xref: /openbmc/linux/arch/x86/kernel/signal.c (revision e9b7b8b3)
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 
31 #include <asm/processor.h>
32 #include <asm/ucontext.h>
33 #include <asm/fpu/signal.h>
34 #include <asm/fpu/xstate.h>
35 #include <asm/vdso.h>
36 #include <asm/mce.h>
37 #include <asm/sighandling.h>
38 #include <asm/vm86.h>
39 
40 #include <asm/syscall.h>
41 #include <asm/sigframe.h>
42 #include <asm/signal.h>
43 #include <asm/shstk.h>
44 
45 static inline int is_ia32_compat_frame(struct ksignal *ksig)
46 {
47 	return IS_ENABLED(CONFIG_IA32_EMULATION) &&
48 		ksig->ka.sa.sa_flags & SA_IA32_ABI;
49 }
50 
51 static inline int is_ia32_frame(struct ksignal *ksig)
52 {
53 	return IS_ENABLED(CONFIG_X86_32) || is_ia32_compat_frame(ksig);
54 }
55 
56 static inline int is_x32_frame(struct ksignal *ksig)
57 {
58 	return IS_ENABLED(CONFIG_X86_X32_ABI) &&
59 		ksig->ka.sa.sa_flags & SA_X32_ABI;
60 }
61 
62 /*
63  * Set up a signal frame.
64  */
65 
66 /* x86 ABI requires 16-byte alignment */
67 #define FRAME_ALIGNMENT	16UL
68 
69 #define MAX_FRAME_PADDING	(FRAME_ALIGNMENT - 1)
70 
71 /*
72  * Determine which stack to use..
73  */
74 void __user *
75 get_sigframe(struct ksignal *ksig, struct pt_regs *regs, size_t frame_size,
76 	     void __user **fpstate)
77 {
78 	struct k_sigaction *ka = &ksig->ka;
79 	int ia32_frame = is_ia32_frame(ksig);
80 	/* Default to using normal stack */
81 	bool nested_altstack = on_sig_stack(regs->sp);
82 	bool entering_altstack = false;
83 	unsigned long math_size = 0;
84 	unsigned long sp = regs->sp;
85 	unsigned long buf_fx = 0;
86 
87 	/* redzone */
88 	if (!ia32_frame)
89 		sp -= 128;
90 
91 	/* This is the X/Open sanctioned signal stack switching.  */
92 	if (ka->sa.sa_flags & SA_ONSTACK) {
93 		/*
94 		 * This checks nested_altstack via sas_ss_flags(). Sensible
95 		 * programs use SS_AUTODISARM, which disables that check, and
96 		 * programs that don't use SS_AUTODISARM get compatible.
97 		 */
98 		if (sas_ss_flags(sp) == 0) {
99 			sp = current->sas_ss_sp + current->sas_ss_size;
100 			entering_altstack = true;
101 		}
102 	} else if (ia32_frame &&
103 		   !nested_altstack &&
104 		   regs->ss != __USER_DS &&
105 		   !(ka->sa.sa_flags & SA_RESTORER) &&
106 		   ka->sa.sa_restorer) {
107 		/* This is the legacy signal stack switching. */
108 		sp = (unsigned long) ka->sa.sa_restorer;
109 		entering_altstack = true;
110 	}
111 
112 	sp = fpu__alloc_mathframe(sp, ia32_frame, &buf_fx, &math_size);
113 	*fpstate = (void __user *)sp;
114 
115 	sp -= frame_size;
116 
117 	if (ia32_frame)
118 		/*
119 		 * Align the stack pointer according to the i386 ABI,
120 		 * i.e. so that on function entry ((sp + 4) & 15) == 0.
121 		 */
122 		sp = ((sp + 4) & -FRAME_ALIGNMENT) - 4;
123 	else
124 		sp = round_down(sp, FRAME_ALIGNMENT) - 8;
125 
126 	/*
127 	 * If we are on the alternate signal stack and would overflow it, don't.
128 	 * Return an always-bogus address instead so we will die with SIGSEGV.
129 	 */
130 	if (unlikely((nested_altstack || entering_altstack) &&
131 		     !__on_sig_stack(sp))) {
132 
133 		if (show_unhandled_signals && printk_ratelimit())
134 			pr_info("%s[%d] overflowed sigaltstack\n",
135 				current->comm, task_pid_nr(current));
136 
137 		return (void __user *)-1L;
138 	}
139 
140 	/* save i387 and extended state */
141 	if (!copy_fpstate_to_sigframe(*fpstate, (void __user *)buf_fx, math_size))
142 		return (void __user *)-1L;
143 
144 	return (void __user *)sp;
145 }
146 
147 /*
148  * There are four different struct types for signal frame: sigframe_ia32,
149  * rt_sigframe_ia32, rt_sigframe_x32, and rt_sigframe. Use the worst case
150  * -- the largest size. It means the size for 64-bit apps is a bit more
151  * than needed, but this keeps the code simple.
