Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
5 *
6 * 1997-11-28 Modified for POSIX.1b signals by Richard Henderson
7 * 2000-06-20 Pentium III FXSR, SSE support by Gareth Hughes
8 * 2000-2002 x86-64 support by Andi Kleen
9 */
10
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13#include <linux/sched.h>
14#include <linux/sched/task_stack.h>
15#include <linux/mm.h>
16#include <linux/smp.h>
17#include <linux/kernel.h>
18#include <linux/errno.h>
19#include <linux/wait.h>
20#include <linux/tracehook.h>
21#include <linux/unistd.h>
22#include <linux/stddef.h>
23#include <linux/personality.h>
24#include <linux/uaccess.h>
25#include <linux/user-return-notifier.h>
26#include <linux/uprobes.h>
27#include <linux/context_tracking.h>
28#include <linux/syscalls.h>
29
30#include <asm/processor.h>
31#include <asm/ucontext.h>
32#include <asm/fpu/internal.h>
33#include <asm/fpu/signal.h>
34#include <asm/vdso.h>
35#include <asm/mce.h>
36#include <asm/sighandling.h>
37#include <asm/vm86.h>
38
39#ifdef CONFIG_X86_64
40#include <asm/proto.h>
41#include <asm/ia32_unistd.h>
42#endif /* CONFIG_X86_64 */
43
44#include <asm/syscall.h>
45#include <asm/syscalls.h>
46
47#include <asm/sigframe.h>
48#include <asm/signal.h>
49
50#define COPY(x) do { \
51 get_user_ex(regs->x, &sc->x); \
52} while (0)
53
54#define GET_SEG(seg) ({ \
55 unsigned short tmp; \
56 get_user_ex(tmp, &sc->seg); \
57 tmp; \
58})
59
60#define COPY_SEG(seg) do { \
61 regs->seg = GET_SEG(seg); \
62} while (0)
63
64#define COPY_SEG_CPL3(seg) do { \
65 regs->seg = GET_SEG(seg) | 3; \
66} while (0)
67
68#ifdef CONFIG_X86_64
69/*
70 * If regs->ss will cause an IRET fault, change it. Otherwise leave it
71 * alone. Using this generally makes no sense unless
72 * user_64bit_mode(regs) would return true.
73 */
74static void force_valid_ss(struct pt_regs *regs)
75{
76 u32 ar;
77 asm volatile ("lar %[old_ss], %[ar]\n\t"
78 "jz 1f\n\t" /* If invalid: */
79 "xorl %[ar], %[ar]\n\t" /* set ar = 0 */
80 "1:"
81 : [ar] "=r" (ar)
82 : [old_ss] "rm" ((u16)regs->ss));
83
84 /*
85 * For a valid 64-bit user context, we need DPL 3, type
86 * read-write data or read-write exp-down data, and S and P
87 * set. We can't use VERW because VERW doesn't check the
88 * P bit.
89 */
90 ar &= AR_DPL_MASK | AR_S | AR_P | AR_TYPE_MASK;
91 if (ar != (AR_DPL3 | AR_S | AR_P | AR_TYPE_RWDATA) &&
92 ar != (AR_DPL3 | AR_S | AR_P | AR_TYPE_RWDATA_EXPDOWN))
93 regs->ss = __USER_DS;
94}
95#endif
96
97static int restore_sigcontext(struct pt_regs *regs,
98 struct sigcontext __user *sc,
99 unsigned long uc_flags)
100{
101 unsigned long buf_val;
102 void __user *buf;
103 unsigned int tmpflags;
104 unsigned int err = 0;
105
106 /* Always make any pending restarted system calls return -EINTR */
107 current->restart_block.fn = do_no_restart_syscall;
108
109 get_user_try {
110
111#ifdef CONFIG_X86_32
112 set_user_gs(regs, GET_SEG(gs));
113 COPY_SEG(fs);
114 COPY_SEG(es);
115 COPY_SEG(ds);
116#endif /* CONFIG_X86_32 */
117
118 COPY(di); COPY(si); COPY(bp); COPY(sp); COPY(bx);
119 COPY(dx); COPY(cx); COPY(ip); COPY(ax);
120
121#ifdef CONFIG_X86_64
122 COPY(r8);
123 COPY(r9);
124 COPY(r10);
125 COPY(r11);
126 COPY(r12);
127 COPY(r13);
128 COPY(r14);
129 COPY(r15);
130#endif /* CONFIG_X86_64 */
131
132 COPY_SEG_CPL3(cs);
133 COPY_SEG_CPL3(ss);
134
135 get_user_ex(tmpflags, &sc->flags);
136 regs->flags = (regs->flags & ~FIX_EFLAGS) | (tmpflags & FIX_EFLAGS);
137 regs->orig_ax = -1; /* disable syscall checks */
138
139 get_user_ex(buf_val, &sc->fpstate);
140 buf = (void __user *)buf_val;
141 } get_user_catch(err);
142
143#ifdef CONFIG_X86_64
144 /*
145 * Fix up SS if needed for the benefit of old DOSEMU and
146 * CRIU.
