
27 #include "qemu.h"
28 #include "user-internals.h"
29 #include "strace.h"
30 #include "loader.h"
31 #include "trace.h"
32 #include "signal-common.h"
33 #include "host-signal.h"
34 #include "user/safe-syscall.h"
35 
36 static struct target_sigaction sigact_table[TARGET_NSIG];
37 
38 static void host_signal_handler(int host_signum, siginfo_t *info,
39                                 void *puc);
40 
41 /* Fallback addresses into sigtramp page. */
42 abi_ulong default_sigreturn;
43 abi_ulong default_rt_sigreturn;
44 
45 /*
46  * System includes define _NSIG as SIGRTMAX + 1,
47  * but qemu (like the kernel) defines TARGET_NSIG as TARGET_SIGRTMAX
48  * and the first signal is SIGHUP defined as 1
49  * Signal number 0 is reserved for use as kill(pid, 0), to test whether
50  * a process exists without sending it a signal.
51  */
52 #ifdef __SIGRTMAX
53 QEMU_BUILD_BUG_ON(__SIGRTMAX + 1 != _NSIG);
54 #endif
55 static uint8_t host_to_target_signal_table[_NSIG] = {
56 #define MAKE_SIG_ENTRY(sig)     [sig] = TARGET_##sig,
57         MAKE_SIGNAL_LIST
58 #undef MAKE_SIG_ENTRY
59     /* next signals stay the same */
60 };
61 
62 static uint8_t target_to_host_signal_table[TARGET_NSIG + 1];
63 
64 /* valid sig is between 1 and _NSIG - 1 */
65 int host_to_target_signal(int sig)
66 {
67     if (sig < 1 || sig >= _NSIG) {
68         return sig;
69     }
70     return host_to_target_signal_table[sig];
71 }
72 
73 /* valid sig is between 1 and TARGET_NSIG */
74 int target_to_host_signal(int sig)
75 {
76     if (sig < 1 || sig > TARGET_NSIG) {
77         return sig;
78     }
79     return target_to_host_signal_table[sig];
80 }
81 
82 static inline void target_sigaddset(target_sigset_t *set, int signum)
83 {
84     signum--;
85     abi_ulong mask = (abi_ulong)1 << (signum % TARGET_NSIG_BPW);
86     set->sig[signum / TARGET_NSIG_BPW] |= mask;
87 }
88 
89 static inline int target_sigismember(const target_sigset_t *set, int signum)
90 {
91     signum--;
92     abi_ulong mask = (abi_ulong)1 << (signum % TARGET_NSIG_BPW);
93     return ((set->sig[signum / TARGET_NSIG_BPW] & mask) != 0);
94 }
95 
96 void host_to_target_sigset_internal(target_sigset_t *d,
97                                     const sigset_t *s)
98 {
99     int host_sig, target_sig;
100     target_sigemptyset(d);
101     for (host_sig = 1; host_sig < _NSIG; host_sig++) {
102         target_sig = host_to_target_signal(host_sig);
103         if (target_sig < 1 || target_sig > TARGET_NSIG) {
104             continue;
105         }
106         if (sigismember(s, host_sig)) {
107             target_sigaddset(d, target_sig);
108         }
109     }
110 }
111 
112 void host_to_target_sigset(target_sigset_t *d, const sigset_t *s)
113 {
114     target_sigset_t d1;
115     int i;
116 
117     host_to_target_sigset_internal(&d1, s);
118     for(i = 0;i < TARGET_NSIG_WORDS; i++)
119         d->sig[i] = tswapal(d1.sig[i]);
120 }
121 
122 void target_to_host_sigset_internal(sigset_t *d,
123                                     const target_sigset_t *s)
124 {
125     int host_sig, target_sig;
126     sigemptyset(d);
127     for (target_sig = 1; target_sig <= TARGET_NSIG; target_sig++) {
128         host_sig = target_to_host_signal(target_sig);
129         if (host_sig < 1 || host_sig >= _NSIG) {
130             continue;
131         }
132         if (target_sigismember(s, target_sig)) {
133             sigaddset(d, host_sig);
134         }
135     }
136 }
137 
138 void target_to_host_sigset(sigset_t *d, const target_sigset_t *s)
139 {
140     target_sigset_t s1;
141     int i;
142 
143     for(i = 0;i < TARGET_NSIG_WORDS; i++)
144         s1.sig[i] = tswapal(s->sig[i]);
145     target_to_host_sigset_internal(d, &s1);
146 }
147 
148 void host_to_target_old_sigset(abi_ulong *old_sigset,
149                                const sigset_t *sigset)
150 {
151     target_sigset_t d;
152     host_to_target_sigset(&d, sigset);
153     *old_sigset = d.sig[0];
154 }
155 
156 void target_to_host_old_sigset(sigset_t *sigset,
157                                const abi_ulong *old_sigset)
158 {
159     target_sigset_t d;
160     int i;
161 
162     d.sig[0] = *old_sigset;
163     for(i = 1;i < TARGET_NSIG_WORDS; i++)
164         d.sig[i] = 0;
165     target_to_host_sigset(sigset, &d);
166 }
167 
168 int block_signals(void)
169 {
170     TaskState *ts = (TaskState *)thread_cpu->opaque;
171     sigset_t set;
172 
173     /* It's OK to block everything including SIGSEGV, because we won't
174      * run any further guest code before unblocking signals in
175      * process_pending_signals().
176      */
177     sigfillset(&set);
178     sigprocmask(SIG_SETMASK, &set, 0);
179 
180     return qatomic_xchg(&ts->signal_pending, 1);
181 }
182 
183 /* Wrapper for sigprocmask function
184  * Emulates a sigprocmask in a safe way for the guest. Note that set and oldset
185  * are host signal set, not guest ones. Returns -QEMU_ERESTARTSYS if
186  * a signal was already pending and the syscall must be restarted, or
187  * 0 on success.