152  */
153 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
154 # define MAX_FRAME_SIGINFO_UCTXT_SIZE	sizeof(struct sigframe_ia32)
155 #else
156 # define MAX_FRAME_SIGINFO_UCTXT_SIZE	sizeof(struct rt_sigframe)
157 #endif
158 
159 /*
160  * The FP state frame contains an XSAVE buffer which must be 64-byte aligned.
161  * If a signal frame starts at an unaligned address, extra space is required.
162  * This is the max alignment padding, conservatively.
163  */
164 #define MAX_XSAVE_PADDING	63UL
165 
166 /*
167  * The frame data is composed of the following areas and laid out as:
168  *
169  * -------------------------
170  * | alignment padding     |
171  * -------------------------
172  * | (f)xsave frame        |
173  * -------------------------
174  * | fsave header          |
175  * -------------------------
176  * | alignment padding     |
177  * -------------------------
178  * | siginfo + ucontext    |
179  * -------------------------
180  */
181 
182 /* max_frame_size tells userspace the worst case signal stack size. */
183 static unsigned long __ro_after_init max_frame_size;
184 static unsigned int __ro_after_init fpu_default_state_size;
185 
186 static int __init init_sigframe_size(void)
187 {
188 	fpu_default_state_size = fpu__get_fpstate_size();
189 
190 	max_frame_size = MAX_FRAME_SIGINFO_UCTXT_SIZE + MAX_FRAME_PADDING;
191 
192 	max_frame_size += fpu_default_state_size + MAX_XSAVE_PADDING;
193 
194 	/* Userspace expects an aligned size. */
195 	max_frame_size = round_up(max_frame_size, FRAME_ALIGNMENT);
196 
197 	pr_info("max sigframe size: %lu\n", max_frame_size);
198 	return 0;
199 }
200 early_initcall(init_sigframe_size);
201 
202 unsigned long get_sigframe_size(void)
203 {
204 	return max_frame_size;
205 }
206 
207 static int
208 setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
209 {
210 	/* Perform fixup for the pre-signal frame. */
211 	rseq_signal_deliver(ksig, regs);
212 
213 	/* Set up the stack frame */
214 	if (is_ia32_frame(ksig)) {
215 		if (ksig->ka.sa.sa_flags & SA_SIGINFO)
216 			return ia32_setup_rt_frame(ksig, regs);
217 		else
218 			return ia32_setup_frame(ksig, regs);
219 	} else if (is_x32_frame(ksig)) {
220 		return x32_setup_rt_frame(ksig, regs);
221 	} else {
222 		return x64_setup_rt_frame(ksig, regs);
223 	}
224 }
225 
226 static void
227 handle_signal(struct ksignal *ksig, struct pt_regs *regs)
228 {
229 	bool stepping, failed;
230 	struct fpu *fpu = &current->thread.fpu;
231 
232 	if (v8086_mode(regs))
233 		save_v86_state((struct kernel_vm86_regs *) regs, VM86_SIGNAL);
234 
235 	/* Are we from a system call? */
236 	if (syscall_get_nr(current, regs) != -1) {
237 		/* If so, check system call restarting.. */
238 		switch (syscall_get_error(current, regs)) {
239 		case -ERESTART_RESTARTBLOCK:
240 		case -ERESTARTNOHAND:
241 			regs->ax = -EINTR;
242 			break;
243 
244 		case -ERESTARTSYS:
245 			if (!(ksig->ka.sa.sa_flags & SA_RESTART)) {
246 				regs->ax = -EINTR;
247 				break;
248 			}
249 			fallthrough;
250 		case -ERESTARTNOINTR:
251 			regs->ax = regs->orig_ax;
252 			regs->ip -= 2;
253 			break;
254 		}
255 	}
256 
257 	/*
258 	 * If TF is set due to a debugger (TIF_FORCED_TF), clear TF now
259 	 * so that register information in the sigcontext is correct and
260 	 * then notify the tracer before entering the signal handler.
261 	 */
262 	stepping = test_thread_flag(TIF_SINGLESTEP);
263 	if (stepping)
264 		user_disable_single_step(current);
265 
266 	failed = (setup_rt_frame(ksig, regs) < 0);
267 	if (!failed) {
268 		/*
269 		 * Clear the direction flag as per the ABI for function entry.
270 		 *
271 		 * Clear RF when entering the signal handler, because
272 		 * it might disable possible debug exception from the
273 		 * signal handler.
274 		 *
275 		 * Clear TF for the case when it wasn't set by debugger to
276 		 * avoid the recursive send_sigtrap() in SIGTRAP handler.
277 		 */
278 		regs->flags &= ~(X86_EFLAGS_DF|X86_EFLAGS_RF|X86_EFLAGS_TF);
279 		/*
280 		 * Ensure the signal handler starts with the new fpu state.