147 */
148 if (unlikely(!(uc_flags & UC_STRICT_RESTORE_SS) && user_64bit_mode(regs)))
149 force_valid_ss(regs);
150#endif
151
152 err |= fpu__restore_sig(buf, IS_ENABLED(CONFIG_X86_32));
153
154 force_iret();
155
156 return err;
157}
158
159int setup_sigcontext(struct sigcontext __user *sc, void __user *fpstate,
160 struct pt_regs *regs, unsigned long mask)
161{
162 int err = 0;
163
164 put_user_try {
165
166#ifdef CONFIG_X86_32
167 put_user_ex(get_user_gs(regs), (unsigned int __user *)&sc->gs);
168 put_user_ex(regs->fs, (unsigned int __user *)&sc->fs);
169 put_user_ex(regs->es, (unsigned int __user *)&sc->es);
170 put_user_ex(regs->ds, (unsigned int __user *)&sc->ds);
171#endif /* CONFIG_X86_32 */
172
173 put_user_ex(regs->di, &sc->di);
174 put_user_ex(regs->si, &sc->si);
175 put_user_ex(regs->bp, &sc->bp);
176 put_user_ex(regs->sp, &sc->sp);
177 put_user_ex(regs->bx, &sc->bx);
178 put_user_ex(regs->dx, &sc->dx);
179 put_user_ex(regs->cx, &sc->cx);
180 put_user_ex(regs->ax, &sc->ax);
181#ifdef CONFIG_X86_64
182 put_user_ex(regs->r8, &sc->r8);
183 put_user_ex(regs->r9, &sc->r9);
184 put_user_ex(regs->r10, &sc->r10);
185 put_user_ex(regs->r11, &sc->r11);
186 put_user_ex(regs->r12, &sc->r12);
187 put_user_ex(regs->r13, &sc->r13);
188 put_user_ex(regs->r14, &sc->r14);
189 put_user_ex(regs->r15, &sc->r15);
190#endif /* CONFIG_X86_64 */
191
192 put_user_ex(current->thread.trap_nr, &sc->trapno);
193 put_user_ex(current->thread.error_code, &sc->err);
194 put_user_ex(regs->ip, &sc->ip);
195#ifdef CONFIG_X86_32
196 put_user_ex(regs->cs, (unsigned int __user *)&sc->cs);
197 put_user_ex(regs->flags, &sc->flags);
198 put_user_ex(regs->sp, &sc->sp_at_signal);
199 put_user_ex(regs->ss, (unsigned int __user *)&sc->ss);
200#else /* !CONFIG_X86_32 */
201 put_user_ex(regs->flags, &sc->flags);
202 put_user_ex(regs->cs, &sc->cs);
203 put_user_ex(0, &sc->gs);
204 put_user_ex(0, &sc->fs);
205 put_user_ex(regs->ss, &sc->ss);
206#endif /* CONFIG_X86_32 */
207
208 put_user_ex(fpstate, (unsigned long __user *)&sc->fpstate);
209
210 /* non-iBCS2 extensions.. */
211 put_user_ex(mask, &sc->oldmask);
212 put_user_ex(current->thread.cr2, &sc->cr2);
213 } put_user_catch(err);
214
215 return err;
216}
217
218/*
219 * Set up a signal frame.
220 */
221
222/*
223 * Determine which stack to use..
224 */
225static unsigned long align_sigframe(unsigned long sp)
226{
227#ifdef CONFIG_X86_32
228 /*
229 * Align the stack pointer according to the i386 ABI,
230 * i.e. so that on function entry ((sp + 4) & 15) == 0.
231 */
232 sp = ((sp + 4) & -16ul) - 4;
233#else /* !CONFIG_X86_32 */
234 sp = round_down(sp, 16) - 8;
235#endif
236 return sp;
237}
238
239static void __user *
240get_sigframe(struct k_sigaction *ka, struct pt_regs *regs, size_t frame_size,
241 void __user **fpstate)
242{
243 /* Default to using normal stack */
244 unsigned long math_size = 0;
245 unsigned long sp = regs->sp;
246 unsigned long buf_fx = 0;
247 int onsigstack = on_sig_stack(sp);
248 int ret;
249
250 /* redzone */
251 if (IS_ENABLED(CONFIG_X86_64))
252 sp -= 128;
253
254 /* This is the X/Open sanctioned signal stack switching. */
255 if (ka->sa.sa_flags & SA_ONSTACK) {
256 if (sas_ss_flags(sp) == 0)
257 sp = current->sas_ss_sp + current->sas_ss_size;
258 } else if (IS_ENABLED(CONFIG_X86_32) &&
259 !onsigstack &&
260 regs->ss != __USER_DS &&
261 !(ka->sa.sa_flags & SA_RESTORER) &&
262 ka->sa.sa_restorer) {
263 /* This is the legacy signal stack switching. */
264 sp = (unsigned long) ka->sa.sa_restorer;
265 }
266
267 sp = fpu__alloc_mathframe(sp, IS_ENABLED(CONFIG_X86_32),
268 &buf_fx, &math_size);
269 *fpstate = (void __user *)sp;
270
271 sp = align_sigframe(sp - frame_size);
272
273 /*
274 * If we are on the alternate signal stack and would overflow it, don't.
275 * Return an always-bogus address instead so we will die with SIGSEGV.
276 */
277 if (onsigstack && !likely(on_sig_stack(sp)))
278 return (void __user *)-1L;
279
280 /* save i387 and extended state */
281 ret = copy_fpstate_to_sigframe(*fpstate, (void __user *)buf_fx, math_size);
282 if (ret < 0)
283 return (void __user *)-1L;
284
285 return (void __user *)sp;
286}
287
288#ifdef CONFIG_X86_32
289static const struct {
290 u16 poplmovl;
291 u32 val;
292 u16 int80;
293} __attribute__((packed)) retcode = {
294 0xb858, /* popl %eax; movl $..., %eax */
295 __NR_sigreturn,
296 0x80cd, /* int $0x80 */
297};
298
299static const struct {
300 u8 movl;
301 u32 val;
302 u16 int80;
303 u8 pad;
304} __attribute__((packed)) rt_retcode = {
305 0xb8, /* movl $..., %eax */
306 __NR_rt_sigreturn,
307 0x80cd, /* int $0x80 */
308 0
309};
310
311static int
312__setup_frame(int sig, struct ksignal *ksig, sigset_t *set,
313 struct pt_regs *regs)
314{
315 struct sigframe __user *frame;
316 void __user *restorer;
317 int err = 0;
318 void __user *fpstate = NULL;
319
320 frame = get_sigframe(&ksig->ka, regs, sizeof(*frame), &fpstate);
321
322 if (!access_ok(frame, sizeof(*frame)))
323 return -EFAULT;
324
325 if (__put_user(sig, &frame->sig))
326 return -EFAULT;
327
328 if (setup_sigcontext(&frame->sc, fpstate, regs, set->sig[0]))
329 return -EFAULT;
330
331 if (_NSIG_WORDS > 1) {
332 if (__copy_to_user(&frame->extramask, &set->sig[1],
333 sizeof(frame->extramask)))
334 return -EFAULT;
335 }
336
337 if (current->mm->context.vdso)
338 restorer = current->mm->context.vdso +
339 vdso_image_32.sym___kernel_sigreturn;
340 else
341 restorer = &frame->retcode;
342 if (ksig->ka.sa.sa_flags & SA_RESTORER)
343 restorer = ksig->ka.sa.sa_restorer;
344
345 /* Set up to return from userspace. */
346 err |= __put_user(restorer, &frame->pretcode);
347
348 /*
349 * This is popl %eax ; movl $__NR_sigreturn, %eax ; int $0x80
350 *
351 * WE DO NOT USE IT ANY MORE! It's only left here for historical
352 * reasons and because gdb uses it as a signature to notice
353 * signal handler stack frames.