188  * If set is NULL, this is guaranteed not to fail.
189  */
190 int do_sigprocmask(int how, const sigset_t *set, sigset_t *oldset)
191 {
192     TaskState *ts = (TaskState *)thread_cpu->opaque;
193 
194     if (oldset) {
195         *oldset = ts->signal_mask;
196     }
197 
198     if (set) {
199         int i;
200 
201         if (block_signals()) {
202             return -QEMU_ERESTARTSYS;
203         }
204 
205         switch (how) {
206         case SIG_BLOCK:
207             sigorset(&ts->signal_mask, &ts->signal_mask, set);
208             break;
209         case SIG_UNBLOCK:
210             for (i = 1; i <= NSIG; ++i) {
211                 if (sigismember(set, i)) {
212                     sigdelset(&ts->signal_mask, i);
213                 }
214             }
215             break;
216         case SIG_SETMASK:
217             ts->signal_mask = *set;
218             break;
219         default:
220             g_assert_not_reached();
221         }
222 
223         /* Silently ignore attempts to change blocking status of KILL or STOP */
224         sigdelset(&ts->signal_mask, SIGKILL);
225         sigdelset(&ts->signal_mask, SIGSTOP);
226     }
227     return 0;
228 }
229 
230 /* Just set the guest's signal mask to the specified value; the
231  * caller is assumed to have called block_signals() already.
232  */
233 void set_sigmask(const sigset_t *set)
234 {
235     TaskState *ts = (TaskState *)thread_cpu->opaque;
236 
237     ts->signal_mask = *set;
238 }
239 
240 /* sigaltstack management */
241 
242 int on_sig_stack(unsigned long sp)
243 {
244     TaskState *ts = (TaskState *)thread_cpu->opaque;
245 
246     return (sp - ts->sigaltstack_used.ss_sp
247             < ts->sigaltstack_used.ss_size);
248 }
249 
250 int sas_ss_flags(unsigned long sp)
251 {
252     TaskState *ts = (TaskState *)thread_cpu->opaque;
253 
254     return (ts->sigaltstack_used.ss_size == 0 ? SS_DISABLE
255             : on_sig_stack(sp) ? SS_ONSTACK : 0);
256 }
257 
258 abi_ulong target_sigsp(abi_ulong sp, struct target_sigaction *ka)
259 {
260     /*
261      * This is the X/Open sanctioned signal stack switching.
262      */
263     TaskState *ts = (TaskState *)thread_cpu->opaque;
264 
265     if ((ka->sa_flags & TARGET_SA_ONSTACK) && !sas_ss_flags(sp)) {
266         return ts->sigaltstack_used.ss_sp + ts->sigaltstack_used.ss_size;
267     }
268     return sp;
269 }
270 
271 void target_save_altstack(target_stack_t *uss, CPUArchState *env)
272 {
273     TaskState *ts = (TaskState *)thread_cpu->opaque;
274 
275     __put_user(ts->sigaltstack_used.ss_sp, &uss->ss_sp);
276     __put_user(sas_ss_flags(get_sp_from_cpustate(env)), &uss->ss_flags);
277     __put_user(ts->sigaltstack_used.ss_size, &uss->ss_size);
278 }
279 
280 abi_long target_restore_altstack(target_stack_t *uss, CPUArchState *env)
281 {
282     TaskState *ts = (TaskState *)thread_cpu->opaque;
283     size_t minstacksize = TARGET_MINSIGSTKSZ;
284     target_stack_t ss;
285 
286 #if defined(TARGET_PPC64)
287     /* ELF V2 for PPC64 has a 4K minimum stack size for signal handlers */
288     struct image_info *image = ts->info;
289     if (get_ppc64_abi(image) > 1) {
290         minstacksize = 4096;
291     }
292 #endif
293 
294     __get_user(ss.ss_sp, &uss->ss_sp);
295     __get_user(ss.ss_size, &uss->ss_size);
296     __get_user(ss.ss_flags, &uss->ss_flags);
297 
298     if (on_sig_stack(get_sp_from_cpustate(env))) {
299         return -TARGET_EPERM;
300     }
301 
302     switch (ss.ss_flags) {
303     default:
304         return -TARGET_EINVAL;
305 
306     case TARGET_SS_DISABLE:
307         ss.ss_size = 0;
308         ss.ss_sp = 0;
309         break;
310 
311     case TARGET_SS_ONSTACK:
312     case 0:
313         if (ss.ss_size < minstacksize) {
314             return -TARGET_ENOMEM;
315         }
316         break;
317     }
318 
319     ts->sigaltstack_used.ss_sp = ss.ss_sp;
320     ts->sigaltstack_used.ss_size = ss.ss_size;
321     return 0;
322 }
323 
324 /* siginfo conversion */
325 
326 static inline void host_to_target_siginfo_noswap(target_siginfo_t *tinfo,
327                                                  const siginfo_t *info)
328 {
329     int sig = host_to_target_signal(info->si_signo);
330     int si_code = info->si_code;
331     int si_type;
332     tinfo->si_signo = sig;
333     tinfo->si_errno = 0;
334     tinfo->si_code = info->si_code;
335 
336     /* This memset serves two purposes:
337      * (1) ensure we don't leak random junk to the guest later
338      * (2) placate false positives from gcc about fields
339      *     being used uninitialized if it chooses to inline both this
340      *     function and tswap_siginfo() into host_to_target_siginfo().
341      */
342     memset(tinfo->_sifields._pad, 0, sizeof(tinfo->_sifields._pad));
343 
344     /* This is awkward, because we have to use a combination of
345      * the si_code and si_signo to figure out which of the union's
346      * members are valid. (Within the host kernel it is always possible
347      * to tell, but the kernel carefully avoids giving userspace the
348      * high 16 bits of si_code, so we don't have the information to
349      * do this the easy way...) We therefore make our best guess,
350      * bearing in mind that a guest can spoof most of the si_codes
351      * via rt_sigqueueinfo() if it likes.