281 		 */
282 		fpu__clear_user_states(fpu);
283 	}
284 	signal_setup_done(failed, ksig, stepping);
285 }
286 
287 static inline unsigned long get_nr_restart_syscall(const struct pt_regs *regs)
288 {
289 #ifdef CONFIG_IA32_EMULATION
290 	if (current->restart_block.arch_data & TS_COMPAT)
291 		return __NR_ia32_restart_syscall;
292 #endif
293 #ifdef CONFIG_X86_X32_ABI
294 	return __NR_restart_syscall | (regs->orig_ax & __X32_SYSCALL_BIT);
295 #else
296 	return __NR_restart_syscall;
297 #endif
298 }
299 
300 /*
301  * Note that 'init' is a special process: it doesn't get signals it doesn't
302  * want to handle. Thus you cannot kill init even with a SIGKILL even by
303  * mistake.
304  */
305 void arch_do_signal_or_restart(struct pt_regs *regs)
306 {
307 	struct ksignal ksig;
308 
309 	if (get_signal(&ksig)) {
310 		/* Whee! Actually deliver the signal.  */
311 		handle_signal(&ksig, regs);
312 		return;
313 	}
314 
315 	/* Did we come from a system call? */
316 	if (syscall_get_nr(current, regs) != -1) {
317 		/* Restart the system call - no handlers present */
318 		switch (syscall_get_error(current, regs)) {
319 		case -ERESTARTNOHAND:
320 		case -ERESTARTSYS:
321 		case -ERESTARTNOINTR:
322 			regs->ax = regs->orig_ax;
323 			regs->ip -= 2;
324 			break;
325 
326 		case -ERESTART_RESTARTBLOCK:
327 			regs->ax = get_nr_restart_syscall(regs);
328 			regs->ip -= 2;
329 			break;
330 		}
331 	}
332 
333 	/*
334 	 * If there's no signal to deliver, we just put the saved sigmask
335 	 * back.
336 	 */
337 	restore_saved_sigmask();
338 }
339 
340 void signal_fault(struct pt_regs *regs, void __user *frame, char *where)
341 {
342 	struct task_struct *me = current;
343 
344 	if (show_unhandled_signals && printk_ratelimit()) {
345 		printk("%s"
346 		       "%s[%d] bad frame in %s frame:%p ip:%lx sp:%lx orax:%lx",
347 		       task_pid_nr(current) > 1 ? KERN_INFO : KERN_EMERG,
348 		       me->comm, me->pid, where, frame,
349 		       regs->ip, regs->sp, regs->orig_ax);
350 		print_vma_addr(KERN_CONT " in ", regs->ip);
351 		pr_cont("\n");
352 	}
353 
354 	force_sig(SIGSEGV);
355 }
356 
357 #ifdef CONFIG_DYNAMIC_SIGFRAME
358 #ifdef CONFIG_STRICT_SIGALTSTACK_SIZE
359 static bool strict_sigaltstack_size __ro_after_init = true;
360 #else
361 static bool strict_sigaltstack_size __ro_after_init = false;
362 #endif
363 
364 static int __init strict_sas_size(char *arg)
365 {
366 	return kstrtobool(arg, &strict_sigaltstack_size) == 0;
367 }
368 __setup("strict_sas_size", strict_sas_size);
369 
370 /*
371  * MINSIGSTKSZ is 2048 and can't be changed despite the fact that AVX512
372  * exceeds that size already. As such programs might never use the
373  * sigaltstack they just continued to work. While always checking against
374  * the real size would be correct, this might be considered a regression.
375  *
376  * Therefore avoid the sanity check, unless enforced by kernel
377  * configuration or command line option.
378  *
379  * When dynamic FPU features are supported, the check is also enforced when
380  * the task has permissions to use dynamic features. Tasks which have no
381  * permission are checked against the size of the non-dynamic feature set
382  * if strict checking is enabled. This avoids forcing all tasks on the
383  * system to allocate large sigaltstacks even if they are never going
384  * to use a dynamic feature. As this is serialized via sighand::siglock
385  * any permission request for a dynamic feature either happened already
386  * or will see the newly install sigaltstack size in the permission checks.
387  */
388 bool sigaltstack_size_valid(size_t ss_size)
389 {
390 	unsigned long fsize = max_frame_size - fpu_default_state_size;
391 	u64 mask;
392 
393 	lockdep_assert_held(&current->sighand->siglock);
394 
395 	if (!fpu_state_size_dynamic() && !strict_sigaltstack_size)
396 		return true;
397 
398 	fsize += current->group_leader->thread.fpu.perm.__user_state_size;
399 	if (likely(ss_size > fsize))
400 		return true;
401 
402 	if (strict_sigaltstack_size)
403 		return ss_size > fsize;
404 
405 	mask = current->group_leader->thread.fpu.perm.__state_perm;
406 	if (mask & XFEATURE_MASK_USER_DYNAMIC)
407 		return ss_size > fsize;
408 
409 	return true;
410 }
411 #endif /* CONFIG_DYNAMIC_SIGFRAME */
412