354 */
355 err |= __put_user(*((u64 *)&retcode), (u64 *)frame->retcode);
356
357 if (err)
358 return -EFAULT;
359
360 /* Set up registers for signal handler */
361 regs->sp = (unsigned long)frame;
362 regs->ip = (unsigned long)ksig->ka.sa.sa_handler;
363 regs->ax = (unsigned long)sig;
364 regs->dx = 0;
365 regs->cx = 0;
366
367 regs->ds = __USER_DS;
368 regs->es = __USER_DS;
369 regs->ss = __USER_DS;
370 regs->cs = __USER_CS;
371
372 return 0;
373}
374
375static int __setup_rt_frame(int sig, struct ksignal *ksig,
376 sigset_t *set, struct pt_regs *regs)
377{
378 struct rt_sigframe __user *frame;
379 void __user *restorer;
380 int err = 0;
381 void __user *fpstate = NULL;
382
383 frame = get_sigframe(&ksig->ka, regs, sizeof(*frame), &fpstate);
384
385 if (!access_ok(frame, sizeof(*frame)))
386 return -EFAULT;
387
388 put_user_try {
389 put_user_ex(sig, &frame->sig);
390 put_user_ex(&frame->info, &frame->pinfo);
391 put_user_ex(&frame->uc, &frame->puc);
392
393 /* Create the ucontext. */
394 if (static_cpu_has(X86_FEATURE_XSAVE))
395 put_user_ex(UC_FP_XSTATE, &frame->uc.uc_flags);
396 else
397 put_user_ex(0, &frame->uc.uc_flags);
398 put_user_ex(0, &frame->uc.uc_link);
399 save_altstack_ex(&frame->uc.uc_stack, regs->sp);
400
401 /* Set up to return from userspace. */
402 restorer = current->mm->context.vdso +
403 vdso_image_32.sym___kernel_rt_sigreturn;
404 if (ksig->ka.sa.sa_flags & SA_RESTORER)
405 restorer = ksig->ka.sa.sa_restorer;
406 put_user_ex(restorer, &frame->pretcode);
407
408 /*
409 * This is movl $__NR_rt_sigreturn, %ax ; int $0x80
410 *
411 * WE DO NOT USE IT ANY MORE! It's only left here for historical
412 * reasons and because gdb uses it as a signature to notice
413 * signal handler stack frames.
414 */
415 put_user_ex(*((u64 *)&rt_retcode), (u64 *)frame->retcode);
416 } put_user_catch(err);
417
418 err |= copy_siginfo_to_user(&frame->info, &ksig->info);
419 err |= setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
420 regs, set->sig[0]);
421 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
422
423 if (err)
424 return -EFAULT;
425
426 /* Set up registers for signal handler */
427 regs->sp = (unsigned long)frame;
428 regs->ip = (unsigned long)ksig->ka.sa.sa_handler;
429 regs->ax = (unsigned long)sig;
430 regs->dx = (unsigned long)&frame->info;
431 regs->cx = (unsigned long)&frame->uc;
432
433 regs->ds = __USER_DS;
434 regs->es = __USER_DS;
435 regs->ss = __USER_DS;
436 regs->cs = __USER_CS;
437
438 return 0;
439}
440#else /* !CONFIG_X86_32 */
441static unsigned long frame_uc_flags(struct pt_regs *regs)
442{
443 unsigned long flags;
444
445 if (boot_cpu_has(X86_FEATURE_XSAVE))
446 flags = UC_FP_XSTATE | UC_SIGCONTEXT_SS;
447 else
448 flags = UC_SIGCONTEXT_SS;
449
450 if (likely(user_64bit_mode(regs)))
451 flags |= UC_STRICT_RESTORE_SS;
452
453 return flags;
454}
455
456static int __setup_rt_frame(int sig, struct ksignal *ksig,
457 sigset_t *set, struct pt_regs *regs)
458{
459 struct rt_sigframe __user *frame;
460 void __user *fp = NULL;
461 unsigned long uc_flags;
462 int err = 0;
463
464 frame = get_sigframe(&ksig->ka, regs, sizeof(struct rt_sigframe), &fp);
465
466 if (!access_ok(frame, sizeof(*frame)))
467 return -EFAULT;
468
469 if (ksig->ka.sa.sa_flags & SA_SIGINFO) {
470 if (copy_siginfo_to_user(&frame->info, &ksig->info))
471 return -EFAULT;
472 }
473
474 uc_flags = frame_uc_flags(regs);
475
476 put_user_try {
477 /* Create the ucontext. */
478 put_user_ex(uc_flags, &frame->uc.uc_flags);
479 put_user_ex(0, &frame->uc.uc_link);
480 save_altstack_ex(&frame->uc.uc_stack, regs->sp);
481
482 /* Set up to return from userspace. If provided, use a stub
483 already in userspace. */
484 /* x86-64 should always use SA_RESTORER. */
485 if (ksig->ka.sa.sa_flags & SA_RESTORER) {
486 put_user_ex(ksig->ka.sa.sa_restorer, &frame->pretcode);
487 } else {
488 /* could use a vstub here */
489 err |= -EFAULT;
490 }
491 } put_user_catch(err);
492
493 err |= setup_sigcontext(&frame->uc.uc_mcontext, fp, regs, set->sig[0]);
494 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
495
496 if (err)
497 return -EFAULT;
498
499 /* Set up registers for signal handler */
500 regs->di = sig;
501 /* In case the signal handler was declared without prototypes */
502 regs->ax = 0;
503
504 /* This also works for non SA_SIGINFO handlers because they expect the
505 next argument after the signal number on the stack. */
506 regs->si = (unsigned long)&frame->info;
507 regs->dx = (unsigned long)&frame->uc;
508 regs->ip = (unsigned long) ksig->ka.sa.sa_handler;
509
510 regs->sp = (unsigned long)frame;
511
512 /*
513 * Set up the CS and SS registers to run signal handlers in
514 * 64-bit mode, even if the handler happens to be interrupting
515 * 32-bit or 16-bit code.