352      *
353      * Once we have made our guess, we record it in the top 16 bits of
354      * the si_code, so that tswap_siginfo() later can use it.
355      * tswap_siginfo() will strip these top bits out before writing
356      * si_code to the guest (sign-extending the lower bits).
357      */
358 
359     switch (si_code) {
360     case SI_USER:
361     case SI_TKILL:
362     case SI_KERNEL:
363         /* Sent via kill(), tkill() or tgkill(), or direct from the kernel.
364          * These are the only unspoofable si_code values.
365          */
366         tinfo->_sifields._kill._pid = info->si_pid;
367         tinfo->_sifields._kill._uid = info->si_uid;
368         si_type = QEMU_SI_KILL;
369         break;
370     default:
371         /* Everything else is spoofable. Make best guess based on signal */
372         switch (sig) {
373         case TARGET_SIGCHLD:
374             tinfo->_sifields._sigchld._pid = info->si_pid;
375             tinfo->_sifields._sigchld._uid = info->si_uid;
376             if (si_code == CLD_EXITED)
377                 tinfo->_sifields._sigchld._status = info->si_status;
378             else
379                 tinfo->_sifields._sigchld._status
380                     = host_to_target_signal(info->si_status & 0x7f)
381                         | (info->si_status & ~0x7f);
382             tinfo->_sifields._sigchld._utime = info->si_utime;
383             tinfo->_sifields._sigchld._stime = info->si_stime;
384             si_type = QEMU_SI_CHLD;
385             break;
386         case TARGET_SIGIO:
387             tinfo->_sifields._sigpoll._band = info->si_band;
388             tinfo->_sifields._sigpoll._fd = info->si_fd;
389             si_type = QEMU_SI_POLL;
390             break;
391         default:
392             /* Assume a sigqueue()/mq_notify()/rt_sigqueueinfo() source. */
393             tinfo->_sifields._rt._pid = info->si_pid;
394             tinfo->_sifields._rt._uid = info->si_uid;
395             /* XXX: potential problem if 64 bit */
396             tinfo->_sifields._rt._sigval.sival_ptr
397                 = (abi_ulong)(unsigned long)info->si_value.sival_ptr;
398             si_type = QEMU_SI_RT;
399             break;
400         }
401         break;
402     }
403 
404     tinfo->si_code = deposit32(si_code, 16, 16, si_type);
405 }
406 
407 void tswap_siginfo(target_siginfo_t *tinfo,
408                    const target_siginfo_t *info)
409 {
410     int si_type = extract32(info->si_code, 16, 16);
411     int si_code = sextract32(info->si_code, 0, 16);
412 
413     __put_user(info->si_signo, &tinfo->si_signo);
414     __put_user(info->si_errno, &tinfo->si_errno);
415     __put_user(si_code, &tinfo->si_code);
416 
417     /* We can use our internal marker of which fields in the structure
418      * are valid, rather than duplicating the guesswork of
419      * host_to_target_siginfo_noswap() here.
420      */
421     switch (si_type) {
422     case QEMU_SI_KILL:
423         __put_user(info->_sifields._kill._pid, &tinfo->_sifields._kill._pid);
424         __put_user(info->_sifields._kill._uid, &tinfo->_sifields._kill._uid);
425         break;
426     case QEMU_SI_TIMER:
427         __put_user(info->_sifields._timer._timer1,
428                    &tinfo->_sifields._timer._timer1);
429         __put_user(info->_sifields._timer._timer2,
430                    &tinfo->_sifields._timer._timer2);
431         break;
432     case QEMU_SI_POLL:
433         __put_user(info->_sifields._sigpoll._band,
434                    &tinfo->_sifields._sigpoll._band);
435         __put_user(info->_sifields._sigpoll._fd,
436                    &tinfo->_sifields._sigpoll._fd);
437         break;
438     case QEMU_SI_FAULT:
439         __put_user(info->_sifields._sigfault._addr,
440                    &tinfo->_sifields._sigfault._addr);
441         break;
442     case QEMU_SI_CHLD:
443         __put_user(info->_sifields._sigchld._pid,
444                    &tinfo->_sifields._sigchld._pid);
445         __put_user(info->_sifields._sigchld._uid,
446                    &tinfo->_sifields._sigchld._uid);
447         __put_user(info->_sifields._sigchld._status,
448                    &tinfo->_sifields._sigchld._status);
449         __put_user(info->_sifields._sigchld._utime,
450                    &tinfo->_sifields._sigchld._utime);
451         __put_user(info->_sifields._sigchld._stime,
452                    &tinfo->_sifields._sigchld._stime);
453         break;
454     case QEMU_SI_RT:
455         __put_user(info->_sifields._rt._pid, &tinfo->_sifields._rt._pid);
456         __put_user(info->_sifields._rt._uid, &tinfo->_sifields._rt._uid);
457         __put_user(info->_sifields._rt._sigval.sival_ptr,
458                    &tinfo->_sifields._rt._sigval.sival_ptr);
459         break;
460     default:
461         g_assert_not_reached();
462     }
463 }
464 
465 void host_to_target_siginfo(target_siginfo_t *tinfo, const siginfo_t *info)
466 {
467     target_siginfo_t tgt_tmp;
468     host_to_target_siginfo_noswap(&tgt_tmp, info);
469     tswap_siginfo(tinfo, &tgt_tmp);
470 }
471 
472 /* XXX: we support only POSIX RT signals are used. */
473 /* XXX: find a solution for 64 bit (additional malloced data is needed) */
474 void target_to_host_siginfo(siginfo_t *info, const target_siginfo_t *tinfo)
475 {
476     /* This conversion is used only for the rt_sigqueueinfo syscall,
477      * and so we know that the _rt fields are the valid ones.