516 *
517 * SS is subtle. In 64-bit mode, we don't need any particular
518 * SS descriptor, but we do need SS to be valid. It's possible
519 * that the old SS is entirely bogus -- this can happen if the
520 * signal we're trying to deliver is #GP or #SS caused by a bad
521 * SS value. We also have a compatbility issue here: DOSEMU
522 * relies on the contents of the SS register indicating the
523 * SS value at the time of the signal, even though that code in
524 * DOSEMU predates sigreturn's ability to restore SS. (DOSEMU
525 * avoids relying on sigreturn to restore SS; instead it uses
526 * a trampoline.) So we do our best: if the old SS was valid,
527 * we keep it. Otherwise we replace it.
528 */
529 regs->cs = __USER_CS;
530
531 if (unlikely(regs->ss != __USER_DS))
532 force_valid_ss(regs);
533
534 return 0;
535}
536#endif /* CONFIG_X86_32 */
537
538static int x32_setup_rt_frame(struct ksignal *ksig,
539 compat_sigset_t *set,
540 struct pt_regs *regs)
541{
542#ifdef CONFIG_X86_X32_ABI
543 struct rt_sigframe_x32 __user *frame;
544 unsigned long uc_flags;
545 void __user *restorer;
546 int err = 0;
547 void __user *fpstate = NULL;
548
549 frame = get_sigframe(&ksig->ka, regs, sizeof(*frame), &fpstate);
550
551 if (!access_ok(frame, sizeof(*frame)))
552 return -EFAULT;
553
554 if (ksig->ka.sa.sa_flags & SA_SIGINFO) {
555 if (__copy_siginfo_to_user32(&frame->info, &ksig->info, true))
556 return -EFAULT;
557 }
558
559 uc_flags = frame_uc_flags(regs);
560
561 put_user_try {
562 /* Create the ucontext. */
563 put_user_ex(uc_flags, &frame->uc.uc_flags);
564 put_user_ex(0, &frame->uc.uc_link);
565 compat_save_altstack_ex(&frame->uc.uc_stack, regs->sp);
566 put_user_ex(0, &frame->uc.uc__pad0);
567
568 if (ksig->ka.sa.sa_flags & SA_RESTORER) {
569 restorer = ksig->ka.sa.sa_restorer;
570 } else {
571 /* could use a vstub here */
572 restorer = NULL;
573 err |= -EFAULT;
574 }
575 put_user_ex(restorer, (unsigned long __user *)&frame->pretcode);
576 } put_user_catch(err);
577
578 err |= setup_sigcontext(&frame->uc.uc_mcontext, fpstate,
579 regs, set->sig[0]);
580 err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
581
582 if (err)
583 return -EFAULT;
584
585 /* Set up registers for signal handler */
586 regs->sp = (unsigned long) frame;
587 regs->ip = (unsigned long) ksig->ka.sa.sa_handler;
588
589 /* We use the x32 calling convention here... */
590 regs->di = ksig->sig;
591 regs->si = (unsigned long) &frame->info;
592 regs->dx = (unsigned long) &frame->uc;
593
594 loadsegment(ds, __USER_DS);
595 loadsegment(es, __USER_DS);
596
597 regs->cs = __USER_CS;
598 regs->ss = __USER_DS;
599#endif /* CONFIG_X86_X32_ABI */
600
601 return 0;
602}
603
604/*
605 * Do a signal return; undo the signal stack.
606 */
607#ifdef CONFIG_X86_32
608SYSCALL_DEFINE0(sigreturn)
609{
610 struct pt_regs *regs = current_pt_regs();
611 struct sigframe __user *frame;
612 sigset_t set;
613
614 frame = (struct sigframe __user *)(regs->sp - 8);
615
616 if (!access_ok(frame, sizeof(*frame)))
617 goto badframe;
618 if (__get_user(set.sig[0], &frame->sc.oldmask) || (_NSIG_WORDS > 1
619 && __copy_from_user(&set.sig[1], &frame->extramask,
620 sizeof(frame->extramask))))
621 goto badframe;
622
623 set_current_blocked(&set);
624
625 /*
626 * x86_32 has no uc_flags bits relevant to restore_sigcontext.
627 * Save a few cycles by skipping the __get_user.