478      */
479     abi_ulong sival_ptr;
480 
481     __get_user(info->si_signo, &tinfo->si_signo);
482     __get_user(info->si_errno, &tinfo->si_errno);
483     __get_user(info->si_code, &tinfo->si_code);
484     __get_user(info->si_pid, &tinfo->_sifields._rt._pid);
485     __get_user(info->si_uid, &tinfo->_sifields._rt._uid);
486     __get_user(sival_ptr, &tinfo->_sifields._rt._sigval.sival_ptr);
487     info->si_value.sival_ptr = (void *)(long)sival_ptr;
488 }
489 
490 static int fatal_signal (int sig)
491 {
492     switch (sig) {
493     case TARGET_SIGCHLD:
494     case TARGET_SIGURG:
495     case TARGET_SIGWINCH:
496         /* Ignored by default.  */
497         return 0;
498     case TARGET_SIGCONT:
499     case TARGET_SIGSTOP:
500     case TARGET_SIGTSTP:
501     case TARGET_SIGTTIN:
502     case TARGET_SIGTTOU:
503         /* Job control signals.  */
504         return 0;
505     default:
506         return 1;
507     }
508 }
509 
510 /* returns 1 if given signal should dump core if not handled */
511 static int core_dump_signal(int sig)
512 {
513     switch (sig) {
514     case TARGET_SIGABRT:
515     case TARGET_SIGFPE:
516     case TARGET_SIGILL:
517     case TARGET_SIGQUIT:
518     case TARGET_SIGSEGV:
519     case TARGET_SIGTRAP:
520     case TARGET_SIGBUS:
521         return (1);
522     default:
523         return (0);
524     }
525 }
526 
527 static void signal_table_init(void)
528 {
529     int host_sig, target_sig, count;
530 
531     /*
532      * Signals are supported starting from TARGET_SIGRTMIN and going up
533      * until we run out of host realtime signals.
534      * glibc at least uses only the lower 2 rt signals and probably
535      * nobody's using the upper ones.
536      * it's why SIGRTMIN (34) is generally greater than __SIGRTMIN (32)
537      * To fix this properly we need to do manual signal delivery multiplexed
538      * over a single host signal.
539      * Attempts for configure "missing" signals via sigaction will be
540      * silently ignored.
541      */
542     for (host_sig = SIGRTMIN; host_sig <= SIGRTMAX; host_sig++) {
543         target_sig = host_sig - SIGRTMIN + TARGET_SIGRTMIN;
544         if (target_sig <= TARGET_NSIG) {
545             host_to_target_signal_table[host_sig] = target_sig;
546         }
547     }
548 
549     /* generate signal conversion tables */
550     for (target_sig = 1; target_sig <= TARGET_NSIG; target_sig++) {
551         target_to_host_signal_table[target_sig] = _NSIG; /* poison */
552     }
553     for (host_sig = 1; host_sig < _NSIG; host_sig++) {
554         if (host_to_target_signal_table[host_sig] == 0) {
555             host_to_target_signal_table[host_sig] = host_sig;
556         }
557         target_sig = host_to_target_signal_table[host_sig];
558         if (target_sig <= TARGET_NSIG) {
559             target_to_host_signal_table[target_sig] = host_sig;
560         }
561     }
562 
563     if (trace_event_get_state_backends(TRACE_SIGNAL_TABLE_INIT)) {
564         for (target_sig = 1, count = 0; target_sig <= TARGET_NSIG; target_sig++) {
565             if (target_to_host_signal_table[target_sig] == _NSIG) {
566                 count++;
567             }
568         }
569         trace_signal_table_init(count);
570     }
571 }
572 
573 void signal_init(void)
574 {
575     TaskState *ts = (TaskState *)thread_cpu->opaque;
576     struct sigaction act;
577     struct sigaction oact;
578     int i;
579     int host_sig;
580 
581     /* initialize signal conversion tables */
582     signal_table_init();
583 
584     /* Set the signal mask from the host mask. */
585     sigprocmask(0, 0, &ts->signal_mask);
586 
587     sigfillset(&act.sa_mask);
588     act.sa_flags = SA_SIGINFO;
589     act.sa_sigaction = host_signal_handler;
590     for(i = 1; i <= TARGET_NSIG; i++) {
591 #ifdef CONFIG_GPROF
592         if (i == TARGET_SIGPROF) {
593             continue;
594         }
595 #endif
596         host_sig = target_to_host_signal(i);
597         sigaction(host_sig, NULL, &oact);
598         if (oact.sa_sigaction == (void *)SIG_IGN) {
599             sigact_table[i - 1]._sa_handler = TARGET_SIG_IGN;
600         } else if (oact.sa_sigaction == (void *)SIG_DFL) {
601             sigact_table[i - 1]._sa_handler = TARGET_SIG_DFL;
602         }
603         /* If there's already a handler installed then something has
604            gone horribly wrong, so don't even try to handle that case.  */
605         /* Install some handlers for our own use.  We need at least
606            SIGSEGV and SIGBUS, to detect exceptions.  We can not just
607            trap all signals because it affects syscall interrupt
608            behavior.  But do trap all default-fatal signals.  */
609         if (fatal_signal (i))
610             sigaction(host_sig, &act, NULL);
611     }
612 }
613 
614 /* Force a synchronously taken signal. The kernel force_sig() function
615  * also forces the signal to "not blocked, not ignored", but for QEMU
616  * that work is done in process_pending_signals().
617  */
618 void force_sig(int sig)
619 {
620     CPUState *cpu = thread_cpu;
621     CPUArchState *env = cpu->env_ptr;
622     target_siginfo_t info = {};
623 
624     info.si_signo = sig;
625     info.si_errno = 0;
626     info.si_code = TARGET_SI_KERNEL;
627     info._sifields._kill._pid = 0;
628     info._sifields._kill._uid = 0;
629     queue_signal(env, info.si_signo, QEMU_SI_KILL, &info);
630 }
631 
632 /*
633  * Force a synchronously taken QEMU_SI_FAULT signal. For QEMU the
634  * 'force' part is handled in process_pending_signals().
635  */
636 void force_sig_fault(int sig, int code, abi_ulong addr)
637 {
638     CPUState *cpu = thread_cpu;
639     CPUArchState *env = cpu->env_ptr;
640     target_siginfo_t info = {};
641 
642     info.si_signo = sig;
643     info.si_errno = 0;
644     info.si_code = code;
645     info._sifields._sigfault._addr = addr;
646     queue_signal(env, sig, QEMU_SI_FAULT, &info);
647 }
648 
649 /* Force a SIGSEGV if we couldn't write to memory trying to set
650  * up the signal frame. oldsig is the signal we were trying to handle
651  * at the point of failure.
652  */
653 #if !defined(TARGET_RISCV)
654 void force_sigsegv(int oldsig)
655 {
656     if (oldsig == SIGSEGV) {
657         /* Make sure we don't try to deliver the signal again; this will
658          * end up with handle_pending_signal() calling dump_core_and_abort().
659          */
660         sigact_table[oldsig - 1]._sa_handler = TARGET_SIG_DFL;
661     }
662     force_sig(TARGET_SIGSEGV);
663 }
664 #endif
665 
666 void cpu_loop_exit_sigsegv(CPUState *cpu, target_ulong addr,
667                            MMUAccessType access_type, bool maperr, uintptr_t ra)
668 {
669     const struct TCGCPUOps *tcg_ops = CPU_GET_CLASS(cpu)->tcg_ops;
670 
671     if (tcg_ops->record_sigsegv) {
672         tcg_ops->record_sigsegv(cpu, addr, access_type, maperr, ra);
673     }
674 
675     force_sig_fault(TARGET_SIGSEGV,
676                     maperr ? TARGET_SEGV_MAPERR : TARGET_SEGV_ACCERR,
677                     addr);
678     cpu->exception_index = EXCP_INTERRUPT;
679     cpu_loop_exit_restore(cpu, ra);
680 }
681 
682 void cpu_loop_exit_sigbus(CPUState *cpu, target_ulong addr,
683                           MMUAccessType access_type, uintptr_t ra)
684 {
685     const struct TCGCPUOps *tcg_ops = CPU_GET_CLASS(cpu)->tcg_ops;
686 
687     if (tcg_ops->record_sigbus) {
688         tcg_ops->record_sigbus(cpu, addr, access_type, ra);
689     }
690 
691     force_sig_fault(TARGET_SIGBUS, TARGET_BUS_ADRALN, addr);
692     cpu->exception_index = EXCP_INTERRUPT;
693     cpu_loop_exit_restore(cpu, ra);
694 }
695 
696 /* abort execution with signal */
697 static G_NORETURN
698 void dump_core_and_abort(int target_sig)
699 {
700     CPUState *cpu = thread_cpu;
701     CPUArchState *env = cpu->env_ptr;
702     TaskState *ts = (TaskState *)cpu->opaque;
703     int host_sig, core_dumped = 0;
704     struct sigaction act;
705 
706     host_sig = target_to_host_signal(target_sig);
707     trace_user_dump_core_and_abort(env, target_sig, host_sig);
708     gdb_signalled(env, target_sig);
709 
710     /* dump core if supported by target binary format */
711     if (core_dump_signal(target_sig) && (ts->bprm->core_dump != NULL)) {
712         stop_all_tasks();
713         core_dumped =
714             ((*ts->bprm->core_dump)(target_sig, env) == 0);
715     }
716     if (core_dumped) {
717         /* we already dumped the core of target process, we don't want
718          * a coredump of qemu itself */
719         struct rlimit nodump;
720         getrlimit(RLIMIT_CORE, &nodump);
721         nodump.rlim_cur=0;
722         setrlimit(RLIMIT_CORE, &nodump);
723         (void) fprintf(stderr, "qemu: uncaught target signal %d (%s) - %s\n",
724             target_sig, strsignal(host_sig), "core dumped" );
725     }
726 
727     /* The proper exit code for dying from an uncaught signal is
728      * -<signal>.  The kernel doesn't allow exit() or _exit() to pass
729      * a negative value.  To get the proper exit code we need to
730      * actually die from an uncaught signal.  Here the default signal
731      * handler is installed, we send ourself a signal and we wait for
732      * it to arrive. */
733     sigfillset(&act.sa_mask);
734     act.sa_handler = SIG_DFL;
735     act.sa_flags = 0;
736     sigaction(host_sig, &act, NULL);
737 
738     /* For some reason raise(host_sig) doesn't send the signal when
739      * statically linked on x86-64. */
740     kill(getpid(), host_sig);
741 
742     /* Make sure the signal isn't masked (just reuse the mask inside
743     of act) */
744     sigdelset(&act.sa_mask, host_sig);
745     sigsuspend(&act.sa_mask);
746 
747     /* unreachable */
748     abort();
749 }
750 
751 /* queue a signal so that it will be send to the virtual CPU as soon
752    as possible */
753 void queue_signal(CPUArchState *env, int sig, int si_type,
754                   target_siginfo_t *info)
755 {
756     CPUState *cpu = env_cpu(env);
757     TaskState *ts = cpu->opaque;
758 
759     trace_user_queue_signal(env, sig);
760 
761     info->si_code = deposit32(info->si_code, 16, 16, si_type);
762 
763     ts->sync_signal.info = *info;
764     ts->sync_signal.pending = sig;
765     /* signal that a new signal is pending */
766     qatomic_set(&ts->signal_pending, 1);
767 }
768 
769 
770 /* Adjust the signal context to rewind out of safe-syscall if we're in it */
771 static inline void rewind_if_in_safe_syscall(void *puc)
772 {
773     host_sigcontext *uc = (host_sigcontext *)puc;
774     uintptr_t pcreg = host_signal_pc(uc);
775 
776     if (pcreg > (uintptr_t)safe_syscall_start
777         && pcreg < (uintptr_t)safe_syscall_end) {
778         host_signal_set_pc(uc, (uintptr_t)safe_syscall_start);
779     }
780 }
781 
782 static void host_signal_handler(int host_sig, siginfo_t *info, void *puc)
783 {
784     CPUArchState *env = thread_cpu->env_ptr;
785     CPUState *cpu = env_cpu(env);
786     TaskState *ts = cpu->opaque;
787     target_siginfo_t tinfo;
788     host_sigcontext *uc = puc;
789     struct emulated_sigtable *k;
790     int guest_sig;
791     uintptr_t pc = 0;
792     bool sync_sig = false;
793     void *sigmask = host_signal_mask(uc);
794 
795     /*
796      * Non-spoofed SIGSEGV and SIGBUS are synchronous, and need special
797      * handling wrt signal blocking and unwinding.