628 */
629 if (restore_sigcontext(regs, &frame->sc, 0))
630 goto badframe;
631 return regs->ax;
632
633badframe:
634 signal_fault(regs, frame, "sigreturn");
635
636 return 0;
637}
638#endif /* CONFIG_X86_32 */
639
640SYSCALL_DEFINE0(rt_sigreturn)
641{
642 struct pt_regs *regs = current_pt_regs();
643 struct rt_sigframe __user *frame;
644 sigset_t set;
645 unsigned long uc_flags;
646
647 frame = (struct rt_sigframe __user *)(regs->sp - sizeof(long));
648 if (!access_ok(frame, sizeof(*frame)))
649 goto badframe;
650 if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
651 goto badframe;
652 if (__get_user(uc_flags, &frame->uc.uc_flags))
653 goto badframe;
654
655 set_current_blocked(&set);
656
657 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, uc_flags))
658 goto badframe;
659
660 if (restore_altstack(&frame->uc.uc_stack))
661 goto badframe;
662
663 return regs->ax;
664
665badframe:
666 signal_fault(regs, frame, "rt_sigreturn");
667 return 0;
668}
669
670static inline int is_ia32_compat_frame(struct ksignal *ksig)
671{
672 return IS_ENABLED(CONFIG_IA32_EMULATION) &&
673 ksig->ka.sa.sa_flags & SA_IA32_ABI;
674}
675
676static inline int is_ia32_frame(struct ksignal *ksig)
677{
678 return IS_ENABLED(CONFIG_X86_32) || is_ia32_compat_frame(ksig);
679}
680
681static inline int is_x32_frame(struct ksignal *ksig)
682{
683 return IS_ENABLED(CONFIG_X86_X32_ABI) &&
684 ksig->ka.sa.sa_flags & SA_X32_ABI;
685}
686
687static int
688setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
689{
690 int usig = ksig->sig;
691 sigset_t *set = sigmask_to_save();
692 compat_sigset_t *cset = (compat_sigset_t *) set;
693
694 /* Perform fixup for the pre-signal frame. */
695 rseq_signal_deliver(ksig, regs);
696
697 /* Set up the stack frame */
698 if (is_ia32_frame(ksig)) {
699 if (ksig->ka.sa.sa_flags & SA_SIGINFO)
700 return ia32_setup_rt_frame(usig, ksig, cset, regs);
701 else
702 return ia32_setup_frame(usig, ksig, cset, regs);
703 } else if (is_x32_frame(ksig)) {
704 return x32_setup_rt_frame(ksig, cset, regs);
705 } else {
706 return __setup_rt_frame(ksig->sig, ksig, set, regs);
707 }
708}
709
710static void
711handle_signal(struct ksignal *ksig, struct pt_regs *regs)
712{
713 bool stepping, failed;
714 struct fpu *fpu = ¤t->thread.fpu;
715
716 if (v8086_mode(regs))
717 save_v86_state((struct kernel_vm86_regs *) regs, VM86_SIGNAL);
718
719 /* Are we from a system call? */
720 if (syscall_get_nr(current, regs) >= 0) {
721 /* If so, check system call restarting.. */
722 switch (syscall_get_error(current, regs)) {
723 case -ERESTART_RESTARTBLOCK:
724 case -ERESTARTNOHAND:
725 regs->ax = -EINTR;
726 break;
727
728 case -ERESTARTSYS:
729 if (!(ksig->ka.sa.sa_flags & SA_RESTART)) {
730 regs->ax = -EINTR;
731 break;
732 }
733 /* fallthrough */
734 case -ERESTARTNOINTR:
735 regs->ax = regs->orig_ax;
736 regs->ip -= 2;
737 break;
738 }
739 }
740
741 /*
742 * If TF is set due to a debugger (TIF_FORCED_TF), clear TF now
743 * so that register information in the sigcontext is correct and
744 * then notify the tracer before entering the signal handler.
745 */
746 stepping = test_thread_flag(TIF_SINGLESTEP);
747 if (stepping)
748 user_disable_single_step(current);
749
750 failed = (setup_rt_frame(ksig, regs) < 0);
751 if (!failed) {
752 /*
753 * Clear the direction flag as per the ABI for function entry.
754 *
755 * Clear RF when entering the signal handler, because
756 * it might disable possible debug exception from the
757 * signal handler.
758 *
759 * Clear TF for the case when it wasn't set by debugger to
760 * avoid the recursive send_sigtrap() in SIGTRAP handler.
761 */
762 regs->flags &= ~(X86_EFLAGS_DF|X86_EFLAGS_RF|X86_EFLAGS_TF);
763 /*
764 * Ensure the signal handler starts with the new fpu state.
765 */
766 fpu__clear(fpu);
767 }
768 signal_setup_done(failed, ksig, stepping);
769}
770
771static inline unsigned long get_nr_restart_syscall(const struct pt_regs *regs)
772{
773 /*
774 * This function is fundamentally broken as currently
775 * implemented.
776 *
777 * The idea is that we want to trigger a call to the
778 * restart_block() syscall and that we want in_ia32_syscall(),
779 * in_x32_syscall(), etc. to match whatever they were in the
780 * syscall being restarted. We assume that the syscall
781 * instruction at (regs->ip - 2) matches whatever syscall
782 * instruction we used to enter in the first place.
783 *
784 * The problem is that we can get here when ptrace pokes
785 * syscall-like values into regs even if we're not in a syscall
786 * at all.
787 *
788 * For now, we maintain historical behavior and guess based on
789 * stored state. We could do better by saving the actual
790 * syscall arch in restart_block or (with caveats on x32) by
791 * checking if regs->ip points to 'int $0x80'. The current
792 * behavior is incorrect if a tracer has a different bitness
793 * than the tracee.
794 */
795#ifdef CONFIG_IA32_EMULATION
796 if (current_thread_info()->status & (TS_COMPAT|TS_I386_REGS_POKED))
797 return __NR_ia32_restart_syscall;
798#endif
799#ifdef CONFIG_X86_X32_ABI
800 return __NR_restart_syscall | (regs->orig_ax & __X32_SYSCALL_BIT);
801#else
802 return __NR_restart_syscall;
803#endif
804}
805
806/*
807 * Note that 'init' is a special process: it doesn't get signals it doesn't
808 * want to handle. Thus you cannot kill init even with a SIGKILL even by
809 * mistake.