798      */
799     if ((host_sig == SIGSEGV || host_sig == SIGBUS) && info->si_code > 0) {
800         MMUAccessType access_type;
801         uintptr_t host_addr;
802         abi_ptr guest_addr;
803         bool is_write;
804 
805         host_addr = (uintptr_t)info->si_addr;
806 
807         /*
808          * Convert forcefully to guest address space: addresses outside
809          * reserved_va are still valid to report via SEGV_MAPERR.
810          */
811         guest_addr = h2g_nocheck(host_addr);
812 
813         pc = host_signal_pc(uc);
814         is_write = host_signal_write(info, uc);
815         access_type = adjust_signal_pc(&pc, is_write);
816 
817         if (host_sig == SIGSEGV) {
818             bool maperr = true;
819 
820             if (info->si_code == SEGV_ACCERR && h2g_valid(host_addr)) {
821                 /* If this was a write to a TB protected page, restart. */
822                 if (is_write &&
823                     handle_sigsegv_accerr_write(cpu, sigmask, pc, guest_addr)) {
824                     return;
825                 }
826 
827                 /*
828                  * With reserved_va, the whole address space is PROT_NONE,
829                  * which means that we may get ACCERR when we want MAPERR.
830                  */
831                 if (page_get_flags(guest_addr) & PAGE_VALID) {
832                     maperr = false;
833                 } else {
834                     info->si_code = SEGV_MAPERR;
835                 }
836             }
837 
838             sigprocmask(SIG_SETMASK, sigmask, NULL);
839             cpu_loop_exit_sigsegv(cpu, guest_addr, access_type, maperr, pc);
840         } else {
841             sigprocmask(SIG_SETMASK, sigmask, NULL);
842             if (info->si_code == BUS_ADRALN) {
843                 cpu_loop_exit_sigbus(cpu, guest_addr, access_type, pc);
844             }
845         }
846 
847         sync_sig = true;
848     }
849 
850     /* get target signal number */
851     guest_sig = host_to_target_signal(host_sig);
852     if (guest_sig < 1 || guest_sig > TARGET_NSIG) {
853         return;
854     }
855     trace_user_host_signal(env, host_sig, guest_sig);
856 
857     host_to_target_siginfo_noswap(&tinfo, info);
858     k = &ts->sigtab[guest_sig - 1];
859     k->info = tinfo;
860     k->pending = guest_sig;
861     ts->signal_pending = 1;
862 
863     /*
864      * For synchronous signals, unwind the cpu state to the faulting
865      * insn and then exit back to the main loop so that the signal
866      * is delivered immediately.
867      */
868     if (sync_sig) {
869         cpu->exception_index = EXCP_INTERRUPT;
870         cpu_loop_exit_restore(cpu, pc);
871     }
872 
873     rewind_if_in_safe_syscall(puc);
874 
875     /*
876      * Block host signals until target signal handler entered. We
877      * can't block SIGSEGV or SIGBUS while we're executing guest
878      * code in case the guest code provokes one in the window between
879      * now and it getting out to the main loop. Signals will be
880      * unblocked again in process_pending_signals().
881      *
882      * WARNING: we cannot use sigfillset() here because the sigmask
883      * field is a kernel sigset_t, which is much smaller than the
884      * libc sigset_t which sigfillset() operates on. Using sigfillset()
885      * would write 0xff bytes off the end of the structure and trash
886      * data on the struct.
887      */
888     memset(sigmask, 0xff, SIGSET_T_SIZE);
889     sigdelset(sigmask, SIGSEGV);
890     sigdelset(sigmask, SIGBUS);
891 
892     /* interrupt the virtual CPU as soon as possible */
893     cpu_exit(thread_cpu);
894 }
895 
896 /* do_sigaltstack() returns target values and errnos. */
897 /* compare linux/kernel/signal.c:do_sigaltstack() */
898 abi_long do_sigaltstack(abi_ulong uss_addr, abi_ulong uoss_addr,
899                         CPUArchState *env)
900 {
901     target_stack_t oss, *uoss = NULL;
902     abi_long ret = -TARGET_EFAULT;
903 
904     if (uoss_addr) {
905         /* Verify writability now, but do not alter user memory yet. */
906         if (!lock_user_struct(VERIFY_WRITE, uoss, uoss_addr, 0)) {
907             goto out;
908         }
909         target_save_altstack(&oss, env);
910     }
911 
912     if (uss_addr) {
913         target_stack_t *uss;
914 
915         if (!lock_user_struct(VERIFY_READ, uss, uss_addr, 1)) {
916             goto out;
917         }
918         ret = target_restore_altstack(uss, env);
919         if (ret) {
920             goto out;
921         }
922     }
923 
924     if (uoss_addr) {
925         memcpy(uoss, &oss, sizeof(oss));
926         unlock_user_struct(uoss, uoss_addr, 1);
927         uoss = NULL;
928     }
929     ret = 0;
930 
931  out:
932     if (uoss) {
933         unlock_user_struct(uoss, uoss_addr, 0);
934     }
935     return ret;
936 }
937 
938 /* do_sigaction() return target values and host errnos */
939 int do_sigaction(int sig, const struct target_sigaction *act,
940                  struct target_sigaction *oact, abi_ulong ka_restorer)
941 {
942     struct target_sigaction *k;
943     struct sigaction act1;
944     int host_sig;
945     int ret = 0;
946 
947     trace_signal_do_sigaction_guest(sig, TARGET_NSIG);
948 
949     if (sig < 1 || sig > TARGET_NSIG) {
950         return -TARGET_EINVAL;
951     }
952 
953     if (act && (sig == TARGET_SIGKILL || sig == TARGET_SIGSTOP)) {
954         return -TARGET_EINVAL;
955     }
956 
957     if (block_signals()) {
958         return -QEMU_ERESTARTSYS;
959     }
960 
961     k = &sigact_table[sig - 1];
962     if (oact) {
963         __put_user(k->_sa_handler, &oact->_sa_handler);
964         __put_user(k->sa_flags, &oact->sa_flags);
965 #ifdef TARGET_ARCH_HAS_SA_RESTORER
966         __put_user(k->sa_restorer, &oact->sa_restorer);
967 #endif
968         /* Not swapped.  */
969         oact->sa_mask = k->sa_mask;
970     }
971     if (act) {
972         __get_user(k->_sa_handler, &act->_sa_handler);
973         __get_user(k->sa_flags, &act->sa_flags);
974 #ifdef TARGET_ARCH_HAS_SA_RESTORER
975         __get_user(k->sa_restorer, &act->sa_restorer);
976 #endif
977 #ifdef TARGET_ARCH_HAS_KA_RESTORER
978         k->ka_restorer = ka_restorer;
979 #endif
980         /* To be swapped in target_to_host_sigset.  */
981         k->sa_mask = act->sa_mask;
982 
983         /* we update the host linux signal state */
984         host_sig = target_to_host_signal(sig);
985         trace_signal_do_sigaction_host(host_sig, TARGET_NSIG);
986         if (host_sig > SIGRTMAX) {
987             /* we don't have enough host signals to map all target signals */
988             qemu_log_mask(LOG_UNIMP, "Unsupported target signal #%d, ignored\n",
989                           sig);
990             /*
991              * we don't return an error here because some programs try to
992              * register an handler for all possible rt signals even if they
993              * don't need it.
994              * An error here can abort them whereas there can be no problem
995              * to not have the signal available later.
996              * This is the case for golang,
997              *   See https://github.com/golang/go/issues/33746
998              * So we silently ignore the error.
999              */
1000             return 0;
1001         }
1002         if (host_sig != SIGSEGV && host_sig != SIGBUS) {
1003             sigfillset(&act1.sa_mask);
1004             act1.sa_flags = SA_SIGINFO;
1005             if (k->sa_flags & TARGET_SA_RESTART)
1006                 act1.sa_flags |= SA_RESTART;
1007             /* NOTE: it is important to update the host kernel signal
1008                ignore state to avoid getting unexpected interrupted
1009                syscalls */
1010             if (k->_sa_handler == TARGET_SIG_IGN) {
1011                 act1.sa_sigaction = (void *)SIG_IGN;
1012             } else if (k->_sa_handler == TARGET_SIG_DFL) {
1013                 if (fatal_signal (sig))
1014                     act1.sa_sigaction = host_signal_handler;
1015                 else
1016                     act1.sa_sigaction = (void *)SIG_DFL;
1017             } else {
1018                 act1.sa_sigaction = host_signal_handler;
1019             }
1020             ret = sigaction(host_sig, &act1, NULL);
1021         }
1022     }
1023     return ret;
1024 }
1025 
1026 static void handle_pending_signal(CPUArchState *cpu_env, int sig,
1027                                   struct emulated_sigtable *k)
1028 {
1029     CPUState *cpu = env_cpu(cpu_env);
1030     abi_ulong handler;
1031     sigset_t set;
1032     target_sigset_t target_old_set;
1033     struct target_sigaction *sa;
1034     TaskState *ts = cpu->opaque;
1035 
1036     trace_user_handle_signal(cpu_env, sig);
1037     /* dequeue signal */
1038     k->pending = 0;
1039 
1040     sig = gdb_handlesig(cpu, sig);
1041     if (!sig) {
1042         sa = NULL;
1043         handler = TARGET_SIG_IGN;
1044     } else {
1045         sa = &sigact_table[sig - 1];
1046         handler = sa->_sa_handler;
1047     }
1048 
1049     if (unlikely(qemu_loglevel_mask(LOG_STRACE))) {
1050         print_taken_signal(sig, &k->info);
1051     }
1052 
1053     if (handler == TARGET_SIG_DFL) {
1054         /* default handler : ignore some signal. The other are job control or fatal */
1055         if (sig == TARGET_SIGTSTP || sig == TARGET_SIGTTIN || sig == TARGET_SIGTTOU) {
1056             kill(getpid(),SIGSTOP);
1057         } else if (sig != TARGET_SIGCHLD &&
1058                    sig != TARGET_SIGURG &&
1059                    sig != TARGET_SIGWINCH &&
1060                    sig != TARGET_SIGCONT) {
1061             dump_core_and_abort(sig);
1062         }
1063     } else if (handler == TARGET_SIG_IGN) {
1064         /* ignore sig */
1065     } else if (handler == TARGET_SIG_ERR) {
1066         dump_core_and_abort(sig);
1067     } else {
1068         /* compute the blocked signals during the handler execution */
1069         sigset_t *blocked_set;
1070 
1071         target_to_host_sigset(&set, &sa->sa_mask);
1072         /* SA_NODEFER indicates that the current signal should not be
1073            blocked during the handler */
1074         if (!(sa->sa_flags & TARGET_SA_NODEFER))
1075             sigaddset(&set, target_to_host_signal(sig));
1076 
1077         /* save the previous blocked signal state to restore it at the
1078            end of the signal execution (see do_sigreturn) */
1079         host_to_target_sigset_internal(&target_old_set, &ts->signal_mask);
1080 
1081         /* block signals in the handler */
1082         blocked_set = ts->in_sigsuspend ?