810 */
811void do_signal(struct pt_regs *regs)
812{
813 struct ksignal ksig;
814
815 if (get_signal(&ksig)) {
816 /* Whee! Actually deliver the signal. */
817 handle_signal(&ksig, regs);
818 return;
819 }
820
821 /* Did we come from a system call? */
822 if (syscall_get_nr(current, regs) >= 0) {
823 /* Restart the system call - no handlers present */
824 switch (syscall_get_error(current, regs)) {
825 case -ERESTARTNOHAND:
826 case -ERESTARTSYS:
827 case -ERESTARTNOINTR:
828 regs->ax = regs->orig_ax;
829 regs->ip -= 2;
830 break;
831
832 case -ERESTART_RESTARTBLOCK:
833 regs->ax = get_nr_restart_syscall(regs);
834 regs->ip -= 2;
835 break;
836 }
837 }
838
839 /*
840 * If there's no signal to deliver, we just put the saved sigmask
841 * back.
842 */
843 restore_saved_sigmask();
844}
845
846void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
847{
848 struct task_struct *me = current;
849
850 if (show_unhandled_signals && printk_ratelimit()) {
851 printk("%s"
852 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
853 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
854 me->comm, me->pid, where, frame,
855 regs->ip, regs->sp, regs->orig_ax);
856 print_vma_addr(KERN_CONT " in ", regs->ip);
857 pr_cont("\n");
858 }
859
860 force_sig(SIGSEGV);
861}
862
863#ifdef CONFIG_X86_X32_ABI
864asmlinkage long sys32_x32_rt_sigreturn(void)
865{
866 struct pt_regs *regs = current_pt_regs();
867 struct rt_sigframe_x32 __user *frame;
868 sigset_t set;
869 unsigned long uc_flags;
870
871 frame = (struct rt_sigframe_x32 __user *)(regs->sp - 8);
872
873 if (!access_ok(frame, sizeof(*frame)))
874 goto badframe;
875 if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
876 goto badframe;
877 if (__get_user(uc_flags, &frame->uc.uc_flags))
878 goto badframe;
879
880 set_current_blocked(&set);
881
882 if (restore_sigcontext(regs, &frame->uc.uc_mcontext, uc_flags))
883 goto badframe;
884
885 if (compat_restore_altstack(&frame->uc.uc_stack))
886 goto badframe;
887
888 return regs->ax;
889
890badframe:
891 signal_fault(regs, frame, "x32 rt_sigreturn");
892 return 0;
893}
894#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 2000, 2001, 2002 Andi Kleen SuSE Labs
5 *
6 * 1997-11-28 Modified for POSIX.1b signals by Richard Henderson
7 * 2000-06-20 Pentium III FXSR, SSE support by Gareth Hughes
8 * 2000-2002 x86-64 support by Andi Kleen
9 */
10
11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12
13#include <linux/sched.h>
14#include <linux/sched/task_stack.h>
15#include <linux/mm.h>
16#include <linux/smp.h>
17#include <linux/kernel.h>
18#include <linux/kstrtox.h>
19#include <linux/errno.h>
20#include <linux/wait.h>
21#include <linux/unistd.h>
22#include <linux/stddef.h>
23#include <linux/personality.h>
24#include <linux/uaccess.h>
25#include <linux/user-return-notifier.h>
26#include <linux/uprobes.h>
27#include <linux/context_tracking.h>
28#include <linux/entry-common.h>
29#include <linux/syscalls.h>
30#include <linux/rseq.h>
31
32#include <asm/processor.h>
33#include <asm/ucontext.h>
34#include <asm/fpu/signal.h>
35#include <asm/fpu/xstate.h>
36#include <asm/vdso.h>
37#include <asm/mce.h>
38#include <asm/sighandling.h>
39#include <asm/vm86.h>
40
41#include <asm/syscall.h>
42#include <asm/sigframe.h>
43#include <asm/signal.h>
44#include <asm/shstk.h>
45
46static inline int is_ia32_compat_frame(struct ksignal *ksig)
47{
48 return IS_ENABLED(CONFIG_IA32_EMULATION) &&
49 ksig->ka.sa.sa_flags & SA_IA32_ABI;
50}
51
52static inline int is_ia32_frame(struct ksignal *ksig)
53{
54 return IS_ENABLED(CONFIG_X86_32) || is_ia32_compat_frame(ksig);
55}
56
57static inline int is_x32_frame(struct ksignal *ksig)
58{
59 return IS_ENABLED(CONFIG_X86_X32_ABI) &&
60 ksig->ka.sa.sa_flags & SA_X32_ABI;
61}
62
63/*
64 * Enable all pkeys temporarily, so as to ensure that both the current
65 * execution stack as well as the alternate signal stack are writeable.
66 * The application can use any of the available pkeys to protect the
67 * alternate signal stack, and we don't know which one it is, so enable
68 * all. The PKRU register will be reset to init_pkru later in the flow,
69 * in fpu__clear_user_states(), and it is the application's responsibility
70 * to enable the appropriate pkey as the first step in the signal handler
71 * so that the handler does not segfault.
72 */
73static inline u32 sig_prepare_pkru(void)
74{
75 u32 orig_pkru = read_pkru();
76
77 write_pkru(0);
78 return orig_pkru;
79}
80
81/*
82 * Set up a signal frame.
83 */
84
85/* x86 ABI requires 16-byte alignment */
86#define FRAME_ALIGNMENT 16UL
87
88#define MAX_FRAME_PADDING (FRAME_ALIGNMENT - 1)
89
90/*
91 * Determine which stack to use..
92 */
93void __user *
94get_sigframe(struct ksignal *ksig, struct pt_regs *regs, size_t frame_size,
95 void __user **fpstate)
96{
97 struct k_sigaction *ka = &ksig->ka;
98 int ia32_frame = is_ia32_frame(ksig);
99 /* Default to using normal stack */
100 bool nested_altstack = on_sig_stack(regs->sp);
101 bool entering_altstack = false;
102 unsigned long math_size = 0;
103 unsigned long sp = regs->sp;
104 unsigned long buf_fx = 0;
105 u32 pkru;
106
107 /* redzone */
108 if (!ia32_frame)
109 sp -= 128;
110
111 /* This is the X/Open sanctioned signal stack switching. */
112 if (ka->sa.sa_flags & SA_ONSTACK) {
113 /*
114 * This checks nested_altstack via sas_ss_flags(). Sensible
115 * programs use SS_AUTODISARM, which disables that check, and
116 * programs that don't use SS_AUTODISARM get compatible.
117 */
118 if (sas_ss_flags(sp) == 0) {
119 sp = current->sas_ss_sp + current->sas_ss_size;
120 entering_altstack = true;
121 }
122 } else if (ia32_frame &&
123 !nested_altstack &&
124 regs->ss != __USER_DS &&
125 !(ka->sa.sa_flags & SA_RESTORER) &&
126 ka->sa.sa_restorer) {
127 /* This is the legacy signal stack switching. */
128 sp = (unsigned long) ka->sa.sa_restorer;
129 entering_altstack = true;
130 }
131
132 sp = fpu__alloc_mathframe(sp, ia32_frame, &buf_fx, &math_size);
133 *fpstate = (void __user *)sp;
134
135 sp -= frame_size;
136
137 if (ia32_frame)
138 /*
139 * Align the stack pointer according to the i386 ABI,
140 * i.e. so that on function entry ((sp + 4) & 15) == 0.
141 */
142 sp = ((sp + 4) & -FRAME_ALIGNMENT) - 4;
143 else
144 sp = round_down(sp, FRAME_ALIGNMENT) - 8;
145
146 /*
147 * If we are on the alternate signal stack and would overflow it, don't.
148 * Return an always-bogus address instead so we will die with SIGSEGV.
149 */
150 if (unlikely((nested_altstack || entering_altstack) &&
151 !__on_sig_stack(sp))) {
152
153 if (show_unhandled_signals && printk_ratelimit())
154 pr_info("%s[%d] overflowed sigaltstack\n",
155 current->comm, task_pid_nr(current));
156
157 return (void __user *)-1L;
158 }
159
160 /* Update PKRU to enable access to the alternate signal stack. */
161 pkru = sig_prepare_pkru();
162 /* save i387 and extended state */
163 if (!copy_fpstate_to_sigframe(*fpstate, (void __user *)buf_fx, math_size, pkru)) {
164 /*
165 * Restore PKRU to the original, user-defined value; disable
166 * extra pkeys enabled for the alternate signal stack, if any.
167 */
168 write_pkru(pkru);
169 return (void __user *)-1L;
170 }
171
172 return (void __user *)sp;
173}
174
175/*
176 * There are four different struct types for signal frame: sigframe_ia32,
177 * rt_sigframe_ia32, rt_sigframe_x32, and rt_sigframe. Use the worst case
178 * -- the largest size. It means the size for 64-bit apps is a bit more
179 * than needed, but this keeps the code simple.
180 */
181#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
182# define MAX_FRAME_SIGINFO_UCTXT_SIZE sizeof(struct sigframe_ia32)
183#else
184# define MAX_FRAME_SIGINFO_UCTXT_SIZE sizeof(struct rt_sigframe)
185#endif
186
187/*
188 * The FP state frame contains an XSAVE buffer which must be 64-byte aligned.
189 * If a signal frame starts at an unaligned address, extra space is required.
190 * This is the max alignment padding, conservatively.
191 */
192#define MAX_XSAVE_PADDING 63UL
193
194/*
195 * The frame data is composed of the following areas and laid out as:
196 *
197 * -------------------------
198 * | alignment padding |
199 * -------------------------
200 * | (f)xsave frame |
201 * -------------------------
202 * | fsave header |
203 * -------------------------
204 * | alignment padding |
205 * -------------------------
206 * | siginfo + ucontext |
207 * -------------------------
208 */
209
210/* max_frame_size tells userspace the worst case signal stack size. */
211static unsigned long __ro_after_init max_frame_size;
212static unsigned int __ro_after_init fpu_default_state_size;
213
214static int __init init_sigframe_size(void)
215{
216 fpu_default_state_size = fpu__get_fpstate_size();
217
218 max_frame_size = MAX_FRAME_SIGINFO_UCTXT_SIZE + MAX_FRAME_PADDING;
219
220 max_frame_size += fpu_default_state_size + MAX_XSAVE_PADDING;
221
222 /* Userspace expects an aligned size. */
223 max_frame_size = round_up(max_frame_size, FRAME_ALIGNMENT);
224
225 pr_info("max sigframe size: %lu\n", max_frame_size);
226 return 0;
227}
228early_initcall(init_sigframe_size);
229
230unsigned long get_sigframe_size(void)
231{
232 return max_frame_size;
233}
234
235static int
236setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
237{
238 /* Perform fixup for the pre-signal frame. */
239 rseq_signal_deliver(ksig, regs);
240
241 /* Set up the stack frame */
242 if (is_ia32_frame(ksig)) {
243 if (ksig->ka.sa.sa_flags & SA_SIGINFO)
244 return ia32_setup_rt_frame(ksig, regs);
245 else
246 return ia32_setup_frame(ksig, regs);
247 } else if (is_x32_frame(ksig)) {
248 return x32_setup_rt_frame(ksig, regs);
249 } else {
250 return x64_setup_rt_frame(ksig, regs);
251 }
252}
253
254static void
255handle_signal(struct ksignal *ksig, struct pt_regs *regs)
256{
257 bool stepping, failed;
258 struct fpu *fpu = ¤t->thread.fpu;
259
260 if (v8086_mode(regs))
261 save_v86_state((struct kernel_vm86_regs *) regs, VM86_SIGNAL);
262
263 /* Are we from a system call? */
264 if (syscall_get_nr(current, regs) != -1) {
265 /* If so, check system call restarting.. */
266 switch (syscall_get_error(current, regs)) {
267 case -ERESTART_RESTARTBLOCK:
268 case -ERESTARTNOHAND:
269 regs->ax = -EINTR;
270 break;
271
272 case -ERESTARTSYS:
273 if (!(ksig->ka.sa.sa_flags & SA_RESTART)) {
274 regs->ax = -EINTR;
275 break;
276 }
277 fallthrough;
278 case -ERESTARTNOINTR:
279 regs->ax = regs->orig_ax;
280 regs->ip -= 2;
281 break;
282 }
283 }
284
285 /*
286 * If TF is set due to a debugger (TIF_FORCED_TF), clear TF now
287 * so that register information in the sigcontext is correct and
288 * then notify the tracer before entering the signal handler.
289 */
290 stepping = test_thread_flag(TIF_SINGLESTEP);
291 if (stepping)
292 user_disable_single_step(current);
293
294 failed = (setup_rt_frame(ksig, regs) < 0);
295 if (!failed) {
296 /*
297 * Clear the direction flag as per the ABI for function entry.
298 *
299 * Clear RF when entering the signal handler, because
300 * it might disable possible debug exception from the
301 * signal handler.
302 *
303 * Clear TF for the case when it wasn't set by debugger to
304 * avoid the recursive send_sigtrap() in SIGTRAP handler.
305 */
306 regs->flags &= ~(X86_EFLAGS_DF|X86_EFLAGS_RF|X86_EFLAGS_TF);
307 /*
308 * Ensure the signal handler starts with the new fpu state.
309 */
310 fpu__clear_user_states(fpu);
311 }
312 signal_setup_done(failed, ksig, stepping);
313}
314
315static inline unsigned long get_nr_restart_syscall(const struct pt_regs *regs)
316{
317#ifdef CONFIG_IA32_EMULATION
318 if (current->restart_block.arch_data & TS_COMPAT)
319 return __NR_ia32_restart_syscall;
320#endif
321#ifdef CONFIG_X86_X32_ABI
322 return __NR_restart_syscall | (regs->orig_ax & __X32_SYSCALL_BIT);
323#else
324 return __NR_restart_syscall;
325#endif
326}
327
328/*
329 * Note that 'init' is a special process: it doesn't get signals it doesn't
330 * want to handle. Thus you cannot kill init even with a SIGKILL even by
331 * mistake.
332 */
333void arch_do_signal_or_restart(struct pt_regs *regs)
334{
335 struct ksignal ksig;
336
337 if (get_signal(&ksig)) {
338 /* Whee! Actually deliver the signal. */
339 handle_signal(&ksig, regs);
340 return;
341 }
342
343 /* Did we come from a system call? */
344 if (syscall_get_nr(current, regs) != -1) {
345 /* Restart the system call - no handlers present */
346 switch (syscall_get_error(current, regs)) {
347 case -ERESTARTNOHAND:
348 case -ERESTARTSYS:
349 case -ERESTARTNOINTR:
350 regs->ax = regs->orig_ax;
351 regs->ip -= 2;
352 break;
353
354 case -ERESTART_RESTARTBLOCK:
355 regs->ax = get_nr_restart_syscall(regs);
356 regs->ip -= 2;
357 break;
358 }
359 }
360
361 /*
362 * If there's no signal to deliver, we just put the saved sigmask
363 * back.
364 */
365 restore_saved_sigmask();
366}
367
368void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
369{
370 struct task_struct *me = current;
371
372 if (show_unhandled_signals && printk_ratelimit()) {
373 printk("%s"
374 "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
375 task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
376 me->comm, me->pid, where, frame,
377 regs->ip, regs->sp, regs->orig_ax);
378 print_vma_addr(KERN_CONT " in ", regs->ip);
379 pr_cont("\n");
380 }
381
382 force_sig(SIGSEGV);
383}
384
385#ifdef CONFIG_DYNAMIC_SIGFRAME
386#ifdef CONFIG_STRICT_SIGALTSTACK_SIZE
387static bool strict_sigaltstack_size __ro_after_init = true;
388#else
389static bool strict_sigaltstack_size __ro_after_init = false;
390#endif
391
392static int __init strict_sas_size(char *arg)
393{
394 return kstrtobool(arg, &strict_sigaltstack_size) == 0;
395}
396__setup("strict_sas_size", strict_sas_size);
397
398/*
399 * MINSIGSTKSZ is 2048 and can't be changed despite the fact that AVX512
400 * exceeds that size already. As such programs might never use the
401 * sigaltstack they just continued to work. While always checking against
402 * the real size would be correct, this might be considered a regression.
403 *
404 * Therefore avoid the sanity check, unless enforced by kernel
405 * configuration or command line option.
406 *
407 * When dynamic FPU features are supported, the check is also enforced when
408 * the task has permissions to use dynamic features. Tasks which have no
409 * permission are checked against the size of the non-dynamic feature set
410 * if strict checking is enabled. This avoids forcing all tasks on the
411 * system to allocate large sigaltstacks even if they are never going
412 * to use a dynamic feature. As this is serialized via sighand::siglock
413 * any permission request for a dynamic feature either happened already
414 * or will see the newly install sigaltstack size in the permission checks.
415 */
416bool sigaltstack_size_valid(size_t ss_size)
417{
418 unsigned long fsize = max_frame_size - fpu_default_state_size;
419 u64 mask;
420
421 lockdep_assert_held(¤t->sighand->siglock);
422
423 if (!fpu_state_size_dynamic() && !strict_sigaltstack_size)
424 return true;
425
426 fsize += current->group_leader->thread.fpu.perm.__user_state_size;
427 if (likely(ss_size > fsize))
428 return true;
429
430 if (strict_sigaltstack_size)
431 return ss_size > fsize;
432
433 mask = current->group_leader->thread.fpu.perm.__state_perm;
434 if (mask & XFEATURE_MASK_USER_DYNAMIC)
435 return ss_size > fsize;
436
437 return true;
438}
439#endif /* CONFIG_DYNAMIC_SIGFRAME */