1083             &ts->sigsuspend_mask : &ts->signal_mask;
1084         sigorset(&ts->signal_mask, blocked_set, &set);
1085         ts->in_sigsuspend = 0;
1086 
1087         /* if the CPU is in VM86 mode, we restore the 32 bit values */
1088 #if defined(TARGET_I386) && !defined(TARGET_X86_64)
1089         {
1090             CPUX86State *env = cpu_env;
1091             if (env->eflags & VM_MASK)
1092                 save_v86_state(env);
1093         }
1094 #endif
1095         /* prepare the stack frame of the virtual CPU */
1096 #if defined(TARGET_ARCH_HAS_SETUP_FRAME)
1097         if (sa->sa_flags & TARGET_SA_SIGINFO) {
1098             setup_rt_frame(sig, sa, &k->info, &target_old_set, cpu_env);
1099         } else {
1100             setup_frame(sig, sa, &target_old_set, cpu_env);
1101         }
1102 #else
1103         /* These targets do not have traditional signals.  */
1104         setup_rt_frame(sig, sa, &k->info, &target_old_set, cpu_env);
1105 #endif
1106         if (sa->sa_flags & TARGET_SA_RESETHAND) {
1107             sa->_sa_handler = TARGET_SIG_DFL;
1108         }
1109     }
1110 }
1111 
1112 void process_pending_signals(CPUArchState *cpu_env)
1113 {
1114     CPUState *cpu = env_cpu(cpu_env);
1115     int sig;
1116     TaskState *ts = cpu->opaque;
1117     sigset_t set;
1118     sigset_t *blocked_set;
1119 
1120     while (qatomic_read(&ts->signal_pending)) {
1121         sigfillset(&set);
1122         sigprocmask(SIG_SETMASK, &set, 0);
1123 
1124     restart_scan:
1125         sig = ts->sync_signal.pending;
1126         if (sig) {
1127             /* Synchronous signals are forced,
1128              * see force_sig_info() and callers in Linux
1129              * Note that not all of our queue_signal() calls in QEMU correspond
1130              * to force_sig_info() calls in Linux (some are send_sig_info()).
1131              * However it seems like a kernel bug to me to allow the process
1132              * to block a synchronous signal since it could then just end up
1133              * looping round and round indefinitely.
1134              */
1135             if (sigismember(&ts->signal_mask, target_to_host_signal_table[sig])
1136                 || sigact_table[sig - 1]._sa_handler == TARGET_SIG_IGN) {
1137                 sigdelset(&ts->signal_mask, target_to_host_signal_table[sig]);
1138                 sigact_table[sig - 1]._sa_handler = TARGET_SIG_DFL;
1139             }
1140 
1141             handle_pending_signal(cpu_env, sig, &ts->sync_signal);
1142         }
1143 
1144         for (sig = 1; sig <= TARGET_NSIG; sig++) {
1145             blocked_set = ts->in_sigsuspend ?
1146                 &ts->sigsuspend_mask : &ts->signal_mask;
1147 
1148             if (ts->sigtab[sig - 1].pending &&
1149                 (!sigismember(blocked_set,
1150                               target_to_host_signal_table[sig]))) {
1151                 handle_pending_signal(cpu_env, sig, &ts->sigtab[sig - 1]);
1152                 /* Restart scan from the beginning, as handle_pending_signal
1153                  * might have resulted in a new synchronous signal (eg SIGSEGV).
1154                  */
1155                 goto restart_scan;
1156             }
1157         }
1158 
1159         /* if no signal is pending, unblock signals and recheck (the act
1160          * of unblocking might cause us to take another host signal which
1161          * will set signal_pending again).
1162          */
1163         qatomic_set(&ts->signal_pending, 0);
1164         ts->in_sigsuspend = 0;
1165         set = ts->signal_mask;
1166         sigdelset(&set, SIGSEGV);
1167         sigdelset(&set, SIGBUS);
1168         sigprocmask(SIG_SETMASK, &set, 0);
1169     }
1170     ts->in_sigsuspend = 0;
1171 }
1172 
1173 int process_sigsuspend_mask(sigset_t **pset, target_ulong sigset,
1174                             target_ulong sigsize)
1175 {
1176     TaskState *ts = (TaskState *)thread_cpu->opaque;
1177     sigset_t *host_set = &ts->sigsuspend_mask;
1178     target_sigset_t *target_sigset;
1179 
1180     if (sigsize != sizeof(*target_sigset)) {
1181         /* Like the kernel, we enforce correct size sigsets */
1182         return -TARGET_EINVAL;
1183     }
1184 
1185     target_sigset = lock_user(VERIFY_READ, sigset, sigsize, 1);
1186     if (!target_sigset) {
1187         return -TARGET_EFAULT;
1188     }
1189     target_to_host_sigset(host_set, target_sigset);
1190     unlock_user(target_sigset, sigset, 0);
1191 
1192     *pset = host_set;
1193     return 0;
1194 }
